Perspectives on Semantic Roles

Typological Studies in Language (TSL)

A companion series to the journal Studies in Language. Volumes in this
series are functionally and typologically oriented, covering specific topics in
language by collecting together data from a wide variety of languages and
language typologies.
For an overview of all books published in this series, please see
http://benjamins.com/catalog/tsl

Editors
Spike Gildea Fernando Zúñiga
University of Oregon University of Bern

Editorial Board
Balthasar Bickel John Haiman Marianne Mithun
Zurich St Paul Santa Barbara
Bernard Comrie Martin Haspelmath Doris L. Payne
Leipzig / Santa Barbara Leipzig Eugene, OR
Denis Creissels Bernd Heine Franz Plank
Lyon Köln Konstanz
William Croft Paul J. Hopper Dan I. Slobin
Albuquerque Pittsburgh Berkeley
Nicholas Evans Andrej A. Kibrik Sandra A. Thompson
Canberra Moscow Santa Barbara
Carol Genetti František Lichtenberk
Santa Barbara Auckland

Volume 106
Perspectives on Semantic Roles
Edited by Silvia Luraghi and Heiko Narrog
Perspectives on Semantic Roles

Edited by

Silvia Luraghi
University of Pavia

Heiko Narrog
Tokohu University

John Benjamins Publishing Company

Amsterdam / Philadelphia
 TM
The paper used in this publication meets the minimum requirements of
8

the American National Standard for Information Sciences – Permanence

of Paper for Printed Library Materials, ansi z39.48-1984.

Library of Congress Cataloging-in-Publication Data

Perspectives on Semantic Roles / Edited by Silvia Luraghi and Heiko Narrog.

p. cm. (Typological Studies in Language, issn 0167-7373 ; v. 106)
Includes bibliographical references and index.
1. Semantics. 2. Typology (Linguistics) I. Luraghi, Silvia, 1958- editor. II. Narrog,
Heiko, editor.
P325.P47 2014
401’.43--dc23 2014016110
isbn 978 90 272 0687 9 (Hb ; alk. paper)
isbn 978 90 272 6985 0 (Eb)

© 2014 – John Benjamins B.V.

No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any
other means, without written permission from the publisher.
John Benjamins Publishing Co. · P.O. Box 36224 · 1020 me Amsterdam · The Netherlands
John Benjamins North America · P.O. Box 27519 · Philadelphia pa 19118-0519 · usa
 Table of content

Perspectives on semantic roles: An introduction 1

Silvia Luraghi & Heiko Narrog
Inducing semantic roles 23
Michael Cysouw
The grammaticalization chain of case functions: Extension and reanalysis
of case marking vs. universals of grammaticalization  69
Heiko Narrog
Plotting diachronic semantic maps: The role of metaphors 99
Silvia Luraghi
The (non-)prototypicality of Direction: The (allative and illative)
case(s) of Finnish 151
Seppo Kittilä
The morphosyntax of the Experiencer in Early Vedic 181
Eystein Dahl
Against the Addressee of speech – Recipient metaphor: Evidence
from East Caucasian 205
Michael Daniel
Semantic roles and word formation: Instrument and Location
in Ancient Greek 241
Eugenio R. Luján & César Ruiz Abad
From semantic roles to evaluative markers: The dative and affected possessors.
A constructional comparison between German, Italian and French 271
Domenico Niclot
Author index 327
Language index 331
Subject index 333
 Perspectives on semantic roles
An introduction

Silvia Luraghi & Heiko Narrog

University of Pavia / Tohoku University

1. Semantic roles

Introduced in the late 1960s by Gruber (1965) and Fillmore (1968, 1971), semantic
roles (or ‘thematic roles’, ‘deep cases’, ‘thematic relations’, ‘theta roles’) have become
as indispensable for linguistic analysis across theoretical frameworks as they are
controversial in multiple respects, as argued among others by Newmeyer (2010).
In general, semantic roles are taken to refer to the roles taken by participants in
an event (see Kittilä, Västi & Ylikoski 2011: 7). Thus, there is general consensus on
the fact that in (1):
(1) Mary cut the rope with a knife.

Mary is the agent (i.e. the participant who brings about an event voluntarily), the
rope is the patient (i.e. the participant which undergoes a change of state brought
about by the agent), and the knife is the instrument (i.e. the participant used by
the agent to bring about the event). This broad generally accepted definition con-
stitutes a common ground for the papers in this volume.
Traditionally, research on semantic roles has centered on the following issues:

i. Determining the number and the kind of semantic roles entailed by the
­argument structures of verbs and construction types, based on the assump-
tion that they are not language-specific but form a cross-linguistically (or, uni-
versally) valid set;
ii. Related to (i) are descriptive and methodological issues such as criteria for
distinguishing semantic roles, and dealing with variation in the expression of
semantic roles across languages and within one language;
iii. The definition of particular semantic roles, for example through a close set
of necessary and sufficient features (cf. e.g. Radden 1989), or as prototypical
categories (cf. e.g. Nishimura 1993 or Schlesinger 1989);
 Silvia Luraghi & Heiko Narrog

iv. Determining the level of linguistic description at which semantic roles apply,
and their relationship to units of linguistic description at adjacent levels (e.g.
Lehmann & Shin 2005);
v. Determining the relationship of semantic roles to case, argument structure
and grammatical relations (e.g. Bornkessel et al. 2006);
vi. Related to (iv) and (v), the level of granularity at which semantic roles are
identified: a coarse level implies the necessity to introduce proto-roles (Dowty
1991) or macro-roles (Foley & Van Valin 1984; Van Valin & LaPolla 1997),
partly as an improvement to the interface with syntax (e.g. Kailuweit 2004);
vii. Determining hierarchies between semantic roles with respect to certain syn-
tactic operations (e.g. Blake 2004: 91–92; Primus 2006).

In contrast, the following two issues are more recent, and have been particularly
motivated by grammaticalization research:

viii. The synchronic and diachronic relationship between semantic roles; i.e. their
semantic similarity and paths of extension from the expression of one seman-
tic role to others;
ix. The systematic representation of this relationship in so-called semantic maps.

To this we can add the following issue, which has not been particularly prominent
in the literature about semantic roles, but which is highlighted in one of the con-
tributions to this volume:

x. The shift from semantically based case marker to syntactic case marker via
constructional reanalysis.

In spite of the vast array of studies especially, or at least partially, devoted to

semantic roles, no treatment has thus far appeared that gives a satisfactory answer
to the issues raised by often conflicting views on the points listed above. For
example, regarding granularity, mismatches at the syntax-semantics interface are
sometimes dealt with by the proto-role approach referred to in (vi), but can also
motivate the opposite approach, i.e. that of a verb-specific level of analysis, which
individuates a number of predicate-dependent microroles (Levin & Rappaport
Hovav 2005: 40–41). Such a fine-grained approach often misses important gener-
alizations, and fails to consider the complex interplay of different factors involved
in the linguistic encoding of events. As a consequence, verb-specific semantic roles
are not especially enlightening if the goal is to understand the reasons for coding
similarities, for polysemy, and for the diachronic extension of coding devices typi-
cal of a certain semantic role to others (cf. (ii) and (viii) above).
 Perspectives on semantic roles 

Indeed, approaches to semantic roles display such a wide variety because their
proponents often have different interests in mind: in other words, they answer dif-
ferent questions, as they target different aspects of the linguistic encoding of events
and event participants. This volume is particularly concerned with the semantic
rather than syntactic aspects of semantic roles. The papers assembled here there-
fore deal with the semantic and diachronic issues (i), (ii), (iv), (viii), and (ix), with
a particular focus on the issues (viii) and (ix) related to diachronic change, and to
the representation of similarity and extension between semantic roles in semantic
maps.

2. Encoding semantic roles

Semantic roles are encoded through a variety of morphosyntactic means cross-

linguistically. Such morphosyntactic coding devices can be seen as exponents of
semantic roles; consequently, it is important to clarify their status. In the first
place, semantic roles can be encoded by morphological cases. In addition, seman-
tic roles are frequently encoded by adpositions, or even by adpositions plus (pos-
sibly different) cases. In head-marking languages, semantic roles can be indicated
by verb affixes, and some basic semantic roles are indicated by word order in
­inflectionally poor languages such as English. In the sentence The nurse gave the
mother the baby, it is word order that indicates that the nurse is the agent, the
mother is the recipient, and the baby is the theme. As papers in this volume do
not discuss the e­ ncoding of semantic roles through verb affixes or word order, we
will concentrate on the role of cases and adpositions in this section.

2.1 Case
The close connection between cases and semantic roles is highlighted by defini-
tions of case such as the following: “an inflectional dimension of nouns that serves
to code the noun phrase’s semantic role.” (Haspelmath 2002: 267). Not everybody
would subscribe to this definition, unless it is somewhat enlarged to include the
role of cases in the encoding of grammatical relations, such as subject and object,
which are not univocally connected with semantic roles. In a language like Eng-
lish, a subject can be an agent, an experiencer, a patient, and so on depending on
the type of predicate. In (2):
(2) John loves Mary.
for example, there is no agent. Still, in many case-marking languages the subject
NP is coded in the nominative in the same way as the subject of action verbs.
 Silvia Luraghi & Heiko Narrog

Thus, a broader and more comprehensive definition of case that takes into account
both its grammatical and its semantic function is the following: “an inflectional
category of nouns that helps identify a NP’s grammatical relation and/or semantic
role.” (Luraghi & Parodi 2010: 69).
It is widely known and has been remarked by several scholars over time that
in specific case systems there are cases that play a bigger role in the encoding of
grammatical relations (so-called grammatical cases), while others play a bigger
role in the encoding of semantic roles (so-called semantic or concrete cases, see
Haspelmath 2009: 508 and Luraghi & Parodi 2010: 71 for this terminology). We
discuss this distinction below, in Section 2.2.
Research on semantic roles, when connected with cases, can be seen as a mod-
ern development of such traditional studies as those devoted to case meaning by
structuralists and even earlier by comparative linguists (see Luraghi 2008). Thus,
studies such as Nikiforidou (1991) or Janda (1993), devoted to the “meaning” of
single cases in a specific language or in a cross-linguistic perspective, investigate
the semantic relations most frequently indicated by case endings. In work by
scholars inspired by Cognitive Linguistics, case polysemy is described as struc-
tured according to the model of radial categories (Janda 1993: 6). This model is
especially useful when one needs to understand the relations among different
meanings, and the way in which semantic extension operates.
Typological research when investigating the functions of cases cross-­
linguistically has also been confronted with the problem of polysemy and com-
patibility of meanings. By comparing similar instances of polysemy and semantic
extension cross-linguistically, typologists have identified a network of relations
among meanings, which can be represented on semantic maps. We will return to
this important tool below in Section 3.
Before proceeding, however, it must be noted that not all papers in this col-
lection approach the issue of semantic roles taking cases (or more in general
grammatical forms) as their starting point: in particular, Michael Cysouw’s paper
Inducing semantic roles takes the reverse approach, and induces semantic roles
from the distribution of case-like affixes cross-linguistically. Domenico Niclot’s
paper takes yet another approach, that is, rather than focus on a specific case form,
it focuses on a range of constructions that utilize the case form. For more discus-
sion on different perspectives and approaches in the volume see Section 4 below.

2.2 Semantic roles and structural case vs. lexical (inherent) case
As remarked in the preceding section, a distinction is often made between gram-
matical and semantic or concrete cases. This distinction reaches back to ­traditional
studies on cases, and has been discussed thoroughly in such classical studies as
 Perspectives on semantic roles 

Kuryłowicz (1949, 1964). According to Kuryłowicz, cases such as the nominative

and the accusative in Indo-European languages are grammatical, as they most
often indicate grammatical relations (subject and direct object), regardless of
semantic roles. On the other hand, cases such as the locative or the ablative are
concrete, as they most often indicate semantic roles. More recently, this distinc-
tion has been referred to as a split between structural cases and lexical or inherent
cases. With respect to semantic roles, this can be translated into the idea that some
cases do not express semantic roles while others do. This idea has been further
adapted and refined in formal frameworks, but it has also stimulated research on
case in functional research, where the emphasis tends to be on the observation
that even with “structural” or “grammatical” cases one can find a core of seman-
tic motivation. Furthermore, the status of specific cases in specific languages as
structural or lexical can be controversial. This is, for example, notoriously the case
with the dative in Indo-European languages, as already discussed in Kuryłowicz
(1949=1960: 146).
In this volume, no paper deals specifically with the distinction between
structural and lexical case. Dahl’s paper deals with a semantically rich semantic
role, the experiencer, which surprisingly is most commonly expressed as struc-
tural case (nominative or accusative) in the language under investigation, Early
Vedic Sanskrit. In contrast, the “addressee of speech act”, if it is to be acknowl-
edged as a semantic role, as argued for by Daniel, is less surprisingly marked by
dative or a clearly Goal-related cases in Caucasian languages. Kittilä’s paper deals
with semantically rich semantic roles that are expressed by locative (mostly Goal-
related) cases, or even adpositions. Niclot’s paper as well targets typical ‘lexical’ or
semantically rich case, namely datives of affectedness. Luraghi’s paper is mostly
concerned with the lexical cases, especially the source role of locative cases for the
expression of other, non-locative semantic roles. Narrog’s paper, like Luraghi’s, is
primarily concerned with case polysemy, and the development and directionality
of case polysemy, but it is probably the one paper in this collection whose con-
tents most directly bears on the distinction between structural and lexical case.
Specifically, he argues that those cases traditionally identified as “structural” or
“grammatical”, namely nominative, accusative, absolutive and ergative, serve as
unique endpoints in the historical change of morphological case expression. One
could loosely speak of them as one of two final “dumps” in the development of
case markers, the other being semantically highly abstract cases marking adverbial
adjuncts. Interestingly, the one case that is most controversial with respect of its
status of being structural or lexical, the dative, is in a pivot position at which cases
either develop into core structural case, or into adverbial adjuncts (remarkably,
however, Narrog writes about them in terms of semantic roles). Cysouw’s paper
also has some direct relation on our understanding of the distinction between
 Silvia Luraghi & Heiko Narrog

structural and lexical case, namely by introducing a bottom-up approach to the

analysis of semantic roles and case which makes no a priori distinction between
different types of case or semantic roles. Cysouw’s approach is equally able to deal
with semantic roles and case whether they have been labeled as “grammatical” or
as “lexical” and thus calls into question the theoretical distinction from a cross-
linguistic empirical perspective.

2.3 Adpositions
Similar to cases, adpositions are also possible markers of semantic roles (cf.
­Kittilä, Västi & Ylikoski 2011: 3). This functional similarity is best exemplified in
languages that have no morphological case: for example, in French the preposi-
tion à is readily comparable with the German dative case in sentences such as
(3) and (4):
(3) Je donne le livre au père
1sg.nom give:prs.1sg art.m book(m) to.art.m father(m)
‘I give father the book.’
(4) Ich gebe dem Vater das Buch
1sg.nom give:prs.1sg art.dat.m father(m) art.acc.n book(n)
(same meaning).

Both the preposition à in French and the German dative case indicate the seman-
tic roles of recipient in these examples.
From the formal point of view, prepositions are different from cases as they
are free, rather than bound morphemes. This difference does not seem to be rel-
evant when we compare the function of a preposition such as French à with the
function of the dative case in many other languages, in examples such as those just
discussed. However, the role of adpositions in the encoding of semantic roles is
somewhat wider than that of cases. In the first place, even in languages with a large
case system, the number of cases is limited, while one can always find a certain
continuum between adpositions and other lexical items (adverbs, nouns, verbs)
that can at least occasionally be used as adpositions. Consider the Italian examples
in (5) and (6):
(5) Vado dentro alla casa.
(6) Vado dentro casa.

While in (5) the adverb dentro ‘inside’ is followed by the primary preposition a,
which reveals the adverbial, rather than adpositional, nature of dentro, in (6) it
functions as a preposition in its own right. Thus, we can see that there is some
overlap between prepositions and adverbs in Italian. In general, we can view this
 Perspectives on semantic roles 

as varying degrees of grammaticalization. The primary preposition a cannot occur

without a noun (*Vado a), while the adverb dentro can (Vado dentro). But even this
difference between primary and adverbial prepositions is not clearcut, as the pri-
mary preposition su ‘on’, ‘over’, ‘up’, can occur alone (Vado su ‘I’m going up[stairs]).
Cases are even more grammaticalized than primary adpositions. While in Italian
some adverbial prepositions can occur without a primary preposition, in a case
marking language such as Finnish, cases occur on nouns even without adposi-
tions, as in (7), but an adposition cannot occur without a case, as shown in (8):
(7) pallo vier-i laatikko-on
ball roll-3sg.pst box-ill
‘The ball rolled into the box.’
(8) Mari laitto-i kirja-n laatiko-n sisälle/sisään
Mary put-3sg.pst book-acc box-gen inside
‘Maria put the book into/inside the box.’
(from Kittilä, this volume)

This is of course because cases are bound morphemes, and are more grammatical-
ized than adpositions.
We are not going to pursue further formal distinctions between cases and
various types of adpositions here. Remarkably, however, the more or less gram-
maticalized nature of these morphemes has a bearing on the extent to which they
encode semantic roles. In the first place, in case systems with a limited number of
cases, the latter typically encode grammatical relations, such as subject and direct
object, while semantic roles are encoded by case-marked nouns plus adpositions.
German case marking is a case in point: in German not only are the nomina-
tive and the accusative case basically limited to encoding grammatical relations,
but also the dative can encode semantic roles usually only to the extent that they
are indicated by the verb. Thus, in example (4) the dative does in fact encode the
role of Recipient, but this role is required for its third argument by the verb ‘give’.
The dative also has a usage not conditioned by the verbal valency (so-called ‘free
dative’), which is, however, basically limited to Beneficiary (or Malefactive; this
usage is treated in Chapter 8 in this volume).
Another important difference between cases and adpositions is a consequence
of the more polysemous nature of cases: adpositions are semantically more spe-
cific, hence more suitable for specifying special features of given roles. Thus, in a
language with a generic locative case, for example, various types of location (inside,
near, on the surface of an entity, etc.) can be specified by adpositions. In addition,
cases often rely on lexical meaning to disambiguate possibly polysemy. It follows
that cases most often indicate semantic roles when the entity involved has a high
chance to take this role in an event. We will discuss this issue in the next section.
 Silvia Luraghi & Heiko Narrog

2.4 Semantic roles and lexical meaning

Consider a highly polysemous case such as the Latin ablative. This case can indi-
cate Source (especially with toponyms and depending on inflectional class),
Instrument, and Cause, among other roles. Consider now Examples (9)–(11):
(9) senex qui huc Athenis
old.man(m):nom rel.nom.m here Athens:abl
exsul venit
exile(m):nom come:pf.3sg
‘An old gentleman, who came here as an exile from Athens.’ (Pl. Rud. 35)

(10) senex qui huc curru

old.man(m):nom rel.nom.m here wagon:abl
exsul venit
exile(m):nom come:pf.3sg
‘An old gentleman, who came here as an exile on a wagon.’

(11) senex qui huc metu

old.man(m):nom rel.nom.m here fear:abl

exsul venit
exile(m):nom come:pf.3sg
‘An old gentleman, who came here as an exile out of fear.’

Athens is a city, and it is highly expected that its involvement in an event is that of
a landmark of spatial relations. Similarly, a vehicle is normally used as an instru-
ment for motion, and an abstract entity such as fear is likely to function as a cause.
Of course, one can imagine a situation in which something that is not a natural
location functions as a landmark for spatial relations. However, as such a situation
is less expected, it is usually encoded with ‘heavier’ morphological means, typi-
cally by means of adpositions, as in (12):
(12) isdem … qui nuntii ab
same:abl.pl rel.nom.pl messenger:nom.pl from
Iccio venerant
Iccius:abl come:ppf.3pl
‘The same (persons) who had come as messengers from Iccius.’
(Caes. Gal. 2.7)

In (12), a man called Iccius functions as the landmark of the same spatial relation
indicated by the ablative Athenis in (9). However, here we find a prepositional
phrase with ab ‘from’, which makes this relation explicit. Nouns referring to par-
ticipants whose involvement in a situation is highly expected require less marking
 Perspectives on semantic roles 

than other nouns. This fact has often been noted for toponyms (see e.g. C ­ omrie
1986: 2–3; Luraghi 2003: 65–66 among many other). Typically, adpositions, though
polysemous, are less polysemous than cases, so they can indicate semantic roles
in a more specific manner (see Kittilä, Västi & Ylikoski 2011: 10). Concerning
Example (12), it can still be added that human beings are highly unexpected as
landmarks of spatial relations (see Luraghi 2011, Creissel & Mounole 2011, and
Kittilä this volume), and often require extra marking, or differential marking (this
is not the case in Latin, where ab indicates Source with all types of noun except, to
a limited extent, toponyms).
Lexical meaning can also indicate that a certain noun refers to an entity which
is likely to take a specific semantic role. This is often achieved through deriva-
tional morphology. Thus, nouns marked as agent nouns, or as toponyms or instru-
ment nouns indicate referents that are more likely to occur as Agents, Locations,
or Instruments in sentences. In very much the same way as cases and adpositions,
derivational affixes may also be polysemous, and mark nouns for neighboring
roles such as e.g. Instrument and Agent, as English -er in (13) and (14):

(13) Paul is a writer.

(14) The printer is jammed.

Polysemy of derivational affixes is discussed in Luján & Ruiz (this volume).

3. Semantic maps

Semantic maps constitute an increasingly popular method of representing the

relationship between linguistic units, especially meanings and functions, in terms
of similarity. They have become an elegant way to deal with the polyfunction-
ality of linguistic elements (affixes, clitics, auxiliaries, or whole constructions)
that conveniently bypasses the often thorny issue of polysemy vs. homonymy,
shedding light on both diachronic and synchronic patterns within and across
languages. Since the introduction of semantic maps by Anderson (1982, 1986),
and stimulated by Haspelmath (1997, 2003), especially in the past decade a rap-
idly growing body of linguistic phenomena has been systematically dealt with in
terms of semantic maps (see for example the recent issue of Linguistic Discovery
8/1). Among them, the semantic roles covered by case morphology and adposi-
tions have been one of the topics most commonly studied (e.g. Haspelmath 2003,
Narrog & Ito 2007, Malchukov & Narrog 2009, and references therein). An often
cited example of such a map is given in Figure 1, from Haspelmath (2003: 219).
 Silvia Luraghi & Heiko Narrog

dative
predicative external
possessor possessor

French á
direction recipient beneficiary judicantis

purpose experiencer

Figure 1. A semantic map of typical dative functions and the borders of french à and
of french dative clitics

In the map in Figure 1, semantic roles are indicated by primitive meanings

connected by lines. The position of such meanings with respect to each other is
taken to represent universally valid relations among them. The meaning of a spe-
cific linguistic form (here the French preposition à and French dative clitics) cov-
ers the area indicated by the dotted lines.
Some classic historical-typological studies on synchronic and diachronic vari-
ation in the expression of semantic roles tried to indicate what are possible and
impossible patterns of polysemy (see e.g. Croft 1991: 185; Heine et al. 1991: 159).
More recently, various issues have been raised, connected with the structure of
semantic space, and the best way to represent relations among semantic roles.
­Narrog (2010), for example, supports a model of semantic maps which include
connections between items, in line with the original proposal in Haspelmath
(1997, 2003). In such ‘classical’ semantic maps, lines between points usually do
not cross each other (cf. Croft & Poole 2008: 22), and semantic space is represented
as being bi-dimensional. Bi-dimensionality is an obvious consequence of graphic
representation; however, according to Haspelmath (1997), keeping low the num-
ber of dimensions, and possibly also of connecting lines, is advantageous, as it
increases predictive power of semantic maps. It turns out that, especially in the
domain that concerns us here, that is, the relations among semantic roles, numer-
ous predictions have been shown to be too strong (see Luraghi 2001 for a dis-
cussion of Croft’s Causal Chain, proposed in Croft 1991). However, the opposite
tendency can also raise problems, since by increasing the number of connections
between points frequent and infrequent semantic extensions are put on the same
plane, as pointed out by Lestrade (2010).
As an answer to the above issues, so-called ‘second generation’ semantic
maps have been proposed, which rely on multi-dimensional scaling (MDS),
and represent the distance among semantic roles based on statistical calculation
 Perspectives on semantic roles 

(see e.g. Cysouw 2010 and this volume). An example is given in Figure 2, from
Wälchli (2010: 348).

French
0.4

Companion
dans
Animate a
Goal
0.2

de
N
Dimension 2

ACC
vers
avec
0.0

aupres=de
Path Inanimate sur
en
Goal
chez
Source
–0.2

–0.4 –0.2 0.0 0.2

Dimension 1

Figure 2. The meaning of french local prepositions in Mark’s Gospel in a mds representation

As pointed out in Croft & Poole (2008: 25), “multidimensional scaling as

used in the social and behavioral sciences is mathematically well understood and
computationally tractable”. As a consequence, arguably MDS has the advantage of
avoiding unwarranted assumptions by linguists regarding the structure of seman-
tic space. Indeed, they do not rely on meanings set up as primitives by linguists.
Rather, they take formal coding as their point of departure (see further Cysouw,
this volume). On the other hand, van den Auwera (2008) and Narrog (2010) point
out that MDS cannot capture diachronic developments in the way that ‘classical’
semantic maps can do if connecting lines and arrows are drawn that are supported
by data on attested developments.
Although semantic maps are certainly not designed in order to answer all
the open questions raised by semantic roles, it is the claim of the present book
that both the analytical decisions made when devising particular semantic maps
and the patterns found within and across languages can help answer ques-
tions related to granularity, the motivation of polyfunctionality and diachronic
change, and the role played by lexical, grammatical, and contextual elements of
meaning. Neighboring semantic roles are most often taken by similar partici-
pants, similarity being reflected by lexical features, unless some other contex-
tual factors intervene. For example, the occurrence of a predicate that requires a
 Silvia Luraghi & Heiko Narrog

v­ olitional subject may induce a metaphorical reading in the case that the subject
is inanimate.
In addition, the semantic range of specific coding devices usually covers a
coherent area on a semantic map, but it may cover two separate ones in connec-
tion with different lexical features. For example, evidence from various languages
shows that different semantic roles may be coded through the same marker in
connection with different semantic features of noun phrases. We have already dis-
cussed various reflexes of lexical features in the coding of spatial relations (see
Section 2.4), but there are more. For example, the Ancient Greek dative codes
typical dative relations with human nouns, but it codes instrument with inani-
mates (cf. Luraghi 1991). It is doubtful that the polysemy of the Ancient Greek
dative is indicative of semantic closeness of instrument and recipient, for example.
Rather, while the merger can be explained in semantic terms if one considers the
diachronic development, synchronically it could well be taken as an instance of
polysemy, disambiguated by animacy (see further Luraghi 2009).

4. Contents of this volume

Semantic Maps are a central issue in four out of the eight papers of this volume.
Michael Cysouw primarily addresses methodological issues. The object of his
investigation is not, strictly speaking, semantic roles, but a more concrete unit of
linguistic analysis that he labels as ‘context-specific roles’. In a corpus of the same
text translated into 15 languages, he isolates 34 contexts in which the same noun
(‘bible’) is used in a large variety of semantic relationships to the clause predi-
cate. He goes on to compare whether the noun is marked identically or differently
within one language for each of the contexts, and by inputting the data from the 15
languages in a MDS calculation, arrives at a map of similarity between the 34 con-
texts (other forms of visual representation such as hierarchical clustering and split
decomposition are also briefly demonstrated by Cysouw but not further pursued).
He further performs flat clustering within the MDS model, dividing the contextual
roles into three mutually exclusive groups of similar roles. One important corol-
lary of his analysis is that it allows the researcher to statistically classify languages
on the basis of their language-specific marking of contextual roles. The resulting
alignment typology is based on concrete contexts and is more fine-grained than
the traditional nominative-accusative vs. ergative-absolutive typology.
Silvia Luraghi’s paper focuses on diachronic aspects of semantic maps. She
aims to demonstrate how specific metaphors are responsible both for patterns
of polysemy and for diachronic extensions among semantic roles as represented
in semantic maps. In doing so, she not only reaffirms the traditional notion of
 Perspectives on semantic roles 

c­ entrality of the domain of spatial relations for the secondary expression of other
meanings and functions, but points out that space may be more fundamental than
was previously thought. Concretely, she argues that it is often the domain of space,
rather than the intermediate domain of human relations, that directly provides
the source for the conceptualization of inanimate, abstract relations. Furthermore,
examining the diachronic relationship between semantic roles in more detail than
previously, Luraghi points out a number of facts that can easily be overlooked in
a more coarse-grained perspective. First, the same semantic role may have more
than two sources, and the extension from each source may be associated with a
different metaphor and a different conceptualization, leading to different patterns
of polysemy. Thus, for example, Instrument can be derived from Comitative via
the Companion Metaphor, from Location via the Container/Support Metaphor
or from Path markers through the Channel Metaphor. Second, some extensions
may result in polyfunctionality while others do not. For example, while there is a
historically documented extension from Location to Comitative, this extension is
usually not retained as a synchronic polysemy, presumably because of an incom-
patibility of these two roles. This means that synchronic polysemy does not neces-
sarily faithfully reflect diachronic extension. Somewhat differently to the case of
Location-Comitative, with respect to the triangle of the three spatial roles Source,
Location, and Direction, Source tends to be maximally distinct from Location and
Direction, diachronic extension from Source to the other roles being rare, as also
noted in previous research (e.g. Noonan 2009).
Heiko Narrog’s paper is likewise concerned with semantic maps as a repre-
sentation of diachronic relationships between semantic roles, and with metaphor.
He specifically examines a purportedly unidirectional chain of grammaticaliza-
tion, from spatial roles to human participants, to participants in human activ-
ity, and finally to inanimate concepts, as proposed by Heine et al. (1991). While
this unidirectional chain has held up well in a number of respects, including the
basic directionality of extension from spatial to non-spatial concepts, it has turned
out to be problematic in other areas, especially concerning the ‘intermediate’
domains of human participants and participants in human activity. For example,
the well-known extension from instrumental to passive agent fits neither the pro-
posed concrete relationship between semantic roles nor the overall relationship
between domains. Narrog analyzes the grammaticalization chain in detail, list-
ing first all extensions that support Heine et al. ’s (1991) hypothesis, which are
the majority, and then all extensions that run counter the hypothesis. The latter
mainly comprise extensions from Instruments, Possessors and Causes to Agents,
and shifts from Possessors, nominatives and datives to Causees, which are often,
but not always, the result of syntactic reanalysis rather than semantic extension.
­Narrog suggests not only that Heine et al.’s chain is untenable but also that a single
 Silvia Luraghi & Heiko Narrog

­ nidirectional chain cannot be upheld. Instead the chain must be split up in at

u
least two directions, one leading to the expression of core (human) participants,
and the other to abstract, inanimate relations. Recipients turn out to be an inter-
mediate point on both chains. Both chains have in common that they start out
with spatial notions and that they represent a shift from more concrete seman-
tic notions towards highly frequent, automatized, and semantically bleached case
functions. While the extension towards abstract, inanimate relations can be fully
covered by extension through metaphor, Narrog suggests that the extension to
core participants has mostly syntactic rather than semantic motivations, and if
viewed in terms of metaphor, it must be explained in terms of grammatical rather
than conceptual metaphor.
Semantic roles are usually associated with grammatical items, and this partly
also holds for the semantic map methodology. That this is not necessarily the
case is shown in the paper by Eugenio Luján and César Ruiz. They explore poly-
functionality in word formation patterns, specifically investigating three produc-
tive suffixes in nominal word formation of Ancient Greek, namely -tron, -thron,
and -tér ̄ion. According to the authors, in word formation a formal pattern can
be related to a semantic role when the meanings of the words built according to
that pattern are overtly construed in such a way that they evoke a state of affairs
in which the referent of the noun plays that semantic role. Compiling exhaustive
lists of the nouns derived by means of the three suffixes named above, they show
that these essentially agentive patterns are not only used to derive instrumental
and locative nouns, but also other nouns that are semantically related, such as
non-prototypical instruments, containers, prizes and rewards, among others. In
conclusion they observe that the semantic organization found in grammar and in
word formation, as represented on semantic maps, is similar. This may point to
common underlying mental structures that are reflected both at the syntactic and
the morphological and lexical level.
The question of the number and kind of semantic roles, and relatedly, varia-
tion in their expression, is the main topic of three papers in this volume (Dahl,
Daniel, Kittilä), and is more peripherally also dealt with in Luján and Ruiz. One
criterion for acknowledging the existence of a semantic role that is agreed upon by
all authors is the existence of case marking solely dedicated to that semantic role
in at least one language. A looser criterion, as for example mentioned by Luján and
Ruiz, is conflation of the candidate for a semantic role with different case functions
in different languages. ‘Means’ for example, may be acknowledged as a semantic
role if it is conflated with Instrument in one language but with Agent in another.
A perennially controversial semantic role has been the Experiencer, which
was established fairly early (Fillmore 1971) and is intuitively basic. However, it is
difficult to find case-marking dedicated just to Experiencers cross-linguistically.
 Perspectives on semantic roles 

An exception may be Caucasian languages from which case markers specifically

for Experiencers have been reported (cf. Comrie 1986, van den Berg 2006). Eystein
Dahl analyzes the encoding of Experiencers in detail in a language where it does
not have a dedicated expression, namely Vedic Sanskrit. He shows that Experienc-
ers can be expressed by three cases, the nominative, the accusative and the dative.
Dahl argues that different patterns instantiate three typologically distinct types of
Experiencer constructions where the Experiencer argument is assimilated to an
Agent (Experiencer-as-Agent), a Patient (Experiencer-as-Patient) and a Benefi-
ciary (Experiencer-as-Beneficiary), respectively. Among these patterns, the con-
flation with the Agent in the nominative case is clearly the most common and
the most diversified, extending to practically all classes of experiential predicates,
while the conflation with Beneficiary in the dative is the most uncommon, being
limited to emotion predicates. Overall, Dahl confirms the common perception
that the Experiencer is a considerably less unitary semantic role than Agent or
Patient, for example.
Almost the opposite argument is developed by Michael Daniel with respect
of the Addressee of verbs of communication. Addressee is often not singled out as
a separate semantic role. In the well-known European languages it is usually cov-
ered by the dative or dative-like prepositions and does not have a dedicated formal
expression. For this reason, it is easily conceived of as a metaphorical extension
of the Recipient. However, as Daniel argues, East Caucasian languages present
quite a different picture. First of all, there are languages in which the Addressee of
speech acts does receive a unique formal expression. Furthermore, the Addressee
is not construed as a Recipient, but as an animate Goal, that is, if it does not have
a unique expression, it is usually not conflated with the dative but with markers
for spatial relations (typically, lative). Daniel also suggests that the common asso-
ciation of Addressees with the dative in European languages may be deceptive.
Although, semantically speaking, Recipient is the core semantic role expressed
by datives, datives are often historical developments from spatial (allative/Goal)
markers. Thus, the extension from spatial Goal marking may be generally more
fundamental to the expression of Addressees than the extension from datives/
Recipients.
Variation in the formal expression of semantic roles is a secondary topic in the
papers of Dahl and Daniel, whose primary purpose is deconstructing an established
role, and establishing a new role, respectively. As discussed above, Dahl and Dan-
iel show in detail the variation in the expression of Experiencers in one language
(Dahl), and in the expression of Addressees across a group of languages (Daniel). In
the contribution by Seppo Kittilä, variation in the coding of Direction (Goal) itself
is the central topic. His language of investigation is Finnish, which is well-known
for its rich inventory of local case markers and adpositions. Kittilä shows that the
 Silvia Luraghi & Heiko Narrog

e­ ncoding of Direction is amenable to features such as animacy, size of the landmark,

and the nature of the reference to the landmark. Landmarks which are more typical
(or unmarked) with respect to these features receive a different coding in Direc-
tion expressions than those which are atypical (marked). As evidenced in previous
research (see e.g. Comrie 1986), the typical instances receive less elaborate coding
and the atypical instances more elaborate coding. In the Finnish data explored by
Kittilä, more elaborate coding means adpositional instead of case marking. Thus Kit-
tilä contributes to the question of the distinction between case marking and adposi-
tional marking in languages where both are available as alternatives.
The paper by Domenico Niclot approaches the issue of similar encoding
of related, though partly different, semantic roles in a Construction Grammar
framework. Following this approach, he does not take the individual seman-
tic roles as his starting point, and does not try to detect the reason for similar
encoding with a bottom up procedure, but rather proceeds top-down, by show-
ing how the meaning of a certain construction can extend to different verbs,
whose meaning is crucially affected by the construction. What could be argued
to be different semantic roles, then, represent different facets of single roles as
used in different constructions. His topic is the functions of so-called datives
of ­affectedness, which are associated with possession, across three continental
European languages (French, German, Italian). In these languages, construc-
tions such as Italian La mia macchina mi si è rotta, which would literally mean
‘My car broke down on me’, that are uncommon in English, are conventional-
ized. Niclot bases his analysis on the concept of Personal Sphere as developed
by Raineri and Evola (2008). He breaks up the dative of affectedness into four
micro-constructions, in which the Possessum is (i) acted on, (ii) changes state,
(iii) acts, and (iv) finds itself in a certain state. In the order of (i), which still has
the property of an intentional event transfer, to (iv), these micro-constructions
become increasingly remote from the original dative construction and can be
considered as more grammaticalized. For each of these constructions, Niclot
shows how far they are entrenched in each language under investigation. Espe-
cially micro-constructions (iii) and (iv), which are the most advanced in terms
of grammaticalization, show that datives of affectedness are more entrenched
and grammaticalized in Italian than in the two other languages. Concerning
(iv), the existence of the construction as such is not yet well-­established. Niclot
concludes that datives of affectedness are essentially evaluative markers that
reflect the perspective of their referent or of a third person, and suggests that the
analysis in terms of constructions and grammaticalization is more accurate than
one in terms of the nature of the Possessum, as previously put forward by König
& Haspelmath (1997).
The question at which level semantic roles apply is dealt with in Cysouw’s
paper in a practical manner. The fundamental idea behind semantic roles is that
 Perspectives on semantic roles 

they exist at a universal, cognitive level, and do not necessarily have a directly
corresponding expression at the morpho-syntactic surface. This idea is, for exam-
ple, developed quite explicitly in Lehmann & Shin (2005), who posit a cogni-
tive level of analysis that is basically independent of language and its structure,
an interlingual or typological level above it, and finally a language-specific level.
The association with a cognitive level independent of specific languages is one
reason for the difficulty in determining a valid set of semantic roles. Cysouw
circumvents this difficulty by introducing the concept of ‘context-specific roles’,
which are bound to concrete linguistic forms. These context-specific roles do not
have cognitive status and therefore cannot replace semantic roles; however, they
­provide a tool to operationalize entities close to semantic roles through surface
linguistic expressions.
Overall, as stated in the beginning of this introduction, among the wide range
of topics and issues related to semantic roles, the synchronic and diachronic rela-
tions among semantic roles as represented on semantic maps is the gravity center
of this volume. The papers by Luraghi and Narrog take primarily a diachronic
typology perspective and investigate relationships of extension among seman-
tic roles on a broader scale than has been common. This has become possible
through the burgeoning interest in grammaticalization and the accompanying
accumulation of historical and comparative data. These papers do not only put
forward their own solutions and hypotheses concerning extant problems but also
point forward to still unresolved issues, such as semantic extensions that leave
no synchronic trace (Luraghi) or the role of grammatical metaphor in the exten-
sion of case functions (Narrog). The papers by Cysouw and Luján & Ruiz take
a synchronic perspective on semantic maps. Both papers open up new avenues
of research, Cysouw through the development of new quantificational method-
ologies, and Luján and Ruiz by extending research on semantic roles from the
morphosyntactic domain to the domain of word formation. Cysouw leads a line
of research that is certain to grow through the increasing availability of electronic
data, and, as Cysouw shows, can shed unexpected light on well-known issues
such as alignment typology. On the other hand, the examination of word forma-
tion patterns opens up the opportunity to test in a different domain the relations
and directionalities of extension among different roles that have been held to be
universal in the grammatical domain.
The two papers that primarily question the number and kinds of semantic
roles (Dahl, Daniel) take almost opposite stances, and show that both perspectives
can be supported by arguments and thus claim legitimacy. Dahl “deconstructs”
the Experiencer in Vedic Sanskrit, demonstrating in detail that it has no unique
expression. According to him, this means that, at least for Vedic Sanskrit, an Expe-
riencer cannot be posited as a semantic role on the same level as an Agent or a
Patient. Daniel, on the other hand, shows that a participant that rarely receives
 Silvia Luraghi & Heiko Narrog

unique expression in the well-known European languages, the Addressee of

speech acts, receives such expression in some Caucasian languages, and thus can
legitimately claim recognition as a universally valid semantic role. Both contribu-
tions will further fuel the debate on the status of semantic roles, whether their
inventory is best conceived of as universal or, at least partly, language-specific.
Kittilä also presents a fine-grained study of the expression of a specific semantic
role in a single language, namely Direction. In this case, the semantic role as such
is uncontroversial, but the details in its expression shed new light on our under-
standing of the choice of expression when competing markers are available. Kittilä
argues that varying degrees of typicality of the landmark primarily determine this
choice. In doing so, he also elucidates the difference between case and adpositions
in Finnish. This language, with its rich inventory of case and adpositional mark-
ing, is especially suitable for such a study, and it is hoped that more such studies
will follow to further deepen our understanding of these issues.
Finally, Niclot’s paper points at new theoretical possibilities to investigate case
and semantic roles by applying a Construction Grammar approach to this area of
research. By distinguishing different constructions, rather than focusing on case
or semantic roles as such, he manages to capture how the function of case varies in
syntactic context, and also to distinguish different functions and their grammati-
calization across these contexts and across languages. This line of research, which
is not widely developed in connection with semantic roles, is has considerable
future potential. The idea that a construction can be extended based on metaphor
and inference to different groups of verbs implies that the semantic roles assigned
to the arguments of these verb, albeit different, must be possibly conceived as simi-
lar at a higher level of schematicity. Thus, possible extension of a construction in
a given language also points toward partial similarity of semantic roles which are
consistently kept distinct by means of encoding in other languages.
In this manner, we hope and believe that the papers assembled in this vol-
ume not only make individual contributions in their specific area but also help to
advance our understanding of semantic roles as a whole and point to avenues of
future research.

Acknowledgments

We wish to thank Spike Gildea for his extremely helpful comments to all the con-
tributions to this volume, including this introduction. Heiko Narrog wishes to
acknowledge the support of the Japanese Society for the Promotion of Science
through Grant-in-Aid for Scientific Research #24520450.
 Perspectives on semantic roles 

References

Anderson, Lloyd B. 1982. The “perfect” as a universal and as a language specific category. In
Tense – Aspect. Between Semantics and Pragmatics [Typological Studies in Language 1],
Paul J. Hopper (ed.), 227–264. Amsterdam: John Benjamins.
Anderson, Lloyd B. 1986. Evidentials, paths of change, and mental maps: Typologically ­regular
asymmetries. In Evidentiality: The Linguistic Coding of Epistemology, Wallace Chafe &
­Marianne Mithun (eds), 273–312. Norwood NJ: Ablex.
Blake, Barry J. 2004. Case, 2nd edn. Cambridge: CUP.
Bornkessel, Ina, Schlesewsky, Matthias, Comrie, Bernard & Friederici, Angela D. (eds). 2006.
Semantic Role Universals and Argument Linking. Theoretical, Typological, and Psycholin-
guistic Perspectives. Berlin: Mouton de Gruyter. DOI: 10.1515/9783110219272
Comrie, Bernard. 1986. Markedness, grammar, people, and the world. In Markedness, Fred R.
Eckman, Edith Moravcsik & Jessica R. Wirth (eds), 85–106. New York NY: Plenum Press.
DOI: 10.1007/978-1-4757-5718-7_6
Comrie, Bernard & van den Berg, Helma. 2006. Experiencer constructions in Daghestanian
languages. In Bornkessel et al. (eds), 127–154.
Creissels, Denis & Mounole, Céline. 2011. Animacy and spatial cases: Typological tendencies,
and the case of Basque. In Kittilä, Västi & Ylikoski (eds), 157–182.
Croft, William. 1991. Syntactic Categories and Grammatical Relations. Chicago IL: Chicago Uni-
versity Press.
Croft, William & Poole, Keith T. 2008. Inferring universals from grammatical variation:
­Multidimensional scaling for typological analysis. Theoretical Linguistics 34(1): 1–37. DOI:
10.1515/THLI.2008.001
Cysouw, Michael. 2010. Semantic maps as metrics on meanings. Linguistic Discovery 8(1):
70–95. DOI: 10.1349/PS1.1537-0852.A.352
Dowty, David. 1991. Thematic proto-roles and argument selection. Language 67(3): 547–619.
Fillmore, Charles J. 1968. The case for case. In Universals in Linguistic Theory, Emmon Bach &
Robert T. Harms (eds), 1–88. London: Holt, Rinehart & Winston.
Fillmore, Charles J. 1971. Some problems for case grammar. Working Papers in Linguistics, Ohio
State University 10: 245–65.
Foley, & William A. & Van Valin Jr., Robert D. 1984. Functional Syntax and Universal Grammar.
Cambridge: CUP.
Gruber, Jeffrey. 1965. Studies in Lexical Relations. Ph.D. dissertation, MIT.
Haspelmath, Martin. 1997. Indefinite Pronouns. Oxford: OUP.
Haspelmath, Martin. 2003. The geometry of grammatical meaning: Semantic maps and cross-
linguistic comparison. In The New Psychology of Language, Vol. 2, Michael Tomasello (ed.),
211–43. Mahwah NY: Lawrence Erlbaum Associates.
Haspelmath, Martin. 2002. Understanding Morphology. London: Arnold.
Haspelmath, Martin. 2009. Terminology of case. In The Oxford Handbook of Case, Andrej
­Malchukov & Andrew Spencer (eds), 505–517. Oxford: OUP.
Heine, Bernd, Claudi, Ulrike & Hünnemeyer, Friederike 1991. Grammaticalization. A C ­ onceptual
Framework. Chicago IL: Chicago University Press.
Janda, Laura. 1993. The Czech Dative and the Russian Instrumental. Berlin: Mouton de Gruyter.
Kailuweit, Rolf. 2004. Protorollen und Makrorollen. In Semantische Rollen, Rolf Kailuweit &
Martin Hummel (eds), 83–103. Tübingen: Gunter Narr.
 Silvia Luraghi & Heiko Narrog

Kittilä, Seppo, Västi, Katja & Ylikoski, Jussi (eds). 2011. Case, Animacy, and Semantic Roles
[Typological Studies in Language 99]. Amsterdam: John Benjamins. DOI: 10.1075/tsl.99
König, Ekkehard & Haspelmath, Martin. 1997. Les constructions à possesseur externe dans
les langues d’Europe. In Actance et valence dans les langues d’Europe, Jack Feuillet (ed.),
525–606. Berlin: De Gruyter.
Kuryłowicz, Jerzy. 1949. Le problème du classement des cas. Biuletyn Polskiego Towarzystwa
Językoznawczego 9: 20–43. Reprinted in: Kuryłowcz, Jerzy. 1960. Esquisses linguistiques.
Wrocław: Polska Akademia Nauk.
Kuryłowcz, Jerzy. 1960. Esquisses linguistiques. Wrocław: Polska Akademia Nauk.
Kuryłowicz, Jerzy. 1964. Inflectional Categories of Indo-European. Heidelberg: Winter.
Lehmann, Christian & Yong-Min Shin. 2005. The functional domain of concomitance. A
­typological study of instrumental and comitative relations. In Typological Studies in Partici-
pation [Studia Typologica 7], Christian Lehmann (ed.), 9–104. Berlin: Akademie Verlag.
Lestrade, Sander. 2010. The best of two maps. Comment on ‘A Diachronic Dimension in Maps
of Case Functions’ by Heiko Narrog. Linguistic Discovery 8(1): 255–256. DOI: 10.1349/
PS1.1537-0852.A.375
Levin, Beth & Rappaport Hovav, Malka. 2005. Argument Realization. Cambridge MA: The MIT
Press. DOI: 10.1017/CBO9780511610479
Luraghi, Silvia & Parodi, Claudia. 2010. Key Terms in Syntax and Syntactic Theory. London:
Continuum.
Luraghi, Silvia. 2001. Some remarks on instrument, comitative, and agent in Indo-European.
Sprachtypologie und Universalienforschung 54(4): 385–401.
Luraghi, Silvia. 2003. On the Meaning of Prepositions and Cases. A Study of the Expression of
Semantic Roles in Ancient Greek [Studies in Language Companion Series 67]. Amsterdan:
John Benjamins.
Luraghi, Silvia. 1991. Paradigm size, possible syncretism, and the use of cases with adposi-
tions in inflectional languages. In Paradigms: The Economy of Inflection, Frans Plank (ed.),
57–74. Berlin: Mouton de Gruyter. DOI: 10.1515/9783110889109.57
Luraghi, Silvia. 2008. Case in Cognitive Linguistics. In The Handbook of Case, Andrej ­Malchukov
& Andrew Spencer (eds), 136–150. Oxford: OUP.
Luraghi, Silvia. 2009. Cases as radial categories: The limits of polysemy. Paper presented at the
SKY Conference on Case, 27–30 August 2009, Helsinki.
Luraghi, Silvia. 2011. Human landmarks in spatial expressions: From Latin to Romance. In
­Kittilä, Västi & Ylikoski (eds), 209–234.
Malchukov, Andrej & Narrog, Heiko. 2009. Case polysemy. In The Handbook of Case, Andrej
Malchukov & Andrew Spencer (eds), 518–534. Oxford: OUP.
Narrog, Heiko & Ito, Shinya. 2007. Reconstructing semantic maps. The comitative-instrumental
area. Sprachtypologie und Universalienforschung 60(4): 273–292.
Narrog, Heiko 2010. A diachronic dimension in maps of case functions. Linguistic Discovery
8(1): 233–254. DOI: 10.1349/PS1.1537-0852.A.352
Newmeyer, Frederick J. 2010. On comparative concepts and descriptive categories: A reply to
Haspelmath. Language 86(3): 688–695. DOI: 10.1353/lan.2010.0000
Nikiforidou, Kiki. 1991. The meaning of the genitive. Cognitive Linguistics 2(2): 149–205. DOI:
10.1515/cogl.1991.2.2.149
 Perspectives on semantic roles 

Nishimura, Yoshiki. 1993. Agentivity in cognitive grammar. In Conceptualization and Mental

Processing in Language, Richard A. Geiger & Brygida Rudzka-Ostyn (eds), 488–530. Berlin:
Mouton de Gruyter. DOI: 10.1515/9783110857108.487
Noonan, Michael. 2009. Patterns of development, patterns of syncretism of relational morphol-
ogy in the Bodic languages. In The Role of Semantics and Pragmatics in the Development
of Case [Studies in Language Companion Series 108], Johanna Barddal & Shobhanna
Chelliah (eds), 261–282. Amsterdam: John Benjamins.
Primus, Beate. 2006. Mismatches in semantic-role hierarchies and the dimensions of role
semantics. In Bornkessel et al. (eds), 89–126.
Radden, Günter. 1989. Semantic roles. In A User’s Grammar of English, René Dirven & Richard
A. Geiger (eds), 421–471. Frankfurt: Peter Lang.
Raineri, Sophie & Evola, Vito. 2008. A Construction Grammar Analysis of the Empathetic
Dative Construction in French and Italian. Ms.
Schlesinger, Izchak M. 1989. Instruments as agents: On the nature of semantic relations. Journal
of Linguistics 25(1): 189–210. DOI: 10.1017/S0022226700012147
van der Auwera, Johan. 2008. In defense of classical semantic maps. Theoretical Linguistics 34(1):
39–46. DOI: 10.1515/THLI.2008.002
Van Valin Jr., Robert D. & LaPolla, Randy J. 1997. Syntax: Structure, Meaning, and Function.
Cambridge: CUP.
Wälchli, Bernhard. 2010. Similarity semantics and building probabilistic semantic maps from
parallel texts. Linguistic Discovery 8(1): 331–271.
 Inducing semantic roles

Michael Cysouw
Philipps Universität Marburg

Instead of defining semantic roles on the basis of the interpretation of lexical

predicates, I will show that it is possible to induce semantic roles from the usage
of case-like markers across a wide variety of languages. The assumptions behind
this proposal are, first, that semantic roles are strongly contextually determined
and, second, that similarity in coding of contextual roles across many different
languages shows which contexts evoke the same (or better: very similar) semantic
roles. This approach to the investigation of semantic roles will be exemplified by
a study of case-like marking in a parallel text across a sample of fifteen languages.
On this basis, a semantic map of contextual roles can be established, and it will
be shown that higher-level abstractions, like semantic roles or even macro-roles,
can be statistically derived from this diversity of marking across many languages.
Further, a typology of alignment systems can be derived statistically.

Keywords: Semantic roles; case; parallel text; linguistic typology; alignment

1. Introduction

The notions of semantic roles or thematic relations (two terms which I will
treat synonymously here for ease of discussion) have a long tradition in linguistic
analysis (cf. Blake 1930; Fillmore 1968 for some early discussion). A semantic role
can be considered an intermediate level of abstraction in between highly abstract
proto-roles like agent or undergoer (Dowty 1991) and concrete verb-­specific
semantic roles, like the giver, knower or walker. Van Valin (2004: 64) visually
displays these levels of abstraction as a hierarchical clustering, in which lexically
specified semantic roles cluster into a smaller set of thematic relations, which in
turn cluster into a few macroroles or proto-roles.
In this paper I will propose to add an even more concrete level below this
hierarchy of roles, namely context-specific roles. The basic idea behind these
roles is that even low-level roles like the giver or knower are abstractions over the
actual occurrences of giving and knowing in context. The ultimate basic entity
is a specific participant in a concrete context in which giving is taking place, the
 Michael Cysouw

details of which are of course different in each concrete context. The verb-specific
role of the giver is a cluster of many such different (though mostly highly similar)
concrete contextual roles.
Further, I will argue that it is possible to induce higher-level role-abstractions
(alike to semantic roles or proto-roles) from the diversity of overt marking across a
wide variety of languages. Basically, the contextual distribution of case-like mark-
ers across a wide variety of languages allows for the specification of a metric on
the contextually specified roles. This metric can be interpreted as a semantic map
of contextual roles (Cysouw 2010). By using various kinds of statistical clustering,
higher-level roles can be induced from this underlying semantic map.
In this paper, I will first summarize some of the underlying assumptions on
which this kind of research is based (Section 2). I will then describe the data that
has been used for the analysis (Section 3), followed by an analysis of the contex-
tual roles in this data, inducing semantic roles (Section 4). Finally, I will discuss
the analysis of the alignment patterns of the languages investigated, arguing that
it is also possible to statistically derive a language typology from the same data
­(Section 5).

2. Using cross-linguistic variation to approach semantics

The research in this paper is based on two theoretical assumptions, namely exem-
plar semantics and the isomorphism hypothesis. These assumptions are
described in more detail in Wälchli and Cysouw (2012) and will only be sum-
marized here.
First, the isomorphism hypothesis claims that given any two meanings and
their corresponding forms in any particular language, more similar meanings are
more likely to be expressed by the same form. Individual languages will of course
dramatically diverge from this general pattern in their coding of specific mean-
ings (i.e. highly similar meanings might be formally distinguished in a specific
language, while highly divergent meanings might be coded identically). However,
by averaging over the structures of many languages, these idiosyncratic patterns
will vanish among the cross-linguistically recurrent patterns. The isomorphism
hypothesis thus implies that cross-linguistically recurrent formal similarities will
be indicative of the meanings expressed. In this interpretation, cross-linguistic
variation of formal encoding provides evidence for semantic similarity of the
encoded events.
Second, exemplar semantics is a cover term for all approaches to semantics
in which the meaning of concrete utterances (‘exemplars’) is considered more fun-
damental than the meaning of abstract concepts. The assumption is that individual
 Inducing semantic roles 

utterances have a very concrete meaning, strongly depending on the context in

which they are uttered. The ‘overall’ meaning of any linguistic formative (be it a
lexeme, morpheme, or construction) is only a coarse summary of the individual,
and highly specific, meaning each individual occurrence of the formative has in
each specific context of utterance.
Individual expressions as they occur in their context of utterance are thus con-
sidered to be the ultimate exemplars. The context of an expression can be defined
in general as the spatio-temporal surrounding of an individual expression. This
notion of context is deliberately left rather vague here because its precise delimita-
tion depends on the practical implementation in a specific empirical study. The
spatio-temporal surrounding of an expression can be defined as the sentence in
which the expression occurs, or as the complete text around the expression, or it
can even include the (real-world) socio-cultural setting, outside of the linguistic
form, in which the expression is uttered.
Translated to the concrete case of semantic roles, the assumptions behind the
current investigation are the following. First, this study is exemplar-based in that
semantic roles are considered to be strongly contextually determined. To a large
extent it is the lexical predicate that determines the roles, but other contextual
factors will further specify the precise role a participant takes in any situation.
In effect, each participant in context is assumed to perform a specific contextual
role. Second, isomorphism is assumed to be the empirical basis of this investiga-
tion. The coding of contextual roles across many different languages shows which
contexts evoke the same (or better: very similar) roles. Basically, given two par-
ticipants in different contexts, the more often these two participants are marked
identically in language after language, the more similar the contextual roles will be.
This similarity can be used to induce higher-level abstraction, like semantic roles
or proto-roles.

3. The data: Case-like marking in parallel texts

The approach to the investigation of semantic roles as described in the previous

section will be exemplified by a study of overtly marked nouns in a parallel text
across many languages. Strictly for reasons of convenience, I will use religious
brochures from watchtower.org, translations of which are available online in very
many languages. Although these are translated texts, the brochures are meant to
convince people, so the translations should be made such as to feel natural to the
readers. Still, there will be influences from translationese in these texts, so they
are probably not suitable for the investigation of the details of role marking in
individual languages.
 Michael Cysouw

However, for the present study these texts are considered to be specific
­doculects, which might, or might not, be the ‘same’ language as described in
other resources (e.g. grammars or spontaneous text collections). The question,
whether the current religious doculects are similar to other known descriptions
of the ‘same’ languages is an interesting and important question, but that is not
the goal of this paper. In this paper, I will mainly use these doculects to make
cross-linguistic comparisons, and not to analyze the details of the structure of the
individual languages. What is necessary is that the current data is a viable expres-
sion of some kind of human language (which I assume the current translation are),
not necessarily the same as other expressions used by other linguists. Further, for
the purpose of language comparison these texts are highly suitable, because they
present a clearly comparable resource across languages, which is often a difficult
point with other resources (grammars, spontaneous texts).
For this paper, I will restrict myself to bound case-like marking only,
likewise purely because of practical reasons. So, languages without bound case-
like marking are uninformative for this paper. I deliberately use the term ‘case-like’
marking, because I define such marking pragmatically for this study on a purely
orthographic basis. Whatever is written as one word together with a nominally used
root is included here as ‘case-like’ marking. For future research, a more linguisti-
cally adequate and more all-encompassing notion of flagging and cross-referencing
of noun phrases should be considered (including e.g. articles, adpositions and word
order). However, even with the currently rather limited notion of linguistic mark-
ing, it turns out that there is still enough information to induce various semantic
roles. In general, it seems to be the case that the rather coarse-grained linguistic
notions are already sufficient to investigate the typological diversity of the world’s
languages, though it should be realized that such rough approximations of linguis-
tic structure are of course not suitable for the study of the structure of individual
languages.
A further restriction of the current investigation is that I will ignore
­diatheses. Although this might seem to introduce a large amount of random-
ness in the data collected (as explained shortly), it turns out that even with this
strong simplification the main goal of the paper can still be reach, namely the
induction of semantic roles. The possible problem with ignoring diatheses is that
there seem to be many possible alternative translations of a specific contextual
expression that might mark participants differently. For example, a translation ‘to
respect [NP]’ or ‘to have respect [for NP]’ use different flagging in English (i.e. a
different appositional marking). So, depending on the decision of a translator, dif-
ferent participant marking might be used with highly similar meanings. Ideally,
one would not only want to add information about apposition and word order (as
proposed above), but also information about the diathesis used. Again, because
 Inducing semantic roles 

of practical ­considerations, I have not done so. Even more importantly, though, is
that the inductions of semantic roles still seems to work fine, as will be argued in
the rest of this paper.
To easily find comparable roles across the various translations, I have inves-
tigated the marked forms of the word ‘bible’. There are various reasons of this
decision. First, this word occurs with a high frequency in the religious brochures
from watchtower.org, so sufficient data can already be found in a rather short text.
A further profitable aspect of using the word ‘bible’ is that the bible takes on a
great variety of roles in the pamphlets. The bible occurs in agent-like roles, as in
“the bible teaches us”, but also in undergoer-like roles, as in “you should study the
bible”, or in various other roles, as in “the bible’s view” or in “to have respect for
the bible”. This variety of roles offers a suitable background for the investigation of
variation in role marking across the world’s languages. Finally, because of its high
frequency and its often rather obvious form, the word ‘bible’ is easily recognizable,
also in languages, which I am not able to read myself.
In practice, I selected 34 contexts in the pamphlets in which the word bible
occurs. Various possible contexts were removed from the selection because the
actual word for ‘bible’ was not used in a sufficient number of languages (only
pronominal cross-referencing was used in some contexts in some languages).
The English and German translations of the chosen 34 contexts are shown in
Appendix A.
Shown in Table 1 are the 15 languages sampled for the current paper. A
map of the geographic locations of these languages is shown in Appendix B. The
information on genealogical affiliation (genus, family) and geographic location
are taken from the World Atlas of Language Structures (WALS, Haspelmath et al.
2005). The languages show a wide variety of alignment structures, as summarized
in the last column of the table. The alignment of Oromo, Khoekhoe, Irish, Korean,
Drehu, Nias, Greenlandic, Aymara are due to Comrie (2005). Albanian, Faroese,
­Estonian, Azerbaijani are relatively straightforward nominative-accusative lan-
guages, like all Indo-European languages in Europe. The ergative alignment of
Akha is discussed in Terrel (2009). Ma’di is normally not considered to have case
marking (Crazzolara 1960: 20), and the nominal suffix -i which caused the inclu-
sion of this language in the sample is commonly analyzed to be some kind of
focus marking. Likewise, the Irish initial consonant mutation (which is the bound
marking attested in the world for ‘bible’) is normally not considered to be role
marking, but its behavior will turn out to be very similar to ‘regular’ case markers
of other Indo-European languages. Again, the restriction to just 34 contexts in 15
languages might be considered not sufficient, similar to the problems of ignoring
adpositions, word order, and diatheses as discussed above. However, it will turn
out to be sufficient to obtain interesting results.
 Michael Cysouw

Table 1. Language sample for the current study

Language Genus Family Alignment

Oromo Eastern Cushitic Afro-Asiatic Marked nominative

Khoekhoe Central Khoisan Khoisan Nominative-accusative
Ma’di Moru-Ma’di Nilo-Saharan Neutral
Albanian Albanian Indo-European Nominative-accusative
Irish Celtic Indo-European Neutral
Faroese Germanic Indo-European Nominative-accusative
Estonian Finnic Uralic Nominative-accusative
Altai Turkic Altaic Nominative-accusative
Azerbaijani Turkic Altaic Nominative-accusative
Korean Korean Korean Nominative-accusative
Akha Burmese-Lolo Sino-Tibetan Ergative-absolutive
Drehu Oceanic Austronesian Active-inactive
Nias Sundic Austronesian Marked absolutive
Greenlandic Eskimo-Aleut Eskimo-Aleut Ergative-absolutive
Aymara Aymaran Aymaran Marked Nominative

4. Analysis of roles

The actual word forms as attested for the word ‘bible’ in the current language
sample are shown in Appendix C. This appendix represents the basic data for the
further analyses to be performed in this paper. There are various calculations that
can be performed based on the distribution of different forms across the contexts.
First, the marking of contextual roles can be compared by investigating their
language-specific encoding. By simply counting how often two contextual roles
are marked differently in the languages sampled (and dividing this by the number
of comparisons made) an average role similarity can be established (cf. the iso-
morphism hypothesis from Section 2). For example, between the first contextual
role (“What important information is contained in the Bible?”) and the second
contextual role (“Who is the Bible’s author?”) there are ten languages that use a
different form (like in English), but five languages actually use the same form,
so the average distance between these two contextual roles is 10/15=0.67. These
computations are performed for all pairs of context, and the resulting distances
are shown in Appendix D. Note that for the establishment of these distances, there
has not been made any typological comparison between the languages. Only forms
within each language have been compared to each other. There was no decision
 Inducing semantic roles 

necessary which forms from language X should be compared to which forms from
language Y.
This distance matrix between the contextual roles represents a semantic
map on these roles, though without a graphical representation yet (for a more
detailed explanation why this really is a semantic map, see Cysouw 2010). There
is a multitude of possibilities to graphically represent the distance matrix. Figure
1 shows the first two dimensions of a multidimensional scaling (MDS) of the dis-
tance matrix. The position of the numbers in the figure is determined by the MDS,
showing similar predicate-specific roles as being close to each other. The circles
and the annotations in this figure have been added manually to indicate the close
approximation of the statistical analysis to the predicate-based notion of roles.
Remember that at no point in the analysis was any information about the lexical
verbs used to determine the positioning of the points in the figure.

7 author/council/teaching
26 24
2 of the Bible

32
live by spoken of
the Bible in the Bible
27
the Bible
what does the says/tells/refers to
Bible say? urges/teaches/
9 21 foretold/reveals
the Bible
31 was copied 15
to have respect
19 28 29
25
the Bible
for the Bible was written/ 22
has been printed
30 12
34 1 8
contained 18
6 14
to learn from in the Bible
10 the Bible 5
4
16
20 the Bible
11 is a gift
to study
23 3 33 13
the Bible
the Bible is
17 the basis

Figure 1. Semantic map of contextual roles, with hand-drawn clusters of approximate

lexically-specified roles

These first two dimensions of the MDS as shown in Figure 1 are actually still
not a particularly good approximation of the variation, as they only represent
 Michael Cysouw

about 50% of the eigenvalues. Other possibilities would be a hierarchical cluster-

ing scheme (like NeighborJoining, Saitou & Nei 1987) or a split decomposition
(Bandelt & Dress 1992, for example NeighborNet, Bryant & Moulton 2004). Such
pictures are shown in Appendix E. For the purpose of this paper, I will only use the
MDS display, as it allows to overlay other information on top of the two dimen-
sional representation.
Another way to analyze the distance matrix of the contextual roles is to per-
form ‘flat’ clustering, i.e. divide the contextual roles into mutually exclusive groups
of similar roles. For such an analysis, one has to pre-set the number of clusters,
and then an optimal division of the contextual roles into those clusters is deter-
mined. However, not every number of clusters is equally adequate. The suitability
of a flat clustering is determined, roughly speaking, by strong internal similar-
ity within each cluster and clear separability between the clusters. I will use here
the ‘partitioning around mediods’ (PAM) clustering approach by Kaufmann and
Rousseeuw (1990), with the associated measure of suitability of the clustering (the
‘average silhouette width’). The suitability of clustering for all number of clusters
from two to thirty is shown in Figure 2. The optimal clustering is found with ten
clusters, while there are suboptimal maxima at seven and three clusters.
0.30
0.25
0.20
Width
0.15
0.10
0.05

5 10 15 20 25 30
2:33

Figure 2. Suitability of the optimal clustering for different number of clusters. Shown on the
x-axis is the number of clusters, while the y-axis shows the suitability of this ‘flat’ clustering in
the form of the ‘average silhouette width.’ The best clustering is attested with 10 clusters, while
7 and 3 clusters are other good choices

The clustering of all 34 contextual roles into three groups as suggested by the
PAM-method is shown in Figure 3 (the clustering into ten and seven groups are
 Inducing semantic roles 

not shown here for reasons of space, and can be found in Appendix F). This figure
uses the same MDS display of the 34 roles as used in Figure 1, only the superim-
posed groups are different. This attested clustering shows a striking parallel to the
intuitive notion of macro-roles. Note that because the MDS and the clustering are
different mathematical methods that focus on slightly different numerical aspects
of the underlying data, the visual impression as shown in Figure 3 looks slightly
inconsistent, especially concerning the placement of contextual role number one.
However, this simply represents a role with rather undetermined correspondence
to other roles, which results in a placement in the middle of the MDS. The English
translation of the sentence in which this role occurs is “What important informa-
tion is contained in the Bible?”, which is in many languages translated without
the passive construction as found in English (e.g. German “Welchen wichtigen
Aufschluß enthält die Bibel?”).

26 2 724

32 Source

27

9 21
31 15
19 2829
25
22
30 12
34 1 8 18
6 14
10 5
16 4
20 Agent
11
233 Undergoer 33 13
17

Figure 3. Clustering of the contextual roles into three clusters, which strongly correlate with
cross-linguistic macro-roles agent, undergoer and source. The clusters are depicted on the
same MDS basis as Figure 1

5. Comparison of languages

As can be seen in the language-summary as presented in Table 1, there is a wide

variety of alignment patterns (of full noun phrase marking) available in the fifteen
sampled languages. The largest group has nominative-accusative alignment (seven
languages). There are also two languages with ‘marked’ n ­ ominative-accusative
 Michael Cysouw

alignment, which are languages in which – unexpectedly from a typological

perspective – the agent roles are formally more marked than the patient roles
(Handschuh 2011). Further, there are three languages with ergative-absolutive
alignment, among which there is a single ‘marked’ ergative-absolutive one. Finally,
there is one language that is analyzed as active-inactive, and two languages that are
normally analyzed as having no case marking, and thus are of neutral alignment.
Although such typological distinctions suggest strict categorical differences
between the languages, the attested differences are mostly much more continuous
in nature. Traditional typology relies heavily on a small set of strictly selected indi-
cators for the establishment of types. Specific characteristics are selected to define
types and to classify languages accordingly. Further, the classificatory decisions are
mostly made on the basis of secondary sources (i.e. descriptions of the languages
in question), and not on the basis of actual comparable examples.

Figure 4. Language-specific coding of the contextual roles, illustrated for albanian and
­faroese. The position of the forms is identical to the numbers in Figure 1. The clusters are
drawn using an interpolation technique called ‘kriging’. The labels are language-specific labels
as used for the description of these languages

The current dataset offers the possibility to perform a much more detailed
typological comparison. To understand how it is possible to make such compari-
sons of complete languages, consider the semantic maps of Albanian and Faroese,
as shown in Figure 4. These figures use the same MDS layout of the 34 contextual
roles as was also used in earlier figures. However, instead of plotting the numbers
referring to the clauses, in these figures the actual case-marked forms as attested
in the text are shown. To show the language-specific structure of this coding, I
have added automatically drawn clusters around identical forms, resulting in a
­special kind of semantic maps. These clusters were established by first making a
 Inducing semantic roles 

3D ­interpolation for each case-marked form, in which the height of the interpola-
tion is established by the density of the occurrence of the form in the MDS base-
map. Basically, the more forms occur close to each other, the higher the ‘mountain’
will become. This mountain is then drawn in the form of height lines at heights
0.45, 0.50 and 0.55 (which results in the slightly fuzzy appearance of the borders).
More details about this approach to draw semantic maps can be found in Cysouw
and Forker (2009). Labels were manually added to identify the clusters. Note that
these labels are capitalized, as they are names for language-specific structures and
not cross-linguistic categories. I have produced such semantic maps for all lan-
guages in the sample, using exactly the same graphical settings so the resulting
pictures can be visually compared to each other. For reasons of space and because
such semantic maps are much easier to interpret when using colors, the pictures
are not included in this printed article, but can be found in Appendix G.
Looking at Figure 4, the case marking structure of the two languages seems
pretty much alike, pace for the addition of a separate Dative in Faroese. This impres-
sion of relative similarity between two languages can be easily formalized into a
general measurement of language similarity. Basically, for each language I consider
all 1122 (= 34 × 33) pairs of contextual roles, which can have either ­identical (= 1)
or different (= 0) case marking (see Cysouw 2010 for more details on the estab-
lishment of such language-specific metrics). Two languages can be compared by
comparing these 1122 pairs between the two languages, e.g. by taking a Pearson
correlation coefficient between them. This similarity between two ­languages can
then be computed for all pairs of languages (see the results in Appendix H), and the
resulting metric on the languages can be interpreted as a ‘typology without types’.
In such a typology, all languages are compared to each other, and the resulting
grouping of languages can be investigated with various statistics techniques, just
as already roughly outlined in Section 4. Shown here in Figure 5 is a NeighborNet
illustrating the structure of the similarities between the languages.
There are various interesting observations to be made on the basis of this
‘typology without types’. First, the languages to the left include two ergative lan-
guages (Nias, Akha), but also Ma’di and Drehu, which are not normally considered
to be ergative. Looking at the semantic maps for these languages (cf. Appendix
G), the characteristic binding these languages together is the existence of a case-
like marker that is used in a wide variety of contextual roles (spanning almost
the complete set of 34 contextual roles sampled), including all typically patient-
like roles (cf. Figure 3). This can be interpreted as that these ‘ergative’ markers
are functionally unmarked (i.e. they occur in most contextual roles). This makes
also sense for the analysis of Ma’di, as the case-like markers in question here are
traditionally analyzed as being markers of ‘focus’ information structure. However,
the unmarked form occurs in a wide variety of context, while the marked ‘focus-
marking’ suffix mainly occurs in transitive agent like contexts, similar to ergatives.
 Michael Cysouw

Greenlandic
Khoekhoe
Aymara
Drehu

Faroese
Nias

Korean

Ma’di Albanian

Akha
Oromo Estonian Irish

Azerbaijani
Altai

Figure 5. Neighbornet of languages according to their similarity in case marking structure.

The languages to the left typically have a widespread marker that is also used for patient-like
contexts. This group includes typically ergative languages. In contrast, the languages to the
right are typically nominative languages, while aymara and oromo are of the ‘marked’ nomina-
tive type. Greenlandic is unexpectedly grouped with the nominative languages

Further note that Greenlandic, which is traditionally defined as being erga-

tive, does not occur close to these languages on the left side. This is basically due
to the fact that the structure of marking is completely different in Greenlandic.
Although there is a specific case also used for transitive agents (the traditional
definition of ergativity), there is no complementary ‘unmarked’ case used for a
large group of the remaining roles.
All languages on the right are basically nominative-accusative languages,
though Azerbaijani and Altai seem to stand out. It is not clear to me why this
happens, as the structure of the case marking in these languages does look very
similar. However, they both seem to have rather different structures from all other
languages, so they might have been grouped together because of their shared dis-
similarity from all other languages considered here.
Finally, Aymara and Oromo are found in between the ergative languages to the
left and the nominative languages to the right. Both these languages are ­considered
to be ‘marked’ nominative in that formally the marking of the nominative is overt,
while the accusative is formally unmarked. For Oromo this formal marking struc-
ture is also reflected in the functional marking structure, as the unmarked patient-
like case marker also shows a wide distribution over the 34 contextual roles. The
reason for the intermediate status of Aymara is not immediately obvious to me.
 Inducing semantic roles 

In summary, it is possible to statistically classify languages on the basis of their

language-specific marking of contextual roles. The resulting alignment typology is
somewhat similar to the traditional nominative-ergative typology, though much
more emphasis is put on the extent of the distribution of the cases. Languages with
the same kind of distribution of cases over roles are grouped together, which in
practice gives a stronger weight to similarity between forms that are widespread
(i.e. functionally unmarked) and does not assign much influence to details of the
highly specific marked structures.

6. Conclusion

Based on an admittedly rather limited data set, this paper has shown the viability
of using contextual roles as a basis for the typological comparison of roles in the
world’s languages. Contextual roles are the actual roles as they occur in context.
Such roles are of course strongly determined by the lexical predicate used in the
sentence, but also other linguistic coding implicitly is included in the determina-
tion of the marking. To be able to compare contextual roles across languages it is
necessary to have access to some kind of parallel text, be it in the form of transla-
tions (as used in the present study) or in the form of more experimentally con-
trolled parallel utterances (e.g. using pictures, films, or other stimuli).
Clustering of the formal marking of these parallel contextual roles offers the
possibility to statistically derive higher-level role abstractions, very much alike to
traditional predicate-based roles or even macro-roles. From the same data it is also
possible to establish similarities between complete languages, resulting in a ‘typol-
ogy without types’, i.e. a measurement of fine-grained similarities between languages
from which more coarse-grained typological clusters of languages (alike to tradi-
tional ‘types’) can be derived. Looking forward, the fine-grained typology does seem
to offer fascinating possibilities to much more accurately capture the real diversity of
the world’s languages, which normally only under protest agree to be classified into
a few broad all-encompassing types. The real challenge for future research is not
only to formulate such fine-grained typologies, but to successfully show how to they
can elucidate correlations and/or restrictions on linguistic structures.

Acknowledgements

This paper was originally written in June 2010 and presented to audiences in
Leipzig (Workshop “The Fine Structure of Grammatical Relations”, December
2010), Zürich (Workshop “Role Complexes”, April 2011), Nijmegen (February
2012), Tübingen (May 2012) and Stockholm (September 2012). I thank the audi-
ences for their input and suggestions to improve the presentation of this paper.
 Michael Cysouw

I also thank one anonymous review from Benjamins for constructive criticism
and spotting many typos. The support of DFG-grant “Algorithmic corpus-based
approaches to typological comparison” is gratefully acknowledged.

References

Bandelt, Hans-Jorgen & Dress, Andreas W.M. 1992. Split decomposition: A new and useful
approach to phylogenetic analysis of distance data. Molecular Phylogenetics and Evolution
1(3): 242–252. DOI: 10.1016/1055-7903(92)90021-8
Blake, Frank R. 1930. A semantic analysis of case. Language 6(4): 34–49. DOI: 10.2307/521984
Bryant, David & Moulton, Vincent. 2004. Neighbor-Net: An agglomerative method for the
­construction of phylogenetic networks. Molecular Biology and Evolution 21(2): 255–265.
DOI: 10.1093/molbev/msh018
Comrie, Bernard. 2005. Alignment of case marking of full noun phrases. In Haspelmath et al.
(eds), 398–405.
Crazzolara, J.P. 1960. A Study of the Logbara (Ma’di) Language. London: OUP.
Cysouw, Michael. 2010. Semantic maps as metrics on meaning. Linguistic Discovery 8(1): 70–95.
DOI: 10.1349/PS1.1537-0852.A.346
Cysouw, Michael & Forker, Diana. 2009. Reconstruction of morphosyntactic function:
­Non-spatial usage of spatial case marking in Tsezic. Language 85(3): 588–617. DOI: 10.1353
/lan.0.0147
Dowty, David. 1991. Thematic proto-roles and argument selection. Language 67(3): 547–619.
Fillmore, Charles J. 1968. The case for case. In Universals in Linguistic Theory, Emmon Bach &
Robert T. Harms (eds), 1–88. New York NY: Holt, Rinehart and Winston.
Handschuh, Corinna. 2011. A Typology of Marked-S Languages. Ph.D. dissertation, University
of Leipzig.
Haspelmath, Martin, Dryer, Matthew S., Comrie, Bernard & Gil, David (eds). 2005. The World
Atlas of Language Structures. Oxford: OUP.
Kaufman, Leonard & Rousseeuw, Peter J.1990. Finding Groups in Data: An Introduction to Clus-
ter Analysis [Series in Applied Probability and Statistics]. Hoboken NJ: Wiley. DOI: 10.1002
/9780470316801
Saitou, Naruya & Nei, Masatoshi. 1987. The neighbour-joining method: A new method for
reconstructing phylogenetic trees. Molecular Biology and Evolution 4(4): 406–425.
Terrell, Jake. 2009. Semantic case marking in Akha. University of Hawai’i Department of Linguis-
tics Working Papers in Linguistics 40(3): 1–11.
Van Valin Jr., Robert D. 2004. Semantic macroroles in Role and Reference grammar. In Seman-
tische Rollen, Rolf Kailuweit & Martin Hummel (eds), 62–82. Narr: Tübingen.
Wälchli, Bernhard & Cysouw, Michael. 2012. Lexical typology through similarity semantics:
Toward a semantic map of motion verbs. Linguistics 50(3): 671–710. DOI: 10.1515/ling-
2012-0021

Appendices

The following appendices with all data and other supplemental material can be accessed online
at 〈Benjamins Online Repository?〉
 Inducing semantic roles 

Appendix A: Sampled contexts

This appendix lists the English and German translations of the 34 contexts chosen for this study.
For reasons of space only a single sentence is presented here as the context, though it is crucial for
the approach described in this paper that these are not translations of sentences in isolation, but
translations of expressions in a larger context. The impetus for a translation is thus not to capture
just this one sentence, but to present this sentence in a suitable way given the surrounding text.

1. What important information is contained in the Bible?

Welchen wichtigen Aufschluß enthält die Bibel?
2. Who is the Bible’s author?
Wer ist der Autor der Bibel?
3. Why should you study the Bible?
Warum sollte man die Bibel studieren?
4. The Bible is a precious gift from God.
Die Bibel ist ein wertvolles Geschenk Gottes.
5. The Bible alone tells us what we must do to please God.
Nur die Bibel sagt uns, was wir tun müssen, um Gott zu gefallen.
6. The Bible was written by some 40 different men over a period of 1,600 years, beginning in
1513 B.C.E.
Beginnend mit dem Jahr 1513 v.u.Z., wurde die Bibel über einen Zeitraum von 1600 jahren
von etwa 40 verschiedenen Männern geschrieben.
7. So God in heaven, not any human on earth, is the Author of the Bible.
Demnach ist der Autor der Bibel der Gott des Himmels und nicht irgendein Mensch auf
der Erde.
8. God made sure that the Bible was accurately copied and preserved.
Gott sorgte dafür, daß genaue Abschriften der Bibel angefertigt und aufbewahrt wurden.
9. More Bibles have been printed than any other book.
Von der Bibel sind mehr Exemplare gedruckt worden als von irgendeinem anderen Buch.
10. Not everyone will be happy to see you studying the Bible, but do not let that stop you.
Nicht jeder wird darüber erfreut sein, wenn wir die Bibel studieren, aber deshalb sollten
wir uns nicht davon abhalten lassen.
11. But the Bible tells us that there is only one TRUE God.
Aber die Bibel sagt uns, daß es nur einen wahren Gott gibt.
12. But when the Bible was written, the name Jehovah appeared in it some 7,000 times
Doch als die Bibel geschrieben wurde, erschien der Name Jehova darin ungefähr 7000mal.
13. “God is a Spirit,” says the Bible.
Die Bibel sagt: “Gott ist ein geist.”
14. The Bible reveals Jehovah’s personality to us.
Die Bibel gibt uns Aufschluß über die Persönlichkeit Jehovas.
15. The Bible tells us that he is also merciful, kind, forgiving, generous, and patient.
In der Bibel wird auch gesagt, daß er barmherzig, freundlich, zum vergeben bereit, großzü-
gig und geduldig ist.
16. We learn about God from creation and from the Bible.
Wir lernen Gott durch die Schöpfung und durch die Bibel kennen.
17. Another way we can learn about God is by studying the Bible.
Zum anderen können wir mehr über Gott lernen, wenn wir die Bibel studieren.
18. By disobeying God’s command, the first man, Adam, committed what the Bible calls “sin.”
 Michael Cysouw

Durch seinen Ungehorsam gegenüber Gottes Gebot beging der erste Mensch, Adam,
“Sünde”, wie die Bibel es nennt.
19. This is what the Bible refers to as the “ransom.”
Er konnte deshalb sein vollkommenes menschliches Leben opfern, um damit für Adams
Sünde zu bezahlen – in der Bibel „Lösegeld“ genannt.
20. Some of your loved ones may become very angry because you are studying the Bible.
 Einige unserer lieben Angehörigen werden über unser Bibelstudium vielleicht sehr
verärgert sein.
21. What is the Bible’s view of separation and of divorce?
Was sagt die Bibel über Trennung und Ehescheidung?
22. The Bible says that a husband is the head of his family.
Nach der Bibel ist der Mann das Haupt der Familie.
23. Parents need to spend time with their children and study the Bible with them, caring for
their spiritual and emotional needs.
Eltern müssen sich für ihre Kinder Zeit nehmen, mit ihnen die Bibel studieren und sich
um ihre geistigen und auch emotionellen Bedürfnisse kümmern.
24. When marriage mates have problems getting along together, they should try to apply Bible
counsel.
Wenn es Ehepartnern schwerfällt, miteinander auszukommen, sollten sie versuchen, den
Rat der Bibel anzuwenden.
25. The Bible urges us to show love and to be forgiving.
In der Bibel werden sie dazu aufgefordert, liebevoll und zum vergeben bereit zu sein.
26. But God does not approve of them if they come from false religion or are against Bible teachings.
Aber wenn sie ihren Ursprung in der falschen Religion haben oder den Lehren der Bibel
widersprechen, werden sie von Gott nicht gebilligt.
27. The only two birthday celebrations spoken of in the Bible were held by persons who did not
worship Jehovah.
Die beiden einzigen in der Bibel erwähnten Geburtstagsfeiern wurden von Personen
veranstaltet, die jehova nicht anbeteten.
28. The Bible teaches that only a few people are on the narrow road to life.
Aus der Bibel lernen wir, daß sich auf dem schmalen Weg, der zum Leben führt, nur
wenige Menschen befinden.
29. The Bible foretold that after the death of the apostles, wrong teachings and unchristian
practices would slowly come into the Christian congregation.
In der Bibel wurde vorausgesagt, daß nach dem Tod der Apostel falsche Lehren und
unchristliches Handeln langsam eingang in die Christenversammlung finden würden.
30. True Christians love one another, respect the Bible, and preach about God’s Kingdom.
Wahre Christen haben Liebe zueinander, haben Respekt vor der Bibel und verkündigen
Gottes Königreich.
31. Another mark of true religion is that its members have a deep respect for the Bible.
Ein anderes Merkmal der wahren Religion ist, daß ihre Anhänger tiefen Respekt vor der
Bibel haben.
32. They try to live by the Bible in their everyday life.
Im täglichen Leben versuchen sie, sich nach der Bibel auszurichten.
33. The Bible is the basis for what is taught.
Die Bibel dient als Lehrbuch.
34. By now you have learned many good things from the Bible.
Inzwischen haben wir aus der Bibel viel gutes gelernt.
 Inducing semantic roles 

Appendix B: Map of languages

Greenlandic (West)
Faroese
Estonian
lrish Altai (southern)
German
Albanian Azerbaijani Korean

Akha

Oromo (Harar)
Ma’di Nias

Aymara
Drehu
Khoekhoe
 Michael Cysouw

Appendix C: Wordforms of the word for ‘bible’ in the 34 contexts

This table gives the complete data that was the basis for the analyses of this paper. It presents
all words for ‘bible’ as they were attested in the specific contexts. Note that the actual language-
specific analysis does not necessarily call these forms case marking. However, they seem to be
strongly correlated with the marking of roles in all languages. To be able to quickly check the
computations, I have conveniently added German to the list, including the case-marked articles.
These forms are not part of the data selection for this paper.
 Inducing semantic roles 

Contexts German Altai Albanian Irish Faroese Estonian Aymara Greenlandic Akha Azerbaijani Nias Khoekhoe Oromo Madi Drehu Korean
1 die Bibel библияда bibla bhíobla bíbliuni piibel bibliaxa biibili jaˇliˆ müqəddəskitabda zurani’amoni’õ bybelsa macaafaqulqulluu ofuoleri tusihmitrötr 성서에는
2 der Bibel NA biblës bhíobla bíbliunnar piibli bibliax biibilimik jaˇliˆ müqəddəskitabın surani’amoni’õ bybels macaafaqulqulluu ofuoleri tusihmitrötr 성서의
3 die Bibel библияны biblën mbíobla bíbliuna piiblit bibliat biibili jaˇliˆ müqəddəskitabı zurani’amoni’õ bybelsa macaafaqulqulluu ofuoleri tusihmitrötr 성서를
4 die Bibel библия bibla bíobla bíblian piibel bibliax biibili jaˇliˆaˇ müqəddəskitab surani’amoni’õ bybels macaafniqulqulluun ofuolerii tusihmitrötre 성서는
5 die Bibel NA bibla bíobla bíblian piibel bibliakiw biibilip jaˇliˆ müqəddəskitabdan surani’amoni’õ bybels macaafaqulqulluu ofuolerii tusihmitrötre 성서만이
6 die Bibel библияны bibla bhíobla bíbliuna piibli bibliax biibili jaˇliˆaˇ müqəddəskitabı surani’amoni’õ bybels macaafniqulqulluun ofuolerii tusihmitrötre 성서는
7 der Bibel библияныҥ biblës bhíobla bíbliunnar piibli bibliaxa biibilimut jaˇliˆaˇ müqəddəskitabın surani’amoni’õ bybels macaafaqulqulluu ofuoleri tusihmitrötr 성서의
8 der Bibel библияны bibla bíobla bíblian piiblit bibliaxa biibilip jaˇliˆaˇ müqəddəskitabın surani’amoni’õ bybelsa macaafniqulqulluun ofuolerii tusihmitrötr 성서가
9 der Bibel библия bibla bhíobla NA piiblit bibliawa biibili jaˇliˆaˇ müqəddəskitabın surani’amoni’õ bybeldi macaafaqulqulluutti ofuolerii tusihmitrötr 성서는
10 die Bibel библия biblën bhíobla bíbliuna piiblit bibliat biibilimik jaˇliˆ müqəddəskitabı zurani’amoni’õ bybel macaafaqulqulluu ofuoleri tusihmitrötr 성서를
11 die Bibel библияда bibla bíobla bíblian piibel bibliaxa biibilimili jaˇliˆ müqəddəskitabda surani’amoni’õ bybelsa macaafniqulqulluun ofuoleri tusihmitrötre 성서는
12 die Bibel библияныҥ bibla bíobla bíblian piibel bibliax biibilili jaˇliˆ müqəddəskitabın zurani’amoni’õ bybels macaafniqulqulluun ofuolerii tusihmitrötr 성서가
13 die Bibel библияда bibla bíobla bíblian piibel biblianxa biibilimi NA müqəddəskitab zurani’amoni’õ bybelsa macaafniqulqulluun ofuoleri tusihmitrötre 성서는
14 die Bibel библия bibla bíobla bíblian piibel bibliax biibilimi jaˇliˆaˇ müqəddəskitab surani’amoni’õ bybels macaafniqulqulluun ofuoleri tusihmitrötre 성서는
15 der Bibel библияда bibla bíobla bíblian piibel NA biibilimi jaˇliˆ müqəddəskitabda surani’amoni’õ bybels macaafniqulqulluun ofuoleri tusihmitrötr 성서는
16 die Bibel библия bibla bíobla bíbliuni piibli bibliampita biibililu jaˇliˆahˇ müqəddəskitabdan zurani’amoni’õ bybelsa macaafaqulqulluurraa ofuoleri tusihmitrötr 성서로부터
17 die Bibel библияны biblën mbíobla bíbliuna piiblit bibliat biibilimik jaˇliˆahˇ müqəddəskitabı zurani’amoni’õ bybelsa macaafaqulqulluu ofuoleri tusihmitrötr 성서를
18 die Bibel библияда bibla bíobla bíblian piiblis biblianxa biibilip jaˇliˆ müqəddəskitabda zurani’amoni’õ bybels macaafniqulqulluun ofuoleri tusihmitrötr 성서에서
19 der Bibel библияда bibla bíobla bíblian piiblis biblianxa biibilimi jaˇliˆ müqəddəskitabda zurani’amoni’õ bybels macaafniqulqulluun ofuoleri tusihmitrötre 성서에서는
20 Bibelstudium библия biblën mbíobla bíbliuna piiblit bibliat biibilimik jaˇliˆ müqəddəskitabı zurani’amoni’õ bybelsa macaafaqulqulluu ofuoleri tusihmitrötr 성서를
21 die Bibel библияныҥ NA bhíobla bíblian piibel bibliax biibilimi jaˇliˆ müqəddəskitabın zurani’amoni’õ bybels macaafniqulqulluun ofuoleri tusihmitrötre 성서의
22 der Bibel библияда bibla bíobla bíbliuni piibel biblianxa biibili jaˇliˆ müqəddəskitabda surani’amoni’õ bybels macaafniqulqulluun ofuoleri tusihmitrötre 성서는
23 die Bibel библияны biblën mbíobla bíbliuna piiblit bibliat biibilimillu jaˇliˆ müqəddəskitabı zurani’amoni’õ bybelsa macaafaqulqulluu ofuoleri tusihmitrötr 성서를
24 der Bibel библиялык biblike bhíobla bíblian piibli biblian biibilimi jaˇliˆ müqəddəskitabın zurani’amoni’õ bybels macaafaqulqulluurratti ofuoleri tusihmitrötr 성서의
25 der Bibel библия bibla bíobla bíblian piibel bibliax biibilimi jaˇliˆ müqəddəskitab surani’amoni’õ bybels macaafniqulqulluun ofuoleri tusihmitrötre 성서는
26 der Bibel библиялык biblës bhíobla bíbliunnar piibli NA biibilimi jaˇliˆ müqəddəskitabın zurani’amoni’õ bybels macaafaqulqulluu ofuoleri tusihmitrötr NA
27 der Bibel библияда bibla bhíobla bíblian piiblis biblian biibilimi jaˇliˆ müqəddəskitabda zurani’amoni’õ bybels macaafaqulqulluu ofuoleri tusihmitrötr 성서에
28 der Bibel библия bibla bíobla bíblian piibel bibliaxa biibilimi jaˇliˆ müqəddəskitab surani’amoni’õ bybels macaafniqulqulluun ofuoleri tusihmitrötre 성서는
29 der Bibel библияда bibla bíobla bíblian piibel bibliaxa biibilimi jaˇliˆ müqəddəskitabda surani’amoni’õ bybels macaafniqulqulluun ofuoleri tusihmitrötre 성서는
30 der Bibel библияны biblën bhíobla bíbliuna piiblist bibliaruxa biibilimik jaˇliˆahˇ müqəddəskitaba zurani’amoni’õ bybelsa macaafaqulqulluu ofuoleri tusihmitrötr 성서를
31 der Bibel библияны biblën bhíobla bíbliuni piibli bibliar biibilimik jaˇliˆahˇ müqəddəskitaba zurani’amoni’õ bybelsa macaafaqulqulluudhaaf ofuoleri tusihmitrötr 성서에
32 der Bibel библия biblës bhíobla bíbliuni piibli NA biibili jaˇliˆ müqəddəskitaba zurani’amoni’õ bybels NA ofuoleri tusihmitrötr 성서에
33 die Bibel библия bibla bíobla bíbliuna piibel bibliat biibilimik jaˇliˆahˇ müqəddəskitabdır surani’amoni’õ bybels macaafaqulqulluurratti ofuoleri tusihmitrötre 성서입니다
34 der Bibel библиянаҥ bibla mbíobla bíbliuni piiblist bibliatxa biibilimeersunik jaˇliˆahˇ müqəddəskitabdan zurani’amoni’õ bybelsa macaafaqulqulluurraa ofuoleri tusihmitrötr 성서에서
 Michael Cysouw

Appendix D: Contextual role distances

This table represents the average coding similarity between all pairs of contexts. Given two lines
in Appendix B, this matrix simply represents for every pair of lines in that table, the relative
number of differences in marking across all 15 languages.

1 2 3 4 5

0 0.67 0.5 0.75 0.67

0.67 0 0.73 0.8 0.73
0.5 0.73 0 0.875 0.8
0.75 0.8 0.875 0 0.4
0.67 0.73 0.8 0.4 0
0.75 0.67 0.6875 0.3125 0.6
0.6875 0.2 0.8125 0.8125 0.8
0.75 0.73 0.75 0.5625 0.6
0.73 0.642857142857143 0.8 0.53 0.785714285714286
0.5625 0.6 0.25 0.875 0.8
0.4375 0.8 0.75 0.4375 0.466666666666667
0.625 0.67 0.75 0.4375 0.466666666666667
0.53 0.928571428571429 0.73 0.4 0.571428571428571
0.75 0.73 0.875 0.125 0.466666666666667
0.53 0.571428571428571 0.8 0.466666666666667 0.5
0.5625 0.8 0.6875 0.75 0.73
0.625 0.73 0.125 0.9375 0.866666666666667
0.5 0.73 0.6875 0.625 0.53
0.625 0.73 0.8125 0.625 0.6
0.625 0.67 0.1875 0.9375 0.866666666666667
0.6 0.5 0.73 0.53 0.571428571428571
0.5 0.67 0.8125 0.4375 0.53
0.5625 0.73 0.0625 0.9375 0.8
0.6875 0.4 0.75 0.875 0.8
0.75 0.6 0.875 0.25 0.466666666666667
0.571428571428571 0.153846153846154 0.642857142857143 0.928571428571429 0.769230769230769
0.4375 0.53 0.6875 0.8125 0.67
0.6875 0.67 0.875 0.3125 0.466666666666667
0.5625 0.67 0.875 0.4375 0.466666666666667
0.625 0.6 0.4375 1 0.933333333333333
0.625 0.6 0.625 1 1
0.5 0.384615384615385 0.642857142857143 0.785714285714286 0.846153846153846
0.75 0.73 0.75 0.5 0.53
0.625 0.8 0.6875 0.9375 0.866666666666667

(Continued)
 Inducing semantic roles 

6 7 8 9 10

0.75 0.6875 0.75 0.73 0.5625

0.67 0.2 0.73 0.642857142857143 0.6
0.6875 0.8125 0.75 0.8 0.25
0.3125 0.8125 0.5625 0.53 0.875
0.6 0.8 0.6 0.785714285714286 0.8
0 0.6875 0.625 0.53 0.75
0.6875 0 0.625 0.6 0.75
0.625 0.625 0 0.466666666666667 0.875
0.53 0.6 0.466666666666667 0 0.73
0.75 0.75 0.875 0.73 0
0.625 0.8125 0.5625 0.8 0.8125
0.625 0.75 0.5 0.73 0.75
0.67 0.933333333333333 0.67 0.857142857142857 0.8
0.4375 0.75 0.625 0.67 0.8125
0.67 0.67 0.53 0.642857142857143 0.8
0.8125 0.8125 0.75 0.8 0.6875
0.75 0.8125 0.75 0.866666666666667 0.25
0.75 0.8125 0.625 0.866666666666667 0.6875
0.75 0.8125 0.6875 0.866666666666667 0.8125
0.875 0.8125 0.8125 0.8 0.1875
0.6 0.6 0.8 0.857142857142857 0.67
0.5625 0.75 0.6875 0.67 0.875
0.75 0.8125 0.75 0.866666666666667 0.25
0.8125 0.5 0.75 0.73 0.6875
0.5625 0.75 0.625 0.67 0.8125
0.785714285714286 0.285714285714286 0.785714285714286 0.692307692307692 0.571428571428571
0.8125 0.625 0.75 0.73 0.625
0.625 0.6875 0.5625 0.67 0.8125
0.625 0.6875 0.5625 0.73 0.875
0.8125 0.6875 0.75 0.8 0.4375
0.8125 0.6875 0.75 0.8 0.625
0.714285714285714 0.5 0.857142857142857 0.615384615384615 0.571428571428571
0.625 0.8125 0.8125 0.8 0.625
0.9375 0.8125 0.75 0.8 0.8125

(Continued)
 Michael Cysouw

11 12 13 14 15

0.4375 0.625 0.53 0.75 0.53

0.8 0.67 0.928571428571429 0.73 0.571428571428571
0.75 0.75 0.73 0.875 0.8
0.4375 0.4375 0.4 0.125 0.466666666666667
0.466666666666667 0.466666666666667 0.571428571428571 0.466666666666667 0.5
0.625 0.625 0.67 0.4375 0.67
0.8125 0.75 0.933333333333333 0.75 0.67
0.5625 0.5 0.67 0.625 0.53
0.8 0.73 0.857142857142857 0.67 0.642857142857143
0.8125 0.75 0.8 0.8125 0.8
0 0.5625 0.266666666666667 0.375 0.266666666666667
0.5625 0 0.53 0.5 0.466666666666667
0.266666666666667 0.53 0 0.266666666666667 0.357142857142857
0.375 0.5 0.266666666666667 0 0.33
0.266666666666667 0.466666666666667 0.357142857142857 0.33 0
0.6875 0.6875 0.6 0.6875 0.73
0.8125 0.8125 0.73 0.875 0.866666666666667
0.4375 0.4375 0.4 0.5625 0.33
0.4375 0.5625 0.33 0.5 0.266666666666667
0.8125 0.8125 0.8 0.875 0.8
0.53 0.33 0.428571428571429 0.4 0.5
0.3125 0.625 0.4 0.4375 0.2
0.75 0.75 0.73 0.875 0.8
0.8125 0.625 0.73 0.75 0.53
0.375 0.5 0.33 0.125 0.2
0.857142857142857 0.642857142857143 0.769230769230769 0.785714285714286 0.571428571428571
0.625 0.625 0.6 0.6875 0.33
0.3125 0.5625 0.33 0.1875 0.2
0.1875 0.5625 0.33 0.3125 0.067
0.875 0.875 0.8 0.9375 0.8
0.875 0.875 0.8 0.9375 0.8
0.857142857142857 0.714285714285714 0.846153846153846 0.785714285714286 0.642857142857143
0.5625 0.6875 0.6 0.4375 0.6
0.8125 0.8125 0.73 0.875 0.73

(Continued)
 Inducing semantic roles 

16 17 18 19 20

0.5625 0.625 0.5 0.625 0.625

0.8 0.73 0.73 0.73 0.67
0.6875 0.125 0.6875 0.8125 0.1875
0.75 0.9375 0.625 0.625 0.9375
0.73 0.866666666666667 0.53 0.6 0.866666666666667
0.8125 0.75 0.75 0.75 0.875
0.8125 0.8125 0.8125 0.8125 0.8125
0.75 0.75 0.625 0.6875 0.8125
0.8 0.866666666666667 0.866666666666667 0.866666666666667 0.8
0.6875 0.25 0.6875 0.8125 0.1875
0.6875 0.8125 0.4375 0.4375 0.8125
0.6875 0.8125 0.4375 0.5625 0.8125
0.6 0.73 0.4 0.33 0.8
0.6875 0.875 0.5625 0.5 0.875
0.73 0.866666666666667 0.33 0.266666666666667 0.8
0 0.625 0.625 0.75 0.6875
0.625 0 0.75 0.875 0.1875
0.625 0.75 0 0.25 0.75
0.75 0.875 0.25 0 0.8125
0.6875 0.1875 0.75 0.8125 0
0.8 0.8 0.53 0.466666666666667 0.8
0.75 0.9375 0.4375 0.3125 0.875
0.6875 0.125 0.6875 0.8125 0.1875
0.75 0.8125 0.625 0.5625 0.75
0.75 0.9375 0.5625 0.375 0.8125
0.714285714285714 0.714285714285714 0.642857142857143 0.571428571428571 0.642857142857143
0.75 0.75 0.375 0.3125 0.6875
0.75 0.9375 0.5625 0.375 0.8125
0.8125 0.9375 0.4375 0.25 0.875
0.6875 0.3125 0.8125 0.8125 0.4375
0.5625 0.5 0.8125 0.8125 0.625
0.571428571428571 0.785714285714286 0.642857142857143 0.642857142857143 0.642857142857143
0.625 0.625 0.6875 0.6875 0.6875
0.4375 0.625 0.6875 0.75 0.6875

(Continued)
 Michael Cysouw

21 22 23 24 25

0.6 0.5 0.5625 0.6875 0.75

0.5 0.67 0.73 0.4 0.6
0.73 0.8125 0.0625 0.75 0.875
0.53 0.4375 0.9375 0.875 0.25
0.571428571428571 0.53 0.8 0.8 0.466666666666667
0.6 0.5625 0.75 0.8125 0.5625
0.6 0.75 0.8125 0.5 0.75
0.8 0.6875 0.75 0.75 0.625
0.857142857142857 0.67 0.866666666666667 0.73 0.67
0.67 0.875 0.25 0.6875 0.8125
0.53 0.3125 0.75 0.8125 0.375
0.33 0.625 0.75 0.625 0.5
0.428571428571429 0.4 0.73 0.73 0.33
0.4 0.4375 0.875 0.75 0.125
0.5 0.2 0.8 0.53 0.2
0.8 0.75 0.6875 0.75 0.75
0.8 0.9375 0.125 0.8125 0.9375
0.53 0.4375 0.6875 0.625 0.5625
0.466666666666667 0.3125 0.8125 0.5625 0.375
0.8 0.875 0.1875 0.75 0.8125
0 0.6 0.73 0.4 0.4
0.6 0 0.875 0.75 0.3125
0.73 0.875 0 0.75 0.875
0.4 0.75 0.75 0 0.625
0.4 0.3125 0.875 0.625 0
0.461538461538462 0.714285714285714 0.642857142857143 0.214285714285714 0.642857142857143
0.53 0.5625 0.6875 0.375 0.5625
0.466666666666667 0.3125 0.875 0.625 0.0625
0.466666666666667 0.1875 0.875 0.625 0.1875
0.8 0.875 0.4375 0.6875 0.875
0.8 0.8125 0.625 0.625 0.875
0.615384615384615 0.571428571428571 0.714285714285714 0.428571428571429 0.642857142857143
0.67 0.5625 0.75 0.8125 0.5
0.866666666666667 0.75 0.6875 0.75 0.8125

(Continued)
 Inducing semantic roles 

26 27 28 29 30

0.571428571428571 0.4375 0.6875 0.5625 0.625

0.153846153846154 0.53 0.67 0.67 0.6
0.642857142857143 0.6875 0.875 0.875 0.4375
0.928571428571429 0.8125 0.3125 0.4375 1
0.769230769230769 0.67 0.466666666666667 0.466666666666667 0.933333333333333
0.785714285714286 0.8125 0.625 0.625 0.8125
0.285714285714286 0.625 0.6875 0.6875 0.6875
0.785714285714286 0.75 0.5625 0.5625 0.75
0.692307692307692 0.73 0.67 0.73 0.8
0.571428571428571 0.625 0.8125 0.875 0.4375
0.857142857142857 0.625 0.3125 0.1875 0.875
0.642857142857143 0.625 0.5625 0.5625 0.875
0.769230769230769 0.6 0.33 0.33 0.8
0.785714285714286 0.6875 0.1875 0.3125 0.9375
0.571428571428571 0.33 0.2 0.067 0.8
0.714285714285714 0.75 0.75 0.8125 0.6875
0.714285714285714 0.75 0.9375 0.9375 0.3125
0.642857142857143 0.375 0.5625 0.4375 0.8125
0.571428571428571 0.3125 0.375 0.25 0.8125
0.642857142857143 0.6875 0.8125 0.875 0.4375
0.461538461538462 0.53 0.466666666666667 0.466666666666667 0.8
0.714285714285714 0.5625 0.3125 0.1875 0.875
0.642857142857143 0.6875 0.875 0.875 0.4375
0.214285714285714 0.375 0.625 0.625 0.6875
0.642857142857143 0.5625 0.0625 0.1875 0.875
0 0.357142857142857 0.642857142857143 0.642857142857143 0.571428571428571
0.357142857142857 0 0.5625 0.4375 0.625
0.642857142857143 0.5625 0 0.125 0.875
0.642857142857143 0.4375 0.125 0 0.875
0.571428571428571 0.625 0.875 0.875 0
0.571428571428571 0.625 0.875 0.875 0.3125
0.307692307692308 0.428571428571429 0.642857142857143 0.714285714285714 0.571428571428571
0.857142857142857 0.8125 0.5 0.5625 0.75
0.714285714285714 0.6875 0.8125 0.8125 0.5625

(Continued)
 Michael Cysouw

31 32 33 34

0.625 0.5 0.75 0.625

0.6 0.384615384615385 0.73 0.8
0.625 0.642857142857143 0.75 0.6875
1 0.785714285714286 0.5 0.9375
1 0.846153846153846 0.53 0.866666666666667
0.8125 0.714285714285714 0.625 0.9375
0.6875 0.5 0.8125 0.8125
0.75 0.857142857142857 0.8125 0.75
0.8 0.615384615384615 0.8 0.8
0.625 0.571428571428571 0.625 0.8125
0.875 0.857142857142857 0.5625 0.8125
0.875 0.714285714285714 0.6875 0.8125
0.8 0.846153846153846 0.6 0.73
0.9375 0.785714285714286 0.4375 0.875
0.8 0.642857142857143 0.6 0.73
0.5625 0.571428571428571 0.625 0.4375
0.5 0.785714285714286 0.625 0.625
0.8125 0.642857142857143 0.6875 0.6875
0.8125 0.642857142857143 0.6875 0.75
0.625 0.642857142857143 0.6875 0.6875
0.8 0.615384615384615 0.67 0.866666666666667
0.8125 0.571428571428571 0.5625 0.75
0.625 0.714285714285714 0.75 0.6875
0.625 0.428571428571429 0.8125 0.75
0.875 0.642857142857143 0.5 0.8125
0.571428571428571 0.307692307692308 0.857142857142857 0.714285714285714
0.625 0.428571428571429 0.8125 0.6875
0.875 0.642857142857143 0.5 0.8125
0.875 0.714285714285714 0.5625 0.8125
0.3125 0.571428571428571 0.75 0.5625
0 0.357142857142857 0.8125 0.5625
0.357142857142857 0 0.785714285714286 0.642857142857143
0.8125 0.785714285714286 0 0.8125
0.5625 0.642857142857143 0.8125 0
 Inducing semantic roles 

Appendix E: Clustering of contextual roles

8
3 23 17
6
20
1 10
33
30
4 31
5

28 16
25
14 34

22
29 15

13 2
11
26
19 27 7
21 24
12 18 32
 Michael Cysouw

to have respect
to learn from for the Bible
the Bible 34
31
30 to study
the Bible
16 3 23 17
live by 20
the Bible
10
32
author of 7
the Bible
2
counsel of
26
the Bible
24 contained
teaching of 1
in the Bible
the Bible
spoken of 27
in the Bible
6
the Bible
the Bible 9 was written
was copied
Bibles have 8
been printed 4 the Bible is a gift
14
12 the Bible reveals
when the Bible 21
was written 5 the Bible alone
18 25 33
what does the 15 28
the Bible is the basis
Bible say? as it is called 19 29
in the Bible 13 the Bible urges
11 22
the Bible the Bible teaches
says
the Bible foretold
the Bible tells
 Inducing semantic roles 

Appendix F: Alternative flat clustering

26 2 724

32

27

9 21
31 15
19 2829
25
22
30 12
34 1 8 18
6 14
10 5
16 4
20
11
233 33 13
17

26 2 724

32

27

9 21
31 15
19 2829
25
22
30 12
34 1 8 18
6 14
10 5
16 4
20
11
233 33 13
17
 Michael Cysouw

Appendix G: Language-specific structures

akha

ja..li.. jaˇliˆaˇ ja..li..

jaˇliˆ ja..li..
jaˇliˆ jaˇliˆ

jaˇliˆ

jaˇliˆ

ja..li..ah.. ja..li..a..
jaˇliˆaˇ jaˇliˆ
jaˇliˆahˇ
jaˇliˆ
jaˇliˆ jaˇliˆ
jaˇliˆ
jaˇliˆ
jaˇliˆ

jaˇliˆahˇ ja..li.. jaˇliˆ

ja..li..ah..
ja..li..ah.. jaˇliˆ jaˇliˆaˇ
jaˇliˆahˇ jaˇliˆ
ja..li.. ja..li..a..
jaˇliˆaˇ ja..li..a..
ja..li.. ja..li..a.. jaˇliˆaˇ
jaˇliˆ
jaˇliˆ
jaˇliˆaˇ
jaˇliˆahˇ
jaˇliˆ
jaˇliˆ
jaˇliˆ
jaˇliˆ jaˇliˆahˇ

ja..li..
jaˇliˆahˇ
 Inducing semantic roles 

albanian
biblës
ës
bibl biblës
biblës
biblës biblike

biblës bibla
bibla
bibla
biblës

bibla

biblën bibla
biblën
bibla
bibla bibla
bibla
bibla
bibla

biblën bibla

bibla bibla bibla

bibla
bibla bibla
biblën
bibla
bibla
bibla
biblën
bibla
biblën
biblën bibla bibla

la bibla
biblën bib
biblën
 Michael Cysouw

altai
 Inducing semantic roles 

aymara

bibliaxa
biblian
bibliaxbiblian

biblian
biblian
biblian
iax
bibliawa bl
bi
bibliax
bibliar
bibli
an
biblianxa xa bibliaxa
bibliax
bibliaxa
nxa
bibliabiblianxa
bibliaxa
bibliax biblianxa
bibliax
bibliaruxa bibliax
bibliaxa bibliaxa bibliax
bibliatxa biblianxa
bibliat bibliaxa bibliaxbliax
liat bi bibliax
bib bibliaxa
bibliat
bibliakiw
bibliax
bib liax
bibliampita
bibliat bibliat
bibliat bibliaxa
bibliat
bibliat bibliat biblianxa

bibliat
bib
liat
 Michael Cysouw

azerbaijani
 Inducing semantic roles 

drehu

tusihmitrÖtr
tusihmitrÖtr
tusihmitrÖtr
tusihmitrÖtr

tusihmitrÖtr

tusihmitrÖtr

tusihmitrÖtr tusihmitrÖtre
tusihmitrÖtr
tusihmitrÖt
tusihmitrÖtr
re
tusihmitrÖtre
tusihmitrÖtre tusihmitrÖtre
tusihmitrÖtre
tusihmitrÖtre

tusihmitrÖtr tusihmitrÖtr
tr
rÖ
tusihmitrÖtr
tusihmitrÖtr tusihmitrÖtr hmit
i
tus
tusihmitrÖtr
tusihmitrÖtre tusihmitrÖtre

tusihmitrÖtr
tusihmitrÖtre
tusihmitrÖtre
tusihmitrÖtr
tusihmitrÖtr
tusihmitrÖtre
tusihmitrÖtr
tusihmitrÖtr tusihmitrÖtre tusihmitrÖtre
rÖtre
tusihmit
tr
tr Ö tusihmitrÖtre
tusihmitrÖtr
ihmi
tus
 Michael Cysouw

estonian

piibli
piibli
piibli piibli

piibli

piiblis
piiblis
piiblis
piibel
piibel
piiblis
piiblit
pii
piiblit piibel bel
piibli
lit piibel
piib piibel
piiblis
piibel
piibel
piibel
piiblist piiblist
piiblist
piiblist
piiblist piibel
piiblit piibel piiblit
piiblist piiblit piiblis
t
piibli
piibli
piibel
piibli
piiblit
piibel piibel
piibel
piibli
piiblit el
piib

el
piibel

ib
piibli

pi
piiblit
piiblit piibli piibel piibel

l
piiblit piibe
 Inducing semantic roles 

faroese
bíbliunnar
bíbliunnar
bíbliunnar
an
bíbliunnar bíblian bíbli
bíbliunnar
bíbliunnar
bíbliuni

bíblian

bíblian bíb
lia
bíbliuni n
bíblian
bíblian bíblian
bíblian
bíblian
bíblian
bíbliuna

bíbliuna bíblian
bí
bl
iu

bíbliuni
na

bíbliuni bíblian
bíblian
bíblian bíblian
bíbli
an
bíbliuna
bíblian
bíblian
bíbliuni bíbliuna
bíbliuna
bíbliuna
bíblian
bíbliuna
bíbliuna bíbliuni bíbliuna bíblian
bíbliuni
bíblian
bíbliuni
bíbliuna
liu
na
 Michael Cysouw

greenlandic

biibilimut
biibilimi biibilimi
biibilimik

biibili

biibilimi
biibilimik
biibili
biibilimik
biibilimik biibilibiibilimi
biibilimi
biibili biibilimi

biibilimi
ik
im

biibilimi
biibilimi
bil

biibilimi
biibilimi
bii
biibili
bil
i
biibilip
biibilimik biibilili
biibili biibilip
biibilimeersunik
biibilip
biibili
biibil
biibilip i biibilimi
biibilip
biibili biibilimi
biibilimik biibili
biibilip biibilip
biibili
biibililu
biibilimik biibilimik
ik
m

biibilimik biibilimili
ili
ib

biibil
bi

biibili
biibilimillu imi
biibilimik biibilimi

biibilimik
 Inducing semantic roles 

irish

bhíobla
bhíobla bhíobla
bhíobla

bhíobla

bhíobla

bíobla
bhíobla bhíobla bíobla
bhíobla
bíobla
bíobla bíobla
bíobla
bíobla
bhíobla
bhíobla bíobla
mbíobla
mbíobla bhíobla bíobla
bhíobla bíobla
bhíobla bhíobla
bíobla
bhíobla bíobla bíobla
mbíobla bíobla
bíobla
mbíobla
bíobla
mbíobla
mbíobla bíobla bíobla

mbíobla
mbíobla
 Michael Cysouw

khoekhoe

bybels
bybels
bybels bybels

bybels

bybels

bybelsa
bybelsa
bybelsa
bybels
bybels bybeldi by
bybels be
ls
bybelsa
bybels
bybels
bybels
bybels
bybels
bybels

bybelsa bybels
bybelsa bybelsa
bybelsa
bybels
bybels
bybels

bybel
bybels
bybels
bybelsa bybels
bybelsa bybelsabybels
byb elsa
bybelsa
bybelsa
bybelsa bybels bybelsa

bybelsa

bybelsa
 Inducing semantic roles 

korean
 Michael Cysouw

madi
ri
ofuole
ofuoleri
ofuoleri
ofuoleriofuoleri

ofuoleri

ofuoleri

ofuolerii ofuoleri
ofuoleri
ofuoleri
ofuoleri
ofuoleri
ofuoleri
ofuoleri
ofuoleri
ofuolerii
ofuoleri ofuolerii
ofuoleri ofuolerii
ofuoleri i
ofuoleri oleri
ofu
ofuolerii
ofuoleri

ofuoleri ofuolerii
ofuolerii
ofuolerii
ofuoleri ofuo
ofuoleri ofuoleri leri
ofuoleri
ofuoleri
ofuoleri
ofuoleri ofuoleri ofuoleri

ofuoleri
ofuo
le ri
 Inducing semantic roles 

nias

surani’amoni’õ
zurani’amoni’õ
zurani’amoni’
surani’amoni’ õzuran õ
i’...am
zuran oni’.
..õ
i’...am
oni’...
zurani’amoni’õ õ

zurani’amoni’õ

õ
zura

i’...
ni’...

on
amo
ni’...

m
zur õ

...a
ani

ni’
surani’amoni’õ ’...a
zurani’amoni’
mo õ

ra
ni’..

zu
zurani’amoni’õ .õ
surani’amoni’õ
surani’amoni’õ
zurani’amoni’õ surani’amoni’õ

surani’amoni’õ

zurani’amoni’õ zurani’amoni’õ

zurani’amoni’õ zurani’amoni’õ surani’amoni’õ ..õ

õ moni.
surani’...a
zurani’amoni’
surani’amoni’ õ oni...õ
surani’...am surani’amoni’õ

zurani’amoni’õ
surani’amoni’õ
surani’amoni’õ
zurani’amoni’õ
zurani’amoni’õ
surani’amoni’
suran i’...amoni’õ...õ
zurani’amoni’õ
zurani’amoni’õ surani’amoni’õ zurani’amoni’õ
zurani’amoni’õ

zurani’amoni’õ
 Michael Cysouw

oromo
macaafaqulqulluu
macaafaqulqulluu
macaafaqulqulluurratti
macaafaqulqulluu
macaafaqulqulluu

macaafaqulqulluu
macaafaqul macaafn
qulluu iqulqull
uun
macaa
faqulq
ulluu
macaafaqulqulluuttimacaafniqulqulluun
macaafaqulqulluudhaaf
macaafniqulqulluun
macaafniqulqulluun
macaafniqulqulluun
macaafniqulqulluun
macaafniqulqulluun
macaafniqulqulluun

macaafa
q
aafaqulqulluuulqulluu macaafniqulqulluun
macaafaqulqulluu macaafniqulqulluun
macaafaqulqulluurraa
ma macaafniqulqulluun
caa macaafniqulqulluun
faq macaafniqulqulluun
ulq
ull
macaafaqulqulluu uu
rra macaafaqulqulluu
a
macaafniqulqull
macaafaqulqulluurraa
macaafaqulqulluu maca
afaqu
lqullu
maca urraa macmacaafniqulqulluun
afaqu aafniq
lqullu ulqull
macaafaqulqulluu urraa uun
macaafaqulqulluu macaa
macaafaqulqulluurratti macaafniqulqulluun
fniqulq
ulluun

caafaqulqulluu
macaafaqulqull
uu
 Inducing semantic roles 

Appendix H: Language distance

Altai Albanian Irish Faroese Estonian Aymara Greenlandic Akha

0 0.454555579 0.734541135 0.662878031 0.523770122 0.721402204 0.829388318 0.840987522

0.454555579 0 0.277027386 0.214015164 0.298738949 0.756323197 0.708335963 1.068484248
0.734541135 0.277027386 0 0.347941317 0.273804364 0.738323503 0.740430881 1.087291201
0.662878031 0.214015164 0.347941317 0 0.333139169 0.420641899 0.305442493 0.814935631
0.523770122 0.298738949 0.273804364 0.333139169 0 0.587264817 0.727472259 0.95022716
0.721402204 0.756323197 0.738323503 0.420641899 0.587264817 0 0.73660816 0.885771028
0.829388318 0.708335963 0.740430881 0.305442493 0.727472259 0.73660816 0 0.796787817
0.840987522 1.068484248 1.087291201 0.814935631 0.95022716 0.885771028 0.796787817 0
0.268159258 0.562611653 0.669077552 0.451956556 0.603643571 0.660876245 0.875205469 0.688264496
0.984228933 0.906016113 0.790240017 0.970799958 0.883076078 0.679158868 1.129209693 0.663022961
0.695487306 0.458800266 0.587307273 0.453834796 0.477489825 0.589736703 0.595743526 0.73538531
0.787615113 0.484458227 0.418059578 0.344925135 0.531457372 0.431785952 0.652962563 0.556432297
0.999100642 1.561467318 1.175034154 1.314967621 1.194382166 0.79904949 0.671401375 0.474521894
1.024562788 0.974518753 0.605873894 0.864045225 0.468545371 0.798025101 0.933084078 1.071045402
0.545910577 0.216943402 0.407304741 0.374488152 0.290423385 0.504709385 0.515954765 0.978112036

(Continued)
 Michael Cysouw

Azerbaijani Nias Khoekhoe Oromo Madi Drehu Korean

0.268159258 0.984228933 0.695487306 0.787615113 0.999100642 1.024562788 0.545910577

0.562611653 0.906016113 0.458800266 0.484458227 1.561467318 0.974518753 0.216943402
0.669077552 0.790240017 0.587307273 0.418059578 1.175034154 0.605873894 0.407304741
0.451956556 0.970799958 0.453834796 0.344925135 1.314967621 0.864045225 0.374488152
0.603643571 0.883076078 0.477489825 0.531457372 1.194382166 0.468545371 0.290423385
0.660876245 0.679158868 0.589736703 0.431785952 0.79904949 0.798025101 0.504709385
0.875205469 1.129209693 0.595743526 0.652962563 0.671401375 0.933084078 0.515954765
0.688264496 0.663022961 0.73538531 0.556432297 0.474521894 1.071045402 0.978112036
0 0.967376892 0.913551306 0.678797328 1.020851253 0.793611408 0.402527169
0.967376892 0 0.793666095 0.627986605 0.583007061 0.410261209 0.440168833
0.913551306 0.793666095 0 0.646754748 1.047158113 0.928912303 0.581111573
0.678797328 0.627986605 0.646754748 0 0.962514104 0.451281879 0.282443004
1.020851253 0.583007061 1.047158113 0.962514104 0 0.886576602 1
0.793611408 0.410261209 0.928912303 0.451281879 0.886576602 0 0.543719133
0.402527169 0.440168833 0.581111573 0.282443004 1 0.543719133 0
 The grammaticalization chain of case
functions
Extension and reanalysis of case marking vs.
universals of grammaticalization

Heiko Narrog
Tohoku University

In the 1980s and 1990s, grammaticalization research brought forth a number of

intriguing proposals concerning the directionality of extension and change between
case functions, i.e. semantic roles. One of these proposals, by Heine et al. (1991),
consisted in a single unidirectional chain of increasing grammaticalization of case
functions, advancing from the spatial domain via an anthropocentric domain
to inanimate and abstract domains, based on metaphorical extension. This bold
proposal is still often referred to, but has proven to be problematic in some points,
especially in the intermediate area of anthropocentric concepts. The present paper
investigates this chain in detail, showing which hypothesized extensions still hold
and which not. It is argued that the single chain is untenable and has to be broken
up into at least two larger directionalities, one leading to the development of the
expression of core participants, often triggered by constructional reanalysis, and the
other leading to highly abstract, inanimate case roles. The application of the concept
of metaphorical extension to the reanalysis cases is much less straightforward than
that to the semantically motivated cases. However, the result of change is similar,
as the semantic weight of the phrase shifts practically completely to the nominal,
and case marking becomes almost purely syntactic marking. Dative case functions
are in an intermediate position leading to extensions both in the direction of core
participants and to inanimate, abstract roles.

Keywords: Grammaticalization; semantic roles; semantic maps;

unidirectionality; case functions; metaphor

1. I ntroduction: Directionality of semantic/functional change

in the domain of case

Within functional and cognitive theories of language change, it is common to

assume unidirectionality in the semantic and functional change of lexical and
 Heiko Narrog

grammatical items. However, there is no universal agreement yet as to the

concrete directionality of these changes and the motivations and mechanisms
behind them.
The area of case functions and their grammaticalization is a case in point.
Three of the most basic and influential proposals in this area are already more than
twenty years old. Lehmann (1982; later editions 1995 and 2002) proposes a num-
ber of so-called “grammaticalization channels between case markers” (1995: 112),
which are limited in number and not very problematic except for some details
about which we have better knowledge by now. On the other hand, they do not
provide much theoretical stimulation either, because they are a matter of descrip-
tion and not based on an underlying idea beyond the empirical observations. Then
there is Croft’s (1991) “causal chain model”, which predicts polysemy of case mark-
ing functions along a chain of action from cause to result. As this model attempts to
map linguistic functions on a specific extra-linguistic conceptualization of ‘event’,
it is theoretically stimulating, but the match with empirical facts is not as good as
with Lehmann (1995(1982)). Furthermore, while this model predicts polysemies,
predictions concerning meaning changes can be obtained only indirectly and are
not very concrete. The third proposal by Heine et al. (1991) combines advantages
(and disadvantages) from both the aforementioned proposals. It is theoretically
stimulating because it is grounded on a specific theoretical assumption, namely,
the idea that semantic change essentially operates on the basis of metaphor. Fur-
thermore, the match with empirical data is better than in the case of Croft (1991)
but not quite as good as with Lehmann (1995(1982)). There are a number of inter-
esting mismatches, which beg the question why they occur and if they can be
resolved.
Figure 1 shows Heine et al. (1991) “chain of increasing grammaticalization of
case functions”.

1 2 3 4 5 6
ablative agent purpose
allative > comitative > instrument > time > condition > manner
locative benefactive dative cause
path possessive

Figure 1. Chain of increasing grammaticalization of case functions (Heine et al. 1991: 159)

In order to facilitate further reference, I have given a number to each of the

six groups of meanings/functions depicted in Figure 1. The leftmost group of case
function, number 1, is spatial. Then follow human participants (2), participants
in human activity (3), time (4), condition and cause (5), and manner (6). This
cline can be further broken up into three larger domains, namely, space (the first
 The grammaticalization chain of case functions 

group), anthropocentric concepts (the second and third group), and inanimate
concepts (the last three groups 4, 5, 6). Thus, Heine et al. (1991: 160) came to the
broad generalization in (1):
(1) spatial relations > human relations > inanimate relations

Relating this generalization to more general concepts of semantic and functional

change, two ideas feed into the interpretation of this chain. This is firstly the ‘local-
ist hypothesis’, an influential line of thought in linguistics from the 1970s on which
claims the primacy of spatial concepts before anything else. This is reflected in
the directionality between the first and the second grouping. Secondly, there is
the metaphorical shift from animate to inanimate concepts between the second
and the third grouping. Interestingly, this order of semantic/functional changes is
different from the one that Heine et al. (1991) found for lexical concepts, which is
represented in (2):
(2) PERSON > OBJECT > ACTIVITY > SPACE > TIME > QUALITY

The chain in (2) operates purely on a metaphorical basis, proceeding from

­anthropocentric concrete concepts to more abstract inanimate concepts. As to
why different directionalities should hold for different domains of language,
Heine et al. (1991: 160–1) suggest that we are dealing with referential units in the
case of (2), “where tangible, visible units form the most “concrete” taxa”, but with
relational concepts in the case of (1), where “space provides the most concrete
domain of concepts”.
Now, while the chain, or hierarchy, in Figure 1 still appears to hold on its two
ends, it has somehow unraveled in the middle part, particularly concerning the
directionality of change of anthropocentric concepts. For example, as Luraghi
(2001b) has pointed out, it appears that in Indoeuropean languages a change
from ­instrumental to agentive case function has taken place, contrary to the rep-
resentation in (1).
In later publications, Heine himself (Heine 2003, 2009) has not returned to
this idea of a single unified chain of grammaticalizations in case functions, but
instead has presented lists of individual changes and individual chains of changes.
Nevertheless, his original idea is interesting enough to warrant a more thorough
investigation of what exactly does and does not work in this chain, and what the
reasons might be for that, and finally, if this chain can be salvaged or must be fun-
damentally revised. This, then, is the goal of this paper. I will first briefly address
a methodological issue that will become relevant in the latter sections of this
paper (Section 2), and then show what actually works in Heine et al.’s (1991) chain
­(Section 3), before proceeding to investigate the problematic domains (Section 4).
In Section 5, I investigate in more detail an area that was subsumed under a single
 Heiko Narrog

category label in the Heine et al.’s (1991) grammaticalization chain, namely that
of ‘Agent’. Section 6 summarizes the results of the preceding sections in the form
of a revised grammaticalization chain. Section 7 discusses the implications of the
findings with respect to presumptive mechanisms underlying the semantic and
functional changes, and Section 8 wraps up the paper.

2. Semantic and constructional change

Heine et al.’s (1991: 159) chain introduced above is labeled as a “chain of increas-
ing grammaticalization of case functions”. Therefore, no explicit claim is made
with respect to the mechanism(s) through which individual changes took place.
However, in the context of the book, it is clear that the authors are taking it
for granted that the main mechanism of change is semantic extension through
metaphor. In contrast, some of the cases of change on this chain are obviously
due to a reanalysis of entire constructions, in which the semantic contribution
of the individual case marker was probably of very limited relevance (such cases
will be presented in Sections 4 and 5). This is something that Heine et al. (1991)
apparently had not been fully aware of. Meanwhile there is a growing agree-
ment in the field that all semantic change in the area of grammar also to some
degree involves constructions. While it may be realistic to assume a continuum
between changes with some saliency of the semantics of the case marker, it is
useful here to make a distinction between semantic extension through metaphor
and metonymy in the original sense of Heine et al. (1991) on the one hand, and
reanalysis of a construction in the sense of Harris and Campbell (1995: 61) on
the other. This distinction corresponds to extension and reanalysis as the two
major mechanisms in language-internal grammatical innovation as laid out by
Gildea (1998, Chapter 3.2). Heine’s concept of metaphorical extension is based
on Willett (1988: 80), who defines it as “[a mechanism] in which the concrete
meaning of an expression is applied to a more abstract context where some of
its (original) cooccurrence restrictions no longer hold.” Harris and Campbell
(1995: 61) define reanalysis as “a mechanism which changes the underlying
structure of a syntactic pattern and which does not involve any immediate or
intrinsic modification of its surface manifestation.” Note that the term exten-
sion is ambiguous, because irrespective of the concrete mechanism it also more
broadly describes change in which the original function or meaning of some
item or construction is retained when a new meaning or function is acquired,
which is presumptively the most common scenario in grammaticalization (cf.
the concept of ‘overlapping categories’ in Heine & Narrog (2010)). On the other
hand, reanalysis of a whole construction often leads to a loss of the original
 The grammaticalization chain of case functions 

f­unction of the ­construction, especially in the cases discussed in this paper.

Therefore, this ambiguity is not a major concern here.

3. Directionalities that fit the chain

While doubts have been raised against some of the directions of meaning change
hypothesized by Heine et al. (1991), others have stood the test of time. Most impor-
tantly, this applies to the hypothesis of spatial functions as the source, and not the
target, for other changes. I am not aware of any research so far seriously challeng-
ing this generalization. Later research as a rule has accepted spatial functions as the
source concepts for change. A good example is Yamaguchi (2004), who builds his
model of extensions of case functions entirely on spatial domains,1 as in Figure 2.

spatial/temporal domain
SOURCE PATH LOCATION GOAL

(spatial/temporal) (spatial/temporal) (spatial/temporal) (spatial/temporal)

ablative path locative allative
comitative

non-spatial/temporal domain

result

agentive instrumental/means possessive benefactive/reci

pient
manner
causal purpose

Figure 2. Domains and extensions of case functions according to Yamaguchi (2004: 150)

Yamaguchi chose exactly the same four spatial concepts as Heine as the start-
ing points for extension, but from there on he depicts more concretely which func-
tions from other domains are built on which spatial function. In contrast to Heine,

. Yamaguchi (2004) uses the term ‘spatial’ synonymously with ‘spatiotemporal’, based on the
observation that temporal meanings practically automatically derive from spatial meanings,
to the extent that it is often difficult to discern historically which one precedes the other (cf.
Yamaguchi 2004: 83).
 Heiko Narrog

he does not posit any other larger conceptual domain except the spatial domains.
Everything outside these domains is simply ‘non-spatial’.
Some of the directionalities posited by Yamaguchi (2004) are problematic (cf.
Narrog & Ito 2007; Narrog 2009), but this does not concern us here much further.
Instead I wish to give a list of actually attested changes from spatial functions as
found in the extant literature. Note that from here on, I replace Heine et al.’s (1991)
terminology, which mixes case labels with semantic role labels, with a terminology
using only semantic role labels, except when quoting others’ research. Also note
that if I write that a marker for semantic role X can “become” a marker for seman-
tic role Y, I assume as a default that the marker retains its older function expressing
the role X, although there are exceptions (cf. Luraghi, this volume).

3.1 From domain 1 (spatial functions) to domain 2 (human participants)

Linguistic expressions marking Location can extend to mark Companions (cf.
Luraghi, this volume, for more detail).2 A well-known case from language history
is English with, which originally had a meaning of spatial opposition and motion
or rest in proximity, and in Middle English extended to the domain of comita-
tive and instrumental meanings previously covered by mid (cf. OED 〈 with prep.,
(adv., conj.)〉). An extension from the locative domain to Companion can also be
assumed for French avec, which is probably derived from apud hoc > avoec with
originally local function (cf. Brachet 1882: 44; Endruschat 2007: 84). The Greek
preposition metá is described to originally have had the spatial meanings ‘in
midst of; between’ before developing comitative and instrumental meanings (cf.
Schwyzer 1950: 483–5).
Markers for Location can also become markers for (Passive) Agents. The
­Chinese verb yú ‘go’ grammaticalized first into a Goal marker, from there into a
Location marker, and that became the source of the latter Passive Agent marker, as
a detailed study by Yan (2003: 147) shows. Heine and Kuteva (2002: 199–200) list
up other possible examples from Albanian, Jeri, Luba and Turkish.
Markers for Sources can extend to cover Agents, namely Ergative Agents (cf.
Palancar 2002: 234 for Tauya, Dani, Athpare), Passive Agents (cf. Heine & Kuteva
(2002); Wiemer 2011), primarily for a number of Indo-European languages) and
Causees (cf. Palancar 2002: 181 naming Kannada, Maithili and P ­ unjabi). The

. From this point on, I will apply the following convention: Semantic roles are capitalized,
cases are not. The only exception is comitative, which is a case label, but is capitalized here,
since in the literature to which I refer, usually no distinction is made between comitative case
on the one hand, and companion as a semantic role on the other hand.
 The grammaticalization chain of case functions 

­ est-known cases may be Passive Agent marking by the Latin ablative preposition
b
a(b), Modern German von, and the modern Romance descendants of Latin dē.
Markers of Path as well can become Agent markers, namely for Causees and
Passive Agents from various Indo-European languages (e.g. door in Dutch, and
durch in German, przez in Polish; cf. Palancar 2002: 183).
Markers of Goal can come to cover Beneficiaries cf. e.g. Luraghi (2001a: 45–6).
For example, according to the OED (<to, prep., conj., adv.>), the meaning of to in
West Germanic can be reconstructed as “direction towards”. It later obtained its
dative-related senses, which, according to the hypothesis by Heine et al. (1991)
would have first involved a Beneficiary sense. Michaelis and Rosalie (2000: 90–95)
provide examples from a number of French Creoles, where this change seems to
have taken place, and Luraghi (2010: 119) cites the development of Beneficiary
uses of the Ancient Greek directive preposition eis ‘to’.

3.2 From domain 1 (spatial functions) to domain 3 (human activity)

Extensions directly from spatial functions to ‘participants in human activity’
appear to be just as common as extensions to human participants. The following
changes have been attested.
A marker for Source can come to cover instrumental meaning. This ­extension
has been reconstructed by Palancar (2002: 234) for Athpare, Dani and Tauya.
­Generally speaking ablative-instrumental syncretism, and more specifically abla-
tive-instrumental-ergative syncretism (with the change probably proceeding just
in this direction) appears to be cross-linguistically very wide-spread (cf. Noonan
& Mihas 2007). Furthermore, ablatives can become markers of an attributive Pos-
sessor as the example of de in French and similar examples of languages in Europe
show (cf. Heine & Kuteva 2002: 34–5).
Path markers can become the source of instrumental markers, as has been
hypothesized by Palancar (2002: 170) for the Russian instrumental inflection. A
historically documented case is presented by Luraghi (2001b: 394; 2003: 322; cf.
also Schwyzer 1950: 451), who shows that this extension is also valid for Ancient
Greek diá. English through had a path sense from Old English on, and developed
an instrumental sense later in the 14th century (cf. OED: through, prep. and adv.).
Markers for Location can become Instrument markers, as is known from
Indo-European languages (e.g. Blake (2001: 173); Luraghi (2003: 35f, 88f, 322))
and Finno-Ugric (Grünthal 2003: 139–141), for example. The best-known case
may be English by, to which an original meaning of position ‘at the side or edge of ’
is ascribed (OED; 〈by, prep., adv.〉). They may also become markers of predicative
possession as the preposition u in Russian. Heine and Kuteva (2002: 205) give a
similar case in an African language (So).
 Heiko Narrog

Markers of Goal can become datives (Recipients) and markers of Purpose.

For the first development, Heine and Kuteva (2002: 37) cite Tamil, Lezgian and
the English preposition to (see above). The Greek preposition eis (cf. Luraghi
2003: 39) is an historical example for the extension. Creissels (2009: 621) cites
Akhvakh as an historical example from a non-Indo-European language. For the
latter development, Heine and Kuteva (2002: 39–40) cite examples from a large
range of languages including Albanian and Basque. Creissels (2009: 624) also con-
siders Purpose as one of the “common metaphorical extensions of the use of spa-
tial cases”.

3.3  rom domain 1 (spatial functions) to domains 4 to 6

F
(inanimate concepts)
Markers of Sources can develop a Cause function. This is known from the Latin
ablative or the German preposition von (cf. Palancar 2002: 157–8), and the Greek
prepositions ek and apó (Luraghi 2003: 323). They can also develop a function
indicating a temporal origin. Examples for that are almost innumerable (cf.
­Haspelmath 1997: 66–68; Heine et al. 2002: 35).
Markers of Location and Path can develop a Cause function as well. For the
former, Heine & Kuteva give examples from Imonda and Albanian, for the latter,
the Ancient Greek preposition diá can be cited (Luraghi 2003: 323).

3.4  rom domain 2 (human participants) to domain 3

F
(participants in human activity)
The shift from Companion to Instrument, especially in European languages, is
one of the best-documented and most discussed cases of extension of case func-
tions (e.g. Stroh (1998, 1999) and Endruschat (2007: 59) for Romance languages,
and Stolz, Stroh and Urdze (2006: 366, 369) for other European languages).
Comitatives can also obtain possessive function, when a construction ‘X is with
Y’ becomes grammaticalized for possession. However, the noun phrase marked
with Comitative (Y) becomes the Possessee rather than the Possessor (cf. Heine
1997: 93). Such constructions are especially common in Africa and Oceania/Aus-
tralia (cf. Stolz 2001b: 328; Heine & Kuteva 2002: 88).
Beneficiaries may extend to Recipients and become datives. Heine and Kuteva
(2002: 54) give an example from Ewe. They may also become Purpose mark-
ers, a development that has been suggested for English for and in a large num-
ber of genetically and areally diverse languages (cf. Heine & Kuteva 2002: 55–6).
­Yamaguchi (2004), comparing Recipients and Beneficiaries as possible sources for
Purpose markers on a synchronic basis, concludes that Beneficiary is the more
likely source. The reason is that “the purposive is almost always expressed by a
 The grammaticalization chain of case functions 

gram that also expresses the benefactive sense, whenever the Beneficiary and
Recipient senses can be expressed by different grams (p. 86).” Another potential
target of benefactive markers is possession. For example, according to Heine and
Kuteva (2002: 55) in English-based creoles, English for is often the source for the
marking of Possessors. See Luraghi (2010) and Luraghi (this volume) for more
detail on paths of development involving Beneficiaries.

3.5 F
 rom domain 2 (human participants) to domain 4, 5, 6
(inanimate concepts)
For some reason, extensions directly from the domain of human participants to
inanimate concepts appear to be rare, at least concerning the inanimate concepts
listed by Heine et al. (1991). The Ancient Greek preposition hupó with genitive
(cf. Luraghi 2003: 323–4) is a lone potential example for extension from Passive
Agent to Cause.3 The extension from Companion to Manner (cf. Heine & Kuteva
2002: 87; Yamaguchi 2004: 124; English with care), seems to be an exception to
this paucity of evidence, but this extension obviously usually takes place via an
instrumental function of the Comitative. In my database of case polysemy (cf.
­Narrog & Ito 2007; Narrog 2010), only 12 out of 42 markers with Comitative-
Manner ­polysemy are not polysemous with Instrument.

3.6  rom domain 3 (participants in human activity) to domains 4, 5, 6

F
(inanimate concepts)
An Instrument sense appears to be the source of a relatively wide range of exten-
sions to the inanimate domain. These are Cause, as can be illustrated by English
with, Spanish con, and markers in many other languages (cf. Palancar 2002: 126,
166; Luraghi 2003: 37), probably Temporal Duration (Narrog 2010: 242), although
I am not aware of any specific historical evidence, and Manner, for which there
are many potential examples from a large variety of languages (Heine & Kuteva
2002: 180–1), including well-known ones such as the English preposition with and
its German counterpart mit. One could also add Material (Narrog 2010: 242), but
this is not part of Heine et al.’s (1991) list of case functions.

. “hupó with the dative or genitive was extended to Agent, relying on the Control meta-
phor, by which physical superiority is reinterpreted as implying control. Further on, Agent
expressions are extended to Cause, based on a metaphor according to which A CAUSE IS
AN AGENT […] hupó with the genitive only encoded Agent in Homer, and was extended to
Cause at a later time.”
 Heiko Narrog

Polysemy between Purpose and Cause is cross-linguistically common. It is

usually assumed that the Purpose sense extends to a Cause sense (e.g. Heine &
Kuteva 2002: 46–7). This is historically documented for the Japanese functional
noun tame (cf. Narrog 2010: 249).

3.7 Summary
In the preceding subsections plenty of evidence corroborating Heine et al.’s (1991)
chain of grammaticalization of case functions has been presented. Although not
all of this evidence has a clear-cut historical foundation, most of it is uncon-
troversial. Also, it is clear that some domains are better sources for other target
domains than others. But this is an issue for a more detailed study going beyond
the scope of this paper. Below in Figure 3, I have inserted the attested direction-
alities into the chain of case functions in order to show how actual directions of
change map onto it. Those that are tentative are drawn with dashed lines. The fig-
ure shows graphically that we are dealing with an impressive amount of changes
confirming Heine et al.’s (1991) chain, and the directionalities represented in this
chain are probably not even exhaustive.

1 2 3 4 5 6

Source

Agent Instrument
Condition
Location
Possessor
Time Manner
Companion
Dative
Path
Purpose Cause
Beneficiary

Goal

Figure 3. Chain of increasing grammaticalization of case functions (Heine et al. 1991: 159),
with concrete directionalities of change4

. In this section, I did not discuss extensions within domain 4, 5 and 6, since they are periph-
eral to the central topic of this paper. Nevertheless, following Heine and Kuteva (2002: 293),
I have inserted an extension from time to condition in order to include an extension leading
to condition.
 The grammaticalization chain of case functions 

4. Directionalities contra the claim

While the majority of changes seem to confirm Heine et al. ’s (1991) hypothesis,
some do not. As will be seen in the following subsections, they concern a very spe-
cific part of the grammaticalization chain, and very specific targets. These changes
can be subdivided again by source and target domain. I will present changes from
domain 3 to 2 in subsection 4.1, changes from domain 4, 5, and 6 into domain 2
in 4.2, and changes from domain 4, 5 and 6 into domain 3 in 4.3. Changes from
domain 3 to 2 are more numerous and important than the others. I will divide
them into changes that are likely due to constructional change and those that are
more likely due to semantic extension.

4.1  rom domain 3 (participants in human activity) to domain 2

F
(human participants)
4.1.1 Changes likely due to constructional change
We are dealing here mainly with four changes, namely, (1) the reanalysis of nomi-
nalized clauses with attributive possessors to ergative or nominative main clause
constructions, (2) the reanalysis of possessive constructions to ergative construc-
tions via tense-aspect functions, (3) the reanalysis from passive to ergative, and
(4) the genesis of passives from nominalizations. Without knowledge of historical
facts, the first three cases may not always be clearly distinguishable, especially if
the agent of the passive or the possessive tense-aspect construction is a genitive.
The semantics of the case marking of the noun phrases that acquire a new function
in the new construction is of no apparent relevance to the change.
From Attributive Possessor to Ergative Agent. The marking of an Ergative
Agent through a genitive morpheme is cross-linguistically not uncommon. It may
be inherited from a passive construction (see below), or it may be an independent
development. Palancar (2002: 229) in his sample of 148 languages finds this in 11
languages and markers, seven of which were not polysemous with instrumental
function, that is, were less likely to be derived from passives or predicative pos-
sessives (both below). The following example illustrates a genitive marker as an
ergative in Ladakhi:

(1) Thug-gu-yi pǝlldǝn-ni kǝ-ne şpe-čhǝ khyers.

boy-abs-gen/erg Paldan-gen from-abl book-abs take-simple-perf
‘The boy took the book from Paldan.’ (Palancar (2002: 230)

The common explanation is that a nominal verb form was reanalyzed as a fully
verbal form and the genitive marker thus extended to a more general agentive
marker (cf. Lehmann 2002: 98; Palancar 2002: 229; Heine 2009: 467).
 Heiko Narrog

From attributive Possessor to nominative. Erstwhile markers of possession

can also become nominatives. This is the case historically in Japanese where the
erstwhile genitive case marker ga generalized to become a nominative marker. The
genitive marker, if not used to relate two nouns with each other, initially marked
the relationship between a noun phrase and a nominalized or adnominal verb, or
in a subordinate predicate without nominal/finite distinction. (2) is an example of
such a construction.5

(2) Suzume=no ko=wo Inuki=ga nigasi-t.uru.

sparrow-gen child-acc Inuki-gen release-pfv-anp
‘Inuki let the baby sparrow escape.’ (11th century; Shirane 2005: 156)

From about the 12th century the distinction between nominal and finite verb
forms in Japanese eroded, with the nominal verb forms taking over the func-
tions of the finite ones. The genitive ga thereupon assumed nominative function,
which was up to then zero-marked in main clauses. The genitive function, in turn,
declined and is nowadays only preserved in a number of idioms. According to
Gildea (p.c.), a similar development also took place in Mayan.
From attributive Possessor to absolutive and accusative. Gildea and ­Gildea &
Alves have documented that attributive Possessors can also be reanalyzed as abso-
lutives (Cariban, Gildea 1998: Chapter 9–10; Jê, Gildea & Alves 2010) and as
­accusatives (Cariban, Gildea 1998: Chapter 12).
From Possessor in a possessive predication to Ergative Agent. Alternatively,
in a possessive construction of usually locative-existential origin with a non-
canonically marked Possessor subject, this subject Possessor may be reanalyzed
as a general subject if the construction is grammaticalized to express temporal-
aspectual distinctions (particularly, perfect or perfective) (Creissels 2008: 23–25).
This has presumably happened in Iranian (Anderson 1977) and possibly in a
wide range of other languages as well (cf. Trask 1979 citing languages in Asia and
North America; according to Gildea (p.c.), examples can also be found in South
American languages such as Canela (Jê) and Bribri (Chibchan)). It is uncommon
in Europe but found in some Northern Russian dialects, as described by Jung
(2009).6 Sentence (3) from the 16th century presumably represents one of the first
examples in this development.

. Note that the examples sentence contains one further genitive particle, no. For the finer
distinctions between these two case particles cf. Frellesvig (2010: 128).
. Note that Jung (2009) does not claim that Northern Russian dialects are generally ergative
but only that the North Russian perfect construction is ergative (cf. Jung 2009: 217–8).
 The grammaticalization chain of case functions 

(3) U carja perelož-en-o na se lěto ratь-Ø

at tsar:ppGEN undertake-part-n.sg for this summer troops-nom.f.sg
svoj-a na moskovskuju ukrajnu posla-ti.
own-nom.f.sg to Moscow region send-inf
‘By the tsar it was undertaken to send his troops to the Moscow region for
this summer.’ (Jung 2009: 214)

Note that the possessive subject in the Russian construction is not a possessor in
terms of case (i.e. a genitive) but is marked by a locative preposition.
From Possessor etc. to Passive Agent. At least in Ancient Greek and
in ­Lithuanian, the Agent of the passive was expressed as a genitive (cf. Blake
2001: 172; Luraghi 2005: 66 for Greek; Sawicki 1991 for Lithuanian). Wiemer
(2011) also cites cases from Armenian and Malagasy, and Palancar (2002: 175)
from Finnish and Toba-Batak. Possible causes are the reanalysis of dependency
relations of a possessive noun (Wiemer 2011), or preceding nominalization of
the predicate (cf. Lehmann 2002: 98). Therefore, we are probably dealing here
with the syntactic reanalysis of whole constructions, rather than a Possessor
being semantically reinterpreted as an actor. Gildea (2013) makes a similar argu-
ment for English get-passives, where the agent is constructionally (rather than
semantically) derived from an earlier causee in a reflexive-causative stage in the
development of the construction.
From Instrument etc. as Passive Agent to Ergative Agent. Ergative Agents
may be expressed as instruments etc. not as the result of a semantic extension but
as the result of a reanalysis of a passive (cf. Lehmann (2002: 98); or sometimes
inverse; cf. Siewierska 1998) construction as an ergative construction. The hypoth-
esis that many ergatives are derived from passives is fairly old and well-established
(e.g. Anderson 1977; Trask 1979), even leading to the claim that all ergatives are
of passive origin (Estival & Myhill 1988). Although recent literature has been busy
with debunking such exaggerated claims (e.g. Dixon 1994, Chapter 7.1; Butt 2006),
the reanalysis of passives to ergatives as such at least in some languages is still
fairly uncontroversial. The development from passive to ergative in a nutshell can
be described as follows: “When [the passive] construction becomes more current
and the Agent becomes increasingly obligatory, it is reinterpreted as a transitive
ergative construction, the instrumental serving also as the case of the transitive
subject” (Lehmann 2002: 98). Thus we are dealing here with the syntactic analysis
of a whole construction in which the instrumental already had agentive function,
rather than a direct semantically-motivated extension from instrumental to erga-
tive function.
From Recipient etc. to Causee. Although not mentioned by Heine et al. (1991),
the Causee in causative constructions is a further type of Agent. Of p ­ articular
 Heiko Narrog

interest is so-called ‘indirect’ causation, where a certain degree of autonomy is

attributed to the Causee. As stated by Shibatani and Pardeshi (2001: 148, 151, 154),
the Causee in such cases can be marked at least as dative (Japanese, Marathi) or
instrumental (Hungarian). Palancar (2002: 181, 187) finds five cases of datives,
two of instrumentals and three of ablative-instrumentals with Causee function in
his language sample of 148. It can be assumed that the Causee-marking function is
an extension of the more general case-marking function, and not vice versa. How-
ever, there could also be merely a common origin, so this change remains specu-
lative, and will therefore not play a role in the overall representation of changes
henceforth in this paper.

4.1.2 Changes likely due to semantic extension

The cases presented in this subsection superficially share sources and targets with
those in the previous subsection. However, in these cases, arguably s­emantics
is involved. One can distinguish (a) the appropriation of suitable case markers
for the expression passive agents in passive constructions which originally do
not ­contain agents, and (b) the appropriation of suitable case markers for the
­expression of ergative agents in ergative constructions that are already estab-
lished, or that ­originally did not have Agents as part of their structure, similar to
passives.
From Instrument to Passive Agent. In many languages Passive agents are
expressed as Instruments. This is, for example, the case in many Slavic languages
including Russian and Czech, in Bantu and Dravidian languages ­(Siewierska
1988: 251; Blake 2001: 172; Keenan & Dryer 2007: 343), and in French Creoles
(Michaelis & Rosalie 2000: 90). The following is an illustration from Russian:

(4) Kalitka byla otkryta Olegom.

gate(nom) was open(past.part) Oleg(ins)7
‘The gate was opened by Oleg.’ (Siewierska 1988: 247)

Indicating an Agent is not an essential part of passive constructions, and most

source constructions of passives do not have agents as part of their structure. In
fact, many passive constructions cross-linguistically do not indicate Agents at all,

. The following abbreviations will appear in the glosses. abl ablative; abs absolutive; acc
accusative; age agent; anp adnominal present; cas causative; cau cause; dat dative; erg
­ergative; f feminine; fnp finite present; gen genitive; ind indicative; inf infinitive; ins instru-
mental; n neuter; nom nominative; pas passive; part participle; pfv perfective; prs present;
ptc particle; sg singular. Examples are taken from various authors and various languages. The
original format of the glosses is retained as far as possible
 The grammaticalization chain of case functions 

leading Keenan and Dryer (2007: 328–9; 342) to postulate that passives with an
overt Agent phrase are “non-basic”. Haspelmath (1990: 56) has also claimed that
diachronically passive constructions only acquire Agents at an advanced stage of
grammaticalization. To give an example, Bostoen and Nzang-Bie (2010) present
the addition of Agent-marking as an advanced stage in the development of passives
in Bantu A70 languages. Now, if an agent phrase actually develops and is appro-
priated from something else, it is fair to assume that semantic match plays a role.
With respect to instrumentals in passives, it is generally assumed in the literature
that marking a Passive Agent is an extension of the instrumental function and not
vice-versa. As seen above, Instrument appears to be a particularly good semantic
fit, since it is found cross-linguistically in this function. In historical terms, there
is good evidence for the extension from Instrument to Passive Agent in Indo-
European, where, according to Luraghi (2001b: 396) passives are relatively new
constructions in the individual language families, while the instrumental express-
ing an Instrument must be traced back much further to Proto-Indo-European. In
English, as in many other Indo-European languages as well, the passive seems to
have developed out of a perfective construction, which originally had a stative-
adjectival character and rarely indicated an Agent.
From dative etc. to Passive Agent. Another common source of Agent mark-
ing in passive constructions are datives. Datives have the expression of Recipient
at the core of their semantics, but are generally less semantically specific than
instrumentals, and are in an ‘in-between’ position between semantic and struc-
tural cases (cf. Introduction to this volume). Palancar (2002: 192) cites seven
languages (Nigerian Arabic, Kayardild, Mansi, Alaskan Yupik, Siberian Yupic,
Labrador, and Western Greenlandic) where, based on internal reconstruction,
a dative has presumably undergone an extension to Passive Agent. Other lan-
guages with dative Passive Agents include Japanese and Ancient Greek (Luraghi
2005: 66). In Japanese, the Passive Agent of the –(r)are passive, which developed
in Old to Late Old Japanese, is marked with the dative. While the suffix verb –(r)
are- itself emerged in historical times, it continues functions of the s­ pontaneous
middle/passive suffix verb –(ur) aye-, in constructions in which the passive agent
was already marked with a dative. The dative case particle ni, in turn, is already
highly generalized at the oldest documented stage of the language, covering a
wide range of functions from Location and Goal to Recipient and Cause. It is
therefore difficult to assign any specific semantic function to the dative from
which the passive use would have emerged. Instead, it may be more appropriate
to view the dative in the Japanese passive as generically marking a peripheral ani-
mate ‘third participant’ in an event. Other options include Source and Possessor.
In English, for example, the marking of Agent in English passives was variable up
to the 16th century, with a preference for from and of (cf. Toyota (2008: 21–2)).
 Heiko Narrog

As expected, agents were rarely inserted in Old English and only later the fre-
quency of their use increased (Toyota 2008: 77).
From Instrument directly to Ergative Agent. Syncretism between instru-
mentals and ergatives is an extremely common phenomenon outside Europe,
especially in the Austronesian and Australian phyla (cf. Stolz 2001a: 167–8), and a
pathway through passive or passive-like constructions cannot always be assumed.
A well-known scenario for the extension from instrumental to ergative without
the intervention of a passive or inverse construction has been proposed by Garrett
(1990). Garrett suggests two major sources of split ergativity, the first of which is
a tense/aspect split, and the second a direct reanalysis of instrumentals as erga-
tives. He considers ergative constructions in Anatolian and in Gorokan (Papuan)
as examples of this presumptive reanalysis.8 In short, the scenario works as fol-
lows: In transitive clauses in null-subject languages, where the subject is actually
not expressed, an instrumental phrase is reinterpreted as an agentive phrase, due
to semantic similarity; i.e. the “functional overlap between Instrument and Agent
in clauses with transitive predicates” (Garrett 1990: 265).
A similar scenario, which does not refer to instrumentals in particular but
to oblique cases and adpositions in general, has been proposed by Creissels
(2008: 4.2), who sees “A ellipsis” in predominantly accusative languages as one
source of alignment shift towards ergativity. In so-called trans-impersonal con-
structions, the original S/A argument, which tends to be impersonal and elliptic,
is replaced by an oblique phrase indicating a cause or stimulus. If this construction
generalizes, it becomes an ergative construction.
From dative to Ergative Agent. The development of ergatives in Indo-Aryan
languages, which were formerly taken as the prime examples of reanalysis of pas-
sive constructions as ergative, has been given divergent interpretations in the past
20 years. However, notwithstanding whether Old Indo-Aryan pre-ergative con-
structions are considered as passives or only as temporal periphrasis with quirky
case marking, the ergative case clitics of many Indo-Aryan languages seem to
be later developments, that is, from a time when the ergative construction was
already established (cf. e. g. Verbeke & De Cuypere 2009: 373). In the case of
Hindi, V­ erbeke and De Cuypere (2009) claim that the ergative clitic ne was bor-
rowed from the accusative/dative clitic ne in neighboring Old Rajasthani in dative
function. According to these authors, this was possible “on semantic grounds”
because of “semantic similarity between dative and ergative” in terms of “control”

. Garrett’s analysis for Hittite cannot be considered as valid anymore (cf. Goedegebuure
2012 for a new analysis in which the ergative marking is not derived from case at all). However,
this does not necessarily invalidate his scenario as such.
 The grammaticalization chain of case functions 

(p. 376). The same claim of a dative origin for the related Urdu ergative clitic ne
is made by Butt (2006: 75), who does not give a concrete historical scenario, but
asserts on conceptual grounds that ne was first introduced into the language “as a
non-nominative subject case marker which was used to reinforce semantic con-
trasts with existing subject markings” (p. 81). Likewise, Poudel (2008) claims for
Nepali ergative le that it has its origins in a Recipient/dative marker, independent
from a passive construction. Note that in the Hindi case, even if the source of the
ergative is a borrowed dative marker, this does not mean that the change falls out-
side common grammaticalization paths, since in grammaticalization research the
current assumption is that borrowings follow the same paths as language-internal
developments (cf. Heine & Kuteva 2005).

4.2  rom domains 4, 5, 6 (inanimate concepts) to domain 2

F
(human participants)
From Cause to Passive Agent. There may be an extension from Cause to Passive
Agent but despite the frequent mentioning of this possible change in the literature,
there are few unambiguous examples. Michaelis and Rosalie (2000: 90) provide
the example of the German marginal Passive Agent marker durch, but because
of the polysemy of durch with other related meanings as well, the connection is
not entirely clear. The best case, if correctly represented, may be the Sotho Cause/­
Passive Agent prefix kē, as described by Palancar (2002: 152). See the examples
below, representing the Cause (5) and the Passive Agent (6) use:
(5) Kē-shŏēle kē-tlala
1sg-die cau/age-hunger
‘I am dead from hunger.’
(6) Rē-ratŏa kē-ntate
3sg-love-pass cau/age-father
‘We are loved by (our) father.’ (Palancar 2002: 152)

According to Palancar (2002), kē has only these two meanings, and “the Initial-
Energizer is construed as a causal force…. When the Autonomous-Energizer (…)
is construed as a human being, it is easily recoverable as an Agent (p. 152).” How-
ever, the loving father in (6) is not a typical agent, nor is it clear that he is construed
as a causal force, so evidence for this extension remains weak.

4.3  rom domains 4, 5, 6 (inanimate concepts) to domain 3

F
(participants in human activity)
From Cause to Purpose. Usually directionality from Purpose to Cause is assumed
but Luraghi has made a point for bidirectionality. The best historically d
­ ocumented
 Heiko Narrog

case may be that of the Ancient Greek preposition diá (Luraghi 2003, Chapter 3.9;
Luraghi 2005). Diá originally had two spatial meanings, one being roughly a perla-
tive being translatable as ‘about’, and the other a transitional Goal. If the Purpose
sense had developed as an extension of the goal reading it would be well-behaved.
However, it developed at a later stage when the Goal reading had already disap-
peared, and can only be thought of as an extension of the Cause sense (cf. Luraghi
2003: 174, 184, 187).

5. Directionalities within domain 2

The preceding section has shown that the problem with Heine et al.’s (1991) scale
lies primarily with directionalities concerning Agents. Furthermore, it is clear that
there are not only one but several types of Agents involved in semantic and func-
tional changes. This section therefore deals briefly with internal directionalities
within the agentive domain, although the functional changes within this domain
do not constitute direct counterexamples.
From Passive Agent to Ergative Agent. As already described above, the
“canonical” scenario of emergence of ergative structures is the reanalysis of passive
structures, in the course of which the Passive Agent is reanalyzed as an ­Ergative
Agent. The reverse directionality of change is unknown. This does not need fur-
ther elaboration here.
From Causee to Passive Agent. It is well-known that causative constructions
can be the source for passive constructions (cf. Haspelmath 1990; Section 4.3). In
these cases the Causee is reinterpreted as the Passive Agent. In Middle Japanese,
for example, a passive use of causative constructions occurred in specific phrases
and contexts such as the following (the Causee/Passive Agent is marked by the
dative):

(7) Teki=ni ut-as.u.

enemy-dat- shoot-cau-fnp
‘Let the enemy shoot oneself (= be shot by the enemy).’

In warriors’ language, ‘allowing the enemy to shoot oneself ’ functioned as a euphe-

mism for (suffering the disgrace of) being shot. In Japanese, this usage was short-
lived, but in other languages it grammaticalized. Wiemer (2011) reports such
shifts from Causee to passive Agents for Turkic Tungusic, Gujarati and Greenlan-
dic Inuit.
From Ergative Agent to nominative. When ergative languages become accu-
sative, the erstwhile Ergative Agent marker, only covering the Agent of transitive
 The grammaticalization chain of case functions 

clauses, may generalize to a nominative marker. Dixon (1994: 202) mentions the
case of Mingrelian, where this development apparently took place. One way in
which such a development may proceed is via P ellipsis, as shown by Creissels
(2008) for another Caucasian language, Akhvakh. This language has a certain
number of transitive verbs that consistently lack P arguments in the absolute
form and only have ergatively-marked A arguments and objects marked with
spatial case. This is a pocket of potential accusative alignment which can lead
to a reanalysis of ergative as nominative if it spreads beyond a limited class of
verbs. Lehmann (2002: 98) mentions a similar case in Sherpa, where an accusa-
tive pattern emerged from object-incorporation in transitive clauses. Note that
the reverse case, shift from nominative to ergative case marking is apparently
not documented, and is also unlikely, since ergatives are characterized by being
morphologically marked, while nominatives are usually unmarked in contrast
to accusatives. A shift from morphologically zero-marking to overt marking is
technically difficult to realize.
From Companion to Passive Agent. Although Heine and Kuteva (2002: 79)
mention an extension from Comitative to Passive Agent, polyfunctionality of
Comitative and Passive Agent excluding instrumental function is exceedingly
rare (cf. Stolz 2001a: 162, 167; Narrog 2010: 241), and it is very likely that a path
from comitative to Passive Agent is always mediated by instrumental function.
In contrast to the preceding three ‘constructional’ cases, this would be a semantic
extension.

6. Result

While many of the directionalities underlying Heine et al.’s (1991: 159) chain of
increasing grammaticalization of case functions can still be confirmed, as was
shown in Section 3, the cases of diachronic change presented in Section 4 pro-
vided evidence for the fact that there is also (counter-directional) change from
the ‘participants in human activity’ domain to the ‘human participants’ domain,
and perhaps even some from the inanimate domains to the animate ones. If we
wanted to retain the original chain, and insert some of the directionalities dis-
cussed in Section 4, we would obtain a map as in Figure 4. The problematic areas
are surrounded by rectangles. The connections that are somewhat tentative, or
where the evidence is limited, are indicated with dashed lines. No distinction is
made between presumptive semantically-motivated extension, and extension (in
a broad sense) through constructional change, since Heine et al.’s (1991) chain is
designed to represent grammaticalization in general, although semantic extension
was assumed as a default.
 Heiko Narrog

1 2 3 4 5 6

Source Agent Purpose

Goal Companion Instrument Time Condition
> > > > > Manner
Location Beneficiary Recipient Cause
Path Possessor

Figure 4. Chain of increasing grammaticalization of case functions (Heine et al. 1991: 159)

It is obvious that the chain originally presented by Heine et al. (1991) is essen-
tially not tenable. Furthermore, in Sections 4 and 5 I have shown some important
directionalities that deserve to be integrated in any representation of a grammati-
calization chain of case functions.
There is certainly more than just one way to represent the development of
case markers and functions, but the idea in this paper is to salvage the original
concept of a chain and adjust it to the empirical facts. That is, the goal is to
retain groupings that may allow for some generalizations instead of showing
only individual functions as is more common in semantic maps research (cf.
Narrog (2009, 2010), Malchukov & Narrog (2009) for detailed maps of case
functions in this area), and also to focus on overall directionality instead of on
individual directionalities. It is immediately clear, though, that a linear, one-
dimensional representation cannot accomplish that. There is no one, singular
chain of change. While Heine et al. ’s (1991) chain of change towards abstract
categories such as time, condition etc. to the right can be preserved, there
appears to be at least one more direction of development diverging from this
linear chain, namely a direction of development towards core or ‘syntactic’ case
functions. Thus the chain has to be split at least into two dimensions. Further-
more, the chain of development towards core case functions, can, and probably
should, be more differentiated, indicating different categories of Agents, which
exhibit different degrees of grammaticalization on the one hand, and Patient/
accusative on the other hand.
Moreover, since the one-dimensional “chain” cannot be preserved, it may be
more appropriate to speak of a semantic map (cf. Haspelmath 2003; Narrog & Ito
2007). Note that the horizontal directionality to the right is mainly due to seman-
tic extension, while the vertical one to the top is mainly due to constructional
change, with the exceptions mentioned in 4.1.2. I don’t see a reason, though, to
separate these two types of change, because both result in grammaticalization,
and although the constructional change is probably not semantically motivated, it
should be at least semantically compatible, unless there is evidence for the oppo-
site. Figure 5 shows my proposal.
 The grammaticalization chain of case functions 

nominative
^
Ergative Agent
accusative,
^
absolutive Passive Agent
Patient ^
Causee

Time,
Source Recipient/dative
Manner
Goal Possessor
Condition
Location Beneficiary Purpose
Path Instrument
Companion Cause

Figure 5. Map of increasing grammaticalization of case functions9

By hypothesis then, there are two overall directions in grammaticalization

of case functions, namely one towards ‘core’ syntactic functions, primarily moti-
vated syntactically, and one towards abstract non-participant concepts, primarily
motivated semantically. The ultimate source concepts are, as assumed in previous
research, spatial. There is one line of development going from spatial concepts to
human participants, to participants in human activity to the abstract concepts,
largely as envisioned by Heine et al. (1991), and another line of development
towards the two core case functions, Agents and Patients, further grammatical-
ized as nominatives/ergatives and accusatives/absolutives. Recipient (further
grammaticalized as dative) seems to be the one function that participates in both
chains.
Note that the map is not equally detailed in every area. There are also exten-
sion relationships between spatial cases, and for lack of space that are not rep-
resented here, I had to c­ onflate Heine et al.’s (1991) abstract groups 4, 5, and 6

. Note that like in Heine et al.’s (1991) chain, a category on the left or bottom that extends
to, or shifts to, a category on the right or the top does not necessarily have to go through all
intermediate steps. Furthermore, the fact that the agentive functions are placed to the right
of the patient functions is a pure convenience solution for graphical representation in a two-
dimensional space. It is not assumed here that one of these two domains is more ‘advanced’
with respect to the horizontal directionality of change than the other.
 Heiko Narrog

into just one group. In ­contrast, I introduced considerable differentiation in the

‘Agent’ domain. What counts here, though, is the overall direction of grammati-
calizations. The next ­section shall discuss implications of these findings.

7. Discussion

The grammaticalization chain hypothesized by Heine et al. (1991) has only direc-
tion, which is motivated by metaphor. However, the data in Sections 3 through 5
have shown that meaning changes in fact go off in two directions, namely towards
syntactic core cases on the one hand, and towards abstract adverbial cases on the
other hand. These two directionalities have arguably one thing in common: they
both represent shift away from more concrete semantic notions towards highly
frequent, automatized, and semantically bleached case functions. That is, in the
construction as a whole, in which the case marker typically marks a nominal ele-
ment, the semantic weight shifts practically completely to the nominal, that is,
away from the case marker, which in turn is bleached out.
We have seen above that changes away from semantic contents towards
purely, or almost purely, syntactic case functions are not always semantically
motivated, but are in some cases the by-product of constructional reanalysis.
The question remains, though, whether Lakoffian conceptual metaphor, which
served to explain the original chain, is still applicable to the development of
these bleached-out functions, regardless of the mechanism of change to which
they are due.
The discussion of individual ‘counter-directional’ cases in Section 4 has shown
that the shifts such as those from Passive to Ergative Agents, and from Ergative
Agents to nominatives may not be semantically motivated at all. For example, a
passive construction, rising in frequency, becomes the default alignment pattern,
or, also through a generalization of a henceforth minor construction, the ­Ergative
Agent (A) generalizes to intransitive subjects (S). Even if the extension is not nec-
essarily the by-product of constructional change, it can still be difficult to see a
clear semantic motivation. This holds for example for the (presumptive) exten-
sion of datives to functions such as Causees, Passive Agents, or Patients. While
it may be possible to conceive of a Causee as a Recipient in some sense (i.e. as
­simultaneously a Recipient in a causing event A, and an Agent in a caused event
B), it is not clear at all why a Passive Agent should be a Recipient. Instead, it is
more likely that the dative simply marks the Passive Agent as a third argument.
Likewise, the change from dative to accusative may as well be a change from a
less general ­grammatical case to a more general one without specific semantics
involved.
 The grammaticalization chain of case functions 

However, in the case of changes from less grammaticalized cases, such as

instrumental and ablative, to Agents, a semantic interpretation is plausible. As was
shown in Section 4.1, Passive Agents are not only often optional, they are also dia-
chronically secondary in the development of passive constructions. That is, they
are not inherited from a previous construction but speakers choose a case (recall
that in earlier stages of the grammaticalization of the English passive, there was
still considerable variation in the case marking of the Agent), and one likely choice
is on the basis of semantic similarity.
One possibility would be to exclude the most grammaticalized case func-
tions from the discussion of possible semantic motivations, and retain only those
involving semantically specific cases in clearly semantically motivated change. A
semantic explanation would have to come up at least for the less grammaticalized,
and less abstract cases. However, this would still include the problematic extension
from Instruments and Sources to Agents. Note that even this extension poses a
real challenge to metaphorically-based, or conceptually-based accounts in gen-
eral. It not only contradicts Heine et al. ’s (1991) hierarchy but also more general
notions of the animate domain serving as a template for the inanimate domain and
not vice versa (e.g. Dahl 2008).
We will focus here on the change from Instrument to Agent, which is r­ elatively
well-known in the literature as a representative of this type of change. As Luraghi
(2001b) in her discussion of the Instrument-Agent problem in Heine et al. ’s (1991)
chain notes, the Lakoffian conceptual metaphors posit that physical objects and
inanimate domains are conceptualized in terms of human domains and not vice
versa. That is, conceptualizing an Agent as an Instrument is counter-directional
to conceptual metaphor. Luraghi tries to give a solution to this problem by posit-
ing metonymic instead of metaphorical extension between Instrument and Agent.
In her words, “[a]gent and instrument are neighboring concepts; in a sense, an
instrument is “part” of an Agent bringing about an event”. Furthermore, “metony-
mies often substitute inanimate entities for animate ones” (Luraghi 2001b: 398).
As also argued elsewhere (Sawicki 1991; Butt 2006: 84), Instruments and Agents
are conceptually closely related, and sometimes hardly distinguishable. Claiming
a semantic extension in terms of metonymy thus makes perfect sense. However,
Heine et al. ’s (1991) and most other models of metaphor do not provide different
directionalities of change for metaphor and metonymy. On the contrary, semantic
change is conceived of as proceeding in small metonymical steps in context, which
on a large scale turn out to look metaphoric. That is, metonymies are the small
stepping-stones which make up bigger metaphorical leaps (cf. Heine et al. 1991;
Chapter 3.3).
In fact, a similar kind of apparent counter-directionality as the one between
Instrument and Agent is already acknowledged in Heine et al. (1991: 160, 188) and
 Heiko Narrog

built into the case function chain, namely the extension from domain 1 (spatial
concepts) to domain 2 (human agents), which is apparently counterdirectional to
PERSON > SPACE metaphors in the lexical domain. The authors give the follow-
ing explanation:
On the one hand, there is the level of “concrete,” referential concepts and kinetic
processes that serve as templates for more abstract functions. On this level,
thing-like entities such as body parts, for example, serve as vehicles to express
spatial concepts. For the sake of convenience, we refer to this level as Level A.
On the other hand, there is the level of more “abstract” entities, the level where
concepts used to express relations within a sentence or clause are located. On
this level, SPACE constitutes the most “concrete” category and is employed for
the expression of more “abstract” relations. …. This level may be referred to as
Level B.(Heine et al. 1991: 188)

In other words, space is conceived of as belonging to a level A of concrete, referen-

tial concepts, while Agents in events belong to an abstract level B, and comparing
spatial and human concepts in the grammatical domain, the spatial concepts are
more concrete.
Although on first sight this explanation may appear to be just an ad-hoc expe-
dient it does make sense. In grammar, spatial objects are more concrete and more
specific than Agents and Patients. For example, depending on language, there can
be a large number of prepositions, spatial nouns, and cases that differentiate vari-
ous spatial relationships. In contrast, the marking for Agents and Patients is usually
limited (often they are not marked at all). Arguably this is also the case between
Instruments and Agents. Instruments are more semantically specific than Agents,
which apply to a larger range of events. If we advance further from specific Agent
roles such as Passive or Ergative Agents to nominative case, this usually covers a
large variety of actual semantic roles. In the case of the extension from Instrument
to Agent, it is entirely reasonable to think of a conceptualization of an Agent as
some sort of Instrument. This is what for example the description of the extension
from Instrument to Agent in the case of the English preposition by, as described
in the OED suggests. In the OED (by, prep., adv.), it is stated that a quasi-­agentive
use of by first appeared with by “introducing the i­ ntermediate or subordinate agent
viewed as the medium or channel of action” in the 14th century,10 that is, part of
the instrumental meaning of by (cf. OED; by, prep., adv., 32a.), and only in the

. The first example given by the OED goes as follows: Darie hit wot bye a spye (K. Alis.;
ca. 1300) ‘Darie knew it through a spy’. This is an instrument in a broader sense, namely as a
medium or channel. As Silvia Luraghi has pointed out to me, agents as instruments in a nar-
rower sense are rarer, and usually limited to military contexts.
 The grammaticalization chain of case functions 

15th century full-fledged agent uses are found. The agent as a medium or channel
would be Heine’s level B. This is something different than conceptualizing one con-
crete entity in terms of another concrete entity on level A. On the level of concrete
entity, it is difficult to conceptualize a person as a thing, as a channel or medium.
At this point it is interesting to recall another line of research that focuses on
semantic extension in derivational suffixes. As Dressler (1986), Booij (1986) and
Luján (2010) for example show, in the derivation of agentive nouns, it is usually
the agentive function that precedes the instrumental function of a derivational
suffix. That is, we expect to get first the serial killer, and only later the bottle opener.
But here we are on the level A of naming concrete entities. The bottle opener is
conceptualized as a willful agent that has the potential to perform a specific action.
In conclusion, the principle of metaphor still holds in the domain of case
functions, to the extent that changes are semantically motivated at all. Cases of
extension that are open to semantic explanation generally can be accounted for
in terms of metaphor. In my view, the problems with Heine et al.’s (1991) chain
are three-fold. First, there was not enough awareness of constructional reanaly-
sis as the source of change in case functions. Second, there was and still is not
enough knowledge on metaphor in grammar. Conceptual metaphor research in
the Lakoffian tradition focused almost entirely on a concrete domain of entities
and events. The grammatical domain has been neglected and is understudied.
Thirdly, Heine et al. (1991), probably influenced by the conceptual metaphor
research on concrete entities, misplace Agents on the cline of grammaticaliza-
tion, thinking of them as concrete entities, when in fact they are among the most
bleached out. Overall, a single chain of development cannot accommodate all
case functions.

8. Conclusion

The data presented in this paper, and their analysis have shown that although a large
number of directionalities in the change of case functions hypothesized by Heine et
al. (1991) can be confirmed, there is no one single chain of grammaticalization into
which they can be integrated. Instead, at least two larger directionalities must be
posited. One of them is a shift towards syntactically-conditioned core case mark-
ing, often triggered by constructional reanalysis, and the other is a shift towards
adverbial marking in highly abstract domains. The shift towards core case mark-
ing is reinforced by the tendency within the same domain of Agent role to move
from the marking of optional participants (e.g. Passive Agent) towards marking of
obligatory, highly generalized participants, such as nominative and accusative. It is
worth-while to note that while the vertical chain towards ­syntactically-conditioned
 Heiko Narrog

core case marking is usually the result of constructional reanalysis (4.1.1), in those
minority cases where we see semantic motivation (4.1.2), the directionality does
not contradict that of changes triggered by constructional reanalysis.
Despite proceeding in completely different directions, the two shifts have argu-
ably in common that they both represent shift away from more concrete seman-
tic notions towards highly frequent, automatized, and semantically bleached case
functions. Furthermore, it is interesting to note that dative/recipient functions are
in a pivot position between the two major directionalities. That is, dative func-
tions, which are on the borderline between peripheral and core participants in
syntax, are also in a transitional position between concrete, and more semantic
case functions on the one hand and abstract case functions on the other hand.
Since the chain of increasing grammaticalization of case functions posited by
Heine et al. (1991) turns to out to be untenable, and can be salvaged only in parts,
the question is whether the proposed cognitive process by which it is supported,
namely metaphor, is also untenable. The answer must be differentiated. Firstly, not
all changes of case functions are meaningfully explained in terms of semantics,
especially if those changes are not the result of a semantic reanalysis or extension
but the syntactic reanalysis of a whole construction. On the other hand, in those
cases where a semantic explanation is called for, e.g. the extension from Instru-
ment to Passive Agent, an account in terms of metaphor is still possible. However,
metaphorical processes in the grammatical domain may operate differently form
metaphorical processes on the level of concrete entities. This is a topic which is still
understudied and deserves further investigation.

Acknowledgments

I am very, very grateful to Spike Gildea and Silvia Luraghi for their immensely
helpful comments on previous versions of this paper. All remaining shortcomings
are my own. Part of this research was supported by the Japanese Society for the
Promotion of Science through Grant-in-Aid for Scientific Research #24520450.

References

Anderson, Stephen R. 1977. On mechanisms by which languages become ergative. In Mecha-

nisms of Semantic Change, Charles N. Li, (ed.), 317–63. Austin TX: University of Texas Press.
Blake, Barry J. 2001. Case, 2nd edn. Cambridge: CUP. DOI: 10.1017/CBO9781139164894
Booij, Geert. 1986. Form and meaning in morphology: The case of Dutch ‘agent nouns’. Linguis-
tics 24: 503–517. DOI: 10.1515/ling.1986.24.3.503
Bostoen, Koen & Nzang-Bie, Yolande. 2010. On how “middle” plus “associative/reciprocal”
became “passive” in the Bantu A70 languages. Linguistics 48(6): 1255–1307. DOI: 10.1515/
ling.2010.041
 The grammaticalization chain of case functions 

Brachet, Auguste. 1882. An Etymological Dictionary of the French Language. Translated by

George William Kitchin, 3rd edn. Oxford: Clarendon Press.
Butt, Miriam. 2006. The dative-ergative connection. In Empirical Issues in Syntax and Semantics
6, Olivier Bonami & Patricia Cabredo Hofherr (eds), 69–92. 〈http://www.cssp.cnrs.fr/eiss6/
index_en.html〉
Creissels, Denis. 2008. Direct and indirect explanations of typological regularities: The case of
alignment variations. Folia Linguistica 42(1): 1–38. DOI: 10.1515/FLIN.2008.1
Creissels, Denis. 2009. Spatial cases. In The Handbook of Case, Andrej Malchukov & Andrew
Spencer (eds), 609–25. Oxford: OUP.
Croft, William. 1991. Syntactic Categories and Grammatical Relations. Chicago: Chicago Uni-
versity Press.
Dahl, Östen. 2008. Animacy and egophoricity: Grammar, ontology and phyologeny. Lingua 118:
141–50. DOI: 10.1016/j.lingua.2007.02.008
Dixon, Robert M.W. 1994. Ergativity. Cambridge: CUP. DOI: 10.1017/CBO9780511611896
Dressler, Wolfgang U. 1986. Explanation in natural morphology: Illustrated with comparative
and agent-noun formation. Linguistics 24: 519–548. DOI: 10.1515/ling.1986.24.3.519
Endruschat, Annette. 2007. Durch ‘mit’ eingeleitete präpositionale Objekte in den romanischen
Sprachen. Bochum: Universitätsverlag Dr. N. Brockmeyer.
Estival, Dominique & John Myhill. 1977. Formal and functional aspects of the development
from passive to ergative systems. In: Shibatani, Masayoshi (ed) Passive and Voice, 441–491.
Amsterdam: Benjamins.
Frellesvig, Bjarke. 2010. A History of the Japanese Language. Cambridge: CUP.
Garrett, Andrew. 1990. The origin of split ergativity. Language 66(2): 261–96. DOI: 10.2307/414887
Gildea, Spike. 1998. On Reconstructing Grammar. Comparative Cariban Morphosyntax. Oxford:
OUP.
Gildea, Spike & de Castro Alves, Flavia. 2010. Nominative-absolutive: Counter-universal split
ergativity in Je and Cariban. In Ergativity in Amazonia [Typological Studies in Language
89], Spike Gildea & Francesc Queixalos (eds), 159–199. Amsterdam: John Benjamins.
Gildea, Spike. 2013. Reflexive causative > passive in English and Cariban. Handout of presenta-
tion at the 21st Symposium About Language and Society-Austin (SALSA), 12 April 2013.
Grünthal, Riho. 2003. Finnic Adpositions and Cases in Change. Helsinki: Societé Finno-ougrienne.
Harris, Alice C. & Campbell, Lyle. 1995. Historical Syntax in Cross-linguistic Perspective.
­Cambridge: CUP. DOI: 10.1017/CBO9780511620553
Goedegebuure, Petra. 2012. Split-ergativity in Hittite. Zeitschrift für Assyriologie und Vorderasi-
atische Archäologie 102(2): 270–303.
Haspelmath, Martin. 1990. The grammaticalization of passive morphology. Studies in Language
14(1): 25–72. DOI: 10.1075/sl.14.1.03has
Haspelmath, Martin. 1997. From Space to Time. Temporal Adverbials in the World’s Languages.
Munich: Lincom.
Haspelmath, Martin. 2003. The geometry of grammatical meaning: Semantic maps and cross-
linguistic comparison. In The New Psychology of Language, Vol. 2, Michael Tomasello (ed.),
211–243. Mahwah NJ: Lawrence Erlbaum Associates.
Heine, Bernd. 1997. Cognitive Foundations of Grammar. Oxford: OUP.
Heine, Bernd. 2003. Grammaticalization. In The Handbook of Historical Linguistics, Brian D.
Joseph & Richard D. Janda (eds), 575–601. Oxford: Blackwell. DOI: 10.1002/9780470756393.
ch18
Heine, Bernd. 2009. Grammaticalization of cases. In The Handbook of Case, Andrej ­Malchukov &
Andrew Spencer (eds). Oxford: OUP.
 Heiko Narrog

Heine, Bernd, Claudi, Ulrike & Hünnemeyer, Friederike. 1991. Grammaticalization. A Concep-
tual Framework. Chicago, IL: Chicago University Press.
Heine, Bernd & Kuteva, Tania. 2002. World Lexicon of Grammaticalization. Cambridge: CUP.
DOI: 10.1017/CBO9780511613463
Heine, Bernd & Kuteva, Tania. 2005. Language Contact and Grammatical Change. Cambridge:
CUP. DOI: 10.1017/CBO9780511614132
Heine, Bernd & Narrog, Heiko. 2010. Grammaticalization and linguistic analysis. In The Oxford
Handbook of Linguistic Analysis, Bernd Heine & Heiko Narrog (eds), 401–423. Oxford: OUP.
Jung, Hakyung. 2009. Possessive subjects, nominalization, and ergativity in North Russia. In
Grammatical Change in Indo-European Languages. Papers Presented at the Workshop on
Indo-European Linguistics at the XVIIIth International Conference on Historical Linguistics,
Montreal, 2007. [Current Issues in Linguistic Theory 305], Vit Bubenik, John Hewson &
Sarah Rose (eds), 207–220. Amsterdam: John Benjamins.
Keenan, Edward L. & Dryer, Matthew S. 2007. Passive in the world’s languages. In Language
Typology and Syntactic Description, Vol. I: Clause Structure, Timothy Shopen (ed), 325–61.
Cambridge: CUP.
Lehmann, Christian. 22002. Thoughts on Grammaticalization, 2nd, revised edn. [Arbeitspapiere
des Seminars für Sprachwissenschaft der Universität Erfurt 9]. Erfurt: Seminar für Sprach-
wissenschaft der Universität. Earlier editions 1982, 1995.
Luján, Eugenio R. 2010. Semantic maps and word formation: agents, instruments, and related
semantic roles. Linguistic Discovery 8(1): 162–175. DOI: 10.1349/PS1.1537-0852.A.349
Luraghi, Silvia. 2001a. Syncretism and the classification of semantic roles. Sprachtypologie und
Universalienforschung 54(1): 35–51.
Luraghi, Silvia. 2001b. Some remarks on instrument, comitative, and agent in Indo-European.
Sprachtypologie und Universalienforschung 54(4): 385–401.
Luraghi, Silvia. 2003. On the Meaning of Prepositions and Cases. The Expression of Semantic
Roles in Ancient Greek [Studies in Language Companion Series 67]. Amsterdam: John Ben-
jamins. DOI: 10.1075/slcs.67
Luraghi, Silvia. 2005. Paths of semantic extension. From cause to beneficiary and purpose. In
Historical Linguistics 2003. Selected Papers from the 16th International Conference on His-
torical Linguistics, Copenhagen, 11–15 August 2003. [Current Issues in Linguistic Theory
257], Michael Fortescue, Eva Skafte Jensen, Jens Erik Mogensen & Lene Schøsler (eds),
141–157. Amsterdam: John Benjamins.
Luraghi, Silvia. 2010. Where do beneficiaries come from and how do they come about? Sources
for beneficiary expressions in Classical Greek and the typology of beneficiary. In Histori-
cal Cognitive Linguistics, Margaret E. Winters, Heli Tissari & Kathryn Allen (eds), 93–131.
Berlin: Mouton de Gruyter.
Luraghi, Silvia, Pompei, Anna & Skopeteas, Stavros. 2005. Ancient Greek. Munich: Lincom.
Malchukov, Andrej & Narrog, Heiko. 2008. Case polysemy. In The Oxford Handbook of Case,
Andrej Malchukov & Andrew Spencer (eds), 518–535. Oxford: OUP. DOI: 10.1093/
oxfordhb/9780199206476.001.0001
Michaelis, Laura & Rosalie, Marcel. 2000. Polysémie et cartes sémantiques: La relateur (av)ek en
créole Seychellois. Études Créoles 23(2): 79–100.
Narrog, Heiko. 2009. Varieties of Instrumental. In The Handbook of Case, Andrej Malchukov &
Andrew Spencer (eds), 593–600. Oxford: OUP.
Narrog, Heiko. 2010. A diachronic dimension in maps of case functions. Linguistic Discovery
8(1): 233–257. DOI: 10.1349/PS1.1537-0852.A.352
 The grammaticalization chain of case functions 

Narrog, Heiko & Ito, Shinya. 2007. Reconstructing semantic maps. The comitative-instrumental
area. Sprachtypologie und Universalienforschung 60(4): 273–292.
Noonan, Michael & Mihas, Elena. 2007. Areal dimensions in case syncretism: Ablatives and
genitives. Paper presented at ALT VII, Paris, 25–28 September 2007.
OED = Oxford English Dictionary, 3rd edn. 2002. CD-Rom.
Palancar, Enrique L. 2002. The Origin of Agent Markers [Studia Typologica 5]. Berlin: Akademie
Verlag.
Poudel, Tikaram. 2008. Nepali ergativity: A historical perspective. Handout at the Workshop on
Case and Alignment in Indo-European. University of Bergen, 10–11 December 2008.
Sawicki, Lea. 1991. Genitive and instrumental in passive constructions in Lithuanian. Indoger-
manische Forschungen 96: 168–74.
Schwyzer, Eduard. 1950. Griechische Grammatik. Auf der Grundlage von Karl Brugmanns
Griechischer Grammatik. Zweiter Band. Syntax und syntaktische Stilistik. München: Beck.
Shibatani, Masayoshi & Pardeshi, Prashant. 2001. The causative continuum. Kobe Papers in Lin-
guistics 3: 136–177.
Shirane, Haruo. 2005. Classical Japanese. A Grammar. New York NY: Columbia University Press.
Siewierska, Anna. 1988. The passive in Slavic. In Passive and Voice [Typological Studies in Lan-
guage 85], Masayoshi Shibatani (ed.), 243–289. Benjamins: John Amsterdam.
Siewierska, Anna. 1998. Passive-to-ergative versus inverse-to-ergative. In Case, Typology and
Grammar. In Honor of Barry J. Blake [Typological Studies in Language 38], Anna ­Siewierska
& Jae Jung Song (eds), 229–46. Amsterdam: John Benjamins.
Stolz, Thomas. 2001a. Comitatives vs. instrumentals vs. agents. In Aspects of Typology and Uni-
versals, Walter Bisang (ed.), 153–74. Berlin: Akademie Verlag.
Stolz, Thomas. 2001b. To be with X is to have X: Comitatives, instrumentals, locative and pred-
icative possession. Linguistics 39(2): 321–350. DOI: 10.1515/ling.2001.013
Stolz, Thomas, Stroh, Cornelia & Urdze, Aina. 2006. On Comitatives and Related Categories. A
Typological Study with Special Focus on the Languages of Europe. Berlin: Mouton de Gruyter.
Stroh, Cornelia. 1998. Die Geschichte der Mit-Relationen im Französischen: Komitativ-­
Instrumental-Synkretismus mit Hindernissen. Sprachtypologie und Universalienforschung
51(2): 131–156.
Stroh, Cornelia. 1999. MIT-Relationen in der Romania: Ein klarer Fall? Sprachtypologie und
Universalienforschung 52(2): 183–195.
Toyota, Junichi. 2008. Diachronic Change in the English Passive. Houndmills: Palgrave Macmil-
lan. DOI: 10.1057/9780230594654
Trask, Robert L. 1979. On the origins of ergativity. In Ergativity: Towards a Theory of Grammati-
cal Relations, Frans Plank (ed), 385–404. London: Academic Press.
Verbeke, Saartje & De Cuypere, Ludovic. 2009. The rise of ergativity in Hindi. Assessing the role
of grammaticalization. Folia Linguistica Historica 30: 367–390.
Wiemer, Björn. 2011. The grammaticalization of passives. In The Oxford Handbook of Gram-
maticalization, Heiko Narrog & Bernd Heine (eds), 535–546. Oxford: OUP.
Willett, Thomas. 1988. A cross-linguistic survey of the grammaticalization of evidentiality. Stud-
ies in Language 12(1): 51–97. DOI: 10.1075/sl.12.1.04wil
Yamaguchi, Kazuyuki. 2004. A Typological, Historical, and Functional Study of Adpositions in
the Languages of the World. Ph.D. dissertation, University of New Mexico.
Yan, Yaoliang. 2003. The Grammaticalization of Yu from a Verb to a Function Word. Ph.D. dis-
sertation, University of Minnesota.
 Plotting diachronic semantic maps
The role of metaphors

Silvia Luraghi
University of Pavia

The paper discusses patterns of polysemy among semantic roles in cross-linguistic

perspective, and the possible way in which such polysemies come into being
through meaning extension of morphemes (cases and adposition) that encode
semantic roles. Semantic extension is described based on commonly assumed
metaphors, which are thought to account for frequent polysemies, such as the
Companion Metaphor in the case of the Comitative-Instrument polysemy.
Taking space as a basic domain, metaphors are shown to provide links among
different cognitive domains, which account for the position of semantic roles
relative to one another in the ensuing semantic map. Such map incorporates
diachronic information, assuming that metaphorical extension reflects diachronic
developments. Although the specific focus of the paper concerns the contribution
of metaphors to semantic extension, it is also shown that other processes are
involved, such as analogy and various kinds of metonymy, which can account for
some specific patterns of polysemy among semantic roles.

Keywords: Metaphor; semantic maps; polysemy; meaning extension

1. Introduction1

This paper describes paths of semantic extension undergone by morphemes (cases

and adpositions) commonly used for coding semantic roles, based on Lakoff and
Johnson’s (1980) theory of metaphor. My specific aim is to show how some meta-
phors, commonly held to explain certain patterns of polysemy, can provide direc-
tions for plotting a semantic map that incorporates diachronic information. In
doing so, I also offer an explanation regarding cross-linguistically frequent and

. I thank Kathryn Allan, Spike Gildea, Bernd Heine and Heiko Narrog for helpful com-
ments on the content and style of this paper. In addition, I am grateful to Tuomas Huumo,
Fabrizio Pennacchietti and Manana Topadze, who provided some of the examples.
 Silvia Luraghi

infrequent polysemies, shedding light on the structure of the conceptual space

which serves as the basis for language specific semantic maps.
By highlighting metaphorical patterns that provide connections among dif-
ferent domains, I suggest that different sources for the same semantic role indicate
different conceptualizations that might account for unexpected polysemies. I also
argue that absence of polysemy may be relevant, especially when it frequently
holds among semantic roles which are often related to neighboring concepts. In
addition, tracing the metaphor responsible for a specific polysemy sheds light on
the causes of merger among semantic roles that should in principle be kept dis-
tinct, as they are differently located in the causal chain of events (Croft 1991).
While this paper specifically concentrates on the role of metaphors in seman-
tic extension and in the development of polysemy, it must be pointed out that
not all polysemies are brought about by metaphors. Different types of metonymy
(referential and predicational) also have a role, as I show especially with reference
to the polysemy of beneficiary and recipient and of instrument and agent. In addi-
tion, I argue that semantic extensions may be based on analogical processes not
necessarily resulting in metaphors, caused by analogical extension of partial sche-
maticity (see Section 2.3.1). By singling out semantic extensions based on different
processes one can avoid the risk of overusing the analysis, and set up unwarranted
metaphors.
As is generally accepted, I assume the domain of spatial relations to be the
basic conceptual domain.2 I argue that spatial relations provide a source for the
encoding of semantic roles typical of human beings, but that they are not the only
source. Human relations can also provide the source domain for metaphorical
extension to other human relations. Crucially, patterns of polysemy seem to be
conditioned by the source domain: polysemies that do not arise, or are extremely
infrequent, when space functions as the source domain, seem to be more easily
brought about when the source domain is at the level of human relations.
The understanding of the course taken by diachronic developments can also
shed light on the frequently neglected issue of possible semantic extensions which
result in polysemy vs. those which result in changes, whereby the original mean-
ing is not preserved. Examples for both tendencies are provided by comitative
coding. When comitative markers extend to instrument based on the Companion
Metaphor, they most often also continue to encode comitative (examples are avail-
able from several European languages, cf. Stolz et al. 2006). On the other hand,

. As noted in Croft (1991: 192) “locative and directional markers are probably the ultimate
historical source of most case marking.” Localistic theories have a long history, reaching back
to late antiquity; see Hjelmslev (1935) and Anderson (1971) for a comprehensive discussion.
 Plotting diachronic semantic maps 

reported instances of extension from location to comitative (see Luraghi 2001b

and below, Section 2.2.3) apparently result in the loss of the original meaning by
the marker that undergoes the change.
The latter type of development might have consequences on what we con-
ceive as frequent vs. infrequent syncretism, as well as on our understanding of
(uni)directionality: since diachronic data are often unavailable and can hardly
be reconstructed for a great number of languages, what looks like an infrequent
semantic extension may in reality only be an infrequent type of synchronic poly-
semy (see further Narrog 2010a). Similarly, seeming unidirectionality of possible
semantic extension can be an effect of some meanings being lost while others are
not. Such changes also raise problems for a theory of gradualness in semantic
change, as I argue in Section 4.
The paper is organized as follows. In Section 2 I discuss conceptual domains
to which different semantic roles belong. Following Heine et al. (1991), I distin-
guish between the basic domain of space, the domain of human relations and the
domain of non-human and non-spatial relations. After discussing basic spatial
relations (2.1), I show how space is mapped onto the domain of human relations
(Section 2.2). I then proceed to non-human relations (Section 2.3). I argue that
it is often the domain of space rather than the putatively intermediate domain
of human relations which provides the source for the conceptualization of non-
human relations. In Section 3 I discuss expected and unexpected patterns of poly-
semy among human relations. Section 4 contains a general discussion regarding
the role of metaphors in semantic extension, polysemy, and lack thereof. Meta-
phors described in the paper as responsible for semantic extension and polysemy
among semantic roles are listed in Appendix (i) for convenience.3

2. Encoding semantic roles: Source and target domains

Space is generally assumed to offer the source for conceptualization of other, more
abstract domains (see Pütz 1996: xi; slightly different views are discussed in Zlatev
2007: 319). According to Langacker (1987), space is a basic domain, as it emerges
directly from experience (Lakoff & Johnson 1980: Chapter 12; see further Croft
2003). More abstract relations, in particular those pertaining to the domain of causa-
tion, are held to be conceptualized as metaphorically derived from spatial relations.
Space, then, provides the source domain for the target domain of ­causation, and

. All examples from secondary literature have been checked with informants; I have revised
and unified the glosses, unless differently specified.
 Silvia Luraghi

this explains polysemy of specific coding devices. According to Croft (1991: 193):
“[t]he most important relation between causative and locative-directional roles is
that of domain shift or metaphor.”
Heine et al. (1991) claim that semantic extension proceeds from the concep-
tual domain of space to other domains, following the Chain of Increasing Gram-
maticalization (reproduced in Figure 1 for convenience):

ablative agent purpose

allative > comitive > instrument > time > condition > manner
locative benefactive dative cause
path possessive

Figure 1. Chain of increasing grammaticalization (from Heine et al. 1991: 159)

This scale presents a number of problems, extensively discussed by Narrog

(this volume). Here, I focus on the more general implication contained in the scale
as to the direction of semantic extension, that is, that spatial roles constitute the
basic source domain for all other relations, and that, among the latter, human rela-
tions, i.e. roles typical of human participants, precede inanimate ones as the first
stage of semantic extension. This is represented in (1), from Heine et al. (1991: 160):
(1) spatial relation > human relation > inanimate relation

In the next sub-sections, I survey the groups of relations in Figure 1 in the light of
the scale in (1).

2.1 Spatial relations

Spatial relations are usually described as involving an asymmetrical relation
between a figure and a ground, or a trajector and a landmark. Different specific
relations between the two can potentially be infinite; in practice, only a small num-
ber of spatial relations, including location, direction, source and path, are unani-
mously regarded as basic, as I discuss below.

2.1.1 Basic spatial relations and the position of path

In the light of cross-linguistic coding tendencies, direction (allative),4 source/­
origin (ablative), and (static) location (locative) seem to be more ‘basic’ than path

. I prefer the label “direction” for the allative relation, rater than the label “goal” due to the
polysemy of the latter term, used to indicate the semantic role Patient in certain approaches
(e.g. in S.C. Dik’s Functional Grammar, see Dik 1997), a usage that goes back at least to Bloom-
field (1933).
 Plotting diachronic semantic maps 

(perlative) among spatial relations. As argued in Stolz (1992: 30), there is a ten-
dency for case marking relative to spatial relations to exhibit ‘Dreigliedrigkeit’, i.e.
a tripartite structure featuring dedicated coding devices for location, direction and
source. Indeed, path can often be coded through cases/adpositions that usually
indicate location, as in English:
(2) Mary walks in the field./The child is running in the street.

Tyler and Evans (2003: 257) argue that typical landmarks of the English path prep-
osition through are “characterized by the spatio-geometric property of having four
sides … and hence possessing an interior and an exterior.” Thus, typical landmarks
of through are conceived as containers, in very much the same way as typical land-
marks of in (see e.g. Vandeloise 1994).
It is outside the scope of this paper to provide a detailed cross-linguistic survey
of possible coding of path as location. However, it can be observed that a path and a
location share the same conceptual space between the starting point and the goal of
a directional motion. This is quite straightforward for paths, as shown in Figure 2, in
which I give a schematic representation of a motion event. As shown in Figure 2, the
trajectory that leads from the source to the goal is located in the span that connects
the two endpoints; in dynamic terms, the trajector moves through the landmark.
However, the same situation can also be conceived in static terms, not focusing on
the trajector’s movement from the source to the goal, but rather focusing on the
space occupied by the trajectory. Under this perspective, path is conceived in terms
of location, that is, the portion of space in which a trajectory is located.

source ----> path ----> direction

Tr
trajectory
location
Lm

Figure 2. A prototypical motion event

Note further that a path does not include the starting point and the end point of
the trajectory. Path is basically atelic, a trajector moving along a path is performing
an ongoing activity, and is not implied in an event which includes a change of state/
position (there is no change of position inasmuch as the end result of the motion
activity is not in the scope of this semantic role). On the other hand, a trajector
reaching the goal of a directional motion or one setting out from a source undergo
telic events that imply change of position. The fact of being atelic is another feature
that path shares with location, which is stative and thus obviously atelic.
 Silvia Luraghi

The conceptual space of spatial relations can thus be represented as in Figure 3.

location
source direction

path

Figure 3. The conceptual space of spatial relations

2.1.2 A merger without polysemy: Location and source

Above, I remarked that many case systems display a tripartite nature. This, how-
ever, is only part of the story: as indicated by several studies carried out on large
language samples, patterns of polysemy among the three basic spatial relations
are not random, as languages which code the three relations with only two coding
devices point toward easy merging of direction with location in contrast to the
extreme rarity of source merging with either location or direction. In other words,
source tends to be maximally distinct from other spatial relations. Pantcheva
(2010) compares her and others’ results, as shown in Table 1:

Table 1. Pattern of syncretism for the lexicalization of location, goal, and source
Blake (1977) Noonan (2009) Pantcheva (2010)

L≠G≠S 91% (77/85) 33% (25/76) 53% (28/53)

L=G≠S 9% (8/85) 58% (44/76) 34% (18/53)
L=G=S 0% (0/85) 4% (3/76) 13% (7/53)
L=S≠G 0% (0/85) 2,5% (2/76) 0% (0/53)
L≠G=S 0% (0/85) 2,5% (2/76) 0% (0/53)
(adapted from Pantcheva 2010: 1046; L = location, G = goal [direction], S = source)

The figures in Table 1 point toward a special status of source with respect to
the other roles, and to a comparatively frequent merger of location and direc-
tion. The loss of a tripartite structure in favor of a bipartite one, in which loca-
tion and direction are expressed in the same way, is also historically attested: for
example, loss of the case system in Late Latin eliminated the distinction between
in+abl (location) and in+acc (direction).5 The Romance languages evolved in

. Remarkably,, the three-way distinction was limited to a small number of prepositions

that took two cases in Latin; with the majority of prepositions, location was not distinct from
 Plotting diachronic semantic maps 

two ­directions: some, as e.g. French and Italian, no longer have distinct coding for
the two relations and exhibit the pattern L=G≠S; others, as for example Spanish,
restored the distinction by specializing different prepositions for the coding of
either relation (en = ­location, a = direction; see further Luraghi 2011).
Pantcheva further notes in passing that she left out of her account the ­so-called
‘ablative-locative transfer’, the “historical process where a (originally) source
marker starts being used as a locative marker,” and remarks that in the framework
outlined in her paper “this is quite unexpected.” In fact, this type of extension, to
which little attention has been paid, seems to be frequent cross-linguistically. In
the only extended study of this type of development, Mackenzie (1978) surveys
evidence from several branches of the Indo-European language phylum, Israeli
Hebrew, and two Austronesian languages, Fijian and Sonsorol-Tobi. Among
numerous examples, consider the adverbs in Table 2:

Table 2. French place adverbs

devant ‘before’ < de + avant
dehors ‘outside’ < de + hors
dessus ‘above’ < de + sus
dessous ‘underneath’ <de + sous
deçà ‘on this side’ < de + çà
delà ‘on that side’ < de + là

The French adverbs above can occur, without further adjunction of primary
prepositions, in location or direction expressions. If they occur in source expres-
sions, though, the preposition de ‘from’ must be added:
(3) Il est/va dehors vs. Il vient de dehors
he is/goes outside he comes from outside
‘He is/goes outside.’/ ‘He comes from outside.’

Etymologically, these adverbs already contain the preposition de and consequently

their original meaning contained the notion of source. After acquiring the locative
meaning, they need to be reinforced again by de in order to express source. Note

­ irection (i.e. most prepositions could indicate both location and direction without case vari-
d
ation of the governed NP). In addition, a lexically restricted usage of cases without preposi-
tions displayed both a tripartite structure (with a locative, an ablative, and an allative, this last
function being fulfilled by the accusative) and a bipartite structure sporting the rare source/
location polysemy. Tri- and bipartite structure depended on inflectional classes of specific
nouns; see Leumann, Hofmann & Szantyr (1965) and Luraghi (2009, 2010a) for details.
 Silvia Luraghi

that, given the extensive polysemy between location and direction in French, the
former source adverbs may well have acquired the allative and the locative mean-
ing at the same.
What seems clear from the evidence adduced by Mackenzie, as well as from
other instances of the semantic extension described above (cf. for example Bennett
1989; Nikitina & Spano forthc., Luraghi 2009 and 2010a), is that once a marker
acquires the locative meaning, it loses the original ablative meaning. Thus, while
the extension from source to location is attested, possibly even more frequently
than commonly believed, polysemy tends to be avoided. As we will see further on,
this situation also holds between other semantic roles.6

2.1.3 Space and time

Before moving on to other types of relation, some remarks about the coding of
time are in order. If one takes Heine et al.’s scale in Figure 1 literally, one would
be led to think that human relations, or other non-spatial relations serve as a
source domain for time. However, this does not seem to be the case, as there is
extensive evidence for direct mapping of the domain of space onto the domain of
temporal relations, see e.g. the many examples mentioned for various temporal
concepts in Heine and Kuteva (2002). As pointed out by Lyons (1977: 718): “The
spatialization of time is so obvious and so pervasive a phenomenon in the gram-
matical and lexical structure of so many of the world’s languages that it has been
frequently noted, even by scholars who would not think of themselves as subscrib-
ing to the hypothesis of localism.” Indeed, the metaphor time is space appears to
be deeply entrenched in human cognition, as shown among other things by the
extent to which temporal relations are coded through primarily spatial markers
(­Haspelmath 1997).7

2.2 Human relations

Among human relations, Heine and his associates keep agent, comitative and ben-
eficiary separate from possessor and recipient/addressee (for which they use the
label ‘dative’), which they include in the domain of human activities. In part, there

. It goes beyond the scope of the present paper to provide a detailed investigation of the
reasons why certain polysemies hold while others do not at the level of spatial relations;
however, I will give some tentative suggestions in the discussion in Section 4. Neither is it pos-
sible here to establish the direction of semantic extension in the case of the location/direction
polysemy.
. In spite of frequent extension from space to time and almost inexistent extension in the
opposite direction, some scholars contend that the cognitive domain of time is as basic as that
of space, see e.g. Evans (2004).
 Plotting diachronic semantic maps 

are diachronic considerations supporting this distinction: for example, recipient

seems to originate from beneficiary rather than the other way around. I discuss
this issue later on in this paper; for the time being, however, I prefer not to create
any partition in the group of human relations. In this section, I survey the most
frequent patterns of extension from spatial to human relations, and explore meta-
phorical mechanisms that provide mapping from the source domain of space to
the target domain of human relations.

2.2.1 Two maximally distinct metaphors

Let us start from the observation that polysemy frequently holds between source
and agent on the one hand, and direction and recipient/beneficiary on the other.
Evidence for such polysemies is readily available from numerous genetically
unrelated languages: polysemy of source and agent is common in several Indo-
European languages, including Romance, Greek, part of the Germanic languages;
in the Semitic languages, passive agent is typically encoded through prepositions
cognate to Arabic mn ‘from’, in Turkish one finds the postposition tarafindan
‘from the side’ (Kornfilt 1997; for more evidence see Heine & Kuteva 2002: 29–30).
In many cases, the diachronic development is also attested, as in the case of the
Romance languages.
Such semantic extension is based on the metaphor Agents are Origins:
agents, as initiators of events, are conceived as the point in space where events
originate (see Luraghi 2000). In the case of polysemy of recipient/beneficiary
and direction, the evidence is overwhelming (see Blansitt 1988; Rice & Kabata
2007); the underlying metaphor is Recipients are Destinations. Hence one
can conclude that two separate metaphors account for different semantic roles
typical of human participants. As the source domain provides for maximal dis-
tinction (source and direction are at the two opposite edges of the conceptual
domain of space as represented in Figure 3 above, and polysemy between the
two is virtually non-existent, as shown in Table 1), it is expected for languages
to avoid merging of these two roles. As we will see in Section 3.3, however, this
is not always the case.

2.2.2 Possession: Direction or location?

A possessive relation prototypically implies some kind of control of the possessor
over the possessee (see Taylor 1989a and Heine 1997a and the discussion therein;
control may not apply in case of non-prototypical possession, e.g. inalienable or
abstract, see Heine 1997a: 3). Below is a partial list of features of prototypical pos-
session, adapted from Taylor (1989b: 202):

a. the possessor is a specific human being …;

b. the possessed is a specific concrete thing …;
 Silvia Luraghi

c. the relation is an exclusive one, i.e for each thing possessed there is only one
possessor;
d. the possessor has the right to make use of the possessed; …
e. …
f. the possessor is responsible for the possessed; …
g. … possessor and possessed need to be in close spatial proximity;
h. the relation of possession is a long-term one…

It has been observed by Heine (1997a, b) that the “Locative Schema” is the most
frequent way of expressing possession cross-linguistically. By this schema, pos-
sessors are metaphorically conceived as locations, and the locative relation is
conceived as implying some sort of control over an entity (a possessor controls a
possessee as a consequence of features such as [d] and [f] above). This complex
relation can be explained by means of two underlying metaphors: Possessors are
Places and Physical Vicinity is Control.
Remarkably, Heine also distinguishes a “Locative Schema” from a “Goal
Schema.”8 Among instantiations of the latter, he includes languages in which pos-
sessor is coded through the dative case, and writes: “As a source for predicative
possession, this schema typically consists of a verb of existence or of location,
where the possessor is encoded as a dative/benefactive or goal case expression and
the possessee typically is a subject. Since dative/benefactive markers are frequently
derived from allative/directional markers, the latter functions may also be part of
the case marking figuring in the Goal Schema” (1997b: 95). Based on the examples
provided in the book, however, there is no evidence for a real Goal Schema sepa-
rate from the Locative Schema. Consider the following examples, given by Heine
(1997b: 95–96; glosses by the author) as instantiations of the Goal Schema:
(4) ur velo c’hlas am eus
a bike blue to: me is
‘I have a blue bike.’ Breton (Celtic, Indo-European)
(5) waska tiya- puwan
rope exist- for. me
‘I have a rope.’ Bolivian Quechua (Quechuan, Andean)
(6) Le livre est à moi
the book is to me
‘The book belongs to me.’ French (Romance, Indo-European)

. Further schemas described by Heine (1997a) are the Source Schema, on which see below
Section 2.2.3, the Companion Schema, which I discuss in Section 3.5, and the Topic Schema.
 Plotting diachronic semantic maps 

Remarkably, in all the above occurrences we find verbs of rest (the verb ‘be’ or
‘exist’), so there is no real evidence for the marker to express direction: this could
only be the case if motion verbs were used. Indeed, frequent polysemy of location
and direction markers accounts for the double nature of many cases/adpositions
that can accordingly indicate both semantic roles. Thus, evidence for a Goal Schema
should be sought in languages which present a rigidly tripartite system of local rela-
tions. Remarkably, as we will see below, one such language, Finnish, only provides
evidence for a connection of possessor with location, rather than with direction.

2.2.3 Comitative and the domain of spatial relations

Comitative markers display a number of frequent polysemies that I discuss further
on in this paper, but they indicate spatial relations only infrequently. This is not to
say that space cannot function as a source domain for comitative: indeed, in several
cases in which the origin of comitative markers can be detected, location seems to
provide the source for the comitative. Unfortunately, evidence for this historical
development is limited, and mostly provided by Indo-European languages. Exam-
ples are various comitative prepositions in the Germanic languages, such as English
with, from O.E. wið ‘against, opposite to’ and German mit, from Proto-Germanic
*medi-, cognate of English middle, Greek me, from Ancient Greek metá, original
meaning ‘among’ (see Luraghi 2001b and 2005c on this development), Catalan amb
from Latin apud ‘(near)by’. In some cases, the spatial origin of comitatives cannot
be traced even over several millennia of attested history: Italian and Spanish con,
for example, are the outcome of Latin cum ‘with’, a preposition which has always
indicated the semantic role of comitative and preserves no traces of spatial mean-
ing even in the earliest sources (Leumann, Hofmann & Szantyr 1965: 260). Only
comparative evidence, which connects Latin cum with Oscan ku ‘(near)by’, allows
one to reconstruct the original spatial meaning of the preposition.
Apparently, this is another instance in which a newly acquired meaning
pushes out the original meaning, as in the case of the extension from source to
location. I discuss some possible motivation for this in Section 4. It must of course
be mentioned that sources for comitatives are not limited to spatial relations. As
shown in Heine and Kuteva (2002: 91), comitatives often derive from lexical items
that do not have spatial reference, as in the case of the Estonian ending -ga of
the comitative case, from Balto-Finnic *kansa ‘people’, or of various comitative
markers that derive from serialized verbs with meaning such as ‘follow’, ‘take’, etc.
(see also Stolz et al. 2006). In addition, comitatives may arise from adverbs mean-
ing ‘together’, which, in turn, often do not originate out of spatial concepts. Such
adverbs often reinforce comitative markers, and a possible development is for
them to take over and substitute the latter, in a sort of cycle described in Stolz et al.
(2006: 363). Evidence for their frequent non-spatial origin is extensive; suffice it
 Silvia Luraghi

to mention English together based on the root of the verb gather, German zusam-
men, based on the Germanic source for the word same, which eventually goes back
to the PIE numeral for ‘one’ (*sem, cf. Ancient Greek hén ‘one(neuter)’), French
ensemble from Latin insimul ‘at the same time’. The fact that sources for comitatives
are available outside the spatial domain may be part of the answer to the ques-
tion why the polysemy of comitative and spatial notions is limited. However, the
relevance of non-spatial sources for comitatives must not be overstated. Markers
of other semantic roles may also have non-spatial origin: recipient markers for
example may arise from the verb ‘give’. But still, when direction markers extend to
recipient, they also retain their original spatial meaning, as in the case of English
to (direction and recipient).
This peculiarity regarding comitative might owe to the complexity of local
relations involving human landmarks. It has been noted by Creissel and Mounole
(2011) that human beings are not good landmarks of spatial relations, as they are
highly mobile entities. As a consequence, location with respect to a human land-
mark often really indicates location in his/her habitual space, as with French chez
‘at’ (only with human landmarks):
(7) Je suis chez mon frère (mais il n’ est pas lá).
I am at my brother    but he not is not there
‘I am at my brother’s (place), but he’s not there.’

Compare an inanimate landmark:

(8) Je suis à la gare (*mais elle [=la gare] n’ est pas lá).
I am at the station    but she [  the station] not is not there
‘I am at the station (*but it’s not there).’

On the other hand, the comitative generally indicates that the accompanee is present:
(9) Je suis avec mon frère (*mais il n’ est pas la).
I am with my brother (*but he not is not there)
‘I am with brother (*but he’s not there).’

Thus one can set up the following stages for the extension from locative to comita-
tive markers:
(10) (generic locative –-> specialized locative for human landmarks –->)
comitative

where the parentheses indicate that the former two meanings stop being active
as soon as the markers are extended to the comitative role. Note that there is no
need to set up any special metaphor here: a specialized locative marker for human
landmarks may come to be understood as comitative simply through pragmatic
inference (if a person is located by another person, then they are also together).
 Plotting diachronic semantic maps 

2.2.4 A spatial metaphor or an extension from other human roles?

A third semantic role typical of human beings that is often conceptualized in terms
of location is the experiencer. As well known, this semantic role can be coded in a
variety of ways. In particular, it is often coded as the agent (as in John loves Mary)
or the patient role (as in The thunder scared him). Perhaps the most frequent asso-
ciation of the experiencer role with a certain coding device cross-linguistically
is the dative (English is quite untypical in not following this tendency). Dative
experiencers are reported from Indo-European, Afro-Asiatic, Tibeto-Burman,
and Niger-Congo languages, among others. Below is an example from Hungarian
(Uralic) of dative encoding of the experiencer (see Kenesei et al. 1997):
(11) Ez tetszik Péternek.
this appeal:3sg Peter:dat
‘This appeals to Peter, Peter likes this.’

Indeed, experiencers have something in common with agents, patients, and recipi-
ents, as summarized in Table 3:

Table 3. Features of semantic roles

experiencer agent recipient patient

humanness + + + -
control ± + + -
change of state ± - - +
volitionality ± + - -
(adapted from Luraghi 2010b)

Experiencers are typically human: in this respect, they are similar to agents, as
well as to recipients and beneficiaries. Contrary to agents, experiencers do not act
voluntarily: in this respect, experiencers are similar to patients. However, there are
crucial differences between prototypical experiencers and prototypical patients: in
the first place, as already remarked, experiencers are human, because, by defini-
tion, they must be sentient beings, while prototypical patients are not. In addition,
prototypical patients necessarily undergo a change of state, while experiencers
may or may not do so. Situations in which experiencers occur are prototypically
states (see Van Valin, LaPolla 1997: 85; Croft 2001: 55–156), however, experiential
predicates can also be inchoative, thus implying a change of state by which the
experiencer enters the situation (as in e.g. Mary fell ill).
For what concerns us here, the fact that experiencers are often static partici-
pants is most interesting, because it makes them good candidates for a metaphor
that conceives of human beings as locations/containers, such as Experiencers
 Silvia Luraghi

are Places (or Containers) for Feelings/Sensations (based on another met-

aphor: Feelings/Sensations are Things). Hence coding through the dative case
might be explained by referring to its link with the locative in the source domain.
On the other hand, coding through the dative may owe to similarities between the
roles of experiencer and recipient: as discussed above, experiencers share relevant
features of recipients, as they are human; in inchoative situations they may be con-
ceived as recipients of sensations or emotions. This similarity can be accounted
for in terms of a metaphor according to which Experiencers are Recipients
of Feelings/Sensations (again based on: Feelings/Sensations are Things).
Moreover, even though neither experiencers nor recipients are real controllers in
the sense agents are, the fact that they are necessarily human implies a certain
degree of control, at least inasmuch as they are consciously involved in situations:
I have indicated this feature of the experiencer role with ± in the control column.9
Evidence for both possible interpretations (i.e. experiencers as locations or
as recipients) comes from several Finnic languages. In Estonian, for example,
experiencer can be encoded through the adessive, a spatial case which can also
encode location; the same, though to a more limited extent, is true of Finnish
(­Examples (12)–(14) are from Erelt & Metslang 2006 and Huumo 1996: 229):
(12) Mul on häbi / piinlik
I:adess be:3sg shame:nom / embarrassing:nom
‘I am ashamed/I feel embarrassed.’
(13) Mul on vaja töötada
I:adess be:3sg need work:inf
‘I need to do some work.’
(14) Hänellä on hyvä maku
s/he:adess be:3sg good taste
‘S/he has good taste.’

. There is an extensive literature concerning the notion of control as related to experi-

encers. In principle, experiencers should not be controllers as they simply experience a situ-
ation. However, the construal of experience typically allows for varying degrees of control.
Depending on the type of experiential predicate, the experiencer can be conceived as more or
less actively involved in the situation, hence more or less capable of exerting control. Gener-
ally speaking, control is absent in case of bodily sensations, whereas perception and volition
are most often conceived as controlled activities. Emotions can be construed in a variety of
different ways even within the same language, as shown by possible usage with imperatives
in many languages. See Verhoeven (2008: 41–51) for discussion and further reference, and
Luraghi & Sausa (forthc.). As for recipients, the transfer event in which they are involved
implies that the entity transferred is in their possession, hence that they can control it at the
end of the transfer.
 Plotting diachronic semantic maps 

Note however that encoding of the experiencer with the adessive can be due to
the extension of the Possessor Schema, as the adessive encodes possessor in these
languages (see below): such an extension is cross-linguistically quite frequent (cf.
French J’ai faim, German Ich habe Hunger, both meaning ‘I’m hungry’, lit. ‘I have
hunger’; similar usages are reposted from various genetically unrelated languages,
such as Kannada cf. Amritavalli 2004, Tibeto-Burman languages cf. Bickel 2004;
Thai cf. Matisoff 1986; Niger-Congo languages cf. Heine 1997a). Thus, it seems
more likely that, rather than depend on a putative metaphor by which Experienc-
ers are Places, this type of coding is explained through two other metaphors
according to which Experiencers are Possessors of Sensations and Sensa-
tions are Things Possessed.
Coding of an experiencer through the allative is also frequent both in ­Estonian
and in Finnish, as shown in (15) and (16):
(15) Anne-le meeldi-b džäss
Anne-all like-3sg jazz.nom
‘Anne likes Jazz.’ (Estonian, from Erelt & Metslang 2006: 255)
(16) Minulla on flunssa
I:all be:3sg flu.nom
‘I got a flu.’ (T. Huumo, p.c.)

Again, one could think of a metaphor that connects the domain of space directly
with the domain of experience,10 but, in light of the common features of experi-
encers and recipients discussed above (and of the widespread tendency for dative
markers to encode recipient cross-linguistically), it looks plausible that experiencers
are re-interpreted as recipients. This does not necessarily imply that the metaphors
indicated above for such polysemy are always activated. Rather, the semantic exten-
sion is based on a gestalt effect, in the sense of Lakoff (1977), whereby c­ ommon fea-
tures of both roles, i.e. animacy and sentience without agency, are magnified. Thus,
even though recipient markers ultimately go back to direction markers, there is no
need to necessarily connect the semantic role of direction with that of experiencer.
I return to the coding of the experiencer role below, S­ ection 3.7.

2.2.5 Competing metaphors: The case of possessor

Location is not the only spatial relation to provide a source for possessor cod-
ing. As well known, possessors are frequently coded through the genitive case or

. For example, one might argue that “the allative is used if the verb indicates directionality
towards the experiencer” (T. Huumo, p.c.); however, the verb ‘be’ is pretty frequent in such
constructions.
 Silvia Luraghi

adpositions with similar meanings, and the typical source for genitives is consti-
tuted by ablative markers. This amounts to saying that source also provides a fre-
quent source domain for possessor: this well-known and frequently investigated
development is based on the metaphor Possessors are Origins (Nikiforidou
1991: 174).
The extension from source to possessor is documented in numerous lan-
guages: virtually all the Indo-European languages that have lost case marking
at least partially attest to this development. When a marker of source extends
to possessor and becomes a genitive marker, it often retains its original spa-
tial meaning, but it may also lose it: evidence for both developments is easily
available from German (von means both ‘from’ in spatial sense and ‘of ’) and
English (of is limited to the expression of genitive relations and retains trace of
the original separative meaning only when governed by certain verbs or adjec-
tives, as for example in independent of). More examples are discussed by Heine
and Kuteva (2002), who caution (p. 35) that “most of these examples relate to
Indo-European languages; more research is required on the genetic and areal
distribution of this process.”

source location direction source domain:space

path

Possessors Possessors
are Origins are Locations
Target domain: Possession

Figure 4. The mapping of spatial relations onto possessor

Thus, in the case of possessor two competing metaphors offer the chan-
nel for the mapping of the source domain onto the target domain, as shown in
­Figure 4. That different metaphors can provide the source for a certain concept
is not surprising: as is well known, for example, agents can originate from vari-
ous spatial concepts as well as from instruments. Remarkably, the two types of
spatial concepts involved here, source and location, are encoded through markers
that generally do not allow polysemy between each other. Thus, the two types of
metaphor account for possessors that are always distinct from the point of view
of encoding.
 Plotting diachronic semantic maps 

2.2.6 Recipient and beneficiary

Thus far, I have mentioned recipient and beneficiary as a pair. Indeed, polysemy
of these two semantic roles is frequent cross-linguistically (cf. Kittilä & Zúñiga
2010: 18 among others). Typical examples involve the dative case of various Indo-
European languages as well as of languages of other families, cases partly covering
the same functions as a dative, as e.g. the Finnish allative, or adpositions that also
cover functions typical of the dative. Examples (17) and (18) are from Finnish
(allative case, from Kittilä & Zúñiga 2010), while (19) and (20) are from Biblical
Hebrew (preposition l- ‘to’, also used in direction expressions):
(17) Vanhempi antoi lapselle lahjan
parent:nom give-3sg.pst child:all present:acc
‘The parent gave the child a present.’
(18) Vanhempi leipoi kakun lapselle.
parent:nom bake-3sg.pst cake-acc child:all
‘The parent baked the child a cake.’
(19) wa- ttittēn gam lǝ- ‘īšāh
and she.gave also to-man.her
‘And she gave [the fruit] to her man too.’ (Gen. 3.6)
(20) kī YHWH ‘ĕlohêḵem hū’ hannilḥām l-āḵem
because Lord god.your he he.fights for-you
‘Because the Lord himself is fighting for you.’ (Deut. 3.22)

Whenever the diachronic development of such markers can be traced back in

time, it provides evidence for extension from beneficiary to recipient, as indicated
by Heine et al. (1991: 159). Evidence for this development is not abundant, per-
haps on account of the fact that several languages with a dative case display the
polysemy as far back as one can trance their history. In Luraghi (2010c) the follow-
ing development for the Ancient Greek preposition eis/es is described:

a. Homeric Greek: direction and purpose:

(21) prôta mèn es Púlon elthé
first ptc to Pylos:acc go:impt.aor.2sg
‘First go to Pylos.’ (direction; Hom. Od.1.284).
(22) hê me mál’ eis átēn koimḗsate
ptc 1sg.acc much to damage:acc put.to.sleep:aor.2pl
nēléï húpnōi
harmful:dat sleep:dat
‘You put me to sleep in order to damage me with harmful sleep.’
(purpose; Hom. Od.12.372)
 Silvia Luraghi

b. Classical Greek: direction, purpose and beneficiary:

(23) ho Kroîsos tò pân es autòn epepoiḗkee
art.nom Croesus:nom art.acc all:acc to 3sg.acc make:plpf.3sg
‘Croesus had done all that he could for him.’ (beneficiary; Hdt. 1.85.1)

In post-Classical Greek, eis slowly extended to recipient, and in Modern Greek, its
outcome s is the common means to indicate recipient (and other functions of the
dative case, which disappeared in Byzantine Greek):
(24) Edosa to vivlio ston Antone
I.gave the book to+the Anthony
‘I gave the book to Anthony.’ (recipient).

The Greek example also attests to the polysemy of direction and purpose
(­Examples (21) and (22)) as preceding the extension to beneficiary and recipient,
thus showing a direct mapping of the spatial domain on the domain of abstract
relations. This development, which is based on the metaphor Purposes are Des-
tinations, explored in Section 2.3.3 below, is apparently cross-linguistically fre-
quent; it contradicts the predictions in (1), by which extension to human roles
should always precede extension to inanimate and abstract roles.

2.2.7 Beneficiaries that are not recipients

In spite of the frequent polysemy outlined above, there is extensive evidence for
other sources for beneficiary markers, which do not display it. Remarkably, these
are markers whose original local function is not allative. In Finnish, beneficiary
can be coded through the postposition vuoksi, which also codes cause, as shown
in (25) and (26):
(25) Henkilo opettel-i suome-a yksilo-n vuoksi.
person:nom learn-3sg.pst Finnish-part individual-gen for
‘A person learnt Finnish for an individual.’ (beneficiary; from Kittilä 2010);

(26) Jaatelo sul-i sahkokatko-n vuoksi.

ice.cream:nom melt-3sg.pst power.failure-gen for
‘The ice cream melted because of the power failure.’ (cause; T. Huumo p.c.).

The local meaning of vuoksi is ‘through’, which may explain its extension to cause
(on the meaning and etymology of this postposition see Itkonen, Joki & Pelotla
1978: 1813–1814; 1818). Without going into the details of this development,
what is remarkable is that, as already highlighted above, this is not a marker
of direction. Similarly, the English preposition for and German für, which
can code beneficiary, cause and purpose, derive from Proto-Germanic *fura,
‘before’, cognate with Latin pro. Interestingly, the latter preposition, too, devel-
oped a ­beneficiary (and ­purpose) meaning out of the spatial meaning ‘before’
 Plotting diachronic semantic maps 

(Luraghi 2005a, 2010a). In the intermediate stage, attested in Early Latin, it indi-
cated an exchange, as in (27):
(27) an tibi malam rem vis pro male
ptc you:dat bad:acc thing:acc want:prs.2sg for bad(adv)
dictis dari?
say:ptcp.abl.pl give:inf.p
‘Do you wish a punishment to be given you for your insult?’ (Pl. Men. 496).
Sporadically, the preposition could indicate beneficiary, especially of the behalf
type, already in Early Latin, as in (28):
(28) ego ibo pro te, si tibi non libet
1sg.nom go:fut.1sg for 2sg.abl if 2sg.dat neg like:prs.3sg
‘I’ll go for you, if you don’t feel like.’ (Pl. Mos. 1130).
Later, the meaning extended to all types of beneficiary, as well as to purpose, a
meaning which was already implicit in the notion of exchange. Still later, the
meaning also extended to cause, as shown in (29), from the New Testament:
(29) quia et Christus semel pro peccatis mortuus est
because even Christ:nom once for sin:abl.pl die:pf.3sg
iustus pro iniustis
righteous:nom for unrighteous:abl.pl
‘Because Christ also suffered for sins once, the righteous for the

unrighteous.’ (1 Pet. 3.18).
Note that the second occurrence of pro in (29) could still have a behalf interpreta-
tion, thus pointing to the original meaning of substitution.
A partly similar development has apparently taken place in the case of the
Ancient Greek preposition hupér ‘over’, which, beside having spatial meaning,
developed into a beneficiary marker. In much the same way as with Latin pro, this
extension started with the behalf type of beneficiary. Similar to Germanic for and
Finnish vuoksi, Latin pro and Greek hupér did not code direction. In addition,
none of them underwent a semantic extension to also include the semantic role
recipient. Another case of a beneficiary marker that does not code recipient and
whose spatial origin is not connected with an allative sense is the Turkish postpo-
sition için, from iç ‘inside’, which I discuss below, Section 2.3.3.11

. Note that in studies regarding sources for beneficiaries, usually such types of semantic
extensions are ignored, or just mentioned in passing, while there is a general tendency to only
stress the relatedness of beneficiary and direction. In this vein, Schmidtke-Bode (2010: 127–128)
remarks that there is cross-linguistic evidence for beneficiary (and purpose) markers that do
not indicate directional motion, but then does not pursue this matter further in his discussion.
See Luraghi (2013) for discussion.
 Silvia Luraghi

2.2.8 Competing metaphors: Beneficiary

From the evidence in Section 2.2.7, one must conclude that the relation between
recipient and beneficiary, though well documented and cross-linguistically very fre-
quent, is not such that the two roles cannot be coded independently from one another.
Apparently, we are dealing with different metaphors based on the source domain of
spatial relations. The first one, which accounts for polysemy of recipient and benefi-
ciary, involves the allative relation; the second, which is not active for recipient but
only for beneficiary, involves a more complex locative or perlative relation. Only in
the former case does polysemy of beneficiary and recipient seem to arise naturally.
This may indicate that, even if beneficiary precedes recipient in semantic
extension, the spatial meaning, i.e. direction, is also active when a specific marker
extends to recipient. This conclusion finds further support in the fact that markers
of recipient/beneficiary for which extension from an original spatial meaning is
historically documented also retain the allative meaning (as for example French à,
English to, the Finnish allative case and many other).
The metaphor that maps direction onto beneficiary depicts the latter as the
endpoint of a prospective trajectory: it could be stated as A Beneficiary is a
Destination. However, it is usually stated in terms of similarity between destina-
tions and recipients: Recipients are Destinations (see e.g. Rice & Kabata 2007).
Indeed, the transfer is based on a feature that the beneficiary role has in common
with the recipient role, and that indeed is even more relevant for the latter than for
the former. Beneficiaries typically benefit from a certain state of affairs, and often
they are the prospective recipients of an entity, even though they might not actu-
ally get hold of it in the end. Consider Example (30):
(30) Mary baked a cake for Paul, but Paula ate it up before he could even see it.

On the other hand, a recipient can be conceived as a special case of beneficiary,

who profits by an event by actually getting hold of an entity as its result (Kittilä
2005; Luraghi 2010c), as shown in (31):
(31) Mary gave a cake to Paul (???but he did not receive it).

Indeed, in spite of the fact that in the prototypical beneficiary situation there
may be no actual transfer (Goldberg 1995: 37), in the default interpretation it is
assumed that the intention of the beneficient is to actually transfer the benefac-
tum to the beneficiary (see Luraghi 2010c). In this sense, the polysemy of recipi-
ent and beneficiary can be described as a predicational metonymy in the sense
of Panther and Thornburg (2007: 246), whereby an intended action is taken as an
actual action. See further the polysemy of possessor and recipient/beneficiary,
Section 3.2.
 Plotting diachronic semantic maps 

Other metaphors available for beneficiary are not based on directional motion.
Just a glance at the adpositions mentioned above indicates that there can be more
types of conceptualization at work; one, which accounts for the extension of Latin
pro ‘in front of ’ and Greek hupér ‘over’ to behalf beneficiary and then to all types of
beneficiary involves a covering relation. Both prepositions can indicate that a tra-
jector is placed in such a position with respect to a landmark to cover it from the
perspective of a possible observer, and thus replaces the landmark, as shown sche-
matically in Figure 5. The relation of replacement is accounted for by the metaphor
which equates existence with being in the observer’s visual field, and can be stated
as follows: existence is visibility (Lakoff et al. 1991). Coverage and replacement
induce the idea of exchange and hence of surrogation, which leads to a behalf ben-
eficiary (for further reference see Luraghi 2005a, 2010c, 2013a).

observer

Tr Lm

Figure 5. Covering relation

2.2.9 A Balto-Finnic perspective on the Indo-European dative

The Indo-European dative, a case which is often described as typical of human
roles, provides little evidence for its ultimate spatial origin. In the earliest sources
of the ancient Indo-European languages, spatial usage of the dative is restricted: a
few occurrences that might indicate an allative meaning are available from Latin,
but it must be noted that in several languages the dative tends to merge with the
locative, as in Greek and Anatolian (see Luraghi 2001b and the references therein).
This might very well be the origin of the dative case: it was observed by Kuryłowicz
(1964: 190–193) that the endings of the dative and the locative singular look very
much like morphophonologically conditioned variants of the same morpheme.
Evidence for the relatedness of datives and locatives, as well as for the close asso-
ciation between dative and animacy in other language families has been provided
in Rodriguez Aristar (1996), based on data from several Australian and Amerin-
dian languages.
A look at the Finnish cases that correspond to various functions of the Indo-
European dative is enlightening. In the Indo-European languages, the dative typi-
cally encodes recipient, beneficiary, possessor, maleficiary, among other roles. The
recipient construction is usually extended to third argument of verbs that ­indicate
 Silvia Luraghi

separation, especially in the case that they have human referents and indicate
maleficiary, as shown in (32) and (33):12
(32) Mi è sparita la borsetta
me:dat is disappeared the purse
‘My purse disappeared.’ lit.: ‘The purse disappeared on me (maleficiary).’
(33) Ten pán nám vzal hodně peněz
dem man us:dat took much money
‘That man took a lot of money from us (maleficiary).’ (Czech, from
Janda 1993: 58)
In Finnish, these roles are coded by three different but related cases, that is, the
adessive (possessor), the allative (recipient) and the ablative (which codes source,
especially with human referents). Examples are (34)–(36) (from Huumo 1996: 74):
(34) Liisalla on kirja
Lisa:adess be:3sg book:nom
‘Lisa (possessor) has a book.’
(35) Liisalle tuli kirje
Lisa:all come:3sg.pst letter:nom
‘Lisa (recipient) received a letter.’
(36) Liisalta katosi kukkaro
Lisa:abl disappear:3sg.pst purse:nom
‘Lisa (maleficiary) lost her purse.’
Remarkably, these three cases are considered local cases more on account of their
names than of their usage: as argued in Huumo (1996), they are mostly used
for human relations, as shown in the examples, while local relations are more
­frequently expressed by the inessive, illative and elative. The Finnish data point
toward a ­difference in the conceptualization of possessor and recipient/­beneficiary/
maleficiary, based on different parts of the domain of space as the source domain.
One might wonder whether polysemy of these two roles is connected with poly-
semy of location and direction across languages that present it. This is certainly the
case in the Indo-European languages.13

. Such usage of the dative, which seems to be most typical of the Indo-European languages
of Europe except English (Haspelmath 1999), is usually described as an external possessor
construction, and often occurs with inalienably possessed entities, even though this is not nec-
essarily the case: consider e.g. the following Italian example: Mi hanno rubato la macchina di
Giovanni ‘Someone stole John’s car from me.’ (lit.: ‘They stole John’s car on me.’). The dative in-
dicates maleficiary, and the possessive relation is clearly temporary. See further Havers (1911).
and Luraghi (forthc.)
. It is outside the scope of the present paper to provide a detailed description that can
support this claim; however, some remarks can illustrate it. Even if a distinction between
 Plotting diachronic semantic maps 

2.2.10 Path as a source for human relations: Intermediary

In Section 2.2.1, I have pointed to a lesser basicness of path among spatial seman-
tic roles. This is not to say that path cannot function as an independent concept in
the source domain of space for metaphors leading to other target domains, includ-
ing that of human relations. In particular, a frequent extension of path markers is
the role of intermediary, as in the case of English through in (37):
They have to speak through an interpreter to be able to communicate
(37) 
effectively.

An intermediary can be viewed as a human instrument. Instrument markers, how-

ever, do not easily encode human instruments: a quick cross-linguistic survey indi-
cates that such markers usually encode different relations when they occur with
human nouns, which prevent them from being interpreted as encoding instru-
ment. Let us consider, for example, languages in which instrument is encoded
through the Companion Metaphor, such as English and many other European
languages. In these languages, the instrumental marker encodes the comitative
with animate nouns, as English with: thus, if a human NP co-occurs with with the
default interpretation is a comitative. Indeed, in (38):
I achieved X with Y [=human noun]/I achieved X with Z [=inanimate
(38) 
concrete noun]

the interpretation of the role taken by the NPs with Y and with Z as comitative
and instrument respectively depends crucially on the lexical meaning of the NPs
involved.
In languages that do not conform to the Companion Metaphor, and conse-
quently do not encode instrument through the comitative marker, on the other
hand, there is a frequent tendency for instrument markers to extend to agent (see
Section 2.3.2 below). Thus again, there appears to exist a default interpretation that

allative and locative markers often exists, a certain degree of polysemy of direction and Loca-
tion is also normally present. In particular, Indo-European languages with a separate locative
case, such as Sanskrit, display constructions in which direction is coded by the locative (see
Delbrück 1867). In addition, the typical situation in which spatial relations are coded through
adpositions with case variation, as in the case of German ‘two-way’ preposition, most often
features a number of prepositions that can code location and direction with different cases,
while source is usually kept distinct by the use of different adpositions. In languages where
such alternation occurs (such as Latin or many Slavic languages, for example), one also finds
adpositions that do not allow for case variation and, if they are polysemic, their meaning
includes location and direction. As for the dative case, as already noted above, it seems to
present more affinities with the locative than with the allative (which is most likely the orig-
inal meaning of the Indo-European accusative).
 Silvia Luraghi

makes it impossible to use the instrumental case (or analogous morphemes) with
such non-prototypical instruments as human beings.
The same seems to happen in the case of the frequent instrument/loca-
tion polysemy: the instrument marker usually has another favored interpreta-
tion when occurring with a human referent. The Finnish adessive, for example,
indicates instrument (or manner) with inanimate nouns and possessor with ani-
mates (Huumo 1996: 76–77). The West Semitic instrumental/locative preposition
b- occurs with human referents mostly in the plural in the meaning ‘among’, as in
Example (39) from Biblical Hebrew (see further Pennacchietti 1974: 175):
(39) hay-yāfā b-annāšīm
the-beautiful in-women
‘Beautiful among women.’ Ct 1.8.

In sum, prototypical instrument markers appear to be exploited for different pur-

poses when they occur with human nouns, and have default interpretations that
rule out human instrument. This comes as no surprise: the fact that a human being
is used as an instrument is comparatively rare and unexpected; human beings take
roles such as comitative, agent or possessor much more frequently. In addition,
an intermediary is not simply an instrument: even if the intermediary acts under
someone else’s instigation, s/he holds control on the event. In states of affairs that
feature intermediaries, one can conceive of agency as being split between the
primary agent, who initiates the state of affairs voluntarily and the intermediary
who holds the responsibility for the actual carrying out of the action (see Luraghi
1995). Hence the intermediary functions as the channel through which the agent
achieves his or her purposes. The notion of path accounts for this conceptualiza-
tion based on the metaphors An Intermediary is a Channel and Agency is a
Thing Transferred.
The literature on intermediary is not especially rich, and there are no specific
studies that investigate sources and polysemies that involve this semantic role.
At least in the Indo-European languages, there seems to exist a frequent poly-
semy of path, intermediary and means, as in English through. Indeed, means is
defined as the role taken by entities used by an agent to achieve his/her goals, with
a lesser degree of manipulation than instrument.14 Typical of means e­ xpressions

. There are no specific studies devoted to this semantic role, which can also be conceived
as a non-prototypical type of instrument. Available definitions are not especially enlightening.
Radden (1989: 442–443), for example, gives the following definition of means: “The means
role denotes physical or abstract entities which, without being instruments themselves, are
essentially involved in bringing about the effects of an agent’s action.” He then concedes that
“the distinction between the notions of means and instrument is not always easy to draw,”
 Plotting diachronic semantic maps 

are abstract nouns, which cannot be manipulated, and consequently can be con-
trolled by agents to a lesser degree than prototypical instruments. This feature
makes means similar to intermediary: as a human being, an intermediary cannot
simply be used and controlled completely by an agent. There are no diachronic
studies regarding the relation between intermediary and means. In a study of the
Ancient Greek preposition diá with the genitive, Luraghi (1989) found that exten-
sion from path to intermediary preceded extension to instrument and means, but
more evidence is needed in this respect. Remarkably, in Ancient Greek the distinc-
tion between instrument and means is not reflected in coding, so the extension
from intermediary involves both roles. In any case, the use of markers specifically
connected with the semantic role means to prototypical instruments seems to be
common (see below, Section 2.3.1).
In several languages, path markers also extend to agent. Possibly, this fur-
ther extension needs an intermediate stage at which perlative morphemes encode
intermediary. An attested example which might provide evidence is French par,
which encodes path and passive agent. In Latin, the preposition per could encode
path and intermediary, but it could be understood as indicating an agent only
under specific semantic constraints, even with passive verbs (Luraghi 2010a).

2.2.11 Interim summary

In Sections 2.2–2.2.10 I have surveyed human relations, and shown how they
are conceptualized in terms of metaphors based on spatial relations. Patterns of
polysemy of semantic roles markers (cases and adpositions) recurrently attested
cross-linguistically provide evidence for these metaphors. In particular, I have
shown that the basic spatial relations of source and direction frequently provide
the source domain for agent and recipient/beneficiary respectively. Location may
provide the source for possessor, but this semantic role can also be encoded as
a source: two competing metaphors are at play here. Similarly, beneficiary can
merge with recipient, and be conceptualized and encoded as a direction, but in
several languages there are alternative means of encoding this role, which point on
the spatial relations of location or path as its conceptual source. Experiencers are
involved in static situations, and can be encoded through locative markers; in cases
of inchoative situations, they can be encoded through allative markers. The exten-

and discusses occurrences in which the same NP, money, can be conceived as an instrument
or as a means in the same contexts. A more elaborated definition of means can be found in
Croft (1991), who, however, focuses on means-clauses (e.g. By doing X somebody achieved Y).
Croft further indicates that the occurrence of a means implies the existence of an agent who
acts intentionally: “the means clause must begin with a VOL[itional] arc – that is, it must be a
volitional action.” (1991: 178).
 Silvia Luraghi

sion of markers of spatial relations to this semantic role does not necessarily point
toward a direct mapping of space on the domain of experience, but it can also be
that experiencers are metaphorically conceived as possessors (which in turn can
be conceived as locations), or as recipients (which are conceived as directions).
Location can also provide the source domain for the comitative role, but this type
of polysemy is synchronically rare: apparently, location markers that develop into
comitative markers lose their original meaning when they acquire the new one.
Finally, the spatial relation of path often provides the source for intermediary.

2.3 Non-human relations

According to the scale in Figure 1, extension to non-human relations should
follow extension to human relations. I have already discussed the connection
between space and time; more discussion on this hypothesis can be found in Nar-
rog (this volume). In what follows, I survey frequent polysemies and metaphors
that account for the extension to non-human relations. As we will see, in some
cases the domain of space is mapped directly onto the domain of non-human rela-
tions, while in other the direction of semantic extension contradicts the scale.

2.3.1 Source domains for instrument

As has already been noted in this paper, the Companion metaphor, according to
which An Instrument is a Companion, accounts for the extension of comitative
markers to instrument. Languages which rely on the Companion Metaphor can
be said to be well-behaved with respect to the scale of increasing grammaticaliza-
tion in Figure 1, since they display the semantic extension location > comitative
> instrument, modeled according to (1). As this metaphor has been exhaustively
described in Lakoff and Johnson (1980), I will not dwell longer on it.
Another frequent source for instrument markers is provides by locatives. In
such cases, even if often human roles are also associated with the same morphemes,
there seems to be a direct connection between space and instrumentality. There
are several reasons why locative relations can be understood as instrumental. For
example, certain types of instruments are also places in or on which human beings
can be located: the obvious example for this is constituted by means of locomotion
(see Lehmann, Shin 2005 with several examples from different language and lan-
guage families). Some instruments are shaped as containers, and other as means of
support: this fact accounts for the extension of different locative markers to instru-
ment, based on various transfers and extensions. As both containers and means
of support are at least in part more or less prototypical instruments (for examples,
a cup can be conceived as an instrument for drinking, a bottle as an instrument
for carrying liquids, etc.), it is perhaps not especially useful to multiply underlying
 Plotting diachronic semantic maps 

metaphors.15 Rather, there seems to be a sort of analogical extension based on a

generalization, which can be stated as in (40):
(40) some containers/supports are instruments → all instruments are
encoded as instantiating a containment/support relation.

This generalization consists in the extraction of a schema (see Langacker 1987),

whereby certain types of instrument have certain spatial dimensions (container,
support). The schema is then analogically extended to all types of instrument (see
Tuggy 2007: 100–101 on schematicity and analogy in word formation). Examples
are provided by Balto-Finnic (Grünthal 2003; Huumo 1996), Semitic, several
Australian languages (which however only provide examples of polysemy, but no
historical evidence for semantic extension and its direction); some of them are
surveyed in Luraghi (2001a).
Interestingly, the portion of the conceptual domain of space that provides a
source for instrument is the same, both in the case that the mapping is mediated
by the Companion Metaphor, and in the case that it is direct: it is the space of
location, i.e. of static relations. Indeed, instrumentality is part of what Lehmann
and Shin (2005) call “the domain of concomitance”, a notion originally proposed
by Coseriu (1970), who describes the meaning of a possible German construction
mit C as indicating “und C is dabei”, or “unter Dabeisein von C”. In other words,
a concomitant is a participant which is crucially indicated by its presence during
a state of affairs, that is, by a static relation. Luraghi (2001a) suggests that both
location and instrument should be grouped together as concomitant roles, that
is, roles that are not oriented either from the starting point of the event or toward
its endpoint.
As already mentioned in Section 2.2.10, path markers can extend to instru-
ment, possibly as a further step in an extension that goes from path to interme-
diary to means and then also reaches prototypical instrument. However, it is by
no means clear that the extention of path markers to the coding of means and
instrument is necessarily mediated by a human role, i.e. intermediary. As this
type of extension has never really been investigated on a large language sample,
I will assume for the time being that, similar to the Companion Metaphor, this
type of extension is also well-behaved, and procedes from path to intermediary
to means and instrument, based on the metaphor An Instrument/Means is
an Intermediary. Note however that this assumption is based in its turn on
another assumption, that is, that human roles must precede non-human ones,

. The polysemy of locative and instrumental morphemes is also frequent in word forma-
tion, see Luján and Ruiz (this volume).
 Silvia Luraghi

thus partly resulting in c­ ircularity.16 Remarkably, path is located close to location

in the domain of space, between source and direction.
To sum up, in this section I surveyed three types of semantic extension
involved for the conceptualization of instrument, two based on metaphors (the
Companion Metaphor and the Channel Metaphor), and one on analogy. By the
latter type of extension, markers of spatial relations are used for certain instru-
ments based on their shape, and then extended through analogy to all types
of instrument independent thereof. This means that the extension proceeds
directly from the domain of space to instrumentality. In the case of the two
metaphors, extension is mediated by human relations, comitative and possibly
intermediary. In all such extensions, the source domain is located in an inter-
mediate position between source and Direction, and points to an analogous
position of instrument in the target domain of causation, between antecedent
and subsequent roles in the terminology of Croft (1991). Croft’s Causal Chain
is represented in Figure 6.

Antecedent Subsequent

cause result
SUBJECT means OBJECT

manner

passive instrument benefactive

agent comitative maleficiary/ (recipient)

Figure 6. The causal chain (Croft 1991: 185)

According to Croft, who divides causal semantic roles into two groups, instru-
ment is an antecedent role. In Luraghi (2001a), it is suggested that semantic roles
should better be divided into three groups based on the type of involvement in
causation, as comitative, instrument, manner and means are better understood
as concomitant rather than antecedent. Spatial sources for ­instrument ­confirm

. Alternatively, if it turned out that extension of path markers to instrument/means is not
always preceded by extension to intermediary, one should assume that the domain of space
can provide the immediate source for the non-human roles of instrument and means. One
should set up another metaphor, by which an instrument (or means) is conceived as a channel
that conveys agency (i.e. volitionality, control, and energy transfer). An alternative version of
the Channel Metaphor could accounts for extension from space without passing through the
stage of intermediary: An Instrument/Means is a Channel (and, as in the case of the inter-
mediary, Agency is a Thing Transferred).
 Plotting diachronic semantic maps 

its consistent tendency to arise from the area of non-directional relations (loca-
tion) or at least of relations that do not indicate the starting or the end point of
motion (path).
As we will see in Section 2.3.3, polysemy can also involve instrument and
cause, that is an antecedent role: indeed, polysemies that cross-cut groups of
semantic roles are not infrequent, as discusses in the next sections.

2.3.2 Instrument as a source domain

A frequent extension of instrument is material: the material of which something is
made is understood as the instrument used to make it. Tentatively, I propose the
metaphor Material is an Instrument for Creating Objects to account for
this extension.17 This semantic role of material also has another frequent source
in the space domain, that is, source, following the metaphor Objects Come out
of Substance, explored in Lakoff and Johnson (1980: 73). Both patterns occur for
example in Turkish, as shown in (41) and (42) (from Kornfilt 1997):
(41) tugladan bir ev
brick:abl a house
‘A house of bricks.’
(42) bu ev tuglayla yapumıştır
this house brick:instr make:pst.ptcp.cop
‘This house is made with bricks.’

An often discussed extension of instrument is agent. This pattern of semantic

extension, whose direction is clearly attested at last in the Indo-European lan-
guages, contradicts Heine’s Scale of Increasing Grammaticalization in Figure 1, as
it implies that a non-human role serves as the basis for a human one. Polysemy of
instrument and agent is also frequent in several Australian languages, both erga-
tive and nominative/accusative (see Blake 1977; Dixon 2002). In such languages,
the instrumental case is most often also the locative (see Luraghi 2001b for discus-
sion). The original meaning of the Indo-European instrumental, instead, is usually
held to be comitative (see Delbrück 1867): to some extent and virtually limited to
Vedic Sanskrit, the extension to the agent role also brought about the infrequent
polysemy of agent and comitative, which I survey in Section 3.4.
It is not clear that extension of instrumental markers to the encoding of agent
can be explained in terms of metaphors. In Luraghi (2001b) it is suggested that

. Remarkably, in this case metonymy, rather than metaphor, could be at play, as material
and instrument are contiguous concepts in the conceptual domain of creation. I am not going
to pursue this explanation further here, but see below regarding extension of instrumental
markers to agent.
 Silvia Luraghi

this type if extension should rather be accounted for in terms of metonymy, as

agent and instrument are contiguous concepts, and an instrument can be seen
as a part of an agent. In addition, it is frequently the case that inanimate entities
substitute for animate ones in cases of metonymy, precisely when a part substitutes
the whole. The occurrence of a metonymic extension here would also explain why
this is virtually the only frequent semantic extension that contradicts Heine’s scale
in (1) by moving contrariwise.
Finally, polysemy of cause and instrument is also frequent, but directionality
is all but easy to gauge: inasmuch as instrumental cases also code cause, one can
assume an extension starting from instrument, but this matter has never been the
topic of any in-depth study, at least to my knowledge.18

2.3.3 Cause and purpose

Following a frequent conceptualization, Causes are Origins of Events
(­Nikiforidou 1991), and Purposes are Destinations (cf. e.g. Lakoff & Turner
1989). Instantiations of these metaphors mentioned in the literature operate by
mapping the domain of abstract relations directly onto the domain of space, with-
out necessarily implying an intermediate stage at which human relations provide a
link between the source and the target domain (see also Section 2.2.6).
The two metaphors map cause and purpose onto the most distant edges of the
conceptual domain of space, in very much the same way as remarked for agent and
recipient/beneficiary. And indeed, polysemy of cause and agent on one side, and of
purpose and recipient/beneficiary on the other side is frequent, as often noted in
the relevant literature. Such mapping of the domain of space onto the domain of
causation is in accordance with Croft’s Causal Chain (see above, Figure 6).
As mentioned in Section 2.3.2, polysemies ensuing from semantic extensions
that cross-cut the Causal Chain, and involve antecedent and subsequent roles are
also well documented, even though polysemies within the two (or three) groups
are much more numerous. However, while polysemy of agent and recipient/
beneficiary is not especially frequent (see below, Section 3.3), polysemy of cause
and purpose is extremely frequent, as with English for and many other similar

. Instrument markers often also indicate manner; this is usually the case in languages
that also feature the comitative/instrument polysemy. According to Heine and Kuteva (2002),
comitative markers extend to instrument and then to manner, thus complying with the scale
in Heine et al. (1991) (see above, Figure 1). However, a cursory look at the Indo-European
languages in which instrument is coded through a bare case and comitative through the same
case plus a preposition (such as Russian) reveals that such languages also feature both instru-
ment/manner and comitative/manner polysemy. Thus the issue seems to be more intricate,
and further research is needed in this area.
 Plotting diachronic semantic maps 

­ arkers. Often, this polysemy also involves beneficiary, but typically not recipient,
m
as shown in (43)–(45):
(43) He ran away for fear. (cause)
(44) I bought a present for Mary. (beneficiary)
(45) Mary went out for dinner. (purpose)

The same polysemy holds for other non-allative beneficiary markers surveyed in
Section 2.2.7 from Indo-European and Balto-Finnic languages. More example are
provided by Turkish için in (46)–(48) (from Kornfilt 1997) and Georgian tvis in
(49)–(51) (M. Topadze p.c.):
(46) Söylemek için geldim
say:inf for come:pst.1sg
‘I came in order to say…’ (purpose)
(47) Bayram olduğu için toplar atıldı
holyday be:pst.3sg for cannon:pl employ:pst.3sg
‘Because of the holyday (cause), cannons were shot.’
(48) sizin için bir kitap getirdim
2pl.gen for one book bring:pst.1sg
‘I took a book for you.’ (beneficiary)
(49) ar scalia pikr-is tvis
neg time think(masdar)-gen for
‘S/he has no time to think.’ (purpose)
(50) ertjeradi gamoq’eneb-is-tvis
single usage-gen-for
‘for a single usage’ (purpose)
(51) es bavshv-is- tvis viq’ide
this child-gen-for I.bought
‘I bought it for the child.’ (beneficiary)

Different languages also provide evidence for non-unidirectionality of the exten-

sion between cause and purpose. Luraghi (2005b) shows that Classical Greek
diá+accusative, which indicated cause, slowly extended to purpose and later to
beneficiary (thus also contradicting Heine et al.’s predictions on the relative order
of human and abstract relations). Its Modern Greek outcome já is the standard way
to indicate purpose and beneficiary, while still also encoding cause and reason. On
the other hand, as mentioned above (Section 2.2.7) Latin pro first extended to ben-
eficiary and purpose, but corresponding prepositions in the Romance languages,
such as French pour, also indicate cause. (see Luraghi 2013c for further discussion)
 Silvia Luraghi

Croft (1991) discusses the cause/purpose polysemy, and remarks that the
notion of reason provides an area for possible contact of these two otherwise
separate roles. He writes: “[e]xpressions of reason, which is a category of inten-
tion, not of causation, can represent events that causally follow the verb segment
(a goal or purpose) or precede (a source or motivation).” (p. 293); he further
adds that reason “is nondirectional” (ib.). Arguably then, the intermediate part
of the domain of space, in which location and path are situated, provides a spa-
tial equivalent to reason, hence also giving rise to expressions of cause and pur-
pose. Similar to reason, it is non-directional: either because it is static (location)
or because it profiles a path, but not its starting and endpoints. This seems to
happen when the polysemy also involves beneficiary, most likely if the latter role
is not marked as recipient: the examples discussed above provide evidence for
this claim.19
In Figure 7 I give a graphic representation of the mapping of the source
domain of spatial relations onto the domain of causation, including the semantic
roles of cause, purpose, and reason.

location source domain: space

source direction

path

Causes The Mind is Purposes

are Origins a Container Are Destinations

target domain: causation

cause reason purpose

Figure 7. The mapping of spatial relations onto causation

Again, there is evidence for an intermediate area between antecedent and

subsequent roles, which has a spatial equivalent in the non-directional relation
of location and partly also of path, a role which, as already remarked, does not
include the starting and endpoints of motion and is not telic. The encoding of the
semantic role reason with locative or perlative markers must be accounted for by a
complex metaphor, based on instantiations of the container metaphor (The Mind

. However, it is remarkable that extension from purpose to cause of allative morphemes is
also attested, see Heine & Kuteva (2002: 246–247) for evidence from African languages.
 Plotting diachronic semantic maps 

is a Container), and other more specific metaphors that explain the conceptu-
alization of a reason in terms of particular spatial configurations (as in the case of
Latin pro ‘before’ described in Section 2.2.8).

3. Expected and unexpected polysemies among human relations

3.1 Agent and possessor

The roles of agent20 and possessor may share the same source domain, drawing
on the spatial relation of source (or origin).21 Indeed, genitives of agent are com-
mon in the Indo-European languages, in which, however, they are mostly limited
to nominal forms of verbs. According to Hettrich (1990: 97), such constructions
started out from a possessive genitive modifying a verbal adjective; later, they
also extended to finite verbs to varying extents in different languages (p. 93–95).
A similar development is found in Finnish. Finnish has no real agented passive,
but infinitives can occur with genitive NPs which get an agentive interpretation, as
shown in Example (52):
(52) kirja on minu-n kirjoitta-ma-ni
book:nom be:prs.3sg i-gen write-inf3-1sg.poss
‘The book has been written by me.’

(See further Section 3.3 on a possible interpretation of some genitive agents in

Finnish as instantiating the polysemy of agent and beneficiary.)
In such construction, too, the infinitive takes a possessive genitive. Thus agents
can be interpreted as possessors of actions. The occurrence of nominal forms of
the verb in such constructions favors this interpretation: an event which is referred
to by means of a nominal form is also conceived as a static entity, a thing, which
can be possessed. Metaphors at play here are Agents are Possessors of Events
and Events are Things Possessed.

. Cross-linguistically, agents are most frequently realized as subjects of transitive verbs.
In nominative-accusative languages, this means that they take the nominative case. Crucially,
many transitive verbs do not have agentive subjects; in addition, subjects of all verbs are
encoded in the same way in such languages. This points toward a different status of the nomi-
native case: it may well be that nominative case markers eventually go back to agent markers
(and possibly ablative markers), but their function as markers of a grammatical relation (i.e.
subject) makes their possible semantic motivation virtually inexistent. For this reason, I am
not going to survey the origin of nominative case markers here.
. On the relation between source and origin see Nikiforidou (1991) and Luraghi (2003).
 Silvia Luraghi

Not only passive agents, but also ergatives often derive from genitives. Accord-
ing to Lehmann (2002: 98), this type of polysemy arises from constructions of
nominative-accusative aligned languages in which a nominal form of the verb
takes a possessive genitive: “Whenever a (passive) predicate is nominalized – this
may occur not only in subordinate, but also in main clauses, namely whenever
there is an (analytic) nominal verb form -, its agent may be in the genitive. When
such a construction is reinterpreted as transitive, the genitive develops into an
ergative. Again, genitive/ergative polysemy is a frequent phenomenon in ergative
languages, e.g. in Lak (Caucasian), Eskimo and Sherpa (Tibeto-Burmese).” In fact,
genitive agents with nominal forms of the verb are usually thought to have pro-
vided the origin for ergative alignment in Old Persian (see Hettrich 1990: 95–97;
Pompeo & Benvenuto 2011).22
Polysemy of agent and possessor may also arise as a by-product of both roles
being ultimately connected with source in the domain of space: for example,
­German von ‘from’ indicates source, possessor and passive agent. However, in such
(frequent) cases, the two developments:

a. source → possessor
b. source → agent

can be shown to have occurred historically independent of one another. Thus,

this type of polysemy is explained through the common source domains of the
two roles, rather than on their own features, once disconnected from their spatial
origin.

3.2 Possessor and recipient/beneficiary

Possessor and recipient/beneficiary is a frequent polysemy, and it typically involves
the dative case, hence not the type of beneficiary which does not merge with
recipient (see Section 2.2.6 and 2.2.7). Indeed, as I will discuss below, it is recipi-
ent, rather than beneficiary, that provides contact with possessor. The direction
of semantic extension leads from recipient/beneficiary to possessor: a recipient
is someone who gets hold of an entity which is being transferred; after the event
of transfer has taken place, the recipient is, at least temporarily, the possessor of
this entity. Remarkably, there is no need to set up a metaphor to account for this

. The Old Persian genitive is the merger of the Proto-Iranic genitive and dative. For this
reason, it has been suggested, most recently in Haig (2008), that the agent function was con-
nected with the beneficiary meaning of the genitive/dative. While this might in principle be
the case, comparison with Avestan and Indo-Aryan suggests that the genitive was the original
case in this construction. See Pompeo & Benvenuto (2011) for a detailed analysis of the texts.
 Plotting diachronic semantic maps 

semantic extension: a recipient is not a metaphorical possessor, rather, the recipi-

ent of an entity in a transaction is assumed by default to become its possessor. For
this reason, this type of extension can be accounted for in the same way as the
extension of beneficiary markers to the recipient role (Section 2.2.8).
Note however that, as this type of polysemy usually involves the dative case,
extension to possessor could result from recipient/beneficiary and possessor
being neighboring roles in the source domain of space. Indeed, even though the
source domain for recipient/beneficiary is provided by direction, while the source
domain for possessor seems to be often provided by location (see the discussion in
Section 2.2.5), it is often the case that location and direction merge already in the
source domain, as shown by frequent polysemy of location and direction markers
(Section 2.1.2). On the other hand, one can observe partial extension of the dative
to non-prototypical cases of possession, such as abstract possession, in languages
that otherwise make use of the Locative Schema for prototypical possession, as for
example in Russian, as shown in (53) and (54):
(53) Mne dvacat’ let
me:dat twenty years
‘I am twenty years old.’
(54) U menja kniga
near me:gen book
‘I have a book.’

Most likely, the dative in (53) can be considered close to an experiencer, given the
fact that this construction is also used when one indicates bodily sensations, and
that experiencer datives are cross-linguistically frequent in languages in which the
dative also expresses possession.23

3.3 Agent and recipient/beneficiary

Polysemy of agent and recipient/beneficiary should in principle be infrequent:
the two semantic roles have their source domain in distinct areas in the domain
of space, as agent is connected with source and recipient/beneficiary with direc-
tion, two roles that tend not to merge in the source domain (see Section 2.1.2).
However, this polysemy exists in languages of different families. As in the case of
polysemy of possessor and recipient/beneficiary, the polysemy of agent and ben-
eficiary also seems to involve exclusively the type of beneficiary that also indicates
recipient, that is, which is typically coded by the dative.

. Note that the construction in (53) is older than the one in (54) as it has cognates in other
Indo-European languages.
 Silvia Luraghi

Many Indo-European languages attest to a dative of agent, most often with

verbal adjectives, as in Example (55) from Latin:
(55) adeundus mihi illic est homo
go:gerve.nom 1sg.dat there be:3sg man:nom
‘I have to go to that man there.’ (lit.: ‘That man there must be approached
by me.’) Pl. Rud. 1298.

Hettrich (1990: 64–77) discusses exhaustively numerous examples from various

Indo-European languages, and convincingly argues that this pattern must be
inherited from Proto-Indo-European, and that is must be viewed as an extension
of the recipient/beneficiary function of the Indo-European dative. Indeed, the
construction typically occurs in passages such as the one quoted above, where the
verbal form expresses some type of agent-oriented modality, most often obligation
or necessity (cf. Bybee, Perkins & Pagliuca 1994: 177–179 for the terminology).
The action must or can be performed by a human participant, which is conceived
as being the target of the obligation or necessity. As recipients are targets of trans-
fer events, they can be understood as being targets of such abstract transfers: this
explains the extension of recipient/beneficiary to agent. An underlying metaphor
can be set up as follows: An Obligation is an Entity Transferred.24
Similar to the Indo-European dative, the Finnish genitive also encodes agent
with the same types of agent-oriented modality (such constructions are called
necessitive in Finnish grammatical descriptions), as shown in (56):
(56) Minu-n täyty-y nyt lähte-ä.
1sg-gen must-3sg now leave-inf
‘I must leave now.’/‘I have to leave now.’ (from Sands & Campbell 2001: 270).

It could be said that this usage is based on the extension of a possessor marker
to the encoding of agent, as in the case of (52) quoted above which is, however,
declarative. It can also be that the genitive of agent in Finnish can be accounted for
in two different ways, depending on sentence modality.25

. In Ancient Greek, too, the dative of agent occurs with verbal adjectives that indicate
obligation (see Schwyzer 1950: 150, 409–410). In addition, it is also frequent with perfect
verb forms, see Luraghi (2003: 65 with examples). The Greek perfect has stative meaning and,
similar to nominal forms of the verb, it refers to an event as to a static entity. This might
explain the extension of the dative from verbal adjectives, which are nominal forms, to other
verb forms, such as the perfect, which at least partly share features of nominal forms (i.e. sta-
ticity). See further Luraghi (2013b).
. Sands and Campbell (2001: 275) remark that the Finnish genitive often covers the func-
tion typical of the Indo-European beneficiary dative, as in: Minu-n on hyvä ol-la koto-na
(­1sg-gen be-3sg good be-inf home-ess) ‘It’s good for me to be home.’ In addition, the
 Plotting diachronic semantic maps 

Extension of the dative marker to passive agent occurs in Japanese. According

to current views, passive voice in Japanese has developed out of an original middle
(Narrog 2010b; Toyota 2011). Passive agents already occurred in Old Japanese, but
limited to poetry (Bentley 2001); apparently passive agent phrases could only con-
tain human nouns (Vovin 2005), while they also extended to inanimates at later
stages. In origin, middle forms indicated uncontrolled events, thus agent phrases
must be a late (though pre-literary) addition. It is possible that a dative could be
added to a middle form to indicate the human being affected by the event, and
that it was later reinterpreted as an agent. This would also explain why only human
agents occur at an early stage. A possible evolution would then be as follows: a ben-
eficiary (dative of interest) is added to a verb form that indicates an uncontrolled
event → the notion of benefaction or interest is reinterpreted as agency when the
verb form acquires passive meaning. In fact, an agent is normally the human being
who primarily profits from an event. Also in this case, as for the extension of ben-
eficiary to recipient and of recipient/beneficiary to possessor, there seems to be a
default interpretation at play (the human being in whose primary interest an event
is brought about is the agent), rather than a metaphor.26

3.4 Agent and comitative

This type of polysemy is also very infrequent, as discussed in Stolz (2001a).
One of the few examples that I am aware of is constituted by the Vedic Sanskrit
instrumental, which indicates instrument, passive agent, and to some extent also
comitative, even though for this function adpositional phrases were also used
already at an early stage (see Luraghi 2001a with examples; Delbrück 1867 and
Wenzel 1879 for a thorough discussion of various usages of this case). Consider
Examples (57)–(59):
(57) hatá vr̥trámˌ … índrena sáhasā yujá̄
smite:impt.2pl V.:acc I.:instr mighty companion:instr
‘Smite Vrtra, with the strong Indra as a companion.’ (Rv. i 239);

a­ dessive case is said to indicate agent in Finnish, as in: Isä teetti tuoli-n puusepä-llä (father
make-caus.past.3sg chair-acc carpenter-ade) ‘Father had the chair made by a carpenter’
(from Huumo 1996: 80), but this is clearly a causee, rather than a passive agent.
. This conclusion is admittedly speculative; remarkably, however, several occurrences
often taken as instances of dative of agent in ancient Indo-European languages can also be
explained in a similar way, as shown in Luraghi (2013b). As Spike Gildea (p.c.) points out,
recipient/agent polysemy is also found in the Cariban language family, in which it might go
back to Proto-Cariban.
 Silvia Luraghi

(58) abhí jahi raksásahˌ párvatena

prev smite:impt.2sg r.:acc.pl stone:instr
‘Hit the Raksasa with the stone club.’ (Rv. vii 10419);

(59) hatá índrenˌ  a panˌ  ayahˌ śayadhve

smite:ptcp.pf.nom.pl Indra:instr P.:nom.pl lie.down:prs.mid.2pl
‘Smitten by Indra you, Panis, will sink into death.’ (Rv. x 1084).

The fact that this type of polysemy is usually avoided can be connected to the
nature of prototypical comitative, which refers to a human entity that accom-
plishes an action together with another agent, as in (60):
(60) Mary prepared dinner with Janet.

The normal interpretation of (60) is that both Mary and Janet are actively involved
in the event of preparing dinner, that is, that they are both agents.
Given the possible co-occurrence of a comitative and an agent, polysemy is
usually avoided. Indeed, the Classical Sanskrit patter, in which the instrumen-
tal case indicates the roles of instrument and passive agent, while the comita-
tive is encoded through an adpositional phrase with the instrumental, is typical
of the Indo-European languages which retain the instrumental case, and display
the instrument-agent polysemy. For example, in Russian and most other Slavic
languages the instrumental case encodes instrument and passive agent, while
­comitative is encoded by the preposition s (or its cognates) and an NP in the

instrumental.

3.5 Comitative and possessor

The roles of comitative and possessor are usually kept distinct. This lack of poly-
semy seems connected with another tendency of comitative markers in possessive
constructions, described in Stolz (2001b), by which they tend to be re-interpreted
as marking the possessee, rather that the possessor. Stolz remarks that such exten-
sion of comitatives is based on a conceptualization by which being with something
equals to possessing something. He gives several examples that involve predicative
possession. Example (61) and (62) are from Swahili and Portuguese (both from
Stolz 2001b):
(61) Hamisi a-na kitabu
Hamisi 3sg-with book
‘Hamsa has a book.’
(62) e sentouse porque estava com medo
and sit:pret.3sg.refl because be:impf.3sg with fear
‘And he sat down because he was afraid.’
 Plotting diachronic semantic maps 

Given the features of prototypical possession reviewed in Section 2.2.2, this type
of extension follows quite naturally as a consequence of the assumption that pos-
sessor and possessee need to be in close spatial proximity: if possession implies
spatial proximity, then it can also be the case that spatial proximity indicates pos-
session. Thus, the same metaphors that account for the Locative Schema for pos-
session (see Section 2.2.2) also account for extension of comitatives to possesse
markers.

3.6 Comitative recipient/beneficiary

In Section 3.4 and 3.5, we have seen that polysemy involving the comitative and
other human roles is infrequent. This is the case also for polysemy involving the
comitative and the recipient/beneficiary, which, according to Stolz et al. (2006)
is virtually inexistent. If we consider the regions of space that serve as the source
domain for the latter roles and for the comitative, lack of polysemy may look quite
surprising: after all, comitatives arise from locatives, and recipients arise from
allatives, two spatial semantic roles that are very frequently encoded by means of
the same, polysemous marker. However, the relation between comitative and its
source domain is completely different with respect to the relation of recipient with
its source domain. As I have remarked in Section 2.2.1, markers of spatial relations
(typically direction) which extend to recipient usually also preserve their spatial
meaning. But this is not the case for markers of location which extend to comita-
tive, and which generally lose any spatial meaning they may have had before. Even
though the comitative relation “being with somebody” implies physical proximity,
comitative markers seem to be disconnected from the domain of space.

3.7 Experiencer
As remarked in Section 2.2.4, the experiencer role is encoded in a variety of ways,
its trademark being that it is always coded as something else. As repeatedly noted
in the literature, cross-linguistic evidence seems to point toward the non-existence
of a dedicated case for experiencer: language after language provide evidence for
case systems that include cases for primarily coding agent, patient, recipient, pos-
sessor, various spatial relations, instrument, and other inanimate relations, but
case systems typically do not include a case whose primary function is to code
experiencer. Virtually the only exception is constituted by the so-called affective
case of some Daghestanian languages, whose primary function is indeed that
of encoding experiencers with a sub-set of experiential verbs (some perception,
modal and mental verbs). This case exists in Andic languages and in one Lezgic
language, Tsakhur (see Comrie & van den Berg 2006; Daniel & Ganenkov 2008).
In origin, the affective was a spatial case, as shown by some vestigial usages with
 Silvia Luraghi

­toponyms (see Ganenkov 2006); in Tsakhur it is also used for the addressee of
speech. The latter is a semantic role typically encoded in the same was as recipi-
ent cross-linguistically, through an extension of allative markers (see Daniel this
volume), based on the Conduit metaphor (Reddy 1979), a complex metaphor that
accounts for communication: Ideas/Meanings are Objects; Words/Sentences
are Containers; Communication is Sending. The occurrence of such a case
with the primary function of encoding the experiencer role points toward a con-
ceptualization of experiencers as locations or endpoints of directional motion.27
In the discussion in Section 2.2.4, I have argued that such spatial metaphors
could indeed be mediated by other uses of the markers involved. Thus, adessive
coding, such as in Finnish, can be mediated by the possessor schema, rather than
depend on direct mapping from the domain of space. This conclusion is supported
by the extension of the Possessor Schema to experiential situations in many lan-
guages. More in general, the large extent to which experiencers can be encoded
through the dative case as recipients/beneficiaries in numerous genetically unre-
lated languages is very likely to owe to a gestalt effect, by which common features
of the two roles (humanness, limited control, being sentient) are focused upon and
account for experiencers being understood as similar to recipients/beneficiaries.
In sum, the fact that dedicated case marking for experiencer is cross-linguistically
so infrequent mirrors a more basic cognitive gap regarding this role, that is, that
there are no basic metaphors that map spatial relations onto experiencer.
Note that languages abound of spatial metaphors connected with the
domain of experience: often, emotions/sensations are conceived as containers,
or as things contained in body organs, for example. In addition, the Daghestan-
ian affective case shows that mapping of the spatial domain onto the domain
of experience is possible. Apparently, however, no path leading from space to
experience is so widespread and arguably entrenched in cognition as metaphors
that relate other roles to space, such as Agents are Origins, Possessors are
Places or Recipients are Destinations. Possibly, this depends on the fact
that metaphors that could account for experiencer (Experiencers are Places/
Containers, Experiencers are Destinations) more frequently map space
onto other semantic roles (possessor, beneficiary, recipient). A reason for this
may well be that the domain of experience is very complex and multifaceted, as
it includes very different types of situations (bodily sensations, emotions, per-
ception, mental processes, intetionality), and that experiential predicates may

. One can also detect some similarity between the experiencer role and that of addressee,
based on the fact that an addressee of speech is normally assumed to hear what s/he is being
told, that is, an addressee is normally also a perceiver.
 Plotting diachronic semantic maps 

be stative or inchoative (see Section 2.2.4): for this reason, the encoding of the
experiencer role could hardly be based on a single metaphor.

3.8 Interim summary 2

In Sections 3.1 through 3.7 I have surveyed some (often unexpected) patterns of
polysemy among semantic roles that are typical of human beings. I started with
possible polysemies involving possessor: this semantic roles frequently merges
with agent on the one hand (Section 3.1), or with recipient/beneficiary on the
other hand (Section 3.2). I have argued that the existence of two competing meta-
phors that account for the mapping of space onto the domain of possession, that
is Possessors are Origins and Possessors are Places (see Section 2.2.5 and
Figure 4), accounts for possible polysemy with the semantic role of agent, often
encoded as source, and the semantic role of recipient/beneficiary, often encoded as
direction (see Section 2.2.1). Based on this latter property of the encoding of agent
and recipient/beneficiary, polysemy between these two roles is unexpected. How-
ever, as I have shown in Section 3.3, such polysemy occurs in various languages, in
which it seems to be connected with modal or aspectual features of certain predi-
cates, or with (possibly earlier) features of the voice system. The semantic role of
comitative is apparently very unlikely to develop any type of polysemy with other
human roles. Polysemy of agent and comitative is attested in some ancient Indo-
European languages, but seems to be infrequent elsewhere (Section 3.4), while
polysemy of comitative and possessor seems to be excluded by the tendency for
(inanimate) comitatives to merge with possessees (Section 3.5). Finally, polysemy
of comitative and recipient/beneficiary is virtually inexistent, as has already been
pointed out in previous studies, in spite of the fact that these two roles are encoded
with markers whose original spatial meanings come from neighboring domains.
However, as I have pointed out in Section 2.2.3, comitative markers do not retain
their original spatial meaning, so connection with recipient/beneficiary or, more
in general, other roles cannot be provided by the source domain. The experiencer
role shows frequent polysemies with several other roles, owing to the multifaceted
nature of the domain of experience (Section 3.7).

4. Discussion

In the above sections, I investigated how space can serve as the source domain for
semantic roles typically taken by human beings, and how these can extend through
metaphor to other semantic roles, typical of inanimate entities such as instru-
ment or involving human activities such as purpose. The basic spatial ­relations
 Silvia Luraghi

of source, location, and direction each offer a preferred source domain for one or
more human relations. In particular, source offers the preferred source domain for
agent, while direction offers the preferred source domain for recipient. Given the
infrequence of a merger involving source and direction, metaphors mapping these
two roles onto agent and recipient remain maximally distinct. Notably, polysemy
of agent and recipient is attested, but it is not manifested in means of encoding
(cases, adpositions) connected with both spatial roles at the same time (i.e. it does
not follow from a merger of Direction and Source in the source domain). For
example, in Section 3.3, I have shown that agents can be encoded as recipients with
verbal forms expressing obligation or necessity. In such cases, it is not the domain
of space to provide the source for metaphorical extension. Rather, extension oper-
ates within the domain of human relations.
Location offers the source domain for several human roles, among which pos-
sessor and comitative. Remarkably, the relation of these two roles to their source
domain is different, in that locative markers that extend to possessors usually
retain their spatial meaning, while this does not happen in the case of locative
markers extending to comitative.
Beside location, possessor has another frequent source in the domain of space,
that is, source/origin. This connects it with agent, thus providing a basis for poly-
semy among such human roles. Notably, source and location are two spatial roles
which are typically kept distinct: as remarked in Section 2.1.2, in spite of relatively
frequent extension of source markers to the coding of location, the original mean-
ing seems to necessarily be lost when the second develops, in very much the same
way as in the case of location and comitative. Thus, languages may rely on both
metaphors for indicating possessor, with the two ensuing constructions special-
ized for different functions (as in the case of external, i.e. dative, possessors in
several languages of Europe, see Haspelmath 1999).
Similar to possessor, beneficiary also has different sources in the domain of
spatial relations, as it may originate from allative markers or from locative or per-
lative markers. In the first case, beneficiary typically exhibits polysemy with recipi-
ent, while in the second it does not, but it tends to entertain polysemic relations
with purpose and/or cause. Remarkably, while purpose is also a frequent meaning
of recipient markers, cause is not: the semantic role of cause tends to be related
with source in the domain of space, rather than with direction. Although unex-
pected, the polysemy of cause and purpose is frequent. More research is certainly
needed on the mapping of space onto these two semantic roles; from the examples
analyzed in Section 2.3.3, the polysemy of cause and purpose seems to arise from
markers that do not encode source or direction in the domain of space, but rather
location or path, such as English for and Finnish vuoksi. Such markers typically
also encode beneficiary, but not recipient (see Section 2.2.6). Location and path
are both located in the same region between source and direction in the domain of
 Plotting diachronic semantic maps 

space. As argued in Section 2.1.1, they are atelic, as they do not indicate a change
of state, contrary to source and direction, which imply that an entity is starting or
ending motion. They provide an intermediate, non-directional area, which is the
source domain for the semantic role reason. The latter role in its turn provides the
area of overlap for source and cause.
Experiencer seems to have no direct source in the source domain of space:
various markers of spatial relations that can code this role in different languages
seem to be motivated by its connection to other human roles, in particular recipi-
ent and possessor. In any case, cases and adpositions that might point to a relation
between the experiencer role and the portions of spatial domain which typically
encode location or direction, but not source. This is not surprising: experiencers
are usually conceived as being affected by experiential situations, rather than ini-
tiators thereof.
The polysemies reviewed above all show that source and the other spatial
semantic roles tend not to merge, and that, accordingly, human roles directly
connected with them remain separated. Unexpected polysemies such as the one
involving agent and recipient do not have their direct origin in the domain of
space, but are based on extensions in the domain of human relations.
The predictions in (1) regarding the direction of semantic extension, from
space to human relations, to inanimate ones are generally borne out, even
though the case that space is mapped directly onto an inanimate relation, with
no intermediate human role, is not infrequent. The most obvious example is
the mapping of space onto time; in addition, abstract roles such as cause and
purpose also often involve direct mapping from the source domain of space.
The semantic role of instrument is most interesting in this respect, as the instru-
mental meaning is often acquired by locative markers, possibly by a general-
ization based on the specific shape of certain instruments, which makes them
equally available for instrument and for location expressions. Instrument is
also involved in the only clear case of semantic extension that contradicts the
scale in (1), as instrumental cases or adpositions often extend their meaning to
the agent role. Tentatively, I explain this development as based on a referential
metonymy rather than a metaphor.
The fact that source tends not to display polysemies that involve the other
spatial roles does not mean that morphemes encoding source cannot undergo
semantic extensions in the domain of space. As I have shown in 2.1.2, extension
from source to location and direction is attested. Contrary to many other semantic
extensions reviewed in this paper, though, it does not result in polysemy, as the
original meaning disappears when the new one(s) develop. The data attesting to
this development do not point to a gradual change: to the contrary, the fact that
polysemy of source and other spatial roles is so infrequent, while the correspond-
ing semantic extension seems to be comparatively frequent, rather indicates that
 Silvia Luraghi

the overlapping of the source and target meanings is avoided. The same happens
for semantic extension of locative markers to the comitative role. In this case as
well, polysemy is virtually inexistent, and the original locative meaning of comita-
tive markers can often only be reconstructed.
In spite of the pervasive role of metaphor, other mechanisms also operate in
semantic extension, and are responsible for semantic role polysemy. Beside the
referential metonymy mentioned above concerning the extension of instrument
to agent, I also described predicational metonymies in the case of beneficiary and
recipient and of recipient and possessor polysemy. In addition, I argued that exten-
sion of various spatial relations, involving containment or support, to the coding
of instrument owes to analogical extension. More precisely, I argued that some
instruments are indeed shaped as containers or means of support. This offers a
conceptualization limited to a set of instruments, which is then analogically
extended to all possible instruments.
In Figure 8 below I give a tentative representation of connections among the
semantic roles surveyed in this paper, on the background of the source domain of
space. The map should better be viewed as being tri-dimensional, thus allowing
for a representation of polysemy among non-contiguous roles, such as cause and
purpose. In addition, the role of experiencer should be viewed as indirectly con-
nected with the background source domain through other human roles.

LOCATION

SOURCE

POSSESSOR
DIRECTION
comitative experiencer

cause beneficiary
agent instrument recipient
Force Purpose
intermediary

PATH

reason

Figure 8. A map of semantic roles

 Plotting diachronic semantic maps 

In Figure 8, semantic roles are arranged in a way as to account for patterns

of polysemy and semantic extension reviewed in the paper. I added arrows in
cases in which directionality is made reasonably clear from attested diachronic
developments, as for example from comitative to instrument. Extension from
possessor to experiencer is documented in the generalization of the possessor
schema for feelings and bodily sensations, while extension from possessor to
agent can be assumed in cases in which passive agent can be encoded through
the genitive. In addition, I tentatively indicated directionality from instrument
to agent and from beneficiary to recipient. Note however that, in the latter case,
diachronic developments involving the replacement of the dative case by vari-
ous prepositions in the Indo-European languages seem to involve extension of a
directional preposition to all meanings of the dative (that is, to beneficiary and
recipient at the same time). As noted above, the four basic spatial relations on
the background provide the source domain for (most) other relations. In this
respect, this map can be seen as a flat representation corresponding to those in
Figures 4 and 7.
The four spatial relations are represented as partly overlapping, as in F­ igure 3,
to account for possible polysemy among them. This includes the polysemy of
source and other spatial relations, in the first place location, which is infrequently
attested: as I have argued in Section 2.1.2, merger of ablative with locative markers
is a possible diachronic development, even if it does not usually result in poly-
semy. Again, diachronic extension not resulting in polysemy is also indicated by
the extension of locatives to comitatives. Most likely, as I argued in S­ ection 2.2.3
regarding comitative and location, such mergers without polysemy may rest on
the special complexity of a specific cognitive domain (in this case, of the domain
of spatial relations with human landmarks). In the case of source and other spatial
roles, it points to an asymmetry regarding source, which has been noted by other
authors. Tentatively, I suggest that this asymmetry might be based on an opposi-
tion between what can be conceived as being in the visual field of an observer (a
location, that is where the observer is, and a goal of motion, that is a point that
lies on a trajectory starting from the position of the observer and developing in
front of them), and what is not in the visual field (a position on the back of the
observer).
Finally, the fact that some diachronic mergers do not result in polysemy
leaves the question open whether certain changes are abrupt, rather than grad-
ual. Unfortunately, extension of locatives to comitatives is not fully documented
in its development. Concerning the extension of ablatives to locatives, available
­evidence indicates that there may (perhaps more often than one would think)
some synchronic overlap at least limited to specific lexemes. For example, in
Classical Latin the ablative preposition ab ‘from’ could indicate location with
 Silvia Luraghi

nouns indicating spatial regions (‘back’, ‘side’, etc. see Luraghi 2010a for examples
and further ­reference, and 2011 on developments in Romance). Even if I can-
not pursue this issue further here, this partial overlap seems to indicate that the
extension of ablative morphemes to the coding of location is indeed a gradual
process.28

Appendix – List of metaphors discussed in the text

A Beneficiary is a Destination
Agency is a Thing Transferred
Agents are Origins
Agents are Possessors of Events
An Instrument is a Companion
An Instrument/Means is a Channel
An Instrument/Means is an Intermediary
An Intermediary is a Channel
An Obligation is an Entity Transferred
Causes are Origins of Events
Communication is Sending
Events are Things Possessed
Existence is Position in Space
Existence is Visibility
Experiencers are Places (or Containers) for Feelings/Sensations
Experiencers are Possessors of Sensations
Experiencers are Recipients of Feelings/Sensations
Feelings/Sensations are Things
Ideas/Meanings are Objects
Material is an Instrument for Creating Objects
Objects Come out of Substance
Physical Vicinity is Control
Possessors are Origins
Possessors are Places
Purposes are Destinations
Recipients are Destinations
Sensations are Things Possessed

. A quantitative study of spatial prepositions in a parallel corpus consisting of the Ancient
Greek, Latin, Gothic and Old Church Slavonic Gospels shows that some degree of overlap of
source with other spatial relations is not limited to Latin, but also attested in the other lan-
guages. This, together with cross-linguistically frequent ablative-locative transfers described
in Section 2.1.2, might indicate that such overlap is more frequent than commonly thought,
see Eckhoff, Thomason & de Swart (2013).
 Plotting diachronic semantic maps 

The Mind is a Container/ Channel

The Visual Field is a Container
Time is Space
Words/Sentences are Containers

References

Amritavalli R. 2004. Experiencer datives in Kannada. In Bhaskararao & Subbarao (eds),

1–24.
Anderson, John M. 1971. The Grammar of Case: Towards a Localistic Theory. Cambridge: CUP.
Bennett, David C. 1989. Ablative-locative transfers: Evidence from Slovene and Serbo-Croat.
Oxford Slavonic Papers 22: 133–154.
Bentley, John R. 2001. A Descriptive Grammar of Early Old Japanese Prose. Leiden: Brill.
Bickel, Balthasar. 2004. The syntax of experiencers in the Himalayas. In Bhaskararao & Subbarao
(eds), 77–112.
Bhaskararao, Peri & Subbarao, Karumuri Venkata (eds). 2004. Non-nomitative Subjects, Vol. 1
[Typological Studies in Language 60]. Amsterdam: John Benjamins.
Blake, Barry J. 1977. Case marking in Australian languages [Linguistic Series 23]. Canberra:
Australian Institute of Aboriginal Studies.
Blansitt Jr., Edward L. 1988. Datives and allatives. In Studies in Syntactic Typology, Michael
Hammond, Edith Moravcsik & Jessica Wirth (eds), 173–191. Cambridge: CUP.
Bloomfield, Leonard. 1933. Language. New York NY: Holt.
Bybee, Joan, Perkins, Revere & Pagliuca, William. 1994. The Evolution of Grammar. Chicago IL:
The University of Chicago Press.
Comrie, Bernard & van den Berg, Helma. 2006. Experiencer constructions in Daghestanian
languages. In Semantic Role Universals and Argument Linking: Theoretical, Typological, and
Psycholinguistic Perspectives, Ina Bornkessel, Matthias Schlesewsky, Bernard Comrie &
Angela D. Friederici (eds), 127–154. Berlin: Mouton de Gruyter.
Coseriu, Eugenio. 1970. Einführung in die Strukturelle Betrachtung des Wortschatzes. Tübingen:
Narr.
Creissel, Denis & Mounole, Céline. 2011. Animacy and spatial cases: Typological tendencies
and the case of Basque. In Kittilä, Västi & Ylikoski (eds), 155–182.
Croft, William. 1991. Syntactic Categories and Grammatical Relations. Chicago IL: Chicago
­University Press.
Croft, William. 2001. Radical Construction Grammar: Syntactic Theory in Typological Perspec-
tive. Oxford: OUP. DOI: 10.1093/acprof:oso/9780198299554.001.0001
Croft, William. 2003. The role of domains in the interpretation of metaphors and metonymies.
In Metaphor and Metonymy in Comparison and Contrast, Rene Dirven & Ralf Pörings
(eds), 161–205. Berlin: Mouton de Gruyter.
Daniel, Michael A. & Ganenkov, Dmitri S. 2008. Case marking in Daghestanian. In Handbook
of Case, Andrej Malchukov & Andrew Spencer (eds). Oxford: OUP.
Delbrück, Berthold. 1867. Ablativ localis instrumentalis im Altindischen. Berlin: Dümmler.
Dik, Simon C. 1997. The Theory of Functional Grammar, 2 Vols., 2nd edn, Kees Hengeveld (ed.).
Berlin: Mouton De Gruyter.
 Silvia Luraghi

Dixon, Robert M.W. 2002. Australian Languages: Their Nature and Development. Cambridge:
CUP. DOI: 10.1017/CBO9780511486869
Eckhoff, Hanne Martine, Thomason, Olga A. & de Swart, Peter. 2013. Mapping out the source
domain. Studies in Language 37(2): 302–355. DOI: 10.1075/sl.37.2.03eck
Erelt, Mati & Metslang, Helle. 2006. Estonian clause patterns – from Finno-Ugric to standard
average European. Linguistica Uralica 42(4): 254–266.
Evans, Vyvyan. 2004. The Structure of Time. Language, Meaning and Temporal Cognition [Human
Cognitive Processing 12]. Amsterdam: John Benjamins. DOI: 10.1075/hcp.12
Ganenkov, Dmitry S. 2006. Experiencer coding in Nakh-Daghestanian. In Case, Valency
and Transitivity [Studies in Language Companion Series 77], Leonid Kulikov, Andrej
­Malchukov & Peter de Swart (eds), 179–202. Amsterdam: John Benjamins.
Geeraerts, Dirk & Cuyckens, Hubert (eds). 2007. The Oxford Handbook of Cognitive. Linguistics.
Oxford: OUP.
Goldberg, Adele. 1995. Constructions. A Construction Grammar Approach to Argument Struc-
ture. Chicago IL: University of Chicago Press.
Grünthal, Riho. 2003. Finnic Adpositions and Cases in Change. Helsinki: Societé Finno-ougrienne.
Haig, Geoffrey. 2008. Alignment change in Iranian languages: A Construction Grammar account.
Berlin: Walter de Gruyter.
Haspelmath, Martin. 1997. From Space to Time. Munich: Lincom.
Haspelmath, Martin. 1999. External possession in a European areal perspective. In External
Possession [Typological Studies in Language 39], Doris Payne & Immanuel Barshi (eds),
109–135. Amsterdam: John Benjamins.
Havers, Wilhelm. 1911. Untersuchungen zur Kasussyntax der indogermanischen Sprachen.
­Strassburg: Trübner.
Heine Bernd. 1997a. Auxiliaries. Oxford: OUP.
Heine, Bernd. 1997b. Cognitive Foundations of Grammar. Oxford: OUP.
Heine, Bernd & Kuteva, Tania. 2002. World Lexicon of Grammaticalization. Cambridge: CUP.
DOI: 10.1017/CBO9780511613463
Heine, Bernd, Claudi, Ulrike & Hünnemeyer, Friederike. 1991. Grammaticalization. A Concep-
tual Framework. Chicago IL: University of Chicago Press.
Hettrich, Heinrich. 1990. Der Agens in passivischen Sätzen altindogermanischer Sprachen. Göt-
tingen: Vandenhoek & Ruprecht.
Hjelmslev, Louis. 1935/37. La catégorie des cas. Acta Jutlandica 7(1): xii–184 & 9(2): viii–78.
Huumo, Tuomas. 1996. Domain shifts and the grammaticalization of case: A case study of the
Finnish adessive. Folia Linguistica Historica 17: 73–95.
Itkonen, Erkki, Joki, Aulis J. & Peltola, Reino. 1978. Suomen kielen etymologinen sanakirja,
Vol. 6. Helsinki: Suomalais-Ugrilainen Seura.
Janda, Laura. 1993. A Geography of Case Semantics: The Czech Dative and the Russian instrumen-
tal. Berlin: Mouton de Gruyter. DOI: 10.1515/9783110867930
Kenesei, István, Vago, Robert M. & Fenyvesi, Anna. 1997. Hungarian. London: Routledge.
Kittilä, Seppo. 2010. Beneficiary coding in Finnish. In Kittilä & Zúñiga (eds), 245–270.
Kittilä, Seppo. 2005. Recipient-prominence vs. beneficiary-prominence. Linguistic Typology
9(2): 269–297. DOI: 10.1515/lity.2005.9.2.269
Kittilä, Seppo & Zúñiga, Fernando. 2010. Introduction: Benefaction and malefaction form a
cross-linguistic perspective. In Zúñiga & Kittilä (eds),1–28.
Kittilä, Seppo, Västi, Katja & Ylikoski, Jussi (eds). 2011. Case, Animacy and Semantic Roles
[Typological Studies in Language 99]. Amsterdam: John Benjamins. DOI: 10.1075/tsl.99
 Plotting diachronic semantic maps 

Kornfilt, Jaklin. 1997. Turkish. London: Routledge.

Kuryłowicz, Jerzy. 1964. Inflectional Categories of Indo-European. Heidelberg: Winter.
Lakoff, George. 1977. Linguistic Gestalts. In Papers from the Thirteenth Regional Meeting of the
Chicago Linguistic Society, 236–287. Chicago IL: CLS.
Lakoff, George & Johnson, Mark. 1980. Metaphors we Live By. Chicago IL: University of Chicago
Press.
Lakoff, George & Turner, Mark. 1989. More than Cool Reason: A Field Guide to Poetic Meta-
phor. Chicago IL: University of Chicago Press. DOI: 10.7208/chicago/9780226470986.001
.0001
Lakoff, George, Espenson, Jane & Schwartz, Alan. 1991. The Master Metaphor List. 〈http://araw.
mede.uic.edu/~alansz/metaphor/METAPHORLIST.pdf〉
Langacker, Ronald W. 1987. Foundations of Cognitive Grammar, Vol. 1: Theoretical Prerequisites.
Stanford CA: Stanford University Press.
Lehmann, Christian. 22002. Thoughts on Grammaticalization, 2nd, revised edn. [Arbeitspapiere
des Seminars für Sprachwissenschaft der Universität Erfurt 9]. Erfurt: Seminar für Sprach-
wissenschaft der Universität.
Lehmann, Christian & Yong-Min Shin. 2005. The functional domain of concomitance. A typo-
logical study of instrumental and comitative relations. In Typological Studies in Participa-
tion, Christian Lehmann (ed.), 9–104. Berlin: Akademie Verlag.
Leumann, Manu, Hofmann, Johann Baptist & Szantyr, Anton. 1965. Lateinische Syntax und
Silistik. München: Beck.
Luraghi, Silvia. 1989. Cause and instrument expressions in Classical Greek. Mnemosyne 43:
294–308.
Luraghi, Silvia. 1995. Prototypicality and agenthood in Indo-European. In Historical Linguis-
tics 1993 [Current Issues in Linguistic Theory 124], Henning Andersen (ed.), 259–268.
Amsterdam: John Benjamins.
Luraghi, Silvia. 2000. Spatial metaphors and agenthood in Ancient Greek. In 125 Jahre Indo-
germanistik in Graz, Christian Zinko & Michaela Offisch (eds), 283–298. Graz: Leykam.
Luraghi, Silvia. 2001a. Syncretism and the classification of semantic roles. Sprachtypologie und
Universalienforschung 54(1): 35–51.
Luraghi, Silvia. 2001b. Some remarks on instrument, comitative, and agent in Indo-European.
Sprachtypologie und Universalienforschung 54(4): 385–401.
Luraghi, Silvia. 2003. On the Meaning of Prepositions and Cases. A Study of the Expression of
Semantic Roles in Ancient Greek [Studies in Language Companion Series 67]. Amsterdam:
John Benjamins. DOI: 10.1075/slcs.67
Luraghi, Silvia. 2005a. Prepositions in cause expressions. Papers on Grammar 12(2): 609–619.
Luraghi, Silvia. 2005b. Paths of semantic extension. From cause to beneficiary and purpose. In
Historical Linguistics 2003 [Current Issues in Linguistic Theory 257], Michael Fortescue,
Eva Skafte Jensen, Jens Erik Mogensen & Lene Schøsler (eds), 141–157. Amsterdam: John
Benjamins.
Luraghi, Silvia. 2005c. The history of the Greek preposition metá: From polysemy to the creation
of homonyms. Glotta 81: 130–159.
Luraghi, Silvia. 2009. The evolution of local cases and their grammatical equivalent in Greek and
Latin. In The Role of Semantics and Pragmatics in the Development of Case [Studies in Lan-
guage Companion Series 108], Johanna Barðdal & Shobhann L. Chelliah (eds), 283–305.
Amsterdam: John Benjamins.
 Silvia Luraghi

Luraghi, Silvia. 2010a. Adverbial phrases. In A New Historical Syntax of Latin, Philip Baldi &
Pierluigi Cuzzolin (eds), 19–107. Berlin: Mouton de Gruyter.
Luraghi, Silvia. 2010b. Experiencer predicates in Hittite. In Ex Anatolia Lux, Ronald I. Kim,
Elizabeth Rieken, Norbert Oettinger & Michael J. Weiss (eds), 249–264. Ann Arbor MI:
Beech Stave Press.
Luraghi, Silvia. 2010c. Where do beneficiaries come from and how do they come about? In
­Historical Cognitive Linguistics, Margaret Winters, Heli Tissari & Kathryn Allan (eds),
93–131. Berlin: Mouton de Gruyter.
Luraghi, Silvia. 2011. Human landmarks in spatial expressions: From Latin to Romance. In
­Kittilä, Västi & Ylikoski (eds), 207–234.
Luraghi, Silvia. 2013a. The mapping of space onto the domain of benefaction: Beneficiaries that
are not Recipients and their sources. Paper read at the 10th International Conference of the
Association for Linguistic Typology (Leipzig, 15–18 August 2013).
Luraghi, Silvia. 2013b. The dative of agent in Indo-European languages. Paper read at the
­Workshop historisch-vergleichende Sprachwissenschaft (Salzburg, 22–24 November 2013).
Luraghi, Silvia. Forthcoming. External possessor constructions in Indo-European. In Recon-
structing Syntax. Jóhanna Barðdal, Eugenio Lujan & Spike Gildea (eds). Leiden: Brill
Luraghi, Silvia & Sausa, Eleonora. Forthcoming. Hate and anger, love and desire: The con-
strual of ­emotions in Homeric Greek. In Historical Linguistics 2013, Dag T. Haug (ed).
­Amsterdam: Benjamins.
Lyons, John. 1977. Semantics, Vols I & II. Cambridge: CUP.
Mackenzie, J. Lachlan. 1978. Ablative-locative transfers and their relevance for the theory of
case-grammar. Journal of Linguistics 14: 129–375. DOI: 10.1017/S0022226700005831
Matisoff, James A. 1986. Heans and minds in South-East Asian languages and Engüsh: An essay
in the comparative lexical semantics of psycho-collocations. Cahiers de Linguistique Asie-
orientale 15: 5–57. DOI: 10.3406/clao.1986.1191
Narrog, Heiko. 2010a. Author’s reply to ‘The Best of Two Maps’ by Sander Lestrade. Linguistic
Discovery 8(1): 257–258.
Narrog, Heiko. 2010b. Voice and non-canonical marking in the expression of event-oriented
modality. A cross-linguistic study. Linguistic Typology 14(1): 71–126. DOI: 10.1515/lity.
2010.003
Nikiforidou, Kiki. 1991. The meaning of the genitive. Cognitive Linguistics 2(2): 149–205. DOI:
10.1515/cogl.1991.2.2.149
Nikitina, Tatiana & Spano, Marianna. Forthcoming. ‘Behind’ and ‘in front’ in Ancient Greek:
A case study in orientation asymmetry. In Ort und Bewegung: Linguistischen Arbeiten
zu Sprachen der Antike, Silvia Kutscher & Daniel Werning (eds). Berlin: Mouton de
Gruyter.
Noonan, Michael. 2009. Patterns of development, patterns of syncretism of relational morphol-
ogy in the Bodic languages. In The Role of Semantics and Pragmatics in the Development
of Case [Studies in Language Companion Series 108], Johanna Barðdal & Shobhanna
Chelliah (eds), 261–282. Amsterdam: John Benjamins.
Pantcheva, Marina. 2010. The syntactic structure of locations, goals, and sources. Linguistics
48(5): 1043–1081. DOI: 10.1515/ling.2010.034
Panther, Klaus-Uwe & Thornburg, Linda L. 2007. Metonymy. In Geeraerts & Cuyckens (eds),
236–263.
Pennacchietti, Fabrizio. 1974. Appunti per una storia comparata dei sistemi preposizionali
semitici. Annali dell’Istituto Orientale di Napoli 34(2): 161–183.
 Plotting diachronic semantic maps 

Pompeo, Flavia & Carmela Benvenuto, Maria. 2011. Il genitivo in persiano antico. Un caso
esemplare di categoria polisemica. Studi e Saggi Linguistici 49: 75–123.
Pütz, Martin. 1996. Introduction: Language and the cognitive construal of space. In The Con-
strual of Language and Thought [Cognitive Linguistics Research 8], Martin Pütz & Rene
Dirven (eds), xi–xxiii. Berlin: Mouton de Gruyter.
Radden, Günter. 1989. Semantic roles. In A User’s Grammar of English, Rene Dirven & Richard
A. Geiger (eds), 421–471. Frankfurt: Peter Lang.
Reddy, Michael. 1979. The conduit metaphor. In Metaphor and Thought, Andrew Ortony (ed.),
Cambridge: CUP.
Rice, Sally & Kabata, Kaori. 2007. Crosslinguistic grammaticalization patterns of the allative.
Linguistic Typology 11: 451–514. DOI: 10.1515/LINGTY.2007.031
Rodriguez Aristar, Anthony. 1996. The relationship between dative and locative: Kurylow-
icz’s argument from a typological perspective. Diachronica 13: 207–224. DOI: 10.1075/
dia.13.2.02ari
Sands, Kristina & Lyle Campbell. 2001. Non-canonical subjects and objects in Finnish. In A.
Aikenvald, R. M. W. Dixon, and M. Onishi (eds.), Non-canonical Marking of Subjects and
Objects, 251–305. Amsterdam: Benjamins.
Schmidtke-Bode, Karsten. 2010. The role of benefactives and related notions in the typology of
purpose clauses. In Zúñiga & Kittlä (eds), 121–146.
Schwyzer, Eduard. 1950. Griechische Grammatik, Band 2: Syntax. München: Beck.
Stolz, Thomas. 1992. Lokalkasussysteme. Wilhelmsfeld: Gottfried Egert Verlag.
Stolz, Thomas. 2001a. Comitatives vs. instrumentals vs. agents. In Aspects of Typology and Uni-
versals, Walter Bisang (ed.), 153–174. Berlin: Akademie Verlag.
Stolz, Thomas. 2001b. To be with X is to have X: Comitatives, instrumentals, locative and pred-
icative possession. Linguistics 39(2): 321–350. DOI: 10.1515/ling.2001.013
Stolz, Thomas, Stroh, Cornelia & Urdze, Aina. 2006. On comitatives and Related Categories. A
Typological Study with Special Focus on the Languages of Europe. Berlin: Mouton de Gruyter.
Taylor, John R. 1989a. Possessive genitives in English. Linguistics 27(4–6): 663–86.
Taylor, John R. 1989b. Linguistic Categorization: Prototypes in Linguistic Theory. Oxford: Clar-
endon Press.
Toyota, Junichi. 2011. The Grammatical Voice in Japanese: A Typological Perspective. Cambridge:
CUP.
Tuggy, David. 2007. Schematicity. In Geeraerts & Cuyckens (eds), 82–116.
Tyler, Andrea & Evans, Vyvyan. 2003. The Semantics of English Prepositions: Spatial Sciences,
Embodied Meaning, and Cognition. Cambridge: CUP. DOI: 10.1017/CBO9780511486517
Van Valin Jr., Robert D. & LaPolla, Randy J. 1997. Syntax. Cambridge: CUP.
Vandeloise, Claude. 1994. Methodology and analysis of the preposition. Cognitive Linguistics
5(2): 157–184. DOI: 10.1515/cogl.1994.5.2.157
Verhoeven, Elisabeth. 2008. Experiential Constructions in Yucatec Maya [Studies in Language
Companion Series 87]. Amsterdam: John Benjamins. DOI: 10.1075/slcs.87
Vovin, Alexander. 2005. A Descriptive and Comparative Grammar of Western Old Japanese.
­Folkestone: Global Oriental.
Wenzel, Heinrich. 1879. Ueber den instrumentalis im Rigveda, Tübingen: Laupp.
Zlatev, Jordan. 2007. Spatial semantics. In Geeraerts & Cuyckens (eds), 318–350.
Zúñiga, Fernando & Kittilä, Seppo (eds). 2010. Benefactives and Malefactives. Case Stud-
ies and Typological Perspectives [Typological Studies in Language 92]. Amsterdam: John
Benjamins.
 Silvia Luraghi

List of abbreviations

1 first person mid middle

2 second person neg negation
3 third person nom nominative
acc accusative p passive
adess adessive part partitive
adv adverb pf perfect
all allative pl plural
aor aorist plpf pluperfect
art article poss possessive
cop copula pret preterite
dat dative prev preverb
fut future prs present
gen genitive pst past
gerve gerundive ptc particle
impt imperative ptcp participle
inf infinitive refl reflexive
instr instrumental sg singular
Lm landmark Tr trajector
 The (non-)prototypicality of Direction*
The (allative and illative) case(s) of Finnish

Seppo Kittilä
University of Helsinki

The present paper examines the coding of Direction in Finnish. It will deal with
both typical and less typical instances of Direction in light of features, such as
animacy, size of the landmark and the nature of the reference to the landmark.
The paper pursues two goals. First, it will show that the coding of Direction in
Finnish is determined by markedness. In other words, the coding of Direction
can be explained by referring to the typical vs. atypical nature of Direction (which
depends on the landmark in question), and more importantly, canonical instances
bear less elaborate coding than non-canonical instances. In so doing, the paper
sheds more light on the semantic role of Direction, i.e. the question what the
features are that should be considered in the studies dealing with Direction.
Second, the paper aims at contributing to our understanding of the differences
between cases and adpositions. Cases and adpositions unarguably share common
features, and both of them can be used to express similar functions (such as
coding of semantic roles). However, they differ crucially from each other as
regards their use as markers of Direction, as already hinted at above.

Keywords: Finnish; allative; illative; directional cases; markedness

1. Introduction

The present paper is concerned with (non-)prototypicality of arguments. In con-

trast to many previous studies, the paper focuses on the (non-)prototypicality of
Direction. (1a) and (1b) from Estonian are illustrative examples of the scrutinized
phenomenon:
(1) Estonian (examples courtesy of Geda Paulsen)
a. Laps läks randa
child went shore.ill
‘The child went to the shore.’

* Due to the numerous different ways in which the label Goal has been defined, I have opted
for using the label direction in this paper.
 Seppo Kittilä

b. Laps läks õpetaja juurde

child went teacher.gen to
‘The child went to the teacher.’

In (1a), where the illative case is used, we are dealing with an inanimate Direc-
tion, while in (1b), in which a postposition is used, the Direction is animate. As
has been shown, for example, by Comrie (1986), Creissels and Mounole (2011)
and Kittilä and Ylikoski (2011), animate entities are less typical Directions than
inanimate entities, due to their mobile (instead of stable) nature and other factors.
Formally, the marked nature of (1b) is manifested in the postpositional coding of
Direction. In other words, Estonian (along with many other languages) resorts to
semantically more specific adpositions whenever this is necessary for assuring the
intended reading of a clause.
Cases like (1) constitute the topic of the present paper. The goal of the paper is
to show how (non-)prototypicality determines the coding of Direction in Finnish.
The argumentation in the paper is based exclusively on Finnish, but I hope that the
results of the study can be later applied to other languages as well. Even though the
label ‘markedness’ has been criticized (and to a large extent justly) by Haspelmath
(2006), I use the label in this paper to refer to formally more elaborate coding of a
semantic role. For example, in (1), (1b) presents a marked case, while (1a) can be
seen as its unmarked counterpart. In Finnish, markedness is manifested similarly
to Estonian: case forms code prototypical instances, while adpositions (mostly
postpositions) are used for coding non-prototypical instances. I will also argue
that the formal markedness has a conceptual basis. The formally more elaborately
coded instances of Direction can also be said to be conceptually marked, i.e. they
constitute less than typical manifestations of Direction, for example, because the
landmark is less stable, or animate (animate entities are not entered as a result
of motion, while many inanimate entities are). Aristar (1997) and Kittilä (2008)
have discussed features similar to those examined in this paper, but these studies
have considered also other roles than Direction. Moreover, these two studies are
cross-linguistic in nature, which has the natural consequence that the number of
features discussed is lower.
Finnish has been chosen as the language of the study for two major reasons.
First, Finnish is my native tongue and I can thus elicit the examples I need for the
study myself and I also have an access to other native speakers for checking less
clear cases.1 Second, and more importantly, Finnish is a language with a rich case

. Corpus studies would probably yield different results, but they would probably not include
any occurrences of the less accepted cases, let alone ungrammatical cases, which are central
to the discussion in this paper.
 The (non-)prototypicality of Direction 

inventory, in addition to which Finnish also uses an array of postpositions for

coding spatial relations. In this study, the focus lies on illative and allative cases,
and postpositions luo ‘to (the vicinity of)’, viereen ‘to nearby’, sisään ‘to inside’ and
päälle ‘onto’. I will only consider postpositions that can express the same meanings
as the examined case forms. In other words, postpositions such as alle ‘to under’,
väliin ‘to between’ and taakse ‘to behind’, or any other postposition, will not be
discussed. This makes it possible to focus exclusively on the features relevant to
this study.
One further note on methodology is in order before proceeding. As noted
above, the examples used in the present study are all elicited. This is necessary
since in order to discuss the limits of the examined mechanisms, it is essential to
examine less felicitous and even ungrammatical cases. However, it is important
to bear in mind that there are clear differences in the motivation of the infelicity.
Some of the cases are outright ungrammatical, and they are consequently marked
with an asterisk. Second, there are examples, which can be regarded as rather
rare, and they consequently (usually) require specific marking. Typical examples
include events such as ‘the child went to the grape’, where the landmark is rather
non-salient due to its small size, which makes the example in question concep-
tually non-prototypical. And finally, some examples describe events that practi-
cally never occur in the non-linguistic world, but that are grammatically felicitous
constructions whenever these rare occurrences need to be referred to. This type
is exemplified by cases such as ‘the teacher put the agronomist on the table’. The
latter two cases are marked with one or two question marks depending on the
(semantic) acceptability of the given construction. Exclamation mark is used for
constructions that are acceptable in specific contexts.
The organization of the paper is as follows. In Section 2, I discuss the notion
of Direction, as it is relevant to the present study, and I will also discuss the fea-
tures under scrutiny in the subsequent sections at a general level. Section 3, which
constitutes the bulk of the paper, examines the coding of Direction in detail. All of
the relevant features will be examined in light of actual data. Section 4 summarizes
the paper and Section 5 discusses the theoretical implications of the paper and its
relevance to our understanding of markedness and coding of Direction.

2. Defining Direction

2.1 Preliminaries
Before proceeding to discussing the linguistic coding of Direction, it is of the
utmost importance to define the examined notion. This is especially relevant for
 Seppo Kittilä

Direction, since the more traditional label Goal has been given many definitions
in earlier studies. It is important to note that despite the differences discussed
below, the basic definition applies to all the examined instances of Direction.
We are thus not dealing with differences between semantic roles, but rather
with a semantic role complex comprising an array of similar, yet subtly different
manifestations.
Basically, Direction comprises in this paper any linguistic element that refers
to the endpoint of a given act of motion including both intransitive (non-caused)
and transitive (caused) motion. Moreover, it is important that we are dealing with
an endpoint of motion only, all instances that involve change of possession are
excluded from further discussion. The same applies to clausal constituents refer-
ring to static location. Basically, this means that cases such as ‘the parent gave a
book to the child’ and ‘the book is/lies on the table’ are not relevant to this study.
The role of Direction can be formally represented as below (the definition is largely
from Kittilä & Ylikoski 2011, see Rappaport Hovav & Levin 2008 for a different
kind of definition):

Direction [+endpoint] [-possession] [± coincidence]

The notion therefore comprises instances, such as those in (2) (Direction in bold
face):

(2) a. Mary ran to the school.

b. Bill sent a letter to Leipzig.
c. Laura put the book on the table.
d. Mary went to Bill.

All the instances of Direction exemplified in (2) have the feature [endpoint] in
common. However, they differ in other respects. For example, in (2a), the Direc-
tion is inanimate, while in (2d) the role of Direction is borne by an animate entity.
As such, the denoted act of motion is very similar in both cases. The features of
Direction considered in this paper include animacy of Direction and Theme, (the
lack of) coincidence, dimensionality, size and the nature of linguistic reference to
Direction (e.g. common vs. proper nouns). The chosen features are not random,
but all of these contribute to the nature of Direction (see, e.g. Comrie 1986). For
example, certain inanimate entities are more typical landmarks due to their stable
nature and it is natural to describe our motion with regard to them. Typical exam-
ples include buildings and natural monuments. Other inanimate entities, such as
smaller artifacts and also vehicles, in turn, are less stable entities. All of these fea-
tures are discussed briefly in what follows. A more detailed discussion follows in
subsequent sections.
 The (non-)prototypicality of Direction 

2.2 Features relevant to the discussion in this paper

First, different manifestations of Direction can be distinguished on the basis of
animacy. As noted above, inanimate entities in general can with reason be claimed
to represent the prototype of Direction, since inanimate entities are stable enti-
ties incapable of changing their location independently. It is therefore rational
to describe motion and location of other entities in relation to these stable land-
marks. Animate entities, for their part, are mobile and active participants, which
makes them typical figures and consequently less prototypical landmarks.
The animacy of the Theme also has consequences for Direction. First of all,
animacy of the Theme usually excludes a recipient reading of a Direction argu-
ment with caused motion verbs. This is not directly relevant to the discussion in
this paper, but this is central to studies on Direction/Recipient coding in general
(see e.g. Kittilä & Ylikoski 2011). Second, the animacy of the Theme has conse-
quences for the coding of Direction especially with verbs such as ‘send’, where cer-
tain kinds of Direction appear as rather odd with inanimate Themes. For example,
cases such as ‘the child sent the doll to the table’ sound odd. In this paper, all
potential combinations of animacy of Theme and Direction will be discussed in
detail. Cases where the animacy of the Theme and the Direction coincide present
the most marked scenarios, even though cases such as ‘the king sent an ambassa-
dor to the president’ are completely natural.
Second, the number of dimensions of the landmark has consequences for
the nature of Direction. Referents of the Direction role can be two- or three-­
dimensional, as shown in (3):
(3) a. Mary put the book on the table/floor. (two-dimensional, flat surface)
b. Mary put the book into the box. (three-dimensional)
c. Mary went to the house. (three-dimensional)
For two-dimensional entities/flat surfaces, such as table and floor, the most natural
reading as a Direction is an ‘on(to)’-reading, an ‘into’-reading is clearly less natural.
On the other hand, the variety of potential readings is greater for three-­dimensional
entities. Three-dimensional entities allow also ‘into’- and ‘to the vicinity of ’-read-
ings quite naturally. These differences are relevant to the coding of Direction in
Finnish (as they probably are to some extent in all languages).
Third, physically sizeable entities (such as buildings, rocks, natural monu-
ments) are salient to us, and they are also therefore better landmarks for cod-
ing motion than significantly less sizeable entities (such as fruits, smaller artifacts
(pens, scissors) and flowers). In other words, examples in (4a) are more natural
than those in (4b):
(4) a. Mary went to the house/waterfall.
b. ??Mary went to the pen/banana/rose.
 Seppo Kittilä

In (4b), the figure is considerably bigger than the ground, which renders these
examples less natural than those in (4a). Examples in (4b) are thus conceptually
less salient, which contributes directly to their linguistic coding. Boundedness is
also relevant in this respect. Entities with salient boundaries are better landmarks
than entities with less clear boundaries.
The fourth feature considered in this paper is illustrated by coincidence. Coin-
cidence means here whether or not the location of figure and ground coincide
after the event has been successfully completed. There is coincidence in cases such
as ‘John went to London’ or ‘John put the book on the table’, while coincidence is
lacking for John and house in ‘John went to the vicinity of the house’, even though
there is coincidence for John and the vicinity of the house. Typically we seem to
focus on the result of events in their description, deviations from this need to
be highlighted somehow linguistically (see also, e.g. Verbeke 2011 for a thorough
discussion on argument marking splits based on tense and aspect). For example,
sentences such as John built the house mean that the house was finished, while
John was building a house means that the house was not necessarily finished. The
same applies to direction: John went to the lamppost means that the landmark was
reached, while deviations from this need to be explicitly marked, e.g. John went
towards/to the vicinity of the lamppost. One reason for the explicit coding of vicinal
directions (landmarks not reached) is probably found in the fact that motion to
the vicinity of a landmark cannot always be inferred directly from the inherent
features of the given landmark, which makes it necessary to highlight this lin-
guistically (with animates, the expected, and usually the only possible reading is,
however, ‘to the vicinity of ’).
Finally, landmarks may be referred to in a variety of ways linguistically with-
out any necessary changes in the referent itself. An important feature in this regard
is represented by the differences between proper and common nouns (e.g. Leipzig
vs. city, man vs. Barack Obama). As has been noted, for example, by Comrie
(1986) and Creissels and Mounole (2011), place names tend to behave differently
from common nouns when they occur as Directions/Locations. In contrast to the
features discussed previously, the differences between proper and common nouns
cannot be explained by referring to non-linguistic factors. Toponyms, as a type of
noun are, as expected, more typical directions than common nouns. Also certain
common nouns typically function as Directions/Locations, a typical example is
provided by ‘home’.
Before proceeding to discussing the linguistic coding of Direction itself, two
more clarifications are in order. First, I have not distinguished between different
instances of Direction based on the transitivity of the denoted act of motion. This
means that I will not make an explicit distinction between cases such as ‘The child
 The (non-)prototypicality of Direction 

went to the house’ and ‘The parent sent the child to the house’. Second, as noted
above, different instances of Direction will not be distinguished based on the exact
resulting location of the denoted motion, but the focus is on the most evident
and non-controversial instances of Direction. In more concrete terms, this means
that cases such as ‘I put the book behind/under/above the table’ lie outside the
scope of this paper, but only ‘I put the book on the table’ is considered. The latter
is regarded as more basic in this context, because these readings are possible with
the discussed cases, while readings such as ‘behind/under/above’ are not, but for
them a semantically more specific adposition is always needed. On the other hand,
the readings ‘in/into’ can be expressed both with cases and adpositions, which
makes variation between them relevant to the goals of this paper. Moreover, with
‘behind/under/above’, the potential differences in coding may be caused by fea-
tures others than those scrutinized, whence they are excluded from further discus-
sion in this paper.

3. Coding of Direction in Finnish

3.1 Preliminaries
I begin this section by briefly discussing the linguistic mechanisms employed for
Direction coding (see, e.g. Lauerma 1990 and Huumo & Ojutkangas 2006 for a
more detailed discussion of the Finnish case system and spatial cases). The mecha-
nisms considered are allative and illative cases and postpositions used for express-
ing similar meanings. This will be followed by a thorough discussion of Direction
coding in Finnish. The goal of this section is to examine which of the discussed
instances of Direction are coded by morphological case, and where Finnish resorts
to adpositional coding. As noted above, coding by case is seen as the unmarked
option, while adpositional marking is seen as marked. Consequently, the goal of
this section is also to shed more light on the notion of markedness.
The means of coding considered in the discussion include the following:

(5) Allative case

a. Mari men-i talo-lle
Mary go-3sg.pst house-all
‘Mary went to (the vicinity of) the house.’
Illative case
b. pallo vier-i laatikko-on
ball roll-3sg.pst box-ill
‘The ball rolled into the box.’
 Seppo Kittilä

Postpositions
c. Mari tul-i opettaja-n luo/viereen
Mary come-3sg.pst teacher-gen to/to(beside)
‘Mary came to the teacher.’
d. Mari laitto-i kirja-n pöydä-n päälle
Mary put-3sg.pst book-acc table-gen on(to)
‘Mary put the book on the table.’
e. Mari laitto-i kirja-n laatiko-n sisälle/sisään
Mary put-3sg.pst book-acc box-gen inside
‘Mary put the book into/inside the box.’

As seen in (5), Finnish uses two cases and an array of postpositions for coding
Direction (only the ones illustrated in (5) will be considered in this paper). Alla-
tive is one of the external local cases of Finnish, used for denoting motion to the
vicinity of or on(to) the top of a landmark, depending, for example, on the num-
ber of dimensions of the landmark. Allative is also used for coding Recipient and
certain instances of Beneficiary (see e.g. Kittilä 2006, 2010 for a more detailed
discussion of these). Illative is the directional internal local case of Finnish. It is
used for more concrete instances of Direction, basically for coding motion into
a landmark. Due to its more concrete nature, illative is usually infelicitous with
human landmarks, when the reference is to a Direction (in addition, it is also used
with animate participants with certain verbs, such as kyllästyä ‘get bored’).
As many other languages, Finnish has a rich inventory of adpositions, some of
which are illustrated in (5c–e). As these examples show, adpositions can be used
for coding the same meanings that can be coded also by cases. There are, however,
differences between cases and adpositions, as will be demonstrated below. As a
generalization, adpositions code semantically more specific instances of Direc-
tion. We should, however, note that under specific conditions, cases and adposi-
tons may code identical meanings. An example is given in (6):
(6) Aino laitto-i kirja-n pöydä-lle /pöydä-n päälle
Aino put-3sg.pst book-acc table-all /table-gen on
‘Aino put the book on the table.’

In the remainder of the paper, the focus lies on the functional differences between
cases and adpositions, and cases such as those in (6) will be used only as a basis of
comparison.
In the subsequent sections, I will illustrate the linguistic coding of all the
instances of Direction discussed briefly in Section 2 in light of actual linguistic
data. The goal is to show that Direction is not a uniform notion, but the cod-
ing varies according to the features of the landmark. Moreover, the discussion
will show that the attested changes in coding can in most cases be explained by
 The (non-)prototypicality of Direction 

(non-)pro­totypicality; Finnish resorts to more elaborate marking if the intended

reading cannot be inferred non-linguistically, while less elaborate marking appears
whenever the reading is expected.

3.2 Effects of animacy

3.2.1 Preliminaries
In this subsection, animacy effects on the coding of Direction will be examined.
Both the animacy of the Direction itself and the Theme will be considered (see
also Haspelmath 2007 on the animacy of Theme). Consequently, we have four pos-
sible combinations (note that in 2 Mary is not seen as a recipient):

1. Inanimate Theme + Inanimate Direction

‘Bill sent a book to London’
2. Inanimate Theme + Animate Direction
‘Bill threw a ball to (the vicinity of) Mary’
3. Animate Theme + Inanimate Direction
‘Lisa sent Bill to London’
4. Animate Theme + Animate Direction
‘Lisa sent Bill to (the vicinity of) Mary (for example, to where Mary is
standing)’

The examples above illustrate the most basic instances of the different combina-
tions. There is a lot of variation, especially in the nature of the Direction with
regard to animacy, but this is not considered in this paper. Below, I will discuss the
different combinations noted briefly above in more detail in light of Finnish data.
The focus lies, as expected, on the coding of Direction, since Theme is invariably
coded by the accusative or the partitive regardless of animacy (for the sake of sim-
plicity, only accusative is used in the examples below). However, the animacy of
the Theme is taken into account below, because it has consequences for the coding
of Direction.

3.2.2 Inanimate Theme + Inanimate Direction

The combination of two inanimate objects2 is rather natural and examples are easy
to find, as shown in (7), where examples are given that illustrate canonical scenes
in which an inanimate Theme enters an inanimate Direction or its vicinity. Both
case forms and all the relevant postpositions are possible here. Consider:

. The label object is in this paper used to refer to Theme and Direction arguments when-
ever it is not necessary to explicitly distinguish between them.
 Seppo Kittilä

(7) a. Ville lähett-i kirja-n Leipzig-iin

Ville send-3sg.pst book-acc Leipzig-ill
‘Ville sent the book to Leipzig.’
b. Ville laitto-i /heitt-i kirja-n pöydä-lle/
Ville put-3sg.pst /throw-3sg.pst book-acc table-all/
pöydä-n päälle
table-gen on
‘Ville put/threw the book on the table.’
c. Ville potkais-i pallo-n laatikko-on /laatiko-n viereen/luo
Ville kick-3sg.pst ball-acc box-ill /box-gen to(beside)/to
‘Ville kicked the ball into the box/to the vicinity of the box.’

Both adpositions and cases can be productively used for coding Direction with
this combination. Under certain conditions, as in (7b), cases and adpositions code
very similar meanings. This is rather frequent with inanimate Directions. The only
important difference between case and adposition in (7b) lies in specificity. In (7c),
case and adposition code different meanings, which follows from the semantic dif-
ferences between the employed elements. Also here, the case form can be used to
code the more expected meaning, since boxes are typically used for storing other
entities, which makes the ‘into’-reading more natural.

3.2.3 Inanimate Theme + Animate Direction

The combination of inanimate Theme and animate Direction is illustrated in (8).
Examples in (8) differ from those in (7) in that there is more variation in the roles
borne by animate Directions, which follows from the inherent nature of animate
Directions:
(8) a. Ville lähett-i kirje-en Kalle-lle
Ville send-3sg.pst letter-acc Kalle-all
‘Ville sent Kalle a letter.’
b. Ville heitt-i pallo-n Kalle-lle
Ville throw-3sg.pst ball-acc Kalle-all
‘Ville threw a ball to Kalle/Kalle a ball.’
c. Ville heitt-i pallo-n Kalle-n luo/viereen
Ville throw-3sg.pst ball-acc Kalle-gen to/beside
‘Ville threw the ball to the vicinity of/near Kalle. (not: threw Kalle
a ball)’

(8a) is a very typical example of sending; an inanimate Theme is transferred to an

animate Direction. Differently from, for Example (7a), Kalle can be viewed as a
Recipient in (8a). However, this reading is only strongly implied; ‘send’ n
­ ecessarily
 The (non-)prototypicality of Direction 

involves a Direction argument, and a Recipient only under certain conditions

(typically only with animate entities). (8b) also describes caused motion involv-
ing an animate Direction. As with ‘send’, allative can code an animate Direction.
However, allative marking of animate Directions is very strongly associated with
the role of Recipient in these cases. (8b) is felicitous only if Kalle is seen as a Recipi-
ent, i.e. Kalle receives/catches the ball successfully. A Direction only-reading is not
possible in (8b). For evoking this reading, adpositions, as in (8c) must be used. In
(8c), Recipient-reading is excluded and Kalle functions necessarily as a (vicinal)
Direction.
The use of allative case and adpositions for coding animate Directions is
expected, because allative case and adpositions, such as luo and viereen (both
meaning ‘to’), code motion to the vicinity of a landmark, which is the natural read-
ing with animate Directions. On the other hand, the use of internal local cases and
adpositions with similar meanings yield less felicitous constructions, as shown
in (9) with the illative case:
(9) a. *Ville lähett-i kirje-en Kalle-en
 Ville send-3sg.pst letter-acc Kalle-ill
‘Ville sent a letter in(to) Kalle.’
b. Ville heitt-i !tika-n /!karki-n/??pallo-n Kalle-en
Ville throw-3sg.pst   dart-acc/candy-acc/ball-acc Kalle-ill
‘Ville threw a dart/candy/??ball in(to) Kalle.’
Example (9a) can be seen as ungrammatical. The illative codes motion into some-
thing, a reading that is excluded for animate participants with ‘send’. The examples
in (9b) are also rather odd. However, they are possible in specific, rather infre-
quently occurring, contexts. Darts and candies allow, due to their physical size,
to be thrown to/into animate participants, which makes the two first instances of
(9b) marginally acceptable. On the other hand, a ball is too sizeable an entity to be
thrown into an animate participant, which makes the sentence Ville heitti pallon
Kalleen semantically odd, but we can imagine a scene where this example would
be possible.

3.2.4 Animate Theme + Inanimate Direction

A canonical example of this combination is given in (10):
(10) Ville lähett-i Kalle-n Oulu-un/eduskunta-an/talo-lle
Ville send-3sg.pst Kalle-acc Oulu-ill/Parliament-ill/house-all
’Ville sent Kalle to Oulu/the Parliament/the house.’
The examples in (10) denote scenes where an Agent causes an animate participant
to change his/her location. These kinds of sending events are very natural and
 Seppo Kittilä

cases are thus very productively used for coding them. Differences between the
Directions illustrated above arise with adpositions. Consider:
(11) Ville lähett-i Kalle-n
Ville send-3sg.pst Kalle-acc
*Oulu-n/??eduskunna-n/talo-n luo
   Oulu-gen/Parliament-gen/house-gen to
‘Ville sent Kalle to *Oulu/??Parliament/house.’

As shown above, luo is not possible with place names, it is marginal with Par-
liament, while house allows the postposition for coding (vicinal) Direction. One
thing that could make a contribution here is boundedness. Houses are clearly
bounded entities and we can thus easily conceptualize their boundaries. Moreover,
houses are not seen as institutions, while other nouns referring to buildings (e.g.
theatre, school, post) can, which makes them less typical examples of bounded
entities. On the other hand, Oulu is a place name, and the boundaries are less
salient to us, while Parliament can in (11) be seen as a group of people, which
makes this entity less bounded.
Another major group of caused motion is illustrated by events of ballistic
motion, such as ‘throw’, ‘toss’ and ‘kick’ and pure placement verbs, such as ‘put’.
Examples follow:
(12) a. Ville heitt-i/laitto-i Kalle-n
Ville throw-3sg.pst/put-3sg.pst Kalle-acc
!pöydä-lle/!laatikko-on!Oulu-un
table-all/box-ill/Oulu-ill
‘Ville threw/put Kalle on the table/into the box/to Oulu.’
b. Ville heitt-i/laitto-i Kalle-n pöydä-n päälle/viereen
Ville throw-3sg.pst Kalle-acc table-gen on/beside
‘Ville threw/put Kalle on the table/beside the table.’

As such, examples in (12) are grammatical constructions. It is possible to imag-

ine cases in which animate entities are caused to move through ballistic motion.
However, these scenes are rare, and in (12) humans are treated similarly to inani-
mate entities in that they are placed on other entities similarly to inanimate enti-
ties (cf. ‘Ville put a book on the table’). This renders these examples clearly less
natural than those in (11), which imply active participation of the animate Theme
(i.e. if Kalle is sent to Oulu, he is actively doing something himself). Moreover,
humans usually lack the ability to cause other humans to change location through
ballistic motion, which also contributes to the unnaturalness of (12). They would,
however, be felicitous in tales about heroes with supernatural powers, or a sen-
tence like Ville laittoi Kallen pöydän päälle in (12b) would be normal, if Ville is
a father, and Kalle a baby. The differences between cases and adpositions are not
 The (non-)prototypicality of Direction 

very relevant in (12), because the examples are semantically odd regardless of the
coding of Direction.

3.2.5 Animate Theme + Animate Direction

The last possible combination discussed here is illustrated by cases in which both
the Theme and the Direction are animate. This can be said to represent the most
marked combination, since, as was noted above, animate participants are less
common as both Themes and Directions. Some examples are given in (13) that
clearly illustrate variation in grammaticality and readings with different cases and
postpositions.
(13) a. Opettaja lähett-i/!heitt-i oppilaa-n rehtori-lle
Teacher sent-3sg.pst/!throw-3sg.pst pupil-acc principal-all
‘The teacher sent/threw the pupil to the principal.’
b. Opettaja lähetti/!heitti oppilaa-n rehtori-n luo
Teacher sent/threw pupil-acc principal-gen to
‘The teacher sent/threw the pupil to the principal.’
c. Isä laitto-i lapse-n ??äidi-lle
Father put-3sg.pst child-acc    mother-all
/äidi-n päälle
/mother-gen on
‘The father put the child on the mother.’
d. *Ville laittoi/heitti/lähetti Liisa-n Kalle-en
    Ville put/threw/sent Liisa-acc Kalle-ill
(For: Ville put/threw/sent Liisa to Kalle.)
There are clear differences in the readings of case forms and adpositions in (13).
The most natural reading of (13a) is that the teacher sent the pupil to the principal’s
office, i.e. the Direction is not genuinely animate here. In order to assure the animate
Direction reading, a postposition must be used, as in (13b). The verb ‘throw’ is odd in
(13a–b) in its literal meaning. However, the verb also has the meaning ‘drive some-
one to someone/somewhere’. This reading is possible in (13a–b). Verbs of placement
differ from verbs of sending and ballistic motion in that they cannot govern allative
case with human Directions. Therefore postpositions must be employed for stress-
ing Direction readings. This probably follows from the fact that ‘to someone’s place’
reading is not possible here and also readings involving a Recipient are excluded.
Finally, Example (13d) is best considered ungrammatical. The reason for this is the
incompatibility of the semantics of illative with animate Directions: animate partici-
pants do not enter other animate participants as a result of caused motion.3

. A possible scenario where this would be possible has been provided for me by the series
editor from the movie Star Trek 2: The Wrath of Khan.
 Seppo Kittilä

3.3 Dimensionality
Entities in the non-linguistic world can be either two- (e.g. table, wall, floor)
or three-dimensional (house, box, natural monuments). Certain entities, such
as ‘table’ and ‘chair’, can be conceptualized as either two- or three-dimensional
depending on context. The effects of dimensionality on Direction coding are dis-
cussed below. I will first discuss two-dimensional landmarks, which is followed
by an examination of three-dimensional entities. Only inanimate entities are
taken into consideration for making it possible to focus on the effects of dimen-
sionality only.

3.3.1 Two-dimensional entities

The case forms discussed in this paper display evident differences with two-dimen-
sional entities. Illative is acceptable only in a limited set of cases, and illative coding
often produces marginal, or even ungrammatical constructions, as shown in (14):
(14) a. Ville !laitto-i/!heitt-i/*lähett-i kirja-n pöytä-än
Ville put-3sg.pst/throw-3sg.pst/send-3sg.pst book-acc table-ill
‘Ville !put/?threw/*sent a book to the table.’
b. Ville ??laittoi/!heitti/*lähetti kirja-n katto-on/ lattia-an
Ville     put/threw/sent book-acc roof-ill/ floor-ill
‘Ville ??put/!threw/*sent a book to the roof/floor.’

The infelicity of the illative with two-dimensional entities is not unduly surprising,
because they lack a natural ‘into’-reading. However, ‘table’, ‘roof ’ and ‘floor’ differ
in the degree of acceptability with the illative. None of them accept the verb ‘send’.
Table allows both ‘put’ and ‘throw’, but only in a specific meaning. The sentences
Ville laittoi/heitti kirjan pöytään do not describe mere instances of putting or send-
ing, but they both approximate the meaning ‘Ville put the book on display for a
specific purpose’. (14a) cannot mean ‘Ville put/threw the book on the table’ only.
With ‘roof ’, ‘put’ is odd, but ‘throw’ is again possible in a specific reading. (14b)
with ‘throw’ is possible if Ville throws a book at the ceiling and the book gets stuck
to the ceiling. With floor, ‘throw’ is again possible, but only in the meaning ‘Ville
threw the book to the floor as a sign of his anger’. Mere throwing without any fur-
ther implications would require an allative.
Allative coding of two-dimensional entities is illustrated in (15):
(15) Ville laitto-i/heitt-i/*lähett-i kirja-n
Ville put-3sg.pst/throw-3sg.pst/send-3sg.pst book-acc
pöydä-lle/kato-lle/lattia-lle
table-all/roof-all/floor-all
‘Ville put/threw/*sent a book on the table/roof/floor.’
 The (non-)prototypicality of Direction 

As shown in (15), verb ‘send’ is not possible with two-dimensional entities (which is
more due to inanimacy than the number of dimensions), but both ‘put’ and ‘throw’
appear productively with two-dimensional entities and the allative case. This is eas-
ily accounted for, since the expected reading of the allative is ‘on top of ’ with two-
dimensional entities, which is a natural reading for two-dimensional landmarks.
Last, adpositional coding with two-dimensional referents is illustrated in (16):
(16) a. Ville laitto-i/heitt-i kirja-n
Ville put-3sg.pst/throw-3sg.pst book-acc
pöydä-n/kato-n/?lattia-n päälle
table-gen/roof-gen/floor-gen on
‘Ville put/threw the book on the table/roof/?floor.’
b. Ville laittoi/heitti kirjan pöydä-n/??kato-n/??lattia-n viereen/luo
Ville table-gen/roof-gen/floor-gen to
‘Ville put/threw the book to/beside the table/??roof/??floor.’
c. *Ville laittoi/heitti kirjan pöydän/katon/?lattian sisään/sisälle
 Ville into
(For: Ville put/threw the book inside the table/roof/floor.)

As expected, päälle ‘on’ in (16a) produces grammatical constructions with two-

dimensional entities; two-dimensional entities consist of a flat surface, on which
the theme is transferred. Semantically, (16a) is very close to (15a). Only ‘floor’
is somewhat odd in (15a), since an ‘onto’-reading is the only possible reading
for floor, which makes päälle perhaps semantically too specific. Other postposi-
tions produce less acceptable, or even ungrammatical constructions, as shown in
(16b–c). Only Ville heitti kirjan pöydän viereen/luo ‘Ville threw a book to the vicin-
ity of/beside the table’ is possible if ‘table’ is conceptualized as a three-dimensional
entity. Two-dimensional entities are not natural landmarks for ‘to the vicinity of ’ –
or ‘into’-readings due to their physical properties.

3.3.2 Three-dimensional entities

The world we live in consists mostly of three-dimensional entities. Animate entities
are (practically) all three-dimensional as are many natural monuments, plants and
buildings. As expected, the higher number of dimensions has consequences for
the linguistic coding of three-dimensional entities. In contrast to two-­dimensional
entities, both illative and allative cases along with an array of postpositions are
possible, but with significant differences, as will become clear below.
Illative coding of three-dimensional referents is exemplified in (17), where
different three-dimensional entities and the use of the illative case for their coding
is illustrated. The entities in (17) differ according to whether they can be seen as
natural containers or not. Consider:
 Seppo Kittilä

(17) a. Ville heitt-i/laitto-i kirjan laatikko-on

Ville throw-3sg.pst/put-3sg.pst book-acc box-ill
‘Ville threw/put/*sent a book to the box.’
b. Ville lähett-i Kalle-n talo-on
Ville send-3sg.pst Kalle-acc house-ill
‘Ville sent Kalle (in)to the house.’
c. Ville heitti tika-n kive-en/puu-hun
Ville threw dart-acc rock-ill/tree-ill
‘Ville threw the dart into the rock/tree.’

Box is a three-dimensional entity with an interior, which makes the examples in

(17a) very natural. The verb ‘send’ is possible if the Theme is animate, as in (17b),
where an animate participant enters a three-dimensional entity. In (17c), the coin-
cidence is only partial. For example, Ville heitti tikan puuhun means that a part of
the dart enters the tree, but the coincidence is not complete as in (17a–b). Despite
this minor difference, examples in (17c) are completely normal, and the intended
reading is best expressed by the illative.
In (18), in turn, the use of the allative case with three-dimensional entities is
examined:

(18) a. Ville heitt-i/laitto-i/*lähett-i kirjan laatiko-lle

Ville throw-3sg.pst/put-3sg.pst/send-3sg.pst book-acc box-all
‘Ville threw/put/*sent a book to the box.’
b. Ville lähett-i Kalle-n talo-lle
Ville send-3sg.pst Kalle-acc house-all
‘Ville sent Kalle (in)to the house.’
c. Ville heitti pallo-n kive-lle/puu-lle
Ville threw ball-acc rock-all/tree-all
‘Ville threw the ball into the rock/tree.’

The main difference between (17) and (18) lies in the case form used for coding
Direction. In (18), the most natural reading for all sentences is ‘to the vicinity of X’.
Similarly to (17), all the examples are felicitous. Only ‘send’ is not allowed in (18).
Finally, the examples in (19) illustrate the use of postpositions with three-
dimensional entities. The differences between (17) and (18) become more evident
with postpositions:
(19) a. Ville heitt-i/laitto-i kirjan laatiko-n
Ville throw-3sg.pst/put-3sg.pst book-acc box-gen
päälle/sisään/luo
on/into/to
‘Ville threw/put a book on top of/into/to the box.’
 The (non-)prototypicality of Direction 

b. Ville lähett-i Kalle-n talo-n päälle/sisään/luo

Ville send-3sg.pst Kalle-acc house-gen on/into/to
‘Ville sent Kalle on top of/into the house.’
c. Ville heitti tika-n kive-n/puu-n päälle/sisään/luo
Ville threw dart-acc rock-gen/tree-gen on/into/to
‘Ville threw the dart on top of/into the rock/tree.’

The examples in (19) are all felicitous. Of the postpositions discussed here, sisään
‘into’ can be seen as the semantically more specific counterpart of the illative case,
while päälle and luo are semantically closer to the allative case. With illative and
sisään, we are primarily dealing with a difference in specificity. For example, both
Ville laittoi kirjan laatikkoon and Ville laittoi kirjan laatikon sisään (19a) express
very similar meanings. However, the differences may also be more significant, as
with Ville heitti tikan puuhun/puun sisään, where the postposition sisään stresses
the fact that the whole dart entered the tree (see Ojutkangas 2008 and Huumo &
Ojutkangas 2010 for a more detailed discussion). With päälle/luo and the alla-
tive case, the differences are generally more substantial. Animate entities are less
than ideal Directions, which has the consequence that special mechanisms are
needed for assuring the intended reading. Examples in (19b) and (19c) manifest
this directly: the postpositions päälle and luo have the function of stressing the
Direction function. With päälle, the reading is invariably ‘on top of ’, while luo
codes vicinal Directions. Semantically specific postpositions are needed, since the
intended reading deviates from the expected one.

3.4 Physical size

Salience makes an important contribution to how naturally an entity is perceived
as a landmark. Physical size is one of the central features of salience. It is rational
to describe motion/transfer between two entities using physically large entities as
landmarks. Because it is easier to conceptualize motion between sizeable landmarks,
constructions, such as those in (20), with sizeable entities are completely normal:
(20) a. Ville men-i talo-lle /metsä-än /Helsinki-in /kive-lle
Ville go-3sg.pst house-all /forest-ill /Helsinki-ill /rock-all
‘Ville went to the house/forest/Helsinki/rock.’
b. Ville heitt-i pallo-n kivelle/metsään /tynnyri-in
Ville throw-3sg.pst ball-acc /barrel-ill
‘Ville threw the ball to the rock/forest/into the barrel.’

In (20), the entities functioning as Direction are all relatively sizeable entities,
which makes it easy to conceptualize them as endpoints of motion. Moreover, in
(20), Theme is physically smaller than Direction. In (20a–b), Direction is coded
 Seppo Kittilä

by case, but adpositional coding would also be possible, with minor semantic
consequences.
In (21), in turn, the role of Direction is borne by physically smaller enti-
ties, which has evident consequences for the acceptability of the constructions in
question:
(21) a. Ville men-i !omena-lle /??saksi-lle /??nuppineula-lle
Ville go-3sg-pst apple-all /scissors-all /pin-all
‘Ville went to the apple/scissors/pin.’
b. ??Ville heitt-i karki-n
 Ville throw-3sg.pst candy-acc
  omenalle/saksille/nuppineulalle
‘Ville threw a piece of candy on/to the apple/scissors/pin.’

The examples in (21) are all semantically odd, which follows from the non-salient
nature of the landmark. Ville meni omenalle (21a) is possible in the meaning ‘Ville
went for an apple’, but not in a normal Direction-reading. In (21b), the reading
that first comes to mind is a Recipient-reading, but this is excluded by the inani-
macy of the Direction.
In (22), the case forms of (21) have been replaced with adpositions:
(22) a. ?Ville men-i omena-n /saksi-en
   Ville go-3sg.pst apple-gen /scissors-gen
/nuppineula-n luo/päälle
/pin-gen to/on top of
‘Ville went to/on the apple/scissors/pin.’
b. Ville heitt-i karki-n
Ville throw-3sg.pst candy-acc
omenan/saksien/nuppineulan luo/päälle
‘Ville threw a piece of candy to/on the apple/scissors/pin.’

The examples in (22) sound more natural than those in (21). Because the only
relevant difference between (21) and (22) is the coding of Direction, we can
conclude that semantically specific postpositions more naturally describe scenes
in which the landmark is physically small, and thus a less salient entity. We may
also add that smaller entities are also less stable landmarks (their location is eas-
ily modified by animate entities), which may also be relevant to their linguistic
coding. Semantically specific postpositions make them marginally accepted as
Directions.
In (22b), the transferred entity is smaller than the landmark, which makes
the given constructions more acceptable than those in (22a). The importance of
 The (non-)prototypicality of Direction 

physical size can further be underlined by manipulating the size of the Agent.
Consider (23):
(23) Hiiri juoks-i omena-n /saksi-en
Mouse run-3sg.pst apple-gen /scissors-gen
/nuppineula-n luo/päälle
/pin-gen to/on top of
‘A/the mouse ran to/on the apple/scissors/pin.’

The moving figure is considerably smaller in (23) than in (22), which has con-
sequences for the acceptability of the construction. It is also important to note
that also in (23) coding by cases is marginal, which further underlines the dif-
ferences between cases and adpositions as regards markedness. In (22) and (23),
this means that adpositions appear for stressing the fact that an expected Theme
functions as Direction. Adpositions are thus used very much in the same fashion
as they are used for stressing the unexpected nature of Direction with animate
participants.

3.5 Coincidence
Gallons of ink have been spilled discussing different facets of Goal/Direction
marking (see e.g. Blansitt 1988; Creissels 2008 and Kittilä 2008 among many oth-
ers). It is common for these studies that they focus on Direction as endpoints of
motion/transfer that the Theme successfully reaches. The focus has this been on
cases such as ‘Bill went to London’ or ‘Jane sent Bill to London’. Less attention
has been paid to cases where the motion/transfer is only to the vicinity of a land-
mark, as in ‘Bill went to the (vicinity of)/towards the lamppost’. An exception is
provided by Kittilä and Ylikoski (2011), who focus on what they have labeled as
‘Vicinal Goal’ in their study. Kittilä and Ylikoski show that European Uralic lan-
guages resort to adpositional coding for vicinal Direction (typically borne by ani-
mate participants) instead of case forms used for Direction and Recipients. One
of the reasons they give for this is markedness; animate participants are atypical
Directions, and the location of Theme and Direction does not coincide after the
event has been completed.
Further examples relevant to the discussion in this context are illustrated in
(24). In these examples, postpositions and case forms differ drastically according
to the readings they receive depending on the nature of the landmark:
(24) a. Ville meni talo-on /talo-lle /talo-n luo
Ville go-3sg.pst house-ill /house-all /house-gen to
‘Ville went into the house/to (the vicinity of) the house.’
 Seppo Kittilä

b. Ville heitt-i pallo-n pöydä-lle /pöydä-n luo luo

Ville throw-3sg.pst ball-acc table-all /table-gen to
‘Ville threw the ball on the table/to the table.’
c. Ville heitti pallon kaivo-on /kaivo-lle
Ville throw-3sg.pst ball-acc well-ill /well-all
kaivo-n luo
well-gen to
‘Ville threw the ball into the well/to (the vicinity of) the well.’

The examples in (24) illustrate different kinds of Direction. In (24a), the differ-
ences between allative case and the postposition luo ‘to’ are not significant, and
both mean ‘Ville went to the house’. The illative, in turn, implies that the house
was entered. In (24b), the differences between the allative and the adposition luo
are more obvious. With the allative, the clause means ‘Ville threw the ball on the
table’, i.e. the location of the ball and of the table coincide after the event. The read-
ing ‘Ville threw the ball to the vicinity of the table’ is theoretically possible, but
unlikely with the allative. For emphasizing the lack of coincidence, luo must be
used. With ‘well’, the differences between allative and illative cases are clear; illative
implies coincidence, the allative lack of it. Similarly to ‘house’, the semantic differ-
ences between allative and luo are less significant.
In (24), the potential differences between the employed mechanisms are due
to coincidence to the lack of it. The semantic differences between cases and adpo-
sitions are more evident with human Directions. This is illustrated by the examples
in (25).
(25) a. Opettaja lähett-i oppilaa-n rehtori-lle
Teacher send-3sg.pst pupil-acc principal-all
/rehtori-n luo
/principal-gen to
‘The teacher sent the pupil to the principal.’
b. Ville men-i Stina-lle /Stina-n luo
Ville go-3sg.pst Stina-all /Stina-gen to
‘Ville went to Stina/Stina’s place.’

In (25a), the allative evokes the reading ‘The teacher sent the pupil to the prin-
cipal’s office’. This reading is also possible with luo ‘to’, but luo may also have the
meaning ‘The teacher sent the pupil to the vicinity of the principal’, for example,
to where the principal is standing. In (25b), the differences between the allative
and luo are very similar to (25a). The main difference lies in the fact that in (25b),
the expected reading with the allative is ‘Ville went to Stina’s place’ instead of her
office. Similarly to (25a), luo can also have the meaning ‘go to Stina’s place’, and it
can also mean ‘Ville went to where Stina is standing’. The lack of coincidence is
 The (non-)prototypicality of Direction 

more obvious than in (24), because animate entities do not enter other animate
participants as a result of sending/going.
Examples in (25) are interesting also in light of the fact that animate Directions
are conceptualized as inanimate (see also Luraghi 2011). This is best accounted
for by the atypicality of animate Directions. In both (25a) and (25b), the animate
Direction can be metonymically viewed as a place where the given animate par-
ticipant resides. In (25a), the most natural interpretation is an office, since an ani-
mate participant is viewed through his/her profession. In (25b), in turn, the most
natural reading is Stina’s place, since a proper noun can be seen as a neutral way of
referring to the given participant. In both cases, the unexpected reading where the
Direction is indeed animate needs to be reinforced by using semantically specific
adpositions (which is typical with animate entities in general). This means that
also in (25) markedness plays an important role and in all the cases discussed in
this section, the lack of coincidence needs to be highlighted linguistically.

3.6 The nature of linguistic reference

Cases discussed in this section differ from those examined so far in that they do
not have a non-linguistic motivation, but they are due to the nature of linguistic
reference of an entity. As noted by Comrie (1986) and Creissels & Mounole (2011),
place names are among those entities that display deviant traits as Directions. Also
animacy/referential hierarchies contribute to the linguistic coding of Direction, as
has been shown by Haspelmath (2007). Both of these are discussed below.
First, some differences between nouns and pronouns are illustrated in (26):
(26) a. Oppilas tulee huomenna rehtori-lle
Pupil come.3sg.prs tomorrow principal-all
/rehtori-n luo
/principal-gen to
‘The pupil comes to the principal tomorrow.’
b. Oppilas tulee huomenna !minu-lle /minu-n luokse-ni
Pupil 1sg-all /1sg-gen to-1sg.poss
‘The pupil comes tomorrow to me.’

Examples in (26) illustrate the coding of nouns and pronouns as Direction. As dis-
cussed above, both case form and adposition are correct in (26a), but with seman-
tic differences. In (26b), in turn, there are evident differences between the allative
case and the adposition luo(kse) ‘to’. Allative is infelicitous in a Direction reading
in (26b), and the only plausible reading for oppilas tulee minulle is ‘this student
will be mine’, e.g. in a school where teachers are competing for the best students.
On the other hand, the use of luokse is completely normal and evokes a directional
reading.
 Seppo Kittilä

In (26), it was shown that pronouns are odd in the function of (vicinal) Direc-
tion. However, this applies only to singular pronouns. If the pronoun in (26b)
is replaced with a pronoun in the plural, the construction becomes felicitous, as
shown in (27):
(27) Oppilas tulee huomenna mei-lle /meidä-n luokse-mme
Pupil come.3sg.prs tomorrow 1pl-all /1pl-gen to-1pl.poss
‘The pupil comes tomorrow to me/us.’
With a first person plural pronoun, (27) means ‘the pupil comes to my/our place
tomorrow’. As the free translations show, the distinction between singular and plu-
ral references of me ‘we/us’ is neutralized with the allative case and (27) can mean
both ‘the pupil comes to my place’ and ‘the pupil comes to our place’. Also here,
adpositions are more specific and meidän luoksemme can only mean ‘to our place’.
Similarly to (26), only adposition can have a vicinal Direction reading in (27).
The differences between nouns and place names are illustrated in (28),
consider:
(28) a. Ville men-i kaupunki-in /kaupungille /*kaupungi-n luo
Ville go-3sg.pst town-ill /town-all /town-gen to
‘Ville went downtown/to the town.’
b. Ville matkust-i ulkomai-lle /kotimaa-han-sa
Ville travel-3sg.pst abroad-all /home.country-ill-3poss
‘Ville traveled abroad/to his home country.’
c. Ville matkust-i Turkuun /*Turu-lle
Ville travel-3sg.pst Turku-ill /Turku-all
/Rovanieme-lle /*Rovanieme-en /Laukaa-lle /Laukaa-seen
/Rovaniemi-all / Rovaniemi-ill /Laukaa-all /Laukaa-ill
‘Ville traveled to Turku/Rovaniemi/Laukaa.’
d. Ville matkusti Saksa-an /*Saksa-lle
Ville travel-3sg.pst Germany-ill /Germany-all
/Venäjä-lle /*Venäjä-än /Kyprokse-lle /Kyprokse-en
/Russia-all /Russia-ill /Cyprus-all /Cyprus-ill
‘Ville traveled to Germany/Russia/Cyprus.’

Examples in (28a–b) illustrate the coding of Direction with common nouns. The
noun kaupunki allows both allative and illative cases, but with a clear semantic dif-
ference (‘go downtown’ vs. ‘to the town’), the postposition luo ‘to’ is not possible. In
(28b), the noun ‘country’ appears either in the allative or the illative depending on
the modifying noun. The illative is ungrammatical with ulkomaat, and the allative
with kotimaa. In sum, with common nouns there is semantically determined vari-
ation with Direction coding. (28c–d), in turn, exemplify the coding of t­ oponyms
 The (non-)prototypicality of Direction 

as Directions. As shown in (28c), with towns/municipalities, there is variation

between the allative and illative cases; Turku governs illative case, Rovamiemi alla-
tive case, while Laukaa can take both. A similar pattern is attested for countries.
The great majority of countries take the illative case, but Russia only the allative
case, while some countries, such as Cyprus (along with some other islands), can
take both in a rather free variation (even though speakers have preferences for
either). However, it is essential that the nature of variation is very different from
that attested for common nouns. Many common nouns may take both allative
and illative cases, and also an array of different adpositions, but the formal varia-
tion has a semantic basis. For example in (24a), both the relevant case forms and
the adposition luo, and the two cases produce clearly different readings; illative
refers to the entering of the house, while allative is used whenever the vicinity of
the house is meant. A similar analysis applies to (24c). Formally, similar variation
exists for place names, but the variation is restricted to form, it has no semantic
basis. It is not clear, why, e.g. Turku governs the illative case, while Laukaa allows
both. Moreover, most place names take either case, while the other one is ruled
out, that is, the semantic differences that are relevant to common nouns disappear.
This may follow, because coincidence is expected for toponyms, which renders the
distinction between allative and illative irrelevant. This is also apparent from the
fact that place names do not usually allow luo ‘to’. Toponyms thus allow variation
between allative and illative as a group, but the marking is idiosyncratic for indi-
vidual toponyms.
A further type of nouns that often receive a deviant formal treatment is pre-
sented by words like ‘home’ that (almost) exclusively express Direction (or loca-
tion in general). Consider:
(29) a. Ville men-i talo-lle /talo-on /talo-n luo
Ville go-3sg.pst house-all /house-ill /house-gen to
‘Ville went into the house/to the house.’
b. Ville men-i koti-in /*kodi-lle /??kodi-n luo
Ville go-3sg.pst home-ill /home-all      home-gen to
‘Ville went home.’

In principle, the sentences Ville meni taloon (29a) and Ville meni kotiin (29b) could
refer to the same situation, in which Ville enters a building. However, there are
clear differences in how the event is conceptualized. With kotiin ‘home’ a specific
house is meant, while with talo ‘house’, any house is a possible referent of the given
noun. Moreover, the nouns also differ according to whether the nature of Direc-
tion can be specified. With koti, allative case is not possible, and luo is very mar-
ginal at best. In the case of talo, in turn, the exact resulting location can be further
specified. These differences in the formal behavior of the two nouns are natural in
 Seppo Kittilä

light of their semantics. Kotiin refers generally to going home, and it is not relevant
which part of the entity is reached. On the other hand, with talo ‘house’ it is often
necessary to specify the endpoint, which makes the use of allative and illative case
along with an array of adpositions expected. The exact resulting location may not
be retrievable non-linguistically, and therefore linguistic means are necessary.
To summarize. Finnish does not employ any specific markers (such as dif-
ferent case forms) for coding Direction with toponyms nor are toponyms less,
or zero, marked, in this function. However, also in Finnish as a type of nouns
toponyms behave differently from common nouns that can basically refer to the
same locations as toponyms. First, semantically specific adpositions are usually
infelicitous with them. Second, the semantically-determined variation attested for
common nouns is lacking. Formally, the variation is basically identical, but it lacks
a semantic basis. It suffices to mark Direction in contrast to static Location or
Source, which is achieved by either of the two directional cases. There is no need
for additional specification of the endpoint. We may say that with place names we
are dealing with ‘reversed markedness’. Marking is less elaborate (semantic dis-
tinctions are neutralized), since specific marking is not functionally necessary.

4. Summary

The data discussed in this paper has shown that the formal variation in Direction
coding is not random, but it has a clear semantic basis. The generalization holds
that morphological cases are used for coding conceptually expected scenes, while
adpositions are used for assuring the intended reading whenever this is otherwise
less easily inferred. In other words, more elaborate coding is employed for disam-
biguation. Adpositional coding appears basically in one of the following cases (see
also Aristar 1997 for a similar remark):

1. The intended reading cannot be assured by using morphological case, which

renders the use of adposition necessary.
2. A certain reading cannot be expressed by a case form.
3. Language lacks a case form for expressing a certain type of Direction.

The first type is most relevant to the discussion in this paper and the type is best
illustrated in (18) and (19), where a case form (i.e. allative) is possible, but the
meaning intended (in this case ‘onto’) cannot be unambiguously expressed by the
given case form, and an adposition is needed for reinforcing the right reading. In
contrast to the two other cases, allative is completely normal and grammatical,
but the meaning expressed differs from that intended. Moreover, the allative can
 The (non-)prototypicality of Direction 

express the same function, but not with animate participants, and the role of Direc-
tion is coded more elaborately with animates. This type is attested also in (25),
where an adposition can be used for stressing the fact that ‘the place of X’ func-
tions as Direction, and not the animate participant him/herself. Examples of the
second type are provided by, e.g. (19c). This type differs from the previous one in
that Finnish completely lacks a case form for expressing the meaning in question.
Put another way, Finnish can express the meaning ‘onto’ by both the allative and
the adposition päälle, but the meaning ‘completely into’ cannot be expressed by a
case form with entities like ‘rock’ and ‘tree’ (see also Lauerma 1990: 128). Finally,
in the third type, a given language lacks a case form for expressing a given function
altogether. This type is not directly relevant to the discussion in this paper, since
cases like ‘the child threw the book behind the table’ were not discussed. However,
they should be mentioned, because in these cases as well more elaborate marking
has the function of assuring the intended meaning. More specific locations cannot
be inferred from the inherent nature of a landmark, which makes adpositional
coding necessary.

5. Theoretical implications

The findings of the paper contribute to our understanding of the differences

between cases and adpositons also at a more general level. Cases are more polyse-
mous also in languages like Finnish with a rich case inventory accompanied by a
large number of adpositions (the polysemy is naturally greater in languages with
smaller case inventories, see also Comrie 1986 for similar remarks). Case forms
depend on their host for their reading. This has the consequence that the use of
cases either produces ungrammaticality (as in (12d) and (21a-b)), or the intended
reading is not directly retrievable from the nature of the landmark (as in (8b) and
(13a)). For example, the allative case expresses in general ‘to the vicinity of/onto’
readings, which with animates translates into the marking of possession (posses-
sion can be conceptualized as a sphere of control in the vicinity of a landmark).
Consequently, languages need additional means for reinforcing the intended read-
ing. On the other hand, this kind of polysemy is not attested with adpositions,
because adpositions have a specific, independent meaning, which renders it pos-
sible to use them with practically any kind of landmark (see also Klavan et al. 2011
for Estonian). They are thus also immune to the kind of variation attested for case
forms. However, it is important to note that adpositions are not felicitous with
toponyms, which further underlines their specific nature. They are not used when
only one reading is possible, or at least highly probable. This has also been noted
by Comrie (1986) and Creissels and Mounole (2011) who both discuss the deviant
 Seppo Kittilä

nature of two kinds of entities, animate participants and place names, when these
functions as Locations or Directions.
The findings of the paper also tell us about the semantic differences between
illative and allative cases in Finnish (see also Lauerma 1990). As has been shown
above, illative is clearly the more concrete of the two case forms, which also makes
it less polysemous. Regardless of the landmark, illative codes ‘into’ readings. This
definitely contributes to its incompatibility with animate landmarks and also with
two-­dimensional entities. For example, in typical sending events, Themes do not
physically enter animate participants. The allative, in turn, is a more polysemous case
in Finnish (see Rice & Kabata 2007 for the polysemy of allative across languages). This
follows, since the allative is semantically less concrete and thus allows more variation
in its meaning. The exact reading always depends on the nature of the landmark, but
Finnish also has means of stressing a given reading whenever this is necessary. The
ability of the allative to code ‘to the vicinity of ’ readings also makes it compatible
with animate entities and Direction, which is a central difference between the case
forms. However, also here, case forms are infrequent and adpositions occur more
often. Moreover, it is worth noting that the semantic differences are neutralized with
place names: it is not necessary to specify the exact resulting location of motion. It is
only important to distinguish Direction from Source and static location, which can
be achieved by using either of the two directional case forms.
Finally, the findings of the paper contribute to our understanding of marked-
ness (see, however, Haspelmath 2006 for critical comments on the notion of
markedness). Markedness is motivated differently in the examined cases, but the
examples have in common that unexpected scenes are coded by adpositions, while
the most natural instances of Direction appear in illative or allative case (see also
Comrie 1986 for similar remarks). Animacy constitutes perhaps the least contro-
versial feature in this regard. Animate entities are mobile in their nature, which
makes them less typical Directions, and adpositional marking is thus expected.
Animate entities can also be seen as less prototypical Themes, especially with verbs
of ballistic motion, which, however, does not have any formal consequences in
Finnish, but merely contributes to the oddity of a given reading.
With all other features examined, the question of (un)markedness is less
straightforward. As regards dimensionality, it is expected that ‘onto’-readings are
expressed by the allative case with two-dimensional entities. Any other reading
is excluded by the inherent nature of these entities, whence the illative does not
occur. Three-dimensional entities allow more readings and there is thus more
variation in their coding. Moreover, three-dimensional entities can be either natu-
ral containers (such as ‘box’ and ‘barrel’) or they may lack this feature. With natu-
ral containers, the illative codes expected ‘into’-readings, while allative typically
codes vicinal Direction. Other readings need to be highlighted by adpositions.
 The (non-)prototypicality of Direction 

With entities lacking a normal container reading, such as ‘tree’ and ‘rock’, allative
codes vicinal Direction while other readings are expressed by adpositions.
With physical size and vicinal Directions, the key feature in explaining the
more elaborate marking of certain cases is (the lack of) salience. First, it is ratio-
nal to use sizeable, conceptually salient, entities as landmarks. Physical size dif-
fers from the other features discussed in this paper also in that neither cases nor
adpositions yield fully felicitous readings with small entities, even though adposi-
tions are somewhat more acceptable. Moreover, as shown in (23), adpositions are
possible with smaller trajectors (mouse in this case), whereas cases are excluded
also here. Second, a salient landmark constitutes a natural endpoint of motion. For
example, in ‘the child put the toy in the box’, the denoted event is completed when
the toy reaches the interior of the box. On the other hand, in ‘the child ran to the
vicinity of the house’ a natural endpoint has not been reached, which makes the
end of the event less salient conceptually (we may expect the motion to continue).
In this regard, vicinal Directions have features in common with physically small
entities. Last, the lack of coincidence is not directly inferable from the inherent
features of the entities in question.4 The lack of coincidence is not expected, and
the less expected reading needs to be highlighted linguistically.
In the last discussed case, that is, differences in linguistic reference, the
motivation and also the manifestation of markedness are very different from the
instances examined earlier. In the last case, the differences are only linguistic in
nature. First, pronouns with animate singular reference behave differently from
nouns and also plural pronouns in that they are not felicitous as Directions. One
of the reasons for the oddity of especially first person pronouns as Directions lies
in the fact that we see ourselves rather as active participants than passive targets
or landmarks (see also Dixon 1994: 23 for a similar remark on first person agents).
Second, the deviant formal treatment of toponyms follows from their inherent
semantics as natural locations. In Finnish, this is manifested in the fact that only
case forms can be used for coding Directions with place names. Zero marking
is, however, not possible (cf. Comrie 1986 for Armenian), since it is necessary to
distinguish Direction from Source and static location. Semantically more specific
marking is superfluous, and adpositions are thus not used. Moreover, the semantic
distinctions between allative and illative cases are neutralized, because it is not
relevant to specify the exact resulting location of the motion. Variation attested
elsewhere is thus neutralized whenever this is not semantically needed.

. Animate entities are an exception in this regard. With them, lack of coincidence is ex-
pected, but the non-prototypicality can be said to be primarily determined by the mobile
nature of the landmark in question.
 Seppo Kittilä

Acknowledgements

I would like to thank the editors for inviting me to contribute to this volume. I also
thank them for valuable comments on an earlier version of this paper. And finally,
I thank Academy of Finland (project number 1127724) for providing funding for
this study. All remaining flaws are, as usual, my sole responsibility.

Abbreviations

acc Accusative case

all Allative case
dat Dative case
ill Illative case
gen Genitive
pl Plural
pn Personal name
poss Possessive
prs Present tense
pst Past tense
sg Singular

References

Aissen, Judith. 2003. Differential object marking: Iconicity vs. economy. Natural Language and
Linguistic Theory 21: 435–483. DOI: 10.1023/A:1024109008573
Aristar, Anthony R. 1997. Marking and hierarchy types and the grammaticalization of case
markers. Studies in Language 21(2): 313–368. DOI: 10.1075/sl.21.2.04ari
Blansitt, Edward L. Jr. 1988. Datives and allatives. In Studies in Syntactic Typology, Michael Ham-
mond, Edith Moravcsik & Jessica Wirth (eds), 173–191. Cambridge: CUP. DOI: 10.1075/
tsl.17.14bla
Comrie, Bernard. 1986. Markedness, grammar, people, and the world. In Markedness, Fred R.
Eckman, Edith A. Moravcsik & Jessica R. Wirth (eds), 85–106. New York NY: Plenum.
DOI: 10.1007/978-1-4757-5718-7_6
Comrie, Bernard. 1989. Language Universals and Linguistic Typology. Chicago IL: University of
Chicago Press.
Creissels, Denis. 2008. Spatial cases. In The Oxford Handbook of Case, Andrej Malchukov &
Andrew Spencer (eds), 609–625. Oxford: OUP.
Creissels, Denis & Mounole, Céline. 2011. Animacy and spatial cases: ­Typological tendencies,
and the case of Basque. In Case, Animacy and Semantic Roles [Typological Studies in Lan-
guage 99], Seppo Kittilä, Katja Västi & Jussi Ylikoski (eds), 157–182. ­Amsterdam: John
Benjamins.
Dixon, Robert M.W. 1994. Ergativity. Cambridge: CUP. DOI: 10.1017/CBO9780511611896
Gair, James W. & Paolillo, John C. 1997. Sinhala. Munich: Lincom.
 The (non-)prototypicality of Direction 

Haspelmath, Martin. 2006. Against markedness (and what to replace it with). Journal of Linguis-
tics 42: 25–70. DOI: 10.1017/S0022226705003683
Haspelmath, Martin. 2007. Ditransitive alignment splits and inverse alignment. Functions of
Language 14: 79–102. DOI: 10.1075/fol.14.1.06has
Huumo, Tuomas & Ojutkangas, Krista. 2006. An introduction to Finnish spatial relations: Local
cases and adpositions. In Grammar from the Human Perspective: Case, Space and Person in
Finnish [Current Issues in Linguistic Theory 277], Marja-Liisa Helasvuo & Lyle Campbell
(eds), 11–20. Amsterdam: John Benjamins.
Huumo, Tuomas & Ojutkangas, Krista. 2010. Mikä erottaa muodot sisällä ja sisässä? “Syn-
onyymisten” muotojen analyysi (What is the difference between sisällä ‘in’ and sisässä ‘in’?
An analysis of two “synonymous” forms). Sananjalka 52: 19–45.
Kittilä, Seppo. 2006. On distinguishing between recipient and beneficiary in Finnish. In Gram-
mar from the Human Perspective: Case, Space and Person in Finnish [Current Issues in Lin-
guistic Theory 277], Marja-Liisa Helasvuo & Lyle Campbell (eds), 129–152. Amsterdam:
John Benjamins.
Kittilä, Seppo. 2008. Animacy effects on differential goal marking. Linguistic Typology 12:
245–268. DOI: 10.1515/LITY.2008.038
Kittilä, Seppo. 2010. Beneficiary coding in Finnish. In Benefactives and Malefactives: Typological
Perspectives and Case Studies [Typological Studies in Language 92], Fernando Zúñiga &
Seppo Kittilä (eds), 245–270. Amsterdam: John Benjamins.
Kittilä, Seppo & Ylikoski, Jussi. 2011. Remarks on the coding of goal, recipient and vicinal goal
in European Uralic. In Case, Animacy and Semantic Roles [Typological Studies in Language
99], Seppo Kittilä, Katja Västi & Jussi Ylikoski (eds), 29–64. Amsterdam: John Benjamins.
Klavan, Jane, Kaisa Kesküla & Laura Ojava. 2011. The division of labour between synonymous
locative cases and adpositions: The Estonian adessive and the adposition peal ‘on’. In Case,
Animacy and Semantic roles, Seppo Kittilä, Katja Västi and Jussi Ylikoski (eds.), 1134–134.
Amsterdam: Benjamins.
Lauerma, Petri. 1990. Spatiaalinen kenttä (spatial tier). In Suomen kielen paikallissijat konseptu-
aalisessa semantiikassa (Finnish local cases in conceptual semantics), Pentti Leino, Marja-
Liisa Helasvuo, Petri Lauerma, Urpo Nikanne & Tiina Onikki (eds), 108–145. Helsinki:
University of Helsinki, Department of Finnish.
Luraghi, Silvia. 2011. The coding of spatial relations with human landmarks: From Latin to
Romance. In Case, Animacy and Semantic roles [Typological Studies in Language 99],
Seppo Kittilä, Katja Västi & Jussi Ylikoski (eds), 209–234. Amsterdam: John Benjamins.
Næss, Åshild. 2003. What markedness marks: The markedness problem with direct objects. Lin-
gua 114: 1186–1212. DOI: 10.1016/j.lingua.2003.07.005
Næss, Åshild. 2007. Prototypical Transitivity [Typological Studies in Language 72]. Amsterdam:
John Benjamins. DOI: 10.1075/tsl.72
Ojutkangas, Krista. 2008. Mihin suomessa tarvitaan sisä-grammeja (What for does Finnish
need sisä- ‘inside’ grams). Virittäjä 112(3): 382–400.
Rappaport Hovav, Malka & Levin, Beth. 2008. The English dative alternation: The case for verb
sensitivity. Journal of Linguistics 44: 129–167. DOI: 10.1017/S0022226707004975
Rice, Sally & Kabata, Kaori. 2007. Crosslinguistic grammaticalization patterns of the allative.
Linguistic Typology 11: 451–514. DOI: 10.1515/LINGTY.2007.031
Verbeke, Saartje. 2011. Ergativity and Aligment in Indo-Aryan. Ph.D. dissertation, University
of Ghent.
 The morphosyntax of the Experiencer
in Early Vedic*

Eystein Dahl
University of Tromsø

This paper examines the encoding of Experiencer arguments in Early Vedic, the
earliest attested stage of Indo-Aryan. Although Experiential predicates show
a broad variety of case-marking patterns in this language, the Experiencer is
primarily expressed by the nominative, the accusative or the dative. In this
respect, this semantic role differs from other roles like Agent, Patient, Recipient
or Benefactive which show a more restricted range of case-marking options.
Examining the distribution of the three Experiencer constructions, I demonstrate
that there is a strong tendency in Early Vedic for the Experiencer to be construed
as subject with Experiential verbs, although some experiential verbs show an
inverted argument realization construction, where the Stimulus is expressed as
subject and the Experiencer as object.

Keywords: Vedic Sanskrit; Experiencer; Proto-Roles; Argument realization;

Case-marking

1. Introduction

This paper explores how the Experiencer role is expressed in Vedic Sanskrit, the
language of the sacred texts of Hinduism.1 The Experiencer has a somewhat less
unitary morphosyntactic behavior than other semantic roles like Agent, Patient or

* I am grateful to Jóhanna Barðdal, Chiara Fedriani, Silvia Luraghi, Heiko Narrog and an
anonymous reviewer for commenting upon earlier versions of this paper. Obviously, none
of them are to blame for any of the numerous remaining theoretical inconsistencies and
­philological weaknesses, for which I alone remain responsible. The Fondation Colette Caillat
of the Institut de France provided me with free housing during a three-month research visit
to Paris from October to December 2011, during which some of the ideas in this paper were
developed, for which I hereby wish to express my sincere gratitude.
. Note that several diachronic stages of Vedic may be discerned. The present paper is con-
cerned with Early Vedic, which is understood as the language of the Rigveda and ­Atharvaveda.
 Eystein Dahl

Recipient in this language, alternately receiving nominative, accusative or dative

case marking. The exact number of semantic roles as well as their individuation
remains disputed (cf. Levin & Rappaport Hovav 2005 for a discussion) and this
paper explores some aspects of this issue. It is organized as follows. Section 2 out-
lines some general assumptions about semantic roles in general and their morpho-
syntactic realization in Early Vedic in particular. Section 3 gives an overview of the
most characteristic Experiencer constructions in Early Vedic. Section 4 discusses
some implications arising from the findings in Section 3. Section 5 summarizes
the main points of the paper.

2. Semantic roles and case-marking in Early Vedic

Before turning to the main topic of this paper a few clarifying remarks are in order.
An important preliminary question concerns the nature of semantic roles and to
what extent they may be regarded as linguistic primitives. Another central ques-
tion pertains to the factors determining how a given semantic role is expressed in
a given language. Each of these questions will be discussed in the following.
In the present context I assume that traditional semantic roles like Agent or
Patient may be analyzed in terms of abstractions over clusters of privative lexical
entailments associated with sets of predicates, an assumption in line with Dowty
(1991) and subsequent work (cf. e.g. Primus 1999; Grimm 2005). For example,
two-place verbs like build or murder imply a volitional causation of a change of
state by one argument, the Agent, in another argument, the Patient. Several lexi-
cal entailments may be isolated which contribute to the relatively high agentivity
of the first argument of such verbs, notably volition, sentience, causation of a
change of state and the presence of another participant in the situation. Together,
these entailments are taken to constitute the characteristic properties of a pro-
totypical Agent. Likewise, a prototypical Patient may be defined as a participant
which undergoes a change of state, is causally affected by the another participant
in the situation etc. (cf. e.g. Dowty 1991; Grimm 2005). Elaborating on these
observations, it is tempting to suggest that the Proto-Agent and Proto-Patient
roles may be conceived of as opposite endpoints on a two-dimensional semantic
map or scale and that more and less prototypical instances of each of these roles
are located more and less close to them, as schematically represented in Figure 1
below (cf. Van Valin 1993: 41, 2005: 54 for a similar idea):

AGENT PATIENT

Figure 1. Prototypical agents and patients as poles on a scale

 The morphosyntax of the experiencer in Early Vedic 

The Agent and Patient roles show two important characteristics across lan-
guages: they both strongly tend to receive unitary morphosyntactic encoding and
they tend to be formally distinguished from each other by morphological and/or
syntactic means. In the Early Vedic monotransitive construction, for instance, the
most agentive argument tends to be realized as subject, while the most patientive
argument tends to be realized as object, as illustrated by the examples in (1).2
(1) a. śíśīte nūnáṃ paraśúm˘˙ svāyasáṃ yéna
sharpen.3sg.prs now axe.acc of.good.metal.acc which.ins
vṛścad étaśo bráhmaṇas pátiḥ //
cut.sbjv.3sg brilliant.nom Brahmaṇaspati.nom
‘Brilliant Brahmaṇaspati is now sharpening the axe of good metal with
which he shall cut asunder.’ (Rigveda X 53.9cd)
b. bhinád valám índro áṅgirasvān //
split.inj.3sg Vala.acc Indra.nom with.the.Aṅgirasas.nom
‘Accompanied by the Aṅgirasas Indra killed Vala.’ (Rigveda II 11.20d)
c. úttaraṃ dvis�ató mam ayáṃ maṇíḥ
superior.acc hating.abl 1sg.acc this.nom amulet.nom
kṛṇotu devajaḥ /
make.imp.3sg born.from.gods:nom
‘This amulet, created by the gods, shall make me superior to he who
hates (me).’ (Atharvaveda X 6.31a)

In these cases, the nominative-marked NPs bráhmaṇas pátis3 ‘Brahmaṇaspati’

(1a), índras ‘Indra’ (1b), and maṇís ‘amulet’ (1c), are represented as causing a
change of state in the accusative-marked NPs paraśúm ‘axe’ (1a), valám ‘Vala’ (1b),
and mam ‘me’ (1c), respectively.4

. Here and in the following, the abbreviations in the glosses follow the Leipzig Glossing
Rules 〈http://www.eva.mpg.de/lingua/resources/glossing-rules.php〉. Additionally, I make use
of the following abbreviations:
INJ Injunctive
PPP Past Passive Participle
QP Quotative Particle
. Here and in the following, I cite standardized forms (e.g. bráhmaṇ as pátis, índras, maṇ ís)
in the discussion, forms such as bráhmaṇ as pátiḥ, índro, maṇ íḥ found in the examples repre-
senting sandhi variants.
. At this point it should be noted that Early Vedic has three gender categories, masculine,
feminine and neuter, and that only the masculine and feminine have distinct nominative and
accusative case forms, neuter nouns having only one, syncretic form which may be used either
as a nominative or as an accusative.
 Eystein Dahl

Apart from case marking, the two sets of arguments have few semantic prop-
erties in common. Whereas the verb ŚĀ- ‘sharpen’ (1a) apparently entails that the
subject argument is volitional and sentient and, more generally, conforms more or
less exactly to the Agent prototype, BHED- ‘split’ in (1b) is compatible with and
seemingly always selects a volitional subject argument. However, it is clear from
(1c) that the verb KAR- ‘make, do’ only presupposes that the subject argument
causes a change of state in the object argument.5 The data in (1) likewise show that
patientive arguments, that is, arguments being causally affected by another partici-
pant in the situation, receive accusative case marking, irrespective of their other
inherent semantic features. With active forms of causative verbs, then, the most
agentive argument tends to be selected as subject and, accordingly, receives nomi-
native case marking, while the most patientive argument appears as the object and
is expressed by the accusative case.
It should be pointed out, however, that certain constructions select other case-
marking patterns. For instance, the so-called past passive participle in -tá- selects
an instrumental-marked agent and a nominative-marked patient, as illustrated by
the example in (2).6
(2) yéna ima víśvā cyávanā kṛtani
who.ins these.nom all.nom shakings.nom make.ppp
‘By whom all these violent shakings have been made.’ (Rigveda II 12.4a)

It should be noted that the morphosyntactic properties illustrated in (1) and (2)
are common to all causative two-place predicates. In other words, the core argu-
ments of predicates of this type show a predictable and unitary morphosyntactic
behavior.
At this point a cautionary note should be added, however. While agentive
arguments regularly select nominative case with verbs in the active voice in Early
Vedic, this is not to say that the nominative always codes Agents. On the con-
trary, Early Vedic is a nominative-accusative language where the nominative case
is the default subject marker and is not restricted to any specific semantic role. For
instance, it may encode a patientive argument of unaccusative verbs, as illustrated
by the examples in (3).

. Note that this verb frequently has the meaning ‘create’ which apparently entails that
the subject argument is volitional and hence sentient. However, I prefer to take the generic
meaning ‘make, do’ as basic and regard the verb KAR- as underspecified with regard to and
hence in principle compatible with sentience and volitionality.
. I refer to Dahl (Forthcoming) for a discussion of the semantics and syntax of the -tá-
participle.
 The morphosyntax of the experiencer in Early Vedic 

(3) a. yá es�āṃ bhr�tyam r�n�ádhat

who.nom 3pl.gen nursing.acc make.prosper.prs.3sg
sá jīvāt
he.nom live.sbjv.3sg
‘He who shall make their nursing prosper, shall live.’ (Rigveda I 84.16d)
b. iyáṃ ta indra girvan�o rātíḥ
this.nom 2sg.dat Indra.voc fond.of.praise.voc oblation.nom
ks�arati sunvatáḥ /
flow.prs.3sg worshipper.gen
‘O Indra, fond of praise, this oblation of the worshipper is flowing for
thee.’ (Rigveda VIII 13.4ab)

In these cases, the one-place verbs JĪV- ‘live’ and KS�AR- ‘flow’ select a nominative-
marked NP, sás ‘he’ and rātís ‘oblation’, respectively.
The nominative represents the generalized subject case category in Early Vedic
and its distribution is not sensitive to semantic factors. Similar considerations
apply to the monotransitive construction illustrated in (1) which is the canoni-
cal argument realization pattern of two-place predicates in Early Vedic and to the
ditransitive construction illustrated in (4), where the Recipient argument receives
dative case marking. In this context, the Recipient role is understood as a sentient
participant to whom something is given or sent (cf. also Van Valin 2005: 54).

(4) a. aháṃ bhu mim adadām aryāya aháṃ vr�s�t�íṃ

1sg.nom earth.acc give.ipf.1sg faithful.dat 1sg.nom rain.acc
dāśús�e mártyāya /
pious.dat man.dat
‘I gave the earth to the faithful, I (gave) rain to the pious man.’
(Rigveda IV 26.2ab)
b. ví dāśús�e bhajati sūnáraṃ vásu /
apart pious.dat distribute.prs.3sg delightful.acc wealth.acc
‘He distributes delightful goods to the pious.’ (Rigveda V 34.7b)

In cases like these, the dative-marked participants aryāya ‘to the faithful’, dāśús�e
mártyāya ‘to the pious man’ in (4a) and dāśús�e ‘pious’ in (4b) are taken to represent
the Recipient role and to have argument status in the sense that they are entailed
by the meaning of the verbs DĀ- ‘give’ and VI-BHAJ- ‘distribute’.
It should be noted, however, that dative case marking is by no means restricted
to Recipients. For instance, the Beneficiary, that is a person or entity that has
advantage or disadvantage of the situation named by the verb, is likewise con-
sistently assigned dative case in Early Vedic. Consider, by way of illustration the
examples in (5).
 Eystein Dahl

(5) a. túbhyaṃ bráhmān�i várdhanā kr�n�omi /

2sg.dat prayers.acc strengthening.acc make.prs.1sg
‘For you I am making strengthening prayers.’ (Rigveda VII 22.7b)
b. jyes�t�há āha camasa dva karā
oldest.nom say.prf.3sg sacrificial.cups.acc two.acc make.sbjv.1sg
íti kánīyān trɩn kr�n�avāma íty āha /
qp younger.nom three.acc make.sbjv.1pl qp say.prf.3sg
‘The oldest (brother) said: “I will make two sacrificial cups”. The
younger brother said “Let us make three!”’ (Rigveda IV 33.5ab)

In Example (5a), the dative-marked NP túbhyam ‘you’ picks out a referent which
is not directly involved in the situation denoted by the verb KAR- ‘make’ but still
enjoys an advantage from the situation. Note, however, that the dative-marked
participant is not strictly speaking an argument of the verb as it is not entailed by
its meaning and contributes additional information, as shown by example (5b)
where the verb KAR- ‘make’ is used as a two-place predicate. Along the lines of
Van Valin (1993: 70–71) I take the Beneficiary role to be an optional non-argu-
ment which may be added to virtually verb type but is not lexically specified. This
property of the Beneficiary role distinguishes it from roles like Agent, Patient and
Recipient. Furthermore, the fact that both Recipients and Beneficiaries may be
expressed by the dative case in Early Vedic indicates that the distribution of the
dative is determined by grammatical rather than lexical factors.
While Agent, Patient, Recipient and Beneficiary represent relatively well-
defined semantic roles, the exact semantic properties of the Experiencer role are
somewhat more elusive. Experiencers are necessarily sentient, a semantic feature
they share with prototypical Agents. At the same time Experiencers tend to be
conceived of as being causally affected by the situation or by another participant
in the situation, thus resembling the Patient role. Unlike beneficiaries, Experienc-
ers are, by definition, arguments of Experiential verbs. The somewhat ambiguous
character of the Experiencer role is reflected in the fact that it generally tends
to be characterized rather vaguely in the literature, such as ‘ambivalent status’
(Lehmann 1991), ‘Volitional Undergoers’ (Næss 2007) or ‘bilateral involvement’
(Smith 1993). Moreover, Næss (2007: 190) notes that feelings and emotions are
notions characterized by not being determinate with regard to the extent to which
the participants in the situation have control over and are affected by the situa-
tion (cf. also Fried 2005; Lehmann 1991). An important corollary of this obser-
vation is that emotion verbs imply that the two lexical entailments in question,
volitionality and affectedness may be taken to represent c­ omplementary rather
than m­ utually exclusive notions. On the other hand, emotion verbs generally do
 The morphosyntax of the experiencer in Early Vedic 

not seem to entail that the subject argument is volitional, even though they may
be compatible with a volitional interpretation. Note, however, that the present
approach where lexical entailments are regarded as privative is perfectly com-
patible with these observations. Specifically, Experiential predicates are taken
to be underspecified with regard to and hence in principle compatible with the
volitionality feature. They are also taken to entail that the Experiencer argument
is sentient and affected by the situation. Given the representation of the Agent
and Patient roles as opposite poles on a two-dimensional semantic map in Fig-
ure 1, it is tempting to suggest that the Experiencer role is somehow located in
between these two roles, as schematically represented in Figure 2 (cf. again Van
Valin 1993: 41).

AGENT EXPERIENCER PATIENT

[+SENTIENT] [+SENTIENT] [+CHANGE OF STATE]

[+VOLITIONAL] [+AFFECTED] [+AFFECTED]
[+CAUSE] [+INCREM. THEME]
[+MOVEMENT] [+STATIONARY]
etc etc.

Figure 2. Experiencer located between agent and patient

However, intuitively appealing as it might appear, various considerations sug-

gest that this representation is somewhat too simplistic. One important problem
concerns the status of the lexical entailments constituting these three role types.
The characteristic entailments of the (prototypical) Agent and Patient roles rep-
resent sufficient conditions which license the classification of a given argument
as Agent or Patient, apart from the [+CAUSE] and [+AFFECTED] entailments
which appear to be constitutive of these two roles. The entailments constituting
the Experiencer role, on the other hand, represent necessary conditions in order
for a given argument to qualify as an Experiencer. From a more general perspec-
tive, it is controversial whether the Experiencer may be regarded as a seman-
tic role on a par with Agent, Patient etc. For instance, Dik (1997: 116–119) does
not recognize the Experiencer as a distinct role but regards it “as a “secondary”
semantic function, a “footnote” to the other first argument functions” (1997: 119).
Other scholars recognize an Experiencer role but define it in rather vague terms,
as noted above.
As regards the morphosyntactic properties of Experiencer constructions,
Haig (2009: 6) notes that Experiencers tend to be ‘parasitic’ on other semantic
roles in the sense that Experiencer constructions tend to be assimilated to the
 Eystein Dahl

­ orphosyntactic expression of other semantic roles (cf. also Bickel 2004). Although
m
some languages have a dedicated case for Experiencer (cf. e.g. Daniel this volume
on Daghestani), most languages appear to allow a broad variety of construction
types to identify the Experiencer role. Bossong (1998) distinguishes two main
strategies or construction types which are found with Experiential verbs, so-called
generalized constructions where the Experiencer receives the same case mark-
ing as Agentive subjects, i.e. nominative case marking in accusative languages like
Early Vedic and so-called inverted constructions where the Stimulus is formally
assimilated to Agentive subjects.
It was noted above that the nominative represents the default subject case, the
accusative the direct object case and the dative the indirect object case in Early
Vedic and these three case marking patterns may be regarded as canonical. In
many languages, Experiencer predicates may select non-canonical subject and/
or object case-marking patterns, as illustrated by the examples from Latin and
Ancient Greek in (6).

(6) a. tui me, uxor, pudet

2sg.gen 1sg.acc wife.voc be.ashamed.prs.3sg
‘I am ashamed of you, wife.’ (Plautus Asinaria 933)
b. theoîsin ei díkēs mélei
gods.dat if justice.gen care.prs.3sg
‘If the Gods care about justice.’ (Sophocles Philoctetes 1036)

The example in (6a) illustrates that certain two-place Experiencer verbs, like pudet
‘be ashamed’ select an accusative-marked Experiencer (me) and a genitive-marked
Stimulus (tui). Example (6b) illustrates that the verb mélei ‘care, be an object
of care’ selects a dative-marked Experiencer (theoîsin) and a genitive-marked
Stimulus (díkēs). Significantly, Experiential constructions of this type, where two
arguments receive non-nominative case-marking, are not found in Early Vedic,
something which is in line with the more general observation that non-canonical
subject constructions are extremely marginal and perhaps even inexistent in this
language (cf., however, below).
In the present context, an Experiential verb is understood as a verb lexically
entailing that there is at least one participant, the Experiencer, which is sentient
and which is affected by the situation or by another participant in the situation, the
Stimulus. In her study of Experiential constructions in Yucatec Maya, Verhoeven
(2007) identifies five distinct subclasses of Experiencer verbs. These are verbs of
bodily sensations, emotion verbs, perception verbs, cognition verbs and volition
verbs. A fairly comprehensive list of representative examples from Early Vedic of
each of these verb classes are given in Table 1.
 The morphosyntax of the experiencer in Early Vedic 

Table 1. Classes of experiential verbs in early vedic

Bodily sensations Emotions Perception Cognition Volition

KS�ODH- ‘become OC- ‘take ĪKS�- ‘observe’ Ā-DHAYI- ‘care ARTHAYA-

hungry’ pleasure in’ for’ ‘strive for’
TAND- ‘be KANI- ‘find KHYĀ- ‘see’ JÑĀ- ‘know’ ES�- ‘wish,
exhausted’ pleasure in’ desire’
TAP- ‘cause pain, KAMI- ‘love’ CAKS�- ‘see’ DHAYI- ‘reflect’ VANI- ‘desire’
burn’
TAMI- ‘be tired’ KRAPI- ‘mourn’ CAY- perceive’ MAN- ‘think, VAYI- ‘strive for’
regard’
TARS�- ‘become KRODH- ‘become CET- ‘find out’ ÁPI-MARS�- VAR- ‘choose,
thirsty’ angry’ ‘forget’ prefer’
TRAS- ‘quiver’ JOS�- ‘enjoy’ DARŚ- ‘see’ MARDH- VAŚ- ‘want’
‘neglect’
BODH- ‘awake’ DVES�- ‘hate’ PAŚ- ‘see’ MOH- ‘be
confused’
RES�- ‘become NED- ‘rebuke’ ŚRAV- ‘hear’ SMAR-
hurt’ ‘remember’
ŚRAMI- ‘grow PAN- ‘admire’ VED- ‘know’
fatigued’
PRAYI- ‘please’
CHAND- ‘seem,
please’
TARP- ‘enjoy’
BHAYI- ‘fear’
MAD- ‘rejoice’
MARD�- ‘be
favourable’
MOD- ‘be pleased’
RAN�- ‘be pleased’
SVAD- ‘taste sweet’
HARI- ‘be angry’
HARṢ- ‘be excited’
HED�- ‘provoke’

In the following section, I distinguish some salient types of Experiential con-

structions that are found in Early Vedic and examine their distribution across the
classes of verbs illustrated in Table 1.

3. Experiential constructions in Vedic

Given the fact that Early Vedic has a relatively broad range of argument realization
options, one would expect to find a considerable variety of Experiencer construc-
tions in this language. Somewhat surprisingly, however, we seem to find only a
restricted number of case-marking options for the Experiencer. First of all, many
verbs select a nominative-marked Experiencer, as illustrated by the examples in (7).
 Eystein Dahl

(7) a. yáṃ jīvám aśnávāmahai ná

who.acc alive.acc reach.sbjv.1pl not
sá ris�yāti purus�aḥ //
he.nom become.hurt.sbjv.3sg man.nom
‘He whom we shall encounter alive that man shall not become hurt.’
(Rigveda X 97.17cd)
b. yó no dvés�t�y ádharaḥ sás padīs�t�a
who.nom 1pl.acc hate.prs.3sg lowest.nom he.nom fall.opt.3sg
yám u dvis�más tám u prān�ó jahātu //
who.acc and hate.prs.1pl he.acc and breath.nom leave.imp.3sg
‘He who hates us shall fall down; him whom we hate the breath of life
shall leave.’ (Rigveda III 53.21)
c. ayáṃ ha túbhyaṃ várun�o hr�n�īte
this.nom indeed you.dat Varun�a.nom be.angry.prs
‘Indeed this Varun�a is angry at you.’ (Rigveda VII 86.3)

In cases like these, the Experiencer argument is formally identified with proto-
typical Agents, i.e. these are paradigm examples of generalized constructions
(cf. Bossong 1998). Example (7a) illustrates that one-place Experiential predi-
cates tend to select a nominative-marked Experiencer (EXPNOM). The verb RES�-
‘become hurt’ belongs to a class of predicates that have a present stem formed
by the suffix -ya- (cf. also Hock 1991: 125–126, Dahl 2010: 110; Kulikov 2012). In
Example (7b), the two-place predicate DVES�- ‘hate’ selects a nominative-marked
Experiencer and an accusative-marked Stimulus (EXPNOM-STIMACC). Example
(7c) illustrate that certain Experiential verbs select a nominative-marked Expe-
riencer and a Stimulus marked by the dative or another oblique (non-canonical)
case (EXPNOM-STIMOBL).
Second, some verbs select an Experiencer argument expressed by the accusa-
tive, as illustrated by the examples in (8).

(8) a. yé mā krodháyanti lapita

who.nom.pl 1sg.acc be.angry.caus.prs.3pl noise.ins
hastínaṃ maśákā iva /
elephant.acc gnat.nom like
‘They who provoke me with hum like gnats provoke an elephant.’
(Atharvaveda IV 36.9ab)
b. stríyaṃ dr�s�t�vaya kitaváṃ tatāpa
woman.acc see.abs player.acc burn.prf.3sg
‘Having seen his wife, the gambler is in pain.’ (Rigveda X 34.11ab after
Hock 1991: 127)
 The morphosyntax of the experiencer in Early Vedic 

Example (8a) illustrates a construction where the Experiencer has accusative

case marking whereas the Stimulus is expressed by the nominative (STIMNOM-
EXPACC). In (8b) we find a marginal construction type containing a third singular
impersonal form of a one-place Experiential verb and an accusative-marked Expe-
riencer (EXPACC).
Third, some verbs select a dative-marked Experiencer, as illustrated by the
example in (9).
(9) cakráṃ yád asya ~ apsv a nís�attam utó
wheel.nom when he.gen waters.loc in rest.in.ppp and
tád asmai mádhv íc cacchadyāt /
that.nom he.dat honey.nom even seem.opt.3sg
‘When his wheel is set fast in the waters, even that would seem like honey to
him.’ (Rigveda X 73.9b after Klein 1985: 453)

It should be noted that the construction type with a dative-marked Experiencer

and a nominative-marked Stimulus (STIMNOM-EXPDAT) illustrated in Example
(7) is relatively uncommon in Early Vedic.
The examples in (7) through (9) illustrate that the Experiencer role may be
expressed by the nominative (7), the accusative (8) and dative (9) in Early Vedic.
Drawing on suggestions by Haspelmath (2001), Bickel (2004) and Haig (2009),
these three case-marking patterns may be taken to instantiate three typologically
distinct types of Experiencer constructions where the Experiencer argument is
assimilated to an Agent (Experiencer-as-Agent), a Patient (Experiencer-as-
Patient) and a Recipient (Experiencer-as-Recipient), respectively. In the following,
I examine the distribution of the three main types of Experiencer constructions
in Early Vedic.

3.1 The EXPNOM construction

Nominative-marked Experiencer arguments are found with all of the above-
mentioned predicate types in Early Vedic. The examples in (7) through (9) show
that one may, roughly speaking, distinguish three distinct instantiations of this
construction: it is found with one-place verbs like RES�- ‘become hurt’, with two-
place verbs like DVES�- ‘hate’ that select an accusative-marked Stimulus and
with two-place verbs like HARI- ‘become angry’ that select a Stimulus argument
marked by an oblique case.7 However, I have chosen to restrict the discussion

. At this point it should be noted that the dative is not the only oblique case which is used
to express the Stimulus argument in this construction. Experiential predicates with a nomina-
tive Experiencer may also select a Stimulus in the genitive, instrumental, locative or ablative
 Eystein Dahl

to the EXPNOM-STIMDAT construction, since it represents the mirror image of

the STIMNOM-EXPDAT construction, that is, the Experiencer-as-Recipient type,
and, interestingly, the latter construction shows a far more restricted distribu-
tion than the former. It is convenient to review each of these construction types
in turn.
The following examples present a representative sample of the kinds of one-
place verbs selecting a nominative-marked Experiencer argument in Early Vedic
(cf. also (5a) above).

(10) a. es�á vāṃ dyāvāpr�thivī upásthe ma

this.nom you.gen heaven.and.earth.voc lap.loc don’t
ks�udhan ma tr�s�at //
hunger.inj.3sg don’t thirst.inj.3sg
‘O heaven and earth, this man shall neither hunger nor thirst in your
lap.’ (Atharvaveda II 29.4cd)
b. múhyantv anyé abhíto jánāsa
be.confused.imp.3pl others.nom everywhere people.nom
‘Other people everywhere else shall be confused!’ (Rigveda X 81.6c)

The examples in (10) illustrate that one-place verbs of bodily states like KS�ODH-
‘be hungry’ and TARS�- ‘be thirsty’ and cognition verbs like MOH- ‘be confused’
select a nominative-marked Experiencer (es�á ‘he’ (10a), anyé ‘others’ (10b)) in
Early Vedic. It should be noted that one-place Experiential verbs are extremely
rare in Early Vedic and the fact that one can identify several types of verbs select-
ing a nominative-marked Experiencer is therefore significant.
Turning now to the second construction type, two-place verbs with a nom-
inative-marked Experiencer and an accusative-marked Stimulus, Examples (11)
a.-h. illustrate the range of verb types participating in this construction. The
examples show that the EXPNOM-STIMACC construction is found with verbs
belonging to the semantic domain of volition (ES�- ‘wish, desire’) (11a), perception
(KHYĀ- ‘see’) (11b), cognition (VED- ‘know’ (11c), MAN- ‘think, regard’(11d),
ÁPI-MARS�- ‘forget’(11e)) and emotion (JOS�- ‘enjoy’ (11f), PAN- ‘admire’ (11g),

and alternation patterns where two or more case categories are used to express the stimulus
are not infrequently met with. However, since Experiencer constructions are the main topic
of this paper and the various oblique Stimulus constructions represent a very complex part of
Early Vedic case grammar, a full discussion of these matters is beyond the scope of the present
paper and will have to be undertaken elsewhere.
 The morphosyntax of the experiencer in Early Vedic 

NED- ‘despise, revile’ (11h)).8 Thus, this construction is compatible with four out
of the five general semantic classes outlined by Verhoeven (2007).9
(11) a. karambhám anyá icchati //
gruel.acc other.nom desire.prs.3sg
‘The other desires gruel.’ (Rigveda VI 57.2c)
b. yás te agne sumatíṃ márto ákhyat
who.nom you.gen Agni.voc kindness.acc man.nom see.aor.3sg
‘O Agni, the mortal man who has seen your benevolence (…)’
(Atharvaveda XVIII 1.24)
c. ná ~ aháṃ veda bhrātr�tváṃ nó svasr�tvám
not 1sg.nom know.prs.1sg brotherhood.acc not sisterhood-acc
‘I do not know neither brotherhood nor sisterhood.’
(Rigveda X 108.10a)
d. tan aháṃ manye dúrhitān
they.acc 1sg.nom think.prs.1sg hostile.acc.pl
‘Them I regard as hostile.’ (Atharvaveda IV 36.9c)
e. ná táṃ pūs�a ápimr�s�yate
not he.acc Pus�an.nom forget.prs.3sg
‘Pus�an does not forget him.’ (Rigveda VI 54.4b)
f. práti na stómaṃ
towards we.gen song.of.praise.acc
tvás�t�ā jus�eta
Tvas�t�ar.nom find.pleasure.opt.3sg
‘May Tvas�t�ar find pleasure in our song of praise.’ (Rigveda VII 34.21a)
g. a tát ta indra ~ āyávaḥ pananta
to this.acc you.gen Indra.voc men.nom admire.inj.3pl
‘Men admire this (deed) of yours, O Indra.’ (Rigveda X 74.4a)
h. camasáṃ yé devapanam ánindis�uḥ /
ladle.acc who.nom serving.the.gods.acc despise.ipf.3pl
‘They who despised the drinking ladle of the gods.’ (Rigveda I 161.5b)

Finally, we find a few examples of verbs selecting a nominative-marked Experiencer

and a dative-marked Stimulus (EXPNOM-STIMDAT). Examples (12) a.-d. illustrate
that the EXPNOM-STIMDAT construction under discussion is primarily found with

. Some readers might object that NED- does not really represent an emotion verb but
rather a verb denoting the expression of an emotion.
. The fact that we do not find any examples of this construction type with bodily and
mental state verbs probably reflects that predicates of this type characteristically denote a
one-place relation and accordingly tend to be lexicalized as one-place verbs in Early Vedic.
 Eystein Dahl

emotion verbs denoting a negative attitude (HARI- ‘be angry’, KRODH- ‘be angry’)
or a positive attitude (MARD�- ‘be gracious’, ÁNU-JÑĀ- ‘assent’) towards an ani-
mate Stimulus argument as well as certain perception verbs, notably ŚRAV- ‘hear,
listen’.10
(12) a. malvó yó máhyaṃ krúdhyati
foolish.nom who.nom 1sg.dat be.angry.prs.3sg
‘The foolish one who is angry at me.’ (Atharvaveda IV 36.10c)
b. bháva rajan yájamānāya mr�d�a
Bháva.voc king.voc sacrificer.dat be.gracious.imp.2sg
‘O King Bhava, be gracious to the sacrificer!’ (Atharvaveda XI 2.28a)
c. túbhyaṃ deva ánu jānantu víśve //
you.dat gods.nom along regard.imp.3pl all.nom
‘The all-gods shall assent to you!’ (Atharvaveda VI 112.1d)
d. a ~ u s�ú svasāraḥ kāráve śr�n�ota
to and well sisters.voc poet.dat listen.imp.2pl
‘O sisters, listen well to the poet!’ (Rigveda III 33.9a)

It should be noted that Example (12d) likewise could be interpreted as a predi-

cate implying that the Experiencer argument has a positive attitude towards the
animate Stimulus, as indicated by the adverbial particle sú ‘well’. Therefore, it is
tempting to suggest that the EXPNOM-STIMDAT construction is restricted to predi-
cates expressing that the Experiencer has a markedly positive or negative attitude
toward the Stimulus argument.11
To sum up, the data discussed in this section indicate that the EXPNOM-­
construction covers all the types of Experiential verbs distinguished by Verhoeven
(2007). The use of the nominative to express the Experiencer may be taken to
directly reflect the fact that Early Vedic is a nominative-accusative language, i.e.
a language where subjects of transitive (A) and intransitive (S) clauses tend to
receive the same morphosyntactic encoding which differs from that of objects of
transitive clauses (P). However, the three subtypes distinguished here cover slightly
different, though overlapping Experiential domains. The EXPNOM-STIMACC

. Note that the 2nd person Imperative forms in Examples (12b) and (12d) do not strictly
speaking instantiate the EXPNOM-STIMDAT construction, as the Experiencer argument is ex-
pressed by the vocative. However, the vocative may in many cases be regarded as a contextu-
ally determined alternative to the nominative, notably in cases where the subject argument
coincides with the addressee, as is, for instance, the case with 2nd person Imperative forms.
. One might go even further and suggest that the use of the dative in cases like these is mo-
tivated by its more general use to encode the Recipient and Beneficiary/Maleficiary roles. In
that case, it would be proper to characterize the construction under discussion as something
like Experiencer-as-Agent, Stimulus-as-Recipient/Beneficiary.
 The morphosyntax of the experiencer in Early Vedic 

c­ onstruction covers the broadest range of predicates, including volition, percep-

tion, cognition and emotion verbs. The simpler EXPNOM construction appears to
be restricted to verbs of bodily or mental states and, possibly, to emotion verbs.
Finally, the EXPNOM-STIMDAT construction appears to be restricted to a semanti-
cally rather specific set of predicates based on emotion verbs, i.e. verbs expressing
that the Experiencer has a markedly positive or negative attitude towards an ani-
mate Stimulus.

3.2 The EXPACC construction

The EXPACC construction is considerably less frequent than the EXPNOM construc-
tion. One may distinguish two distinct subtypes of this construction. First, there
are a few isolated cases of accusative-marked Experiencers with impersonal forms
of one-place predicates (cf. also Delbrück 1888; Hock 1991). Second, the Experi-
encer argument is expressed by the accusative with causative verbs derived from
primary emotion verbs. Each of these will now be discussed in turn.
Accusative-marked Experiencers are sporadically found with impersonally
used 3sg forms of experiential predicates in Early Vedic, as illustrated by the exam-
ples in (13) where the verbs TAM- ‘be exhausted’ in (13a) and TAP- ‘be in pain’
(13b) select an accusative-marked Experiencer (mā ‘me’ (13a), kitavám (13b)):
(13) a. ná mā taman ná śraman ná ~
not 1sg.acc be.exhausted.sbjv.3sg not be.weary.sbjv.3sg not
utá tandran ná vocāma ma sunota ~
and grow.fatigued.sbjv.3sg not say.sbjv.1pl don’t extract.imp.2pl
íti sómam /
qp soma.acc
‘I will not become exhausted, not become weary and not grow fatigued.
We will not say: “Don’t extract soma!”’ (Rigveda II 30.7ab)
b. stríyaṃ dr�s�t�vaya kitaváṃ tatāpa
woman.acc see.abs player.acc pain.prf.3sg
‘Having seen his wife, the gambler is in pain (lit. ‘[it] pains the
gambler).’ (Rigveda X 34.11a)

As Early Vedic is a pro-drop language, the examples cited here might in principle
be interpreted as cases where a nominative-marked Stimulus argument has been
omitted. However, elsewhere in the Vedic literature the verb TAM- (13a) invari-
ably has the meaning ‘be exhausted’ rather than ‘make exhausted’.12 Therefore a
causative reading is rather unlikely in this case, especially because the following

. Cf. e.g. Maitrāyan�ī Saṃhitā IV 2.1, Jaiminiyabrāhman�a III 79, Śatapathabrāhman�a IV
2.2.11.
 Eystein Dahl

verbs ŚRAM- ‘be weary’ and TANDR- ‘grow fatigued’ denote similar bodily states.
The verb TAP- ‘burn, pain’ in the second example (13b), on the other hand, repre-
sents a somewhat less compelling example, as this verb can also be used as a two-
place predicate.13 However, as the preceding context does not provide any likely
antecedent for an omitted Stimulus argument, it is reasonable to take example
(13b) at face value as a case of an accusative-marked Experiencer with an imper-
sonal verb. Although this construction type is infrequent in Early Vedic, it appears
fairly clear that it is restricted to Experiential predicates based on verbs denoting
a bodily or mental state.
Turning now to the inverted STIMNOM-EXPACC construction, it should be
noted that it is mainly, though not exclusively found with derived causative forms
of verbs which are also attested with a nominative-marked Experiencer in their
simplex form. The examples in (14) illustrate that the STIMNOM-EXPACC construc-
tion is found with causative verbs derived from verbs of bodily states or processes
(BODH- ‘awake’ (14a)) and emotion verbs (MAD- ‘rejoice’ (14b) TRAS- ‘fear’ (14c)).
(14) a. prá bodhaya jaritar jārám índram /
forth arouse.imp.2sg singer.voc friend.acc Indra.acc
‘O singer, arouse (our) friend Indra!’ (Rigveda X 42.2)
b. té tvā mádā indra mādayantu
these.nom you.acc draughts.nom Indra.voc exhilarate.imp.3pl
‘These draughts (of soma) shall exhilarate you, o Indra!’
(Rigveda VII 23.5a)
c. párā amítrān dundubhínā harin�ásya ~ ajínena ca /
away enemies.acc war.drum.ins deer.gen skin.ins and
sárve deva atitrasan
all.nom gods.nom scare.aor.3pl
‘The All-Gods have scared away our enemies with war drums and deer
skin.’ (Atharvaveda V 21.7a-c)

Moreover, example (14d) illustrates that this construction is not restricted to

derived causative verbs but is occasionally found with simple causative emotion
verbs like HED�- ‘provoke’ as well. Note that cases of the latter type appear to be
rather rare in Early Vedic:
d. táṃ tvā manyo akratúr
this.acc 2sg.acc Manyu.voc powerless.nom
jihīl�a ~ ahám˘˙
provoke.prf.1sg 1sg.nom
‘O Manyu, I have offended you unwillingly.’ (Rigveda X 83.5)

. Cf. e.g. Rigveda I 162.20a.

 The morphosyntax of the experiencer in Early Vedic 

Before concluding the discussion of the EXPACC construction I would briefly like
to draw attention to another related construction type in Early Vedic. Consider
the examples in (15) where verbs like VED- ‘find’, GAM- ‘come’ or VI-AD- ‘gnaw’
select a nominative-marked abstract noun like bhɩ- ‘fear’, kama- ‘love’ or ādhɩ-
‘anxiety’ and an accusative-marked Experiencer tvā ‘you’, mā ‘me’:
(15) a. ná tvā bhɩr iva vindati ~ īm˘˙ //
not you.acc fear.nom like find.prs.3sg ever
‘Does not fear, as it were, ever find/befall you.’
(Rigveda X 146.1cd after Hock 1991: 127)
b. yamásya mā yamyàṃ kama agan
Yama.gen 1sg.acc Yamī.acc love.nom come.aor.3sg
‘Over me, Yamī a love for Yama has come.’ (Rigveda X 10.7a)
c. mus�o ná śiśna vy àdanti mā ādhyà
mice.nom like tails.acc apart eat.prs.3pl 1sg.acc anxieties.nom
stotaraṃ te śatakrato
singer.acc you.gen having.hundred.fold.power.voc
‘Like mice gnawing their tails, anxieties are gnawing at me, your singer,
O you of hundredfold power.’ (Rigveda I 105.8c)

While these and similar constructions in many respects resemble the constructions
that were discussed earlier, they are distinguished by two characteristic features.
First, the nominative-marked element does not represent a Stimulus argument but
rather what Verhoeven (2007) refers to as ‘expertum’, i.e. the content of the Expe-
rience. Second, the base verb does not represent a proper Experiential verb but
has a more general meaning which is metaphorically extended to the Experiential
domain. On the basis of these and similar constructions, one might distinguish yet
another subtype of the EXPACC construction, containing a non-Experiential verb
and an Expertum argument case-marked by the nominative. As indicated by the
examples in (15), this construction type appears to be restricted to the semantic
field of emotions.14
In this section we have seen that the EXPACC construction primarily or
exclusively is attested with verbs of bodily states and causative emotion verbs
in Early Vedic. It was observed that there are some examples of accusative-
marked Experiencers in constructions containing a non-Experiential predicate
and an Expertum argument in the nominative and that this construction type
appears to be restricted to emotion verbs.

. It should be noted that I have found one example containing a non-Experiential verb and
an Expertum argument in the accusative (Atharvaveda VIII 8.18), something which might
be taken to indicate yet another subtype of the EXPNOM construction. However, this highly
interesting possibility will have to be pursued elsewhere.
 Eystein Dahl

3.3 The EXPDAT construction

The EXPDAT construction is not very frequently attested in Early Vedic and its dis-
tribution seems to be subject to considerably stricter lexical constraints than the
EXPNOM or the EXPACC constructions. In all of the cases I have found of this con-
struction it is accompanied by an expressed or unexpressed nominative-marked
Stimulus. Representative examples include the following:
(16) a. sómasya ~ iva maujavatásya bhaks�ó
soma.gen like from.Mujavat.gen portion.nom
vibhɩdako jagr�vir máhyam acchān //
Vibhīdikaberry.nom animating.nom 1sg.dat seem.aor.3sg
‘Like a draught of soma from the mountain Mujavat the animating
Vibhīdikaberry appeared to me.’ (Rigveda X 34.1cd)
b. svádasva ~ índrāya pávamāna indo
taste.sweet.imp.2sg Indra.dat flowing.clear.nom drop.voc
‘Taste sweet to Indra while flowing clear, O drop!’ (Rigveda IX 97.44c)

These examples suffice to illustrate that the STIMNOM-EXPDAT construction is

restricted to predicates denoting a positive emotional state in Early Vedic.15

4. Morphosyntactic variation in the Experiential domain

In the previous section, we have seen that Early Vedic has three distinct types
of Experiencer constructions, EXPNOM, EXPACC and EXPDAT and that one may
further distinguish two or more subtypes of the former two constructions. From
a typological perspective, the three main types of Experiential constructions cor-
respond to the Experiencer-as-Agent, Experiencer-as-Patient and Experiencer-
as-­Recipient, respectively. Significantly, none of the subconstructions we have
identified are compatible with all of the Experiential verb classes but rather
show different patterns of distribution. However, together the various EXPNOM
­constructions cover all of the verb classes as shown in Table 2. The semantic map
in Figure 3 illustrates the distribution of the various subtypes of the EXPNOM
­construction identified above in Early Vedic.

. Indirect evidence for this conclusion is provided by the fact that parallel predicates de-
noting a negative state of mind, albeit of extremely rare occurrence, have a genitive-marked
Experiencer argument in Early Vedic, cf. e.g. the verb ROS�- ‘displease’ (Rigveda VIII 4.8).
However, this construction type is too sporadically attested in Early Vedic to state anything
general about its distribution across the various classes of Experiential predicates.
 The morphosyntax of the experiencer in Early Vedic 

Table 2. Verbs selecting the EXPNOM construction in early vedic

EXPNOM EXPNOM-STIMACC EXPNOM-STIMDAT

RES�- ‘become hurt’ (7a) DVES�- ‘hate’ (7b) HARI- ‘be angry’ (7c)
KS�ODH- ‘become hungry’ (10a) ES�- ‘wish, desire’ (11a) KRODH- ‘be angry’ (12a)
TARS�- ‘become thirsty’ (10a) KHYĀ- ‘see’ (11b) MARD�- ‘be gracious’ (12b)
MOH- ‘be confused’ (10b) VED- ‘know’ (11c) ÁNU-JÑĀ- ‘assent’ (12c)
HARS�- ‘be excited’ MAN- ‘think, regard’ (11d) ŚRAV- ‘hear, listen’ (12d)
MOD- ‘be pleased’ ÁPI-MARS�- ‘forget’ (11e)
RAN�- ‘be pleased’ JOS�- ‘enjoy, find pleasure’ (11f)
PAN- ‘admire’ (11g)
NED- ‘despise, revile’ (11h)
OC- ‘take pleasure in’
KANI- ‘find pleasure in’
KAMI- ‘love’
KRAPI- ‘mourn’
PRAYI- ‘please’
MARD�- ‘be gracious’
MARDH- ‘neglect’
ARTHAYA- ‘strive for’
DHAYI- ‘reflect’
ĪKS�- ‘observe’
KHYĀ- ‘see’
CAKS�- ‘see’
CAY- ‘perceive’
CET- ‘find out’
DARŚ- ‘see’
PAŚ- ‘see’
ŚRAV- ‘hear’
ARTHAYA- ‘strive for’
VANI- ‘desire’
VAYI- ‘strive for’
VAR- ‘choose, prefer’
VAŚ- ‘want’

The semantic map in Figure 3 is intended to illustrate that the domain of the
general EXPNOM construction (marked in boldface and italics), covers the entire
semantic field of Experiential predicates and that the various subtypes of this
construction, i.e. EXPNOM, EXPNOM-STIMACC and EXPNOM-STIMDAT cover dif-
ferent subfields of this domain, the EXPNOM construction being restricted to one-
place predicates denoting a bodily or mental state and one-place emotion verbs,
the EXPNOM-STIMACC construction being found with emotion verbs, cognition
verbs, perception verbs and volition verbs and the and EXPNOM-STIMDAT con-
struction being found with emotion verbs, cognition verbs and perception verbs.
Significantly, however, the EXPNOM-STIMDAT construction shows a much more
restricted lexical distribution than the EXPNOM-STIMACC construction in the rel-
evant semantic fields, as is evident from Table 2.
 Eystein Dahl

EXPERIENTIAL
PREDICATES

EXPNOM

Bodily or
Emotion Perception Cognition Volition
Mental State
EXPNOM

Positive Negative

EXPNOM-STIMDAT
EXPNOM-STIMACC

Figure 3. The distribution of the various subtypes of the EXPNOM construction in Early Vedic

In a similar vein, the semantic map in Figure 4 illustrates the distribution of

the two subtypes of the EXPACC construction in Early Vedic.

EXPERIENTIAL
PREDICATES

EXPACC
Bodily or
Emotion Perception Cognition Volition
Mental state

EXPACC

Positive Negative

STIMNOM-EXPACC

Figure 4. The distribution of the various subtypes of the EXPACC construction in Early Vedic
 The morphosyntax of the experiencer in Early Vedic 

This semantic map is intended to illustrate that the general EXPACC construc-
tion covers a considerably smaller part of the semantic field of Experiential predi-
cates than the EXPNOM construction, as is also evident from Table 3.

Table 3. Verbs with accusative Experiencers

EXPACC STIMNOM-EXPACC

TAND- ‘be exhausted’ (13a) HED�- ‘provoke’ (14d)

TAP- ‘cause pain, burn’ (13b) PRAYI- ‘please’
TAMI- ‘be tired’ (13a) BODH- ‘awake’; derived causative ‘arouse’ (14a)
ŚRAMI- ‘grow fatigued’ (13a) MAD- ‘rejoice’; derived causative ‘exhilarate’ (14b)
TRAS- ‘fear’; derived causative ‘scare’ (14c)
HARS�- ‘be excited’; derived causative ‘impel’
BHAYI- ‘fear’; derived causative ‘scare’

Another difference between the two constructions is that no subtype of the

EXPNOM construction covers its entire domain, whereas the scope of the STIM-
NOM-EXPACC construction coincides with that of the EXPACC construction. These
observations invite the conclusion that the EXPACC construction is semantically
more specific than the EXPNOM construction. Moreover, although a number of
two-place verbs indeed select the STIMNOM-EXPACC construction, it mainly seems
to appear with derived causative verbs where the case frame may be hypothesized
to be triggered by the causative morphology. I have only been able to identify
two simplex two-place experiential predicate with this case frame, the verbs HED�-
‘provoke’ and PRAYI- ‘please’. These considerations suggest that simplex verbs tend
not to prefer the inverted construction in Early Vedic, that is, Experiential con-
structions where the Stimulus is construed as Agent.
Finally, the semantic map in Figure 5 illustrates the semantic domain of the
EXPDAT construction in Early Vedic.
This figure illustrates that the semantic domain of the EXPDAT construc-
tion is considerably less extensive than those of the two other Experiencer con-
structions just discussed, the construction being restricted to emotion verbs
denoting a positive emotion. It should be noted, however, that I have only
been able to identify two Experiential verbs selecting the STIMNOM-EXPDAT
construction, that is, CHAND- ‘seem, please’ (9/16a) and SVAD- ‘taste sweet’
(16b). The fact that so few verbs select this case frame ties in well with the
observation that very few simplex verbs select the STIMNOM-EXPACC con-
struction and that, more generally, inverted Experiential constructions are far
less productive than generalized constructions. In other words, Experienc-
ers strongly tend to be construed as subjects and to receive nominative case
­marking in Early Vedic.
 Eystein Dahl

EXPERIENTIAL
PREDICATES

Bodily or
Emotion Perception Cognition Volition
Mental State

EXPDAT Positive Negative

STIMNOM-EXPDAT

Figure 5. The distribution of the EXPDAT construction in Early Vedic

5. Concluding remarks

In this paper I have explored the morphosyntactic realization of the Experiencer

role in Vedic Sanskrit. It was noted that this role is less well-defined than other
semantic roles like Agent or Patient and that it tends to have a more variegated
morphosyntactic encoding across languages. Having examined pertinent data
from Early Vedic, it was shown that Experiencers may be expressed by the nomina-
tive, the accusative and the dative and that it receives nominative case marking in
the majority of cases and that a rather limited number of verbs select an accusative-
marked or dative-marked Experiencer. Moreover, while the EXPNOM construction
is attested with all types of Experiential predicates, the EXPACC and EXPDAT con-
structions have a very limited distribution, the EXPACC construction only being
found with verbs of bodily or mental states and emotion verbs, and the EXPDAT
construction only being found with emotion verbs denoting a positive emotion.16
Thus, Early Vedic Experiental predicates tend to select a g­ eneralized construction
pattern, that is, they prefer an Experiencer subject to a Stimulus subject.

. Significantly, Dahl and Fedriani (2012) show that there are striking similarities between
Early Vedic, Early Latin and Homeric Greek as to how the various construction types dis-
cerned above are distributed over the different classes of Experiential verbs. This fact may be
taken to suggest that these distribution patterns are inherited and possibly represent a charac-
teristic feature of the Indo-European language family. However, more comparative research is
needed in order to substantiate a claim along these lines.
 The morphosyntax of the experiencer in Early Vedic 

References

Bickel, Baltasar. 2004. The syntax of experiencers in the Himalayas. In Non-nominative Subjects
[Typological Studies in Language 60–61], Peri Bhaskararao & Karumuri Venkata Subbarao
(eds), 77–112. Amsterdam: John Benjamins.
Bossong, Georg. 1998. Le marquage de l’experient dans les langues de l’Europe. In Actance et
valence dans les langues de l’Europe [Empirical Approaches to Language Typology 20–2],
Jack Feuillet (ed.), 259–294. Berlin: Mouton de Gruyter.
Dahl, Eystein. 2010. Time, Tense and Aspect in Early Vedic Grammar. Exploring Inflectional
Semantics in the Rigveda. Leiden: Brill.
Dahl, Eystein. Forthcoming. ‘Aspects of the morphosyntax and semantics of the Early Vedic -tá
participle and its prehistory’. In Ergativity in Indo-Aryan. Selected papers from the workshop
on Ergativity in Indo-Aryan at the SLE 43rd annual meeting [Typological Studies in Lan-
guage], Eystein Dahl and Krzysztof Stroński (eds). Amsterdam: John Benjamins.
Dahl, Eystein & Fedriani, Chiara. 2012. The argument structure of experience: Experiential con-
structions in Early Vedic, Homeric Greek and Early Latin. Transactions of the Philological
Society 110(3): 1–21. DOI: 10.1111/j.1467-968X.2012.01313.x
Delbrück, Berthold. 1888. Altindische Syntax. Halle: Verlag der Buchhandlung des Waisenhauses.
Dik, Simon C. 1997. The Theory of Functional Grammar, Part 1: The Structure of the Clause, 2nd
edn. Berlin: Mouton de Gruyter.
Dowty, David. 1991. Thematic proto-roles and argument selection. Language 67(3): 547–619.
Fried, Miriam. 2005. A frame-based approach to case alternations: The swarm-class verbs in
Czech. Cognitive Linguistics 16(3): 475–512. DOI: 10.1515/cogl.2005.16.3.475
Grimm, Scott. 2005. The Lattice of Case and Agentivity. MSc Dissertation, University of
Amsterdam.
Haig, Geoffrey. 2009. Non-canonical subjects and alignment change – where’s the connection?
Handout from the Workshop ‘Reconstructing Alignment Systems’ May 14–15, 2009.
Haspelmath, Martin. 2001. Non-canonical marking of core arguments in European languages.
In Non-canonical Marking of Subjects and Objects [Typological Studies in Language 46],
­Alexandra Y. Aikhenvald, Robert M.W. Dixon & Masayuki Onishi (eds), 53–83. ­Amsterdam:
John Benjamins.
Hock, Hans H. 1991. Oblique Subjects in Sanskrit? In Experiencer Subjects in South Asian Lan-
guages, Manindra K. Verma & Karuvannur Puthanveettil Mohanan (eds), 119–139. Stan-
ford CA: CSLI.
Klein, Jared. 1985. Toward a Discourse Grammar of the Rigveda, Vol. 1: Coordinate Conjunctions,
Part 1. Heidelberg: Carl Winter.
Kulikov, Leonid. 2012. The Vedic -ya-presents. Passives and Intransitivity in Old Indo-Aryan.
Amsterdam: Rodopi.
Lehmann, Christian. 1991. Predicate classes and PARTICIPATION. In Partizipation: Das
sprachliche Erfassen von Sachverhalten, Hansjakob Seiler & Waldfried Premper (eds),
­183–239. Tübingen: Gunter Narr.
Levin, Beth & Rappaport Hovav, Malka. 2005. Argument Realization. Cambridge: CUP.
Næss, Åshild. 2007. Prototypical Transitivity [Typological Studies in Language 72]. Amsterdam:
John Benjamins. DOI: 10.1075/tsl.72
Primus Beatrice. 1999. Cases and Thematic Roles: Ergative, Accusative and Active. Tübingen:
Niemeyer. DOI: 10.1515/9783110912463
 Eystein Dahl

Smith, Michael B. 1993. Cases as conceptual categories: Evidence from German. In Conceptu-
alizations and Mental Processes in Language [Cognitive Linguistic Research 3], Richard A.
Geiger & Brygida Rudzka-Ostyn (eds), 531–566. Berlin: Mouton de Gruyter.
Van Valin Jr., Robert D. 1993. A synopsis of role and reference grammar. In Advances in Role and
Reference Grammar [Current Issues in Linguistic Theory 82], Robert D. Van Valin Jr. (ed),
1–164. Amsterdam: John Benjamins.
Van Valin Jr., Robert D. 2005. Exploring the Syntax-Semantics Interface. Cambridge: CUP.
Verhoeven, E. 2007. Experiential Constructions in Yucatec Maya [Studies in Language Compan-
ion Series 87]. Amsterdam: John Benjamins. DOI: 10.1075/slcs.87
 Against the addressee of speech – Recipient
metaphor
Evidence from East Caucasian*

Michael Daniel
Higher School of Economics** / Moscow State University

In Standard Average European (SAE), addressees of speech verbs are marked with
dative or, in languages lacking cases, with dative-like prepositions. This merger
is commonly explained through a metaphor: the information transferred in a
speech act is said to be construed as the object being transferred, or Theme, and
the addressee as its Recipient. This status of the addressee as a derived concept,
a metaphor of the Recipient, and its dative marking in many languages rather
than in SAE alone, is the reason why the addressee is usually not considered
to be a separate semantic role. Based on data from East Caucasian languages
that use different marking for Recipients and addressees of speech, I argue that
speech addressees constitute a separate semantic role, also an animate Goal,
but not a metaphor of the Recipient. Focusing on case marking assigned by the
main speech verb, speech acts are shown to be construed in East Caucasian as
spatial configurations: the crucial component is their directedness towards the
addressee. In the conclusion, I come back to SAE and question the status of the
dative addressees. Taking into account that the dative often develops from lative
markers, it is suggested that, in the languages with dative addressees, one should
also consider an alternative to the conventional explanation: merging the Recipient
and the addressee in one marking may result not from a metaphorical extension
but from a purely formal under-specification of two different animate Goals.

Keywords: semantic roles; speech verbs; case functions; metaphor; dative;

allative; Recipient; Goal; East Caucasian languages

* I am grateful to Gilles Authier, Denis Creissels, Diana Forker, Zaira Khalilova, Timur Maisak,
Zarina Molochieva, Johanna Nichols and Sergei Tatevosov for reading and greatly i­mproving
the manuscript of this paper and for sharing their data with me; and to all members of the
Dargwa seminar (Russian State University for Humanities, Moscow) for both general com-
ments and important corrections of Dargwa data.

** Support from the Basic Research Program of the National Research University Higher
School of Economics (TZ65, year 2014) is gratefully acknowledged.
 Michael Daniel

1. Introduction

Human language conceptualizes some participants – animate beings or inanimate

objects in their relation to the real world situation – as more or less similar. While
situations are usually expressed by predicates, participants are represented by
verbal arguments, primarily noun phrases (NPs). A semantic role comprises par-
ticipants whose resemblance is linguistically relevant (Comrie 1981; Van Valin &
LaPolla 1997; Levin & Rappoport Hovav 2005). But what counts as being linguis-
tically relevant varies from one approach to the other. Thus, Role and Reference
Grammar (RRG; e.g. Van Valin & LaPolla 1997; VanValin 2005), isolates seman-
tic roles (thematic roles in terms of RRG) by means of semantic decomposition
of verbal meanings: the set of the semantic roles of a predicate is determined by
what simpler predicates are used in its semantic analyses. Dowty (1991) suggests
that verbs have similar arguments if the semantics of these verbs lead to similar
entailments about the participants. These approaches may be grouped together as
predicate-centered as they focus on the semantics of the verb.
Other approaches are clearly argument-centered, in a sense behavorial. In this
case, semantic roles include participants that are coded by noun phrases that are
linguistically similar; as Levin and Rappoport Hovav write in their overview of the
field (2005: 36): “Semantic role assignments are meant to bring out similarities and
differences in verb meaning that are reflected in argument expression”. Argument-
centered approaches primarily address noun phrase marking, whether it is a case
suffix or an adposition (flagging); but also take into account additional properties
that are of more (morpho)syntactic nature – indexing on the verb, (non)involve-
ment into various valency derivation processes etc. That was the starting point
of the whole discussion of semantic roles by (Fillmore 1968). The fact that spe-
cific verbal arguments receive similar treatment across languages is understood as
an indication that the respective participants are conceived as similar and corre-
spond to one semantic role. (Levin & Rappoport Hovav 2005) discuss weaknesses
of this approach, indicating the fact that semantic categorization cannot be fully
expressed in a limited inventory of the means of flagging, on the one hand, and, on
the other, that no consensus seems ever to be reachable about additional diagnos-
tics and the imminent risk of role fragmentation – splitting classes of arguments
into always smaller groups basing on their sensitivity to additional tests (Dowty
1991) – or overgeneralization.
A way to overcome the shortcomings of the argument-based approaches
highlighted by Levin and Rappoport Hovav is to shift one’s focus from extracting
semantic roles inventory from individual languages to cross-linguistic variation in
argument morphosyntax – something that to a certain extent is implicitly present
in many approaches, and that is an explicit objective of the present volume. An
 Against the addressee of speech – Recipient metaphor 

example of this is the discussion of semantic roles in Comrie (1981), where the
universal inventory of semantic roles depends on what arguments show special
morphosyntactic behavior cross-linguistically. Any participant that is conveyed
by a specialized marker on the noun phrase (dedicated marking) should be con-
sidered an element of the universal semantic role inventory. Recently, the Leipzig
Valency Classes project consistently applies a very similar but more flexible ideol-
ogy. It represents arguments of individual verbs (microroles) as a continuum in a
conceptual space where this arguments are mapped as more or less similar to each
other depending on the degree of similarity of flagging (case/adpositional mark-
ing), indexing on the verb and availability for valency alternations in a sample of
some 30 languages (Malchukov et al. in prep). Such approach also visualize the
idea of the level of granularity of analysis: the less granular is the approach, the
more microroles are lumped together.
In the present paper, I will focus on an unusual way of marking of one specific
participant in one linguistic family. I will consider coding the addressee of speech
in a speech act situation in East Caucasian languages. The fact that the strategy
used here is strikingly different from what we observe in the languages of Europe
requires positing a separate semantic role for the addressees with verbs of speech.
I will primarily rely on noun phrase marking; obviously, this is a clear case of
argument-centered approach. I will briefly come back to the discussion of the dif-
ference between the predicate-centered and argument-centered approaches in the
conclusion.
Obviously, for many there is no direct mapping between semantics and mor-
phosyntax; this was in fact the original source of the notion of deep cases (­Fillmore
1968) that later evolved into semantic roles. The same role may be expressed by dif-
ferent cases and the same case may express different roles, and this is true even in
role-oriented languages (especially for the core, or structural, cases). The variation
is also cross-linguistic: for instance, Experiencers may be marked by the nomina-
tive (e.g. many European languages), by the dative (e.g. Slavic) or by a dedicated
case (e.g. in some East Caucasian languages). However, in the argument-centered
approach, it is clusters of participants typically merging in one marking that pri-
marily determine the universal inventory of the semantic roles. This seems to fol-
low from the outline of the empirical typological approach to semantic roles in
Comrie (1981). Participants that are marked differently in many languages repre-
sent different semantic roles. Participants that are repeatedly marked in the same
way in many languages are assumed to represent the same or related semantic
roles.
Distinct but similar participants, corresponding to different but related
semantic roles, may be conflated. In argument-centered approach that means
that different semantic roles may be combined in one marking. There is ­probably
 Michael Daniel

no language which uses special marking for natural forces, which either go with
Agents (even though natural forces lack intentionality) or with Instruments
(even though natural forces do not require the presence of an Agent). Yet, many
approaches isolate them as a separate semantic role, including both Comrie (1981)
and Van Valin and LaPolla (1997: 118 ff.), even if methodological grounds vary
greatly. Natural forces are not conceptually identical to Agents or Instruments.
However, they are not central to the linguistic construal of the world, and human
language does not afford a special marking for them. Their sharing some impor-
tant properties with doers and some other properties with tools allows construing
natural forces as either Agent or Instrument, the two more central and universally
distinguished semantic roles.
There are varying degrees of participant distinctions, which is also reflected
in their linguistic treatment. Some differences are more and some are less fine-
grained. Runners and walkers are probably never differentiated in any language;
they are so similar that they never diverge in a morphosyntactically relevant way –
requirement put forth in, e.g. Van Valin and LaPolla (1997). Distinguishing nat-
ural forces from other ‘doers’ is obviously more important. In Russian, natural
forces show their intermediate position between Agents (by appearing as subjects
in transitive predications) and Instruments (by appearing as instrumental NPs in
intransitive Agent-less impersonal constructions with a DO). In RRG, the dis-
tinction is made at the level of different thematic roles. However, they are not
cross-linguistically salient, since they never receive dedicated marking. Finally,
distinguishing between those who act and those who are affected by acting on
them is a universal pivot of linguistic construal of participancy (the macrorole
level of Undergoer and Actor in RRG; consistently different case marking in many
languages).
Linguistic relations between participants are further obscured by (assumed)
linguistic metaphors. At the lexical level, metaphors may seem to be uncontrover-
sial. Thus, hand in the sense of ‘help’ in give a hand or need a hand is an obvious
metaphor of supporting someone with your hand(s). But marking natural forces
as Agents or Instruments may also be interpreted as metaphorical. Conceiving a
natural force as an Agent could be considered an animistic metaphor, and conceiv-
ing it as an Instrument could be considered a metaphor of a supernatural being
acting through a natural force.
Another example is the cross-linguistically widespread marking of one of the
participants in situations designated by predicates such as ‘be afraid’ and ‘avoid’
with an ablative (elative) case or adposition. On the other hand, no one would
claim that the situational roles of the participants corresponding to X in ‘be afraid
of X’ and ‘come from X’ are identical. We could include such ‘unpleasant’ and
 Against the addressee of speech – Recipient metaphor 

‘avoidable’ stimuli as a separate item in the inventory of semantic roles (let us say,
Flee’ee). But at a higher level of granularity, we could also include these participants
into the role of the Source (as with motion verbs) by saying that the whole situa-
tion is construed metaphorically as a situation of avoidance and flight. Whether
the metaphor is called for or a separate role is argued for in a sense depends on
the purposes of the analysis. There is usually no empirical evidence in favor of the
metaphor other than common flagging; for instance, no psycholinguistic evidence
is provided.
This paper is concerned with the marking of the addressee of speech verbs. This
participant is cross-linguistically often marked by the dative, i.e. in the same way as
Recipients. This is commonly explained by a reference to a metaphor of information
transfer (see e.g. Janda 1993 for Czech, Dąbrowska 1997 for Polish; Goldberg 1995
for English). The addressee (→Recipient) receives the information (→Theme) that
is transmitted (→given, transferred) to him or her by the speaker (Agent).
(1) English (constructed): dative coding on the Recipient
I gave all my bread to the poor fellow.

(2) English (constructed): dative coding on the addressee

And I said to the poor fellow: that was all the bread I had.

The present paper investigates the universal validity of the assumed metaphor.
Dative marking is not the only way for a language to construe the addressee
of a speech verb. Some speech verbs in European languages allow constructions
with coding other than the dative, introducing the addressee by means of a prepo-
sitional phrase with clearly spatial semantics which may be a marked alternative to
the dative case marking with the same verb, as in (3a–b):
(3) Acts 12: 21–22: spatial preposition instead of the dative in Russian and
German:
a.  V naznačennyj den’ Irod, odevšis’ v carskuju odeždu, sel na vozvyšennom
meste i govoril k nim; a narod vosklical: golos Boga, a ne čeloveka.
(­Synodal Bible)
Aber auf einen bestimmten Tag tat Herodes das königliche Kleid an,
b. 
setzte sich auf den Richtstuhl und tat eine Rede zu ihnen. (Luther)
‘And upon a set day Herod, arrayed in royal apparel, sat upon his
throne, and made an oration unto them.’ (King James)

These examples show alternatives to the dative coding of the addressee. E ­ xample (3)
is highly marked stylistically, archaic. In Russian this option is not available for the
verb skazat’ ‘say’, but in English some contexts also allow say towards or say unto.
Similar examples are available German (sagen zu + Dat).
 Michael Daniel

(4) English (random search in Google)

In cross-examination, he spoke towards the Judges rather than the witnesses.
(5) English (random search in Google)
Has a racist comment been said towards you before?
(6) German (random search in Google)
Was hat der Gastwirt zu mir gesagt?

In all these examples, the addressee of speech appear to be construed as a partici-

pant towards whom the speech act is directed. The spatial construal and the nature
of marking is best compatible with viewing the addressee as a Goal. Note that the
Recipient in (1) is also often considered to be a Goal, as verbs of giving are in many
ways similar to caused motion verbs. In this sense, the addressee in (2) is in a way a
Goal anyway, but it is presumably a metaphorical Goal derived from the Recipient,
because it uses not just any Goal marking but specifically the marking used for the
Recipient. In (3) through (6), on the other hand, the addressee is a Goal which is
independent of the metaphor of transfer.

a. b.
Goal Goal

Recipient Recipient Addressee

Addressee

Figure 1. Alternative models of Recipient~Goal~Addressee relations

The Examples (3) to (6) from Russian, English and other European lan-
guages show that marking of addressees by means other than the means used for
Recipients in ditransitive constructions (below I will refer to such means, whether
affixal or adpositional or other, as Recipient marking) is peripheral in these lan-
guages; and (3) sound clearly archaic. Rare case frames with specific verbs may be
argued to be motivated, among other, by the individual life stories of the lexical
items. Thus, Russian obratits’ja k ‘address someone’ (lit. ‘to(wards)’) goes back to
the mediopassive verb ‘turn’ which naturally introduces the direction (lit. ‘turn
towards’). But what is peripheral in some languages may be salient in some other.
The evidence for a consistently spatial construal of speech acts is abundant in East
Caucasian languages.
Below, I will provide a brief overview of those aspects of East Caucasian nomi-
nal morphology that are relevant for the discussion (Section 2) and an analysis of
 Against the addressee of speech – Recipient metaphor 

addressee marking with the basic verb of speech in East Caucasian (Section 3).
Section 4 opens with a claim that the main way to construe the addressee in East
Caucasian languages is spatial, related not to the Recipient but directly to the Goal
(see Figure 1 above). Then I consider several objections against this claim and
discard them. Section 5 is a conclusion which discusses again the choice between
argument-centered and predicate-centered approaches. It shows that, under the
argument-centered approach, East Caucasian addressees are clearly a separate
semantic role, and not merely a metaphorical extension of the Recipient. I then
come back to dative addressees in SAE and suggest that the syncretism of the two
functions (Recipient and addressee of speech) should not necessarily be consid-
ered as a metaphor in European languages, either. It can be a merger of two dis-
tinct functions at a lower level of semantic detail. Roughly, the addressee merges
with the Recipient not because giving is similar to transferring information but
because they are both animate Goals. In Section 6 I discuss some possible paths
one could follow to probe the hypothesis that the addressee of speech verbs is a
role independent from the Recipient in a wider typological perspective.

2. East Caucasian languages: Background information

East Caucasian is a language family spoken in the Northern Caucasus. It includes

the Nakh, Tsezic, Avar-Andic, Lak, Dargwa, Lezgic and Khinalug branches.
The Lak and Khinalug branches are represented only by one language each, and
Dargwa is traditionally considered to be a dialect cluster which only has one stan-
dard (written) language, though the linguistic divergence between these dialects
is quite considerable. Below I indicate branch affiliations for all languages that I
mention except those that form their own branches (Lak, Khinalug, and, for the
sake of convenience, also Avar); and consider Dargwa idioms to be separate lan-
guages of the Dargwa branch. Only references to papers treating specific questions
and to the sources of language examples are provided in the text. The references
to basic grammars of the East Caucasian languages, which are the source of some
language specific data, are given in the bibliography list at the end of the paper.
The East Caucasian languages are consistently ergative. They typically have a
rich system of noun classes but lack personal agreement, and widely use converbs
in polypredicative constructions (see Daniel & Lander 2011 for a brief overview).
Their most salient typological property is an abundant nominal declensional
morphology which may include many dozens of forms. The nominal paradigm
falls into two separate and by and large independent subsystems. One subsystem
includes what is called syntactic cases and is similar to case inventories of many
languages of the world, typically from four to seven distinct cases. The inventory
 Michael Daniel

usually includes nominative,1 ergative, genitive and dative, sometimes extended

by additional cases such as comitative, instrumental or comparative, but this is
subject to strong variation across the family. Typologically rare is the affective case,
present in Tsakhur (Lezgic) and several Andic languages. Its central function is
marking the Experiencer with some experiential verbs; other experiential verbs
as well as most experiential verbs in other East Caucasian languages use dative
marking for the Experiencer. Experiential verbs in East Caucasian are thus dative-
oriented and only rarely aligned with the transitive prototype.
The other subsystem may be called spatial (or locative) subparadigm. The
forms in the spatial subparadigm contain two separate morphological categories,
localization and orientation. Localization designates a spatial domain with respect
to the landmark. Thus, in Archi (Lezgic) the meaning ‘on the rock’ is expressed
by bound morphology – qʷˁen-ni-t rock-obl-super – where the rock is the land-
mark, obl is a semantically empty suffix forming the stem which is used in oblique
cases, and super is the localization marker meaning ‘on’. There are few East Cau-
casian languages that lack the category of localization (almost) completely – Udi
(Lezgic) and Chechen and Ingush (Nakh). Other languages count from two up to
six localizations, including the following, as shown on Table 1:

Table 1. Localizations in East Caucasian

In – location within a hollow landmark, such as a box or a house
Inter – location inside a landmark which is a mass, such as liquid or powder, or among
multiple landmarks, such as among the trees
Super – location on a (typically horizontal) surface of the landmark: e.g. on the roof
Sub – location under the landmark: e.g. under the bridge
Post – location behind the landmark: e.g. behind the fence
Apud or Ad – location in the vicinity of the landmark: e.g. by the river or near the wall
(some languages distinguish two ‘vicinity’ localizations)
Cont – location on the surface of the landmark to which the object is secured, such as a
picture on the wall

The second morphological slot in a spatial form is the marker of orientation.

This category indicates the movement of the main participant (the nominative
noun phrase) with respect to the spatial domain specified by the localization
marker. The most usual orientation markers include elative (leaving the spatial

. See (Kibrik 2003a) for a justification of the use of the term nominative (rather than abso-
lutive) in ergative languages.
 Against the addressee of speech – Recipient metaphor 

domain) and lative (movement into it). Less frequent are allative alias versative
(movement towards the spatial domain) and translative alias prolative (movement
through the spatial domain). Essive (state of quiescence) is the unmarked member
of all oppositions, both semantically (absence of movement) and morphologically.
(7) Orientation hierarchy in East Caucasian2
essive vs. elative > vs. lative > vs. allative and/or translative

As the hierarchy in (7) shows, the minimal distinction is that between the essive
and the elative. In most languages, the essive is the formally unmarked member.
In Dargwa languages, however, the essive form is different from the lative in that
it additionally contains an agreement slot (controlled by the nominative noun
phrase) and the unmarked form is thus apparently the lative:
(8) standard Dargwa (Musaev 1999): lative unmarked, essive contains an
­agreement marker
stol-li-či vs. stol-li-či-b
table-obl-super(lat) table-obl-super-3(ess)
‘(lay) onto the table’ ‘(lie) on the table’

When there is no special lative marker, the unmarked form is used both in essive
and lative function. The languages of the Tsezic branch show internal variation
with respect to this parameter. They share some cognate localization markers, e.g.
the localization variously labeled Cont, Ad or Poss (-qo in Hinuq and Tsez, -qa
in Khwarshi and Bezhta, -go in Hunzib). East Tsezic languages Bezhta and Hunzib
do not have any lative morphology, and the essive is used in this function. West
Tsezic languages Tsez, Khwarshi and Hinuq have introduced the lative marker -r
(-l in Khwarshi). This marker is optional in at least some lative contexts, and the
unmarked localization (essive) may be used instead. But at least for Khwarshi,
in the majority of textual occurrences the lative meaning is conveyed by a lative-
marked form, and in elicitation consultants clearly prefer explicitly lative forms.
A plausible conclusion would be that the marker is a shared innovation in West
Tsezic and that the use of the essive for the lative in West Tsezic is an archaism.

. Note that the term essive is used in a sense different from the sense it has in Finnic and,
to a certain extent, general linguistics, designating the function of a co-predicative noun as in
He acted as a real warrior and similar – the function for which the term functive was recently
suggested by Denis Creissels (see Creissels in prep.). Also note that, for the sake of brevity, the
terms elative, lative, essive etc. are used with reference both to markers and functions. When
this may lead to a confusion, the two meanings are explicitly disambiguated (e.g. in Khinalug
the essive and lative functions combine in one marker).
 Michael Daniel

Some spatial forms of Lak are given in Table 2 as examples of a typical East
Caucasian spatial paradigm:

Table 2. A subset of the spatial subparadigm from Lak from the total of about 30
q̄at-lu-w q̄at-lu-wu-n q̄at-lu-w-a
house-obl-in house-obl-in-lat house-obl-in-el
‘in the house’ ‘into the house’ ‘from inside the house’
q̄at-lu-j q̄at-lu-j-n q̄at-lu-j-a
house-obl-super house-obl-super-lat house-obl-super-el
‘on the top of the house’ ‘onto the house’ ‘from the top of the house’
q̄at-lu-χ q̄at-lu-χu-n q̄at-lu-χ-a
house-obl-post house-obl-post-lat house-obl-post-el
‘behind the house’ ‘to the back side of the house’ ‘from behind the house’
(adapted from Žirkov 1955: 36–37)

Several languages do not distinguish between lative and essive meanings at

all, e.g. Khinalug, or in some localizations, e.g. cont or inter in Bagvalal (Avar-
Andic). The unmarked form is used for both essive and lative meanings, but is
conventionally called essive.
(9) Bagvalal (Kibrik et al. 2001): essive proper
kort-i-ɬi ek’ʷa-ba-lu-r b-as’i-m-o
gathering.place-obl-inter be-ptcp.hpl-obl-erg n-ask-n-cvb
ek’ ʷa: “he-b-i, ʁaziħand-e, χ:eχ:-l-u-w?”
be    what-n-intrg Hazi.Mohammad-voc quick-vblz-ptcp-m
‘Those who were at the gathering place were asking: Hazi-Mohammad, why
are you in such a hurry?’

(10) Bagvalal (Kibrik et al. 2001): essive in the lative function (covert lative)
šamil-i-r w-alli-w-o ek’ʷa čãws kort-i-ɬi.
Shamil-obl-erg m-call-m-cvb aux Chaus gathering.place-obl-inter
‘Shamil invited Chaus to (come to) the gathering place.’

Tsezic (except Hinuq) and Lak use the same marker for the lative in the spatial
subparadigm and for the dative in the inventory of syntactic cases. The marker is
thus shared by the two otherwise independent morphological subsystems of the
nominal declension. Still, the word forms are different – the ‘spatial’ lative marker
obligatorily follows one of localization markers, while in its function of a dative it
is attached to the nominal stem:
 Against the addressee of speech – Recipient metaphor 

(11) Lak: same morphological marker used in the dative and lative forms
(Žirkov 1955: 39)
duš-ni-n vs. duš-ni-x:-un
girl-obl-lat(=dat) vs. girl-obl-apud-lat
(both can be translated as ‘to the girl’, depending on the context)

The dative and one of the lative forms from the spatial subparadigm may be con-
flated completely, as in Itsari (Dargwa) (Sumbatova & Mutalov 2003). In Itsari, the
dative and SuperLative functions are syncretically conveyed by the form which
was, by all appearance, originally SuperLative. Only two personal pronouns, ‘I’ and
‘you.sg’, have a syncretic dative/SuperLative form that goes back to the original
dative (otherwise lost). From the East Caucasian perspective, the dative use of this
apparently ‘lative’ marking in Tsezic, Lak and Itsari is independent from its spatial
use. It has all the dative functions typical of an East Caucasian dative: not only the
spatial function of marking the Goals with verbs of contact (‘touch’, ‘hit’, both with
animate and inanimate nouns) conveyed by the dative in many East Caucasian
languages (cf. 12), but also clearly abstract dative functions such as Recipient and
Experiencer (cf. 13). Though by no means incidental, this merger may be disre-
garded for the purposes of the present paper.
(12) Khwarshi (Khalilova 2009: 87): spatial use of lative marker
abaxar-i m-oc-un iɬe-s kode γono-qo-l.
neighbor-erg 3-tie-pst.uw that.obl-gen1 hair tree.obl-cont-lat
‘The neighbor tied her hair to the tree.’
(13) Khwarshi (Khalilova 2009: 87): dative use of lative marker (Experiencer)
hobože isual bulh-un idu himon.
now that.lat understand-pst.uw this thing
‘Now he understood this (thing).’

Although the spatial subparadigm is in most cases formally delimited from the
inventory of syntactic cases, many forms have uses that bear little similarity to
spatial semantics. Thus, in Agul (Lezgic), Apud-Essive marks the Causees with
ditransitive causatives, Super-Lative marks Instruments, Super-Elative marks
the benchmark of comparison (as in ‘higher than X’), Apud-Elative marks unin-
tentional Agents etc. Certain forms may lose their (assumedly) original spatial
functions and shift to the abstract domain completely. Thus, in Bagvalal (Andic),
Cont is very rare in spatial contexts; instead, it expresses temporary Possessor
or transitive Causee (Cont-Essive) or retrospective Possessor or unintentional
Agent (Cont-Elative). In Khinalug, the spatial subparadigm includes four forms
only, with two localizations both distinguishing between essive and elative (lative
meaning is conveyed by the essive). But the first localization is primarily used
 Michael Daniel

for ­benchmarks of comparison, both as essive and elative (the difference between
these two forms is unclear). The second mostly combines with human nouns to
designate temporary Possessors and retrospective Possessors, and is thus labeled
Poss. Even if some of these ‘abstract’ uses preserve more or less strong spatial con-
notations (cf. Daniel et al. 2010 on temporary Recipients as Goals), the Khinalug
spatial subparadigm is thus constituted of two localizations whose primary func-
tions are not spatial but rather abstract. I will come back to this point in Section 4.
One final preparatory point concerns the functional range of the dative case in
East Caucasian. Datives often develop from latives, and indeed we have seen above
that in Lak and in Tsezic the dative marker is identical to the lative morpheme.
More generally, the dative is the point where the partition between the spatial and
syntactic subparadigms tends to leak. While in some languages, e.g. Bagvalal, the
dative seems to be deprived of any kind of spatial meaning, many East Caucasian
languages use the dative in contexts related to the lative. An example of a restricted
spatial dative is attested in Archi, a Lezgic language, where the dative is used with
a limited set of contact verbs (14) and with some spatial postpositions (15). Some
other Lezgic languages, e.g. Agul and Lezgian, in addition to the verbs of contact,
also uses the dative in the lative function with some names of locations or other
nouns that may serve as landmarks (16–17); this fact will become important below.
(14) Archi (Arkhipov et al. unpublished)
nac’-a aˁnš caχ-u-li ju-w-mi-s.
bird-obl(erg) apple 4.drop-pfv-evid this-1-obl-dat
‘The bird dropped an apple on him.’
(15) Archi (Arkhipov et al. unpublished)
tij-maj ju-w l’inčas̄ i‹w›χmu-s bo-li
that.pl-obl(erg) this-1 afraid ‹1›lv-inf say.pfv-cvb
kammu-li ju-w-mu-s harak os noc’.
throw.pfv-evid this-1-obl-dat in.front.of one bird
‘To scare him, they threw a bird in front of him.’
(16) Agul (Ganenkov et al. unpublished)
ajč’.u-na ʕʷ.a-j-e te ti-č, q:irʁist:an.di-s.
go.out.pfv-cvb go.ipfv-cvb-cop that that-lat Kyrgystan-dat
‘He leaves and goes there, to Kirghistan.’
(17) Lezgian (Haspelmath 1993: 89)
pačah.di-n rik’ ala-j dewe müq-w.e-laj
king-gen heart be.on-ptcp camel bridge-super.el
wac’.u–z awat-na.
river-dat fall.off-aor
‘The king’s favorite camel fell from the bridge into the river.’
 Against the addressee of speech – Recipient metaphor 

After this brief introduction to the East Caucasian nominal morphology, I can
proceed to the different strategies of addressee marking present in the languages
of the family. For a survey of East Caucasian nominal morphology, see (Kibrik
2003b; Daniel & Ganenkov 2009); and for a more detailed analysis of the spatial
subparadigm, see (Testelets 1980; Ganenkov 2010).

3. Addressee marking in East Caucasian

Not only in SAE, but also in a great many languages of the world, the addressee
of verbs of speech may be marked with the dative. This type is attested in various
Lezgic languages, e.g. in Agul or Udi:
(18) Agul (Ganenkov et al. unpublished): dative addressee
aχpːa ħabaw.a uč.i-as sa ʡaji
then granny(erg) self-(in)el one very
šüt’ atː.u-ne ʁaj uč.i–s hage gaf p.u–f,
pinch {in}take.out.pfv-pf quot self-dat this word say.pfv-a
uč.i-s sa ʡaji itːar-aq’.u=ne ʁaj, ħabaw.a.
self-dat one very ache-do.pfv=pf quot granny(erg)
‘Then, Granny said, this (woman) pinched me, when they told (her) this,
and this pinch really hurt, Granny (says).’
(19) Udi (Ganenkov et al. unpublished): dative addressee
me išqːar-en čoban-a p-i=ne=ki dirispːaka,
this man-erg shepherd-dat say-aor=3sg=cmpl thank.you
gele iräzi=zu va-χun.
much pleased=1sg you.sg-abl
‘This man says to the shepherd – thank you, he says, I’m very pleased
(by you).’

Other languages that follow the same model are Lezgian (Haspelmath 1993: 87–88)
and Tabassaran (e.g. in the texts from Magometov 1965), which, together with
Agul, form a closely related sub-unit of the Lezgic branch. Let us call this a dative
strategy of addressee marking. Other Lezgic languages (Archi, Rutul, Tsakhur,
Budukh, Kryz) and all other branches of the family use other marking patterns.
The major pattern, however, is expressing the addressee by a lative form of one
of the localizations; which localization is used varies across languages (see Table 3
below). Languages that show this lative strategy include all Nakh languages, vari-
ous Dargwa idioms, West Tsezic languages (Khwarshi, Hinuq, Tsez; but see below
on the variation) and several languages of the Andic branch (Bagvalal, Karata,
Akhvakh). Cf. ex. from Chechen (Nakh), Bagvalal (Andic), Itsari (Dargwa) and
Khwarshi (Tsezic):
 Michael Daniel

(20) Chechen (Molochieva et al. unpublished)

hwai k’ant-ie zuda c’a-j-a-iita aala aella.
you.sg.gen son-all woman home-j-come-caus.imp say.imp quot
‘Tell your son to send his wife (back) home, they say.’
(21) Bagvalal (Kibrik et al. 2001)
qoča-s’ok b-uh-ē-r-ō-b dan ek’ʷa” —
sheep.skin-goat.skin n-gather-caus-ipfv-ptcp-n thing be
hel’i šamil-i-r čãws-i-lā.
say Shamil-obl-erg Chaws-obl-super.lat
‘“Skins are objects that are reducible”, said Shamil to Chaws.’ [A pun-threat:
skins are easily shrunk/taken away from the owner]
(22) Itsari (Sumbatova & Mutalov 2003)
du-l cēl b-uχ:-a-t:a,
I-erg something n-know:ipfv-progr-1
ic:i b-ursan-nejk:ʷi.
you:in.lat n-say:ipfv-oblig-cvb.neg.1
‘I know something, but I won’t tell you.’
(23) Khwarshi (Khalilova, p.c.)
uža ešet’-qo-l bit’t’ura-l himon il-in.
boy.erg mother.obl-cont-lat true-4 thing say-pst.uw
‘The boy told the truth to his mother.’

The dative and the lative strategies above form a parallel to the choice of addressee
marking in SAE. The dative strategy, present only in Lezgic of all branches of the
family, corresponds to the ‘mainstream’ SAE strategy which identifies addressees
and Recipients (Example 2). The lative strategy in Archi or Itsari, even though con-
veyed by inflection rather than by a prepositional phrase, corresponds to the non-
Recipient spatial prepositional strategy (say towards, parler envers etc. – Example 3
to 6). Apparently peripheral in SAE, it is the main pattern throughout the East
Caucasian.
Now, what happens in other East Caucasian languages? Many use one of the
essive forms. The presence of the lative component in addressee marking is central
to the argument of this paper, so I will look into the use of the essive, which clearly
splits into two different types.
The first type is apparently different from the lative strategy above but will be
argued to represent the same underlying motivation. It is attested in East Tsezic
(Hunzib and Bezhta), few Andic languages (Andi, Tindi, Chamalal, Godoberi)
and in Khinalug, and also in West Tsezic as an alternative to the lative strategy
(Diana Forker, p.c., Zaira Khalilova p.c.):
 Against the addressee of speech – Recipient metaphor 

(24) Khinalug: essive addressee (Arkhipov et al. unpublished)

χinˌimk’ir t -uza-r la.t�ɨr-zɨ-ʁ-sɨn, ɨst’armejramˌ-i
woman court-loc move-2-go-cvb witch-erg
misi riši-š l-i=ki…
little girl-poss say-ptcp=cmpl
‘The woman went out into the court, and the witch said to the girl…’
(25) Chamalal (Bokarev 1949: 49): essive addressee
wah-u-d il’ im-u-q…
son-obl-erg say.pst father-obl-ad
‘The son said to (his) father…’
(26) Hunzib (van den Berg 1995: 185): essive addressee
ball’o boɬu ʕadam-li-l bulii deno j-āq’-oɬ
in.the.end this.obl person-obl-erg home back 2-come-cvb.temp
bet’erhan-li-g nɨsən li,
husband-obl-ad say-cvb be.5
“ičil suku-l dibira-l di-go bədaa-s rαže nɨs.”
old man-erg mullah-erg I.obl-ad so-gen word say.prs
‘In the end, the woman said to her husband when she came back home,
“That old mullah says such words to me.”’ (Text 18: 7)

These cases may be interpreted in the same way as the lative strategy above, because
the localizations Ad in Chamalal and Hunzib and Poss in Khinalug simply do not
distinguish between the essive and the lative. The same form is used to convey
both meanings (see 7 above for a discussion of the essive-lative syncretism):
(27) Chamalal (Bokarev 1949: 49):
a. essive use of the ad-essive
woh-u-q j-ak’u-da ida bac’-e.
wood-obl-ad npl-be-fut.cvb cop wolf-pl
‘There are wolves in the forest.’
b. lative use of the ad-essive (covert lative)
sā b-uc’u-da ida woh-u-q.
fox 3-run-fut.cvb cop wood-obl-ad
‘A fox runs to the forest.’

For the purposes of this study, it is irrelevant how consistent is the conflation of
the essive and lative functions in the spatial system of the language in question. In
Chamalal, at least some other localizations formally distinguish the two meanings
(e.g. Sub-Lative vs. Sub-Essive forms); or the essive and lative functions may be
combined in one “essive” form throughout the whole spatial subparadigm, as in
Khinalug or Hunzib. The important thing is that the lative and essive ­functions are
 Michael Daniel

not distinguished in the localization which is used for the marking of the addressee
(poss in Khinalug, ad in Chamalal and Hunzib: cf. 24–26). Such uses of the essive
for the addressee of speech may be interpreted as covert latives in the same way as
the lative function is covert in (27b).
For East Tsezic languages this interpretation is indirectly supported by a com-
parison with the West Tsezic contexts which use explicitly lative forms (see 23
above), sometimes as an alternative to the less frequent essive marking of the same
cognate localization (-qV). Above, I have suggested that this preferred (at least
in Khwarshi) but optional use of the lative in West Tsezic is a sign of innovation,
while essive contexts reflect an earlier stage of the system where the essive was the
main means to convey lative semantics.
Avar, Botlikh (Andic) and Lak also use an essive form, but represent a much
more problematic case. The localization in question distinguishes the essive and
lative form; and yet, these languages choose the essive form. I do not have a satis-
factory explanation of this choice, and the issue will be only briefly addressed in
Section 4.
(28) Lak (Žirkov 1955: 43): essive addressee
duš-ni-x uss-i-l barčallah u‹w›k-una.
girl-obl-ad brother-obl-erg thank.you ‹1›say-pst
‘(My) brother said ‘thank you’ to the girl.’

Finally, three Lezgic languages use a form which is not a dative, on the one hand,
but does not have any purely spatial functions, on the other. Budukh and Kryz,
two closely related Lezgic languages in Northern Azerbaijan, use such forms:
(29) Kryz (Authier 2009: 270)
zin zi-va-van li-di-p-ǯi-m-ni ki
I I-you.sg-adr prvb-neg-say-pf.intrg-intrg-pst cmpl
čara yaʕ ?
solution exist
‘Didn’t I tell you there was a solution?’

The Kryz case marker is termed addressive by Authier (2009). It is also used as a
marker for Causee, but is rare in this function, where it competes with the dative
and has clearly lost the competition. Occasional dative marking on the addressee
of speech verbs in the texts and examples in the grammar may be due to Azer-
baijani influence. In the Kryz of the village of Alik, which is the dialect described
by Gilles Authier, the addressive, now unanalyzable and limited to abstract uses,
seems to contain the original lative marker -n. This is supported by the data from
a dialect spoken in the village of Kryz, where the form still preserves some lative
uses (Gilles Authier, p.c.).
 Against the addressee of speech – Recipient metaphor 

Kryz’ closest sister language, Budukh, marks addressees in the same way as
benchmarks of comparison and unintentional Agents (a special function reported
in East Caucasian for modal constructions such as ‘He did (not) manage to P’ or
‘He caused P unintentionally’), the latter two functions expressed by elatives in
various East Caucasian languages (with a notable exception of the Tsezic branch –
Cysouw & Forker 2009).
(30) Budukh (Gilles Authier, p.c.)
dide-ri a-n-ovon/a-n-ovor jɨpa-ǯi.
mother-erg that-h-ad.lat/that-h-ad.el say.pf-narr
‘Mother said to him 〈…〉.’

At first glance, Budukh seems to counter-evidence the lative strategy as defined

above. However, according to Gilles Authier, this is a result of a functional con-
tamination of the former ad‑lative marker -von (cognate to the Kryz addressive
in 29) and the former ad‑elative marker -vor into a syncretic ad(e)lative marker.
As the Example (30) shows, these are in fact two forms, that are interchangeable in
this (and all other) contexts.
The most problematic is Tsakhur which uses the affective case (a case other-
wise used for Experiencers):
(31) Tsakhur (Kibrik et al. 1999: 788–9)
gi‹w›xʷ-īnɢaˁ, manke sudje-j-k’le iwho wo-d:
‹3›put.pfv-temp then judge-obl-aff say.pfv be-4
“ma-ni xunašː-e-k-e qidɣɨn heʔ-e, ȝ̌e-ni
   this-aobl woman-obl-cont-el question 4.do-imp self.obl.2-aobl
sumk’ˌ-ē nimā-n pɨl-nī wo-d-ɨn?”
bag-in how.much-atr money.4-intrg be-4-atr
‘He puts (the purse) down and says to the judge: ask this woman how much
money was there in her purse.’

The solution of the problem of Tsakhur marking may be similar to – though even
more speculative than – the one suggested for Kryz and Budukh, i.e. to be based
on the history of the marking. Affectives are likely to originate from latives. Thus,
in Bagvalal, it preserves some (though very limited) lative functions, e.g. latives
of local place names (Kibrik et al. 2001). More generally, East Caucasian Experi-
encers are rather dative-oriented (Kibrik 2003a; Daniel & Ganenkov 2009). On
the other hand, a comparison of the Tsakhur affective marker -k’le with Rutul, a
related Lezgic language, shows an at least visible similarity with an elative form
in one of the spatial series. The connection between the lative marking and the
addressee of speech in these Lezgic languages remains to various degrees (depend-
ing on the language) speculative and, in the case of Tsakhur, clearly problematic.
 Michael Daniel

The East Caucasian data discussed above are summarized in Table 3:

Table 3. Strategies of addressee marking in East Caucasian3

Strategy Language Branch Marker Source

1. Lative Ingush Nakh [Allative] -ga (Nichols 2011: 421)

2. Lative Chechen Nakh [Directional] -(g)a (Zarina Molochieva,
p.c.)
3. Lative Bats Nakh [Directional I] -go (Zarina Molochieva,
(Tsova-Tush) p.c.)
4. Lative (or Hinuq West [At-Lat] -qo-r (Forker 2013, p.c.)
covert) Tsezic ([At] -qo)
5. Lative (or Khwarshi West [Cont-Lat] -qa-l (Khalilova 2009)
covert) Tsezic ([Cont] -qa)
6. Lative (or Tsez West [Poss-Lat] -qo-r (Zaira Khalilova,
covert) Tsezic ([Poss] -qo) p.c.)
7. Lative (covert) Hunzib East Tsezic [Ad-essive] -go (van den Berg
1995: 46)
8. Lative (covert) Bezhta East Tsezic [“-qa series” ~ (Zaira Khalilova,
Poss] -qa p.c.)
9. Essive Avar (Avar) [“Loc I” ~ Super] (Bokarev 1949: 192)
-da
10. Essive Botlikh Andic [“Loc III” ~ (Gudava 1967: 299,
Apud] -χi 306)
[“Loc IV” ~
Apud] -qi
11. Lative (covert) Andi Andic -qi (Salimov 2010: 111)
12. Lative (covert) Chamalal Andic [Ad-essive] -q (Bokarev 1949: 49)
13. Lative (covert) Godoberi Andic [Ad-essive] -qi (Kibrik 1996)
14. Lative (covert) Tindi Andic [“Loc III” ~ (Magomedova
Loc?] -qa 2003: 162, 549)
15. Lative Bagvalal Andic [Super-lative] -lā (Kibrik et al.
2001: 222)
16. Lative Karata Andic [“CFG3” ~ (Pasquereau
‘abstract’ allative] 2010: 41)
-a-r
17. Lative Akhvakh Andic [General allative] (Creissels,
-g-a manuscript)

(Continued)

. Light shading in the language name and branch columns branch is used to visually delimit
languages of the various branches. The second column shows the type of addressee marking;
here, shading shows problematic cases.
 Against the addressee of speech – Recipient metaphor 

Table 3. (Continued)

Strategy Language Branch Marker Source

18. Lative Akusha Dargwa [In-lative ~ (Musaev 1999)

Inter-Lat?] -zi
19. Lative Itsari Dargwa [In-lative ~ (Sumbatova
Inter-Lat?] -c:i & Mutalov
2003: 29, 184)
20. Lative Mehweb Dargwa [In-lative] -ze (Magometov 1982: 46)
21. Essive Lak (Lak) [Locative VII ~ (Žirkov 1955: 43)
Ad] -x
22. Dative Lezgian Lezgic [Dative] -s (Haspelmath
1993: 87–88)
23. Dative Tabasaran Lezgic [Dative] -s (Magometov 1965)
24. Dative Agul Lezgic [Dative] -s (Ganenkov et al.
unpublished)
25. other Tsakhur Lezgic [Affective] -k’le (Kibrik et al. 1999)
26. Lative Archi Lezgic [Cont-allative] (Kibrik 1977: 181)
-r-ši
27. Lative (covert) Rutul Lezgic [Ad] -da (Makhmudova 2001)
28. (*Lative?) Budukh Lezgic [Ablative IV] (Talibov 2007: 97)
-vor/-von/-van
29. (*Lative?) Kryz Lezgic [Addressive] -van (Authier 2009: 222)
30. Dative Udi Lezgic [Dative] -a (Ganenkov et al.
unpublished)
31. Lative (covert) Khinalug (Khinalug) [Poss-essive] -š (Kibrik et al.
1972: 141)

For Dargwa languages, most of which seem to use the lative strategy, I only
quote the standard variety (Akusha), Mehweb and Itsari; see also Map 1 in the
appendix. It should be emphasized that Table 3 is not intended to show an exhaus-
tive list of case frames available for all verbs of speech in East Caucasian. This
paper is an attempt to present alternatives to the SAE construal of speech verbs, so
it was convenient to leave out lexical diversity and concentrate on one basic verb
of speech in each language (cf. Cysouw & Forker 2009 who draw conceptual space
of the use of spatial markers of many verbs in all languages of the Tsezic branch).
The basic verb considered here roughly corresponds to the English say. It is rela-
tively easy to identify, because this is the most frequent verb of speech across texts,
capable of introducing reported speech clauses, often the diachronic source of
quotative particles, etc. The examples show that this verb is often ‘visibly’ cognate
within branches (comparison between branches needs etymological research) and
shows similar case frames.
 Michael Daniel

Other verbs of speech may behave similarly to or differently from the basic
verb of speech (cf. say vs. tell in English). For ‘tell’, Khwarshi and Hinuq use the
essive form of the same localization as with the basic verb of speech (see Table 3
and Example 23). Both languages are reported to allow essive also with the basic
verb of speech (Diana Forker, p.c.), but with ‘tell’ the essive is the preferred or the
only option (32, 33), while it is only an alternative with the basic verb of speech.

(32) Khwarshi (Khalilova 2009: 84): essive addressee with non-basic verb of speech
me is-o di-qo dubo heč’č’e nucaha-r
you.sg.erg tell-imp I.obl-cont you.sg.gen1 very tasty-4
lil na-sa-k goli lɨn il-in boc’i.
meat where-def-intrg be.prs quot say-pst.uw wolf.obl.erg
‘You tell me, where is your most tasty meat? – the wolf said.’ [Hajj.029]

(33) Hinuq (Forker, p.c.): essive addressee with non-basic verb

šajt’an-za-j hajɬo-qo mol-a-ɬ ʔeli-n: ħaži
devil-obl.pl-erg he.obl-poss sleep-obl-inter speak-pst.uw Hadzhi
caχ-om sabaw-be debez zaral b-uw-a goɬ.
write-proh amulet-pl you.sg.dat harm 3-do-inf be
‘The devils told him in his sleep: Hadzhi, do not write amulets, they will do
you harm.’

Similarly, in Chechen, the basic verb of speech combines only with the allative
(except some secondary uses such as ‘call someone names’), while the verb which
roughly corresponds to the English tell allows both the allative and the dative.

(34) Chechen (Zarina Molochieva, p.c.): dative vs. allative with a non-basic
speech verb
Zaara-s tyyra suu-na/söö-ga d-iic-ira.
Zara-erg tale I.obl-dat/I.obl-all d-tell-pst.w
‘Zara told me a tale.’

Various interpretations of these alternative case frames may be suggested. Thus,

the use of the essive forms in West Tsezic may reflect a relatively recent emergence
of the lative marking in this languages, while the use of the dative in (34) may
indeed be the emerging SAE-style metaphor of transfer. Certainly, all this needs a
more extensive comparative investigation which lies outside the scope of the pres-
ent paper. The conclusion so far is that, in most languages of the family, the basic
verb of speech combines with the addressee marked as a lative rather than suffixed
with the dative. The next section shows that, notwithstanding several complica-
tions, this is enough to say that East Caucasian addressees with the basic speech
verb are different from the Recipient.
 Against the addressee of speech – Recipient metaphor 

4. Discussion

Table 3 suggests that East Caucasian languages show a tendency towards spatial
construal of the situation of speech. The speech is directed – in the literal sense –
and the speaker is oriented towards the addressee. The addressee is construed
as a Goal, separate from the Recipient. Although there are some languages that
apparently follow the SAE strategy (all of them Lezgic), and some other languages
show a totally divergent pattern, this construal seems on the whole quite consis-
tent. Recipients are known to be similar to Goals. Thus, it is not so important that
addressees are marked as Goals but that they are marked as Goals different from
Recipients.
At first glance, it may seems that whenever addressees are expressed in East
Caucasian by a means other than by the dative (mostly the lative strategy above)
they may not be identified with Recipients and that, on the contrary, East Cauca-
sian addressees must be identified with Recipients whenever they are marked by a
dative (the dative strategy).
In fact, on the one hand, East Caucasian Recipients are not always marked
with dative. On the other hand, I will argue that the use of the dative strategy, in
East Caucasian, is not necessarily motivated by identifying the addressee with the
Recipient. The present section will discuss these issues in turn. I will also suggest a
tentative solution of the problem of the languages marked as using essive strategy
on Table 3.
The first complication is that in all East Caucasian languages ‘give’ verbs allow
two constructions, two different strategies of marking the Recipient – the dative
strategy and the lative strategy (the only exception is Udi, Lezgic). In terms of
Van Valin and LaPolla (1997: 127), the two constructions roughly correspond to
the difference between the property transfer verb ‘give’ and the caused motion
verb ‘put’. The peculiar property of East Caucasian languages is that they use the
same verb for both meanings. In (Daniel et al. 2010), this is explained by admit-
ting the two-fold nature of the situation of giving. Giving is motion and transfer
at the same time. The dative strategy in East Caucasian is based on the construal of
the situation of giving as transfer (‘offer’, i.e. transfer of possession); the Recipient
bears a more abstract, dative marking. The lative strategy uses one of the latives
and is associated with the idea of the motion of an object from one person to
another without transfer of possession (‘hand over’), as when lending an object
temporarily. Such lative Recipients have been sometimes called ‘temporary’ Recip-
ients as opposed to dative ‘permanent’ Recipients (e.g. Daniel & Ganenkov 2009).
Although this is not quite precise (see Daniel et al. 2010 for discussion), I will use
these labels for the sake of terminological simplicity.
 Michael Daniel

Cf. the following example from Chechen (Nakh):

(35) Chechen (Molochieva et al. unpublished)

a. juxa v-uod-i i šliapa cynga dʔa-lo.
then v-go-cvb.ant this hat he.all away-give.prs
‘Then (he) goes back and gives the hat back to him.’
b. t’aaq hwoquo qo quor b-ella hoq qa’ k’ant-ana cwacca.
then this.erg 3 pear b-give.pf this 3 boy-dat one.by.one
‘Then he gave each of these boys one more pear.’

Here, the temporary Recipient is marked by the allative, in exactly the same way
as the addressee in Example (20) – note that in giving something back to its true
owner the latter is consistently interpreted as temporary, or lative Recipient in East
Caucasian – see (Daniel et al. 2010) – while the permanent Recipient is marked by
the dative. Similar examples come from Mehweb (Dargwa branch):

(36) Mehweb, Dargwa (personal notes, elicited)

a. aga di-ze eli,
give.pfv.imp I.obl-inter(lat) baby
aħinnanu ħa-ze-la il ar-b-ik-es.
else you.sg-inter-el that prvb-3-fall.pfv-inf
‘Give me your baby, lest you drop it.’
b. ħu-ni di-ze bet’a iši-ni hel
you.sg-erg I.obl-inter(lat) say.pfv.imp this-erg that
ħa-b-aq’-ib ile.
neg-3-do.pfv-aor quot
‘Tell me it wasn’t him who did it!’

Basing on (35a), as compared to (20), and (36), (a) as compared to (b), it can be
argued that there is nothing special about East Caucasian addressees as compared
to SAE. The use of the lative for the addressee is another way to identify it with
the Recipient; it is just that these languages use the lative, not the dative Recipient
for this purpose. East Caucasian then would not represent a deviation from the
well-known tendency to mark addressees in the same way as Recipients. However,
this argument does not hold for several reasons: I argue that, in East Caucasian,
the addressee is a Goal different from both permanent and temporary Recipient.
The most straightforward argument is that the lative used in the lative strategy
for temporary Recipients is not the same as the lative strategy for addressees in all
languages. The two forms are identical in Nakh, Tsezic and Dargwa languages and
in Khinalug, but not in Lak, Andic or Lezgic languages. Those Lezgic languages
that use the dative for the addressees also have their own lative ‘temporary’ Recipi-
ents. These are indications that there is no necessary connection between the two
 Against the addressee of speech – Recipient metaphor 

functions. Cf. (37) where the Super-Lative is used for the lative Recipient with
(18) where the dative is used for the addressee; and Hum.Loc is used in (38) while
SuperLative is used for the addressee in (21):
(37) Agul (Ganenkov et al. unpublished) – lative Recipient in a language with
dative addressees
zun gada qaje-guna, fajš-u-na c’-a-a
I(nom) son post.be-temp fetch-pfv-cvb give-ipfv-prs
ze gada malla-jar-i-l-di mi-s
I.gen son molla-pl-obl-super-lat this.obl-dat
dars-ar ħarq’-a-s.
lesson-pl teach-ipfv-inf
‘And I have a son, so I fetch (him) and give my son to the mullahs, for them
to teach him.’
(38) Bagvalal (Kibrik et al. 2001: 775)
hatu-b hil’aχ awtobus-la b-iʁ-ē-b-o, ibraška,
a.little-n under bus-and n-stop-caus-n-cvb ibrashka
w-aχ̄a-la w-ā-w-o, gaʔišnik-s� ̌u-ɬā
m-outside-and m-come-m-cvb road.policeman-obl.m-humloc
ins� ̌tu-ra azar šard-āla w-eɬi.
five-card thousand hand-caus.pot.inf m-go
‘Some way down the road Ibrashka stopped the bus, got out and went
(back) to the road policeman to give him five thousand roubles.’

Obviously, however, this is only a partial argument – for those languages where
the addressee is expressed by the same form as the lative (temporary) Recipient,
i.e. Nakh, Tsezic, Dargwa and Khinalug, one could maintain that co-expression
is indicative of conceptualization. However, if the speech act is considered to be a
metaphor of information transfer, the addressee seems to be more like the dative
‘permanent’ Recipient (construal of giving as transfer) than the lative ‘temporary’
Recipient (construal of giving as motion). When the information is ‘transferred’,
it is transferred permanently (at least in the sense that it cannot be given back/
moved back to its source). In other words, one would expect the addressee to be
marked in the same way as the permanent Recipient, which is only true in Lezgian,
Agul, Tabassaran and Udi (in the latter, there is no distinction between the per-
manent and temporary Recipients). In other languages the addressee bears either
the lative Recipient marking (Example 20, 36b) or a marking different from both.
When, on the other hand, information transfer is described by a construction with
the verb ‘give’ – as in ‘give news’ or ‘give greeting’, where the construal of commu-
nication as transfer cannot be questioned – East Caucasian languages tend to use
the dative ‘give’ construction (Example 39).
 Michael Daniel

(39) Archi (Arkhipov et al. unpublished): dative metaphor

jeb l’annu-m-mul-če-s̄-u iznu bo-Lo
this.pl beloved-1-pl-obl-dat and permission 3-give.pfv
qesbarti-li-t ̄i-k a‹b›li-s.
dance-obl-super-lat ‹hpl›come-inf
‘They granted these two who loved each other the permission to come to
the dancing place.’
The use of the lative Recipient in the Itsari Example (40) seems to be an exception.
For Itsari, one would expect here the Super-Lative, which is used as the dative
with ‘give’; instead, one finds In-Lative, as with the verb ‘say’ (see Table 3). This
exception may be explained as a re-analysis of the complex ‘give word’ as a regular
speech verb.
(40) Itsari (Sumbatova & Mutalov 2003: 204): ‘give word’ reanalyzed
as a speech verb
u-l=k’unajla di-c:i ʁaj b-īč:-ib-di!
you-erg=but i-in(lat) word n-give.pfv-praet-2sg
‘But you gave me your word!’

The possible conflations of the three functions – dative Recipient, lative Recipient
and addressee – are shown in the following Table 4 (metaphorical uses like that in
(39) are not taken into account because the data are insufficient):

Table 4. Mergers of the addressee, dative ‘permanent’ Recipient and lative ‘temporary’
Recipient
Mergers attested Languages or branches

identical marking in all three constructions Udi

‘permanent’ Recipient = addressee of speech
vs. Agul, Lezgian, Tabassaran
‘temporary’ Recipient
‘permanent’ Recipient
vs. Nakh, Tsezic, Dargwa, Khinalug
‘temporary’ Recipient = addressee of speech
‘permanent’ Recipient
vs.
‘temporary’ Recipient Lak, Avar, Andic
vs.
addressee of speech

It thus seems that East Caucasian construes addressees and ‘permanent’ and
‘temporary’ Recipients as three different Goals with all possible mergers attested
between them.
 Against the addressee of speech – Recipient metaphor 

The second potential objection is connected to the presence of the essive strategy
in Table 3. Indeed, most languages use one of the lative forms, which is quite natu-
ral from the point of view of the construal of the speech act as directed towards the
addressee (lative strategy). Some use the essive form, presumably because it covers
both essive and lative meanings (covert lative strategy). But why would Avar, Botlikh
and Lak use the essive form of the localization when there is a lative form available?
These (few) cases may be considered as exceptions. Nonetheless, they do not comply
with the dative SAE-type, which seems to be dominant. This makes it plausible to
attempt to explain essive marking as a variation on the lative marking. Assuming that
the idea of the spatial metaphor is by and large valid for East Caucasian, one pos-
sible explanation is diachronic. Of the three main spatial meanings, the essive is the
least and the lative is the most marked orientation, with the elative in between (see
7 above). Not all of East Caucasian languages distinguish essive and lative functions:
there are some that merge these two functions in one form. There is no reason not
to believe that this merger also has a diachronic dimension, as discussed above for
Tsezic. The essive may be the old syncretic form, preserving some residual lative uses
including the addressee of speech verbs, while a new lative arrived later. This sugges-
tion is supported by some oblique evidence from languages within the same branch.
Thus, the two localizations used for the addressee in Botlikh, an Avar-Andic
language, employ their essive forms though they also have lative forms. This use of
the essive cannot qualify as a covert lative strategy and contradicts the claim that
East Caucasian languages treat addressees as Goals – the main claim of this paper.
At least one of the localizations in question is cognate to the localization markers
used for the addressee in those Andic languages that lack lative marking and rep-
resent the covert lative strategy (see Table 3). In other words, the Botlikh strategy
is diachronically the same as the one which still does not distinguish essive and
lative functions in Andi, Chamalal, Godoberi and Tindi, the lative morphology
might have been introduced at a later stage – cf. a more clear case of the optional
use of the lative marker in West as compared to East Tsezic languages. If these
cases are interpreted as representing diachronic shift, deviations from the lative
strategy may be explained as transitional stages. Cf. also the ambiguity of the essive
form between the essive and lative functions in the presence of a dedicated essive
in Avar in (Charachidze 1981: 55). This is summarized in the following Table 5:

Table 5. Essive strategy as a transitional stage

Andic Tsezic

archaic Andi, Chamalal etc. covert lative East covert lative

innovative Botlikh essive West essive/lative
 Michael Daniel

I conclude this second part of the discussion by saying that, in the absence of
a robust comparative grammar of the essive ~ lative distinction in East Caucasian,
these conjectures, although not improbable, remain speculative.
Finally, let us consider the third issue, which is the presence of the dative strat-
egy of addressee marking. Of course, the data from Lezgian, Agul, Tabasaran and
Udi can be considered as evidence that these languages are similar to SAE and
showing the expected transfer metaphor, in contrast to the other languages of the
family. But is this evidence unambiguous? Indeed, all these languages commonly
use datives in the lative function:
(41) Udi (Ganenkov et al. unpublished): dative functions as a lative
me šeir k:azet-a taʁ-at:an, zu al slučajno
this poem newspaper-dat go-temp I and occasionally
bäkü-n-a=z taʁ-ala baj.
Baqu-obl-dat=1sg go-deb aux+aor
‘When this poem went to the newspaper, it so happened that I was to go to
Baku.’
(42) Agul (Ganenkov et al. unpublished): dative functions as a lative
dad.a gada dukan.i-s hat.u-ne.
father(erg) son shop-dat send.pfv-pf
‘Father sent the boy to the shop.’

This use is unknown to other languages of the Lezgic branch and to other branches,
including Nakh (cf. Ingush in Nichols 2011: 418–420), Tsezic, Khinalug (Kibrik et al.
1972), Dargwa, Lak and Avar-Andic. Peripheral lative functions that are retained by
the dative in some languages (e.g. in Archi or Nakh) are very limited, including
e.g. the locus of contact with verbs of physical impact. The distribution seems to be
strict: only those languages that have free (i.e. non-subcategorized) spatial datives as
in (41) and (42) conflate addressees with the dative (permanent) Recipients.
Based on this distribution, and with a view to the strategy dominant in the
other languages of the family, I propose that the use of the dative for addressees
in Lezgian, Agul and Tabassaran is not motivated by a metaphor of transfer but
rather by the lative functions of the dative present in these languages. Udi may be
similar or may represent a special case, because its dative is functionally very wide
(possibly under Turkic influence). Combining the addressee with ‘permanent’
Recipient in dative marking in these Lezgic languages is incidental in the same
way combining the addressee with the ‘temporary’ Recipient in Nakh, Tsezic,
Dargwa and Khinalug. The data from these four languages the perfectly fit into
the spatial construal of speech acts, with the only difference that here the direction
of the speech towards the addressee is expressed by a (lative) dative.
 Against the addressee of speech – Recipient metaphor 

5. Conclusion

According to Comrie (1981), a typological approach to semantic roles requires

that any type of participant that is conveyed by a specialized marker on the noun
phrase (dedicated marking) should be considered an element of the universal
semantic role inventory, even if this dedicated marker only exists in one language
of the world. From the point of view of case marking typology, this argument-­
centered approach to semantic roles is empirical and inductive, because gener-
alizations are based on observed morphosyntactic properties of the argument
marking, or the common linguistic treatment of different participants.
This is not so with some formal approaches, as in the approach of Van Valin
and LaPolla (1997) based on semantic decomposition of verbal meanings. For the
purposes of investigating case marking, the RRG approach to the same problem
is rather deductive. Van Valin and LaPolla continuously underscore this by giving
priority to the logical structure of the verb and arguing that semantic (in their
terms, thematic) roles are automatically derived from the logical structure. Differ-
ences and commonalities in linguistic treatment of noun phrases come in at a dif-
ferent level of granularity, when macroroles are introduced (Van Valin & LaPolla
1997: 139 ff.) However, at this level of generalization only core arguments are taken
into account, so that Recipients (and, eventually, addressees of speech verbs) are
considered only when they behave like direct objects, which is certainly not the
case in East Caucasian.
The present study is focused on case marking of one specific peripheral argu-
ment of a single verbal meaning. Semantic decomposition could not possibly
provide a clue to the variation of case marking, which is the main source of my
empirical data. It is obvious that it must proceed by assuming Comrie’s stance on
the notion of semantic roles, with its emphasis on dedicated case marking.
Kryz, Archi or Bagvalal are rather close to having a dedicated addressee
marker. The forms in question (see Table 3) are used exclusively or almost exclu-
sively for the addressees of speech verbs, even though, morphologically, they are
clearly of spatial origins and still a formal element of the spatial subparadigm,
as in Archi (Kibrik 1977) or Bagvalal (Kibrik et al. 2001), or may preserve some
additional peripheral uses, as in Kryz (Authier 2009). Thus, the addressee of verbs
of speech is clearly a separate semantic role in terms of the universal semantic role
inventory and henceforth deserves capitalization as a semantic role label – the
Addressee of Speech.
Comrie’s requirement that marking of a semantic role must be dedicated
seems to be stronger than necessary. Most East Caucasian languages have a mark-
ing for the Addressee that is separate from the Recipient. And even though this
 Michael Daniel

marking in many languages of the family also has other uses, such as movement
towards someone or a ‘temporary’ Recipient, its use for the Addressee clearly is a
typologically distinct case function. If a semantic role is combined with the Recipi-
ent in some languages of the world (SAE and many other) and distinct from it in
other languages (East Caucasian) this is a typologically separable semantic role, in
very much the same way as the role of (Natural) Force which is identified either
with the Agent or with the Instrument but never bears dedicated marking. The
question is, at what level of semantic generalization, and why, is the Addressee
combined with the Recipient in (admittedly) many other languages of the world?
If we assume that Recipients are a specific type of (animate) Goal, what is the rela-
tion of the Addressee to the concepts of the Recipient and Goal?
I repeat here Figure 1 that appeared at the end of Section 1 above:

a. b.
Goal Goal

Recipient Recipient Addressee

Addressee SAE East Caucasian

Figure 1. Alternative models of Recipient~Goal~Addressee relations

East Caucasian languages consistently treat the Addressee separately from the
Recipient. The most plausible semantic motivation behind the marking attested in
most East Caucasian languages is that the Addressee of Speech is construed as a
person towards whom the speech act is directed (though six languages on Table 3,
marked as “essive” and “other”, remain problematic). The Addressee is thus a sepa-
rate type of animate Goal. Recipients, which are marked by the dative, also are
animate Goals (also for Van Valin & LaPolla 1997); but most East Caucasian lan-
guages formally distinguish Addressees and Recipients as different animate Goals
(Figure 1, b).
I also argued, for those East Caucasian languages that do use the dative for the
Addressee (several languages in the Lezgic branch), that combining the Recipient
and the Addressee in one marking (cf. Table 4) does not necessarily mean a meta-
phorical extension of the latter to the former. Cross-linguistically, dative marking
is often a heterogeneous cluster of different Goal-like functions. For instance, in
some languages of the family, the dative also marks the locus of impact with verbs
of physical contact (‘hit something/someone on something’). We have seen that, in
the other languages of the family, the Addressee is a special type of animate Goal.
We may then suspect that the Addressee is a lative use of the dative rather than
 Against the addressee of speech – Recipient metaphor 

metaphorical Recipient of information in those languages which use the dative as

the Addressee marking. And indeed, this provides a perfect explanation why of the
whole family only the languages that use the dative in lative function also use it for
the Addressee.
In East Caucasian, the Addressee is thus not a Recipient to whom the infor-
mation is given but an animate Goal towards whom the speech act is directed. In
SAE, however, there is no direct evidence that any NP marking other than that
of the Recipient in a ditransitive construction can be central to the (understand-
ing of) linguistic construal of the Addressee. Some European languages do show
a possibility for spatial marking non-identical to the dative case (cf. 3a and b in
­Section 1) or the ‘dative prepositions’ for languages lacking cases (4–5 ibidem),
but this evidence is too peripheral and thus weak to consider the Addressee to be
separate from the Recipient. (Rather, for these rare cases one would want to posit
a separate spatial metaphor of the speech act.) Hence the widespread idea of the
metaphor of transfer of information.
But one should remember that datives ultimately go back to (al)latives in many
languages and that, conversely, of all case markers, (al)latives primarily develop
into datives (Heine & Kuteva 2002: 38). Cf. ‘prepositional datives’ in some modern
Indo-European languages that have lost case marking, including various Romance
languages, English or Bulgarian. In these languages at least, the dative might be
considered as a general marker for (animate) Goals that are divided into Recipients,
Addressees and possibly some other functions. While East Caucasian supports ver-
sion (b) of Fig. 1, SAE remains ambiguous between (a) and (b). As with any meta-
phor, it is hard to provide any palpable evidence that the use of the dative marking
of the Addressee is a metaphor of giving. Most people who support the claim seem
to rely on the fact the two functions are combined in one marking, on the one hand,
and on considering the Recipient to be the core function of the dative, on the other
(Janda 1993). This may be the case of classical Indo-European languages that show
no evidence of datives developing from (al)latives, but modern SAE are different
and probably heterogeneous. The situation in SAE requires further investigation,
but it is not so clear a case of metaphorical extension as often considered.

6. Research outlook

The present study is limited to considering the evidence of Addressee marking in

East Caucasian, including some contrastive evidence from SAE. This is certainly
not enough to build a typology of expressing the Addressee as a semantic role, and
expanding this analysis to a wider typological perspective is beyond the scope of
this paper. However, a plan of a possible research may be outlined.
 Michael Daniel

First of all, other languages that use non-dative marking for the Addressee
should be identified. If the point made in this paper is correct, the presence of
non-dative Addressees can be expected in languages which have separate (al)lative
forms, especially if they are dedicated to or used for human or animate Goals.
Judging from East Caucasian data, as soon as there are two or more separate mark-
ing for animate Goals, the Recipient and the Addressee may easily split. If they do
not, this may be evidence in favor of the metaphorical extension approach.
Second, various morphosyntactic properties of dative Addressees should be
investigated for those languages whose ditransitive constructions follow the indi-
rect object pattern (indirective alignment in terms of Malchukov et al. 2010). These
properties may include indexing on the verb, voice and other (morpho)syntactic
parameters of constructions involving Recipients (cf. presentation of Fillmore’s
data in Levin and Rappoport Hovav discussing Fillmore’s distinguishing Place vs.
Object roles for superficially similar arguments of hit and break in Levin & Rappo-
port Hovav 2005: 37). For English, the metaphor of information transfer to explain
the use of the same marking for the Addressee as for the Recipient is usually taken
for granted. However, consider the following evidence that contrasts marking with
syntactic status and passivization – see Table 6:

Table 6. Morphosyntactic difference between two verbs of speech in English

PP DO Passive

a. verb of transfer ‘give’ I gave it to him I gave him He was given

b. verb of speech ‘tell’ (?)I told it to him I told him this He was told
c. verb of speech ‘say’ I said it to him *I said him *He was said

In English, dative Recipients may be expressed by a prepositional phrase or

a direct object and may be passivized (a). So can Addressees with tell – notably,
however, for some speakers the availability of Recipient marking to such address-
ees is questionable, as in (b). On the other hand, addressees with say, for which
only Recipient marking is available, may not be promoted to DO and be passiv-
ized, which shows, in principle, that it might be a different situation construal. In
other words, in English, the Addressee of say is syntactically different from the
Recipient even if it is marked in the same way as the Recipient (to-PP).
Third, evidence against the Recipient ~ Addressee metaphor can be sought
in the languages with the ‘secondary object’ pattern in terms of Dryer (1986) or
secundative alignment in terms of Malchukov et al. (2010) in ditransitive con-
structions. For Van Valin and LaPolla (1997: 145–146), this is a competition of
thematic roles location and theme for the undergoer macrorole, in which the
location (the underlying thematic role for Goal under ‘become’ predicate in the
 Against the addressee of speech – Recipient metaphor 

logical structure) is the winner. In a sense, in RRG, whatever happens, Recipi-

ent remains a Goal. This paper, however, adopts the argument-centered approach
to semantic roles. Within this approach, in case they have a special marking for
Goal, secundative languages construe the Recipient not as a Goal but as a Patient.
If these languages treat the Addressee in the same way as the Recipient (i.e. as a
Patient of a transitive verb), this would provide a strong support in favor of the
metaphor analysis. If, however, the Addressee in such languages chooses a lative,
we may conclude that there is no inherent connection between the Recipient and
the Addressee. They are two different animate Goals that merge in one marking
in those languages that, unlike East Caucasian, do not possess a rich inventory of
(animate) lative markers.

Abbreviations

1, 2, 3, 4 and 5 are used as glosses for class agreement markers in the singular. For reasons not
to be discussed here, in Nakh languages, the glosses for class agreement (J, D, V, B etc.) repeat
the class marker itself.

abl ablative hpl human plural agreement marker

ad localization ‘near the landmark’ humloc localization ‘at someone’s (place)’
adr addressive (case) imp imperative
aff affective (case) in localization ‘inside a hollow
all allative object’
ant precedence in time inf infinitive
aobl attributive in the oblique inter localization ‘inside a mass’ or
aor aorist ‘between’
apud localization ‘near the landmark’ intrg interrogative
atr attributive ipfv imperfective stem
aux auxiliary lat lative
card cardinal (numeral) loc locative
caus causative lv light verb
cmpl complementizer m masculine agreement marker
cont localization ‘on the surface of the n non-human agreement marker
landmark’ narr re-narrative
cop copula neg negation
cvb converb nom nominative
dat dative npl non-human plural agreement
deb debitive marker
def definite obl oblique stem
el elative oblig obligative
erg ergative pf perfect
ess essive pfv perfective stem
evid evidential pl plural
fut future poss possessive localization
gen genitive post localization behind the landmark
 Michael Daniel

pot potential tense/mood sub localization ‘under the landmark’

praet praeterit super localization on the ‘upper surface
progr progressive of the landmark’
proh prohibitive temp temporal
prs present uw unwitnessed
prvb preverb vblz verbalizer
pst past voc vocative
ptcl particle w witnessed
ptcp participle

References

Arkhipov, Alexandre, Daniel, Michael, Kibrik, Aleksandr & Sandro, Kodzasov. Unpublished.
Archi Electronic Corpus. An electronic collection of glossed texts. Partly available at 〈www.
philol.msu.ru/~languedoc〉.
Arkhipov, Alexandre, Daniel, Michael, Kibrik, Aleksandr & Sandro, Kodzasov. Unpublished.
Khinalug Electronic Corpus. Partly available at 〈www.philol.msu.ru/~languedoc〉.
Authier, Gilles. 2009. Grammaire Kryz, langue caucasique d’Azerbaïdjan, dialecte d’Alik. Leuven:
Peeters.
van den Berg, Helma. 1995. A Grammar of Hunzib. Munich: Lincom.
Bokarev, Anatolij. 1949. Očerki grammatiki čamalinskogo jazyka. Moscow – Leningrad: AN SSSR.
Bokarev, Anatolij. 1949. Sintaksis avarskogo jazyka. Moscow – Leningrad: AN SSSR.
Charachidze, Georges. 1981. Grammaire de la langue avar. Langue du Caucase Nord-Est. Paris:
Dunod.
Comrie, Bernard. 1981. Language Universals and Linguistic Typology: Syntax and Morphology.
Oxford: Blackwell & Chicago: University of Chicago Press.
Creissels, Denis. A Grammar of Akhvakh. Ms.
Creissels, Denis. In preparation. Functive phrases in typological and diachronic perspective.
Cysouw, Michael & Diana, Forker. 2009. Reconstruction of morphosyntactic function: Non-
spatial usage of spatial case marking in Tsezic. Language 85(3): 588–617. DOI: 10.1353/
lan.0.0147
Dąbrowska, Ewa. 1997. Cognitive Semantics and the Polish Dative. Berlin: Mouton de Gruyter.
DOI: 10.1515/9783110814781
Daniel, Michael & Yuri, Lander. 2011. The Caucasian languages. In Bernd Kortmann & Johan
van der Auwera (eds.), The Languages and Linguistics of Europe: A Comprehensive Guide,
125–158. Berlin: De Gruyter Mouton. DOI: 10.1515/9783110220261.125
Daniel, Michael & Dmitri, Ganenkov. 2009. Case marking in Daghestanian. In Andrej
­Malchukov & Andrew Spencer (eds.), Handbook of Case, 668–685. Oxford: OUP.
Daniel, Michael, Khalilova, Zaira & Zarina, Molochieva. 2010. Ditransitive constructions in
East Caucasian. In Andrej Malchukov, Martin Haspelmath & Bernard Comrie (eds.), Stud-
ies in Ditransitive Constructions: A Comprehensive Handbook, 277–315. Berlin: De Gruyter
Mouton.
Dowty, David R. 1991. Thematic proto-roles and argument selection. Language 67: 547–619.
Dryer, Matthew S. 1986. Primary objects, secondary objects, and antidative. Language 62:
808–845. 〈http://www.acsu.buffalo.edu/%7Edryer/DryerPrimarySecondaryObjects.PDF〉
 Against the addressee of speech – Recipient metaphor 

Fillmore, Charles J. 1968. The case for case. In Emmon Bach & Robert T. Harms (eds.), Univer-
sals in Linguistic Theory, 1–88. New York: Holt, Rinehart, and Winston.
Forker, Diana. 2013. A Grammar of Hinuq. Berlin: De Gruyter.
Ganenkov, Dmitri, Lander, Yuri & Timur, Maisak. Unpublished. Udi Electronic Corpus. An
electronic collection of glossed texts.
Ganenkov, Dmitri, Maisak, Timur & Solmaz, Merdanova. Unpublished. Agul Electronic Cor-
pus. An electronic collection of glossed texts.
Ganenkov, Dmitry. 2010. Topological relations in Daghestanian languages: An overview.
­Linguistics 48(5): 1011–1041. DOI: 10.1515/ling.2010.033
Goldberg, Adele E. 1995. A Construction Grammar Approach to Argument Structure. Chicago:
UCP.
Gudava, Togo E. 1967. Botlixskij jazyk. In Jazyki narodov SSSR, Vol. 4: Iberijsko-kavkazskije
jazyki, 293–306. Moscow: AN SSSR.
Haspelmath, Martin. 1993. A Grammar of Lezgian. Berlin: Mouton de Gruyter. DOI: 10.1515
/9783110884210
Heine, Bernd & Tania, Kuteva. 2002. World Lexicon of Grammaticalization. Cambridge: CUP.
DOI: 10.1017/CBO9780511613463
Janda, Laura. 1993. A Geography of Case Semantics: The Czech Dative and the Russian
Instrumental. Berlin: Mouton de Gruyter. DOI: 10.1515/9783110867930
Khalilova, Zaira. 2009. A Grammar of Khwarshi. Utrecht: LOT.
Kibrik, Aleksandr & Jakov, Testelets (eds). 1999. Elementy caxurskogo jazyka v tipologičeskom
osveščenii. Moscow: Nasledie.
Kibrik, Aleksandr (ed.). 1996. Godoberi [Lincom Studies in Caucasian Linguistics 02] Munich:
Lincom.
Kibrik, Aleksandr, Kazenin, Konstantin, Ljutikova, Ekaterina & Sergej, Tatevosov (eds). 2001.
Bagvalinskij jazyk: Grammatika, teksty, slovari. Moscow: Nasledie.
Kibrik, Aleksandr, Kodzasov, Sandro & Irina, Olovjannikova. 1972. Fragmenty grammatiki
xinalugskogo jazyka. Moscow: Moscow State University.
Kibrik, Aleksandr. 1977. Opyt strukturnogo opisanija arčinskogo jazyka. Vol. 2. Moscow: Mos-
cow State University.
Kibrik, Aleksandr. 2003a. Konstanty i peremennyje jazyka. St. Petersburg: Aleteia.
Kibrik, Aleksandr. 2003b. Nominal inflection galore: Daghestanian, with side glances at Europe
and the world. In Frans Plank (ed.), Noun Phrase Structure in the Languages of Europe,
37–112. Berlin: De Gruyter.
Levin, Beth & Malka, Rappoport Hovav. 2005. Argument Realization. Cambridge: CUP.
Magomedova, Patimat T. 2003. Tindinkso-russkij slovarj. Maxačkala: Institute of Language and
Literature (Russian Academy of Sciences).
Magometov, Aleksandr. 1965. Tabasaranskij jazyk. (Issledovanija i teksty). Tbilisi: Mecniereba.
Magometov, Aleksandr. 1982. Megebskij dialect darginskogo jazyka. (Issledovanija i teksty).
Tbilisi: Mecniereba.
Makhmudova, Svetlana. 2001. Morfologija rutul’skogo jazyka. Moscow: IJa RAN.
Malchukov, Andrej, Hartmann, Iren, Haspelmath, Martin, Comrie, Bernard & Søren, W ­ ichmann
(eds.). In preparation. Valency Classes in the World’s Languages.
Malchukov, Andrej, Haspelmath, Martin & Bernard, Comrie. 2010. Ditransitive constructions:
A typological overview. In Andrej Malchukov, Martin Haspelmath & Bernard Comrie,
­Studies in Ditransitive Constructions: Comparative Handbook, 1–64. Berlin: De Gruyter
Mouton.
 Michael Daniel

Molochieva, Zarina, Witzlack-Makrevich, Alena, Bickel, Balthasar & Johanna, Nichols. Unpub-
lished. Chechen Pears Stories Collection. Electronic Corpus. University of Leipzig &
­University of Berkeley.
Musaev, Magomed-Salam. 1999. Darginskij jazyk. In Jazyki mira: Kavkazskije jazyki. Moscow:
Nauka.
Nichols, Johanna. 2011. Ingush Grammar [UC Publications in Linguistics]. Berkeley CA: Uni-
versity of California Press.
Pasquereau, Jérémy. 2010. Formes spatiales des noms en karata: Emplois spatiaux et non-­spatiaux.
Lyon: University of Lyon dissertation (Memoire de master 1).
Salimov, Hangerej S. 2010. Gagatlinskij govor andijskogo jazyka. Maxačkala: Institute of Lan-
guage and Literature (Russian Academy of Sciences).
Sumbatova, Nina & Rasul, Mutalov. 2003. A Grammar of Icari Dargwa [Languages of the World,
Materials 92]. Munich: Lincom.
Talibov, Bukar. 2007. Buduxskij jazyk. Moscow: Academia.
Testelets, Jakov. 1980. Imennyje lokativnyje formy v dagestanskix jazykax. Moscow: Moscow
State University.
Van Valin, Robert. 2005. Exploring the Syntax-semantic Interface. Cambridge: CUP. DOI: 10.1017
/CBO9780511610578
Van Valin Jr., Robert D. & Randy J. LaPolla. 1997. Syntax: Structure, Meaning, and Function.
Cambridge: CUP.
Žirkov, Lev. 1955. Lakskij jazyk: Fonetika i morfologija. Moscow: AN SSSR.
Against the addressee of speech – Recipient metaphor 
 Semantic roles and word formation
Instrument and Location in Ancient Greek*

Eugenio R. Luján & César Ruiz Abad

Universidad Complutense de Madrid (Spain)

In functional-typological approaches to syntax the identification of semantic

roles and the grammatical markers associated with them has been an important
goal. In this paper it is argued that a similar approach to word formation
patterns is possible and some criteria for the identification of the semantic roles
in the domain of derivational morphology are provided. Agent, Instrument,
and Location can be considered as different semantic roles in morphology, but
polysemy between two or more of them is also frequent cross-linguistically.
Deverbal suffixes employed in Ancient Greek for deriving instrumental and
locative nouns are surveyed as a case study. The semantic analysis of the
derivative nouns shows that those suffixes are also employed for building other
nouns that are semantically related, but cannot be considered Instruments and
Locations proper, such as non-prototypical instruments, prizes and rewards,
and others. In general, when there is no specific suffix for a semantic role but a
noun expressing it needs to be coined, suffixes for neighbouring semantic roles
will be employed, depending on the available metaphor or metonymy that allows
for the semantic extension. The comparison of these findings to the conceptual
organization of grammatical morphemes in the same language (Greek) and in
other languages in general reveals that both in syntax and morphology a similar
semantic organization is found. However, the number of semantic roles that can
be identified in syntax is higher than the number of semantic roles reflected in
word formation patterns, so that semantic maps of word formation patterns are
similar in shape but more reduced.

Keywords: Semantic roles; word formation patterns; nominal morphology;

derivation; Instrument; Location; Agent; Ancient Greek; semantic maps

* This paper is part of the research project “Estudios de morfología nominal: lenguas paleo-
hispánicas e indoeuropeas antiguas” (“Studies on noun morphology in Palaeohispanic and
Old Indo-European languages”), which has the financial support of the Spanish Ministry of
Economy and Competitiveness (FFI2012-36069-C03-02). The final version has greatly ben-
efited from the comments and suggestions made by the editors and an anonymous reviewer,
for which we are very grateful.
 Eugenio R. Luján & César Ruiz Abad

1. Introduction

Multifunctionality in word formation patterns can be researched in regard to the

semantic roles that the nouns built by means of those patterns express.1 Some
nominal and adverbial formations appear to be related to various semantic roles
according to their meanings. Typical examples include the following:2
(1) English suffix -er
a. killer (Agent)
b. lighter (Instrument)
(2) French suffix -oir
a. débouchoir ‘plunger’ [from déboucher ‘unlock’] (Instrument)
b. dortoir ‘dormitory’ [from dormir ‘sleep’] (Location)
(3) Diola suffix -a
a. ɛliba ‘knife’ [from -lib ‘make slices’] (Instrument)
b. atɛba ‘builder’ [from -tɛb ‘build’] (Agent)

This is also the case with so called Manner adverbs, such as English adverbs in
-ly (quickly, eagerly, etc.), Spanish adverbs in -mente (educadamente ‘gently’, apro-
piadamente ‘aptly’, etc.) and Ancient Greek adverbs in -ōs (alēthôs ‘truly’, kalôs
‘beautifully’, etc.), to mention just a few cases. Their traditional label, “adverbs of
manner”, just underlines their main uses, but it does not necessarily imply that
they can only have that semantic value. As a matter of fact, Crespo (1997: 18)
remarked that Greek adverbs in -ōs can be used not only for the semantic function
Manner-Modality, but also for the functions Means-Intermediary and Reference,
according to his classification.
These types of semantic constelations are frequent cross-linguistically, as
shown in previous literature on this subject. Things being so, it is possible to apply
to word formation patterns the kind of semantic analysis that has been employed
with grammatical morphemes in functional-typological research. However, there
will be some differences, too, and some adjustments will be needed. In this paper
we will be concerned on how to approach such a problem from a typological
perspective.

. For the discussion of the concept of “multifunctionality” as a useful alternative to poly-

semy and homonymy see Haspelmath (2003: 212–213). For a general overview of polysemy in
word formation patterns see Booij (2007: 220–225).
. Further examples in other languages can be found in Luján (2010: 162–163) and
­Luschütsky & Rainer (2011 and 2013). The data of Diola have been taken from Comrie &
Thompson (1985: 354).
 Semantic roles and word formation 

Although word formation patterns have not been systematically analyzed from
this perspective, in principle there is no reason to suppose that derivational mor-
phemes behave so differently from grammatical morphemes (cases and adposi-
tions). In fact, the findings of the studies on grammaticalization carried out in the
last thirty years have shown that lexical and grammatical morphemes constitute a
continuum and their meanings are organized in similar ways. They have core and
peripheral meanings and the borders between those meanings are synchronic-
ally fuzzy.3 As we have argued elsewhere (Luján 2010: 163–164), word formation
patterns can be considered to be midway between lexical and grammatical mor-
phemes: in contrast to the lexicon, there is a limited set of derivational morphemes
and word formation patterns in any given language; in contrast to grammatical
morphemes, these patterns are not mandatory, since word formation patterns can-
not be appplied to every word of the same class and frequently it is not even easy
to determine the criteria governing the distribution. Moreover, the same semantic
content can be expressed by means of grammatical morphemes (cases and adposi-
tions) or affixes in different languages.
Besides those mentioned above, further types of nominalizations can be ana-
lyzed from this perspective.4 Comrie and Thompson (1985: 335–336) provide
several cases of Objective, Manner and Reason nominalizations. Objective ones
denote “the result or the typical or ‘cognate’ object of the action”. Thus in Zulu, a
prefix for nouns in one of the non-human noun classes plus the suffix -o allows for
such derivations, as in the following examples (Comrie & Thompson 1985: 356):
(4) -cabanga ‘think’ → mu-cabang-o ‘thought’; (Object)
(5) -cula ‘sing’ → i-cul-o ‘congregation, hymn’. (Object)
In most Indo-European languages no differentiated marker exists for that pur-
pose. Instead, it is very common to use action nominalizations in a metonymic
sense, and the output is then often lexicalized. Compare, e.g. creation in job cre-
ation (Action) vs. the creation of an artist (Result), with the evolution Action >
Result. In Ionic-Attic Greek, however, there are two different suffixes: result nouns
are expressed by -ma, while action nouns by -sis, as we can see in the following
examples (cf. Chantraine 1933: 190):
(6) didáskō ‘teach, learn’ → dídagma ‘what is learned’ (Result)
→ dídaksis ‘teaching, instruction’ (Action)

. See the general frame and the case studies in Geeraerts (1997).
. This does not involve that every type of nominalization can be linked to one or various
­semantic roles, as is obviously the case with action nouns, which would semantically corre-
spond to the verb of a sentence and not to any of the participants.
 Eugenio R. Luján & César Ruiz Abad

A Manner nominalization expresses “way of ‘verb-ing’” (Comrie & Thompson

1985: 354). In Turkish, the suffix -(y)iš, with vowel harmony, is used in this sense:
yürü- ‘to walk’ → yürü-yüs ‘way of walking’, ye- ‘to eat’ → ye-(y)iš ’way of eating’.
Again metonymic shifts from action nouns are current in Indo-European lan-
guages: cf. Walking is a great way to keep fit vs. his walking (Comrie & Th
­ ompson
1985: 354–355). In Gĩkũyũ, a Bantu language, the circumfix mũ-…-ĩre has the
same meaning: mũ-ina-ĩre ‘manner of singing’, mũ-thom-ere ‘manner of reading’
(Mugane 1997: 56). Accompaniment and Time nominalizations can be illustrated
by Kabardian, a Northwest Caucasian language: lež′e-ɣw ‘workmate’ and laaža-
ɣwa ‘worktime’, from laž′e ‘work’ (Mel’čuk 1997). The Reason nominalization is a
less common pattern: it denotes ‘the reason for “verb-ing”’. Sundanese is the only
example provided by Comrie & Thompson (1985: 356–357): paŋ-dataŋ ‘reason for
arrival’, paŋ-daek ‘reason for being willing’.

2. Semantic roles in word formation

If we adopt that perspective, it will be possible to apply to word formation patterns

the kind of semantic analysis that has been used with grammatical morphemes in
functional-typological research. As we said above, there will be some differences,
so one of our main concerns in this part of the paper will be to discuss the type
of adjustments needed in the methodology. In what follows we will leave aside
adverbs and concentrate mainly on nominal word formation because this is the
focus of our current research.

2.1 Identifying semantic roles in word formation

Semantic roles can be defined according to the state of affairs or the event or action
in which the participants take part. For instance, let us consider (7):
(7) John hit the door with a hammer.

The sentence expresses an action in which John can be analyzed as the Agent, door
as the Patient and hammer as the Instrument. However, a fundamental difference
to syntax is that in word formation there is no explicit state of affairs, event or
action. This leads to the question how it is then possible to approach the semantic
analysis of word formation from this perspective. Although we will specify the
definitions of various semantic roles more precisely in § 3 below, it will be conve-
nient to discuss the general methodology now.
In word formation we can relate a pattern to a semantic role when the meanings
of the words built according to that pattern are overtly construed in such a way
 Semantic roles and word formation 

that they evoke an action, event or state of affairs in which the referent of the noun
plays that semantic role. For instance, a runner is someone who runs, that is, it
is the Agent in the action evoked by its meaning. And a cooker is an object with
which one can cook food, that is, the Instrument in the action referred to. Obvi-
ously, this does not mean that, from a syntactic point of view, runner or cooker
cannot appear in a sentence in other semantic roles, as sentences (8)–(10) show.
(8) John killed the runner (Patient) with his car.
(9) Give it to the runner (Dative).
(10) Put it on the cooker (Location).

What matters is that, regardless of the syntactic function that they may have in
every specific occurrence, the meaning of runner is construed in such a way that
its referent is thought of as the Agent of the action “running” and that of cooker as
the Instrument of the action “cooking”.
The meaning of some nouns can evoke an event or state of affairs in which
they are participants even if they lack an overt marker. For instance, the meaning
of a noun like road is construed in such a way that it is prototypically conceived
as a Perlative and that of pen as an Instrument.5 However, they are not overtly
marked as such from a morphological perspective. At first glance this would seem
to be a problem for applying this kind of semantic analysis to word formation pat-
terns, but, in fact, the situation is not much different from the syntactic domain. In
some cases, no grammatical morpheme (preposition or case) is needed to express
the specific semantic role that a given element plays in a sentence. Compare the
following sentences:
(11) I went to the hospital.
(12) I went home.

Both hospital and home are Adlatives in their sentences; however, hospital needs
a grammatical morpheme (preposition to) to mark its role in the sentence, while
home does not. In the same way, English gardener or player are overtly marked as
Agents by means of the suffix -er, while blacksmith or nurse are not. It is thus only
in cases like gardener, sailor, shoemaker, hairdryer, or hanger, which are formed by

. This can be easily proved by searching for the definitions of such words in the dictionaries.
E.g., the definition of those two words in the Webster’s Dictionary (1989 edition, New York) is
as follows: road, “a strip of smoothed, cleared land, usually provided with a hard surface, for
the passage from place to place of vehicles, riders, pedestrians etc.”; pen, “an instrument for
writing in ink”.
 Eugenio R. Luján & César Ruiz Abad

means of affixes or through composition, that a given word formation pattern can
be related to a semantic role.
Multifunctionality in word formation patterns reveals that semantic roles in
morphology, like in the syntactic domain, are not clear-cut categories, but form a
continuum. As a consequence, the researcher has to face two main problems: the
first one is the interpretation of the empirical data, i.e. which semantic role cor-
responds to a given derivative noun (cf. § 3.2); the second one is the identification
of the different semantic roles in morphology, i.e. which ones are actual linguistic
entities.

2.2 Criteria for differentiating semantic roles

The problem of how to tell apart different semantic roles has been, in fact, a crux of
this approach since Fillmore’s (1968) seminal paper “The case for case”. Even if the
terms “semantic role” and “semantic function” frequently overlap in functionalist
analyses, some scholars have tried to keep them separate. Crespo (1997: 6–7), in
his comprehensive analysis of the semantic functions of Ancient Greek, defined a
semantic role as the role played by a term (entity, state of affairs, possible fact or
utterance) in the relation referred to by the predicate of the sentence. He stressed
the fact that the number of semantic roles is unlimited, since the actual role
depends on the lexical context both of the term and the predicate and also of the
other participants in the event or state. Semantic functions, instead, express bun-
dles of notions, and they can be characterized by their grammatical morphemes,
showing the properties of the morphemes as regards coordination, apposition,
juxtaposition, and answer to partial questions (Crespo 1997: 8–9).
The situation becomes even more complicated when we compare different
languages and try to delimitate semantic functions cross-linguistically. In his study
of temporal adverbs in the world’s languages Haspelmath (1997: 10–13), when dis-
cussing the criteria to identify semantic functions, states:
A semantic function has been isolated when there is a significant number of
languages which clearly distinguish this type from related ones in their means of
expression […] It would be very difficult to base such a list on semantic criteria
alone, because there would be no way of constraining the possible proliferation
of functions.

In a more recent paper, he states (Haspelmath 2003: 217):

A function is put on the [semantic] map if there is at least one pair of languages
that differ with respect to this function. […] In order to justify this distinction
on our semantic map, we need at least one language that has different formal
expressions for the two functions. […] This procedure is repeated as more
languages are taken into account until no new functions are encountered.
 Semantic roles and word formation 

We can thus rely on this criterion for word formation, too. No different semantic
role should be distinguished unless there is at least one dedicated word formation
pattern in a language that is used for a given semantic role but not for other.
Nevertheless, we must be conscious that a strict differentiation among seman-
tic roles or functions is not possible. Givón (2001: 106–107) warns:
The distinction between a major type and a minor sub-type [of semantic role] is
not a principled one, but rather a matter of pragmatic judgement. [… A] major
semantic feature (or type) is simply one that has more extensive grammatical
consequences. In defining each semantic role, we only define a prototype.
Most members of a natural class tend to conform, more or less, to the class’s
prototype. But every natural population also has less prototypical members that
fit the ­prototype less well. Fortunately, such less prototypical members are – by
definition – a minority.

Along the same lines, Luraghi (2003: 18), in her study of the Greek cases and
prepositions, remarked:
In this book, I assume that SRs [semantic roles] are prototypical categories. This
means that besides prototypical exponents of each SR one must allow for the
existence of non-prototypical ones. Assuming prototypicality as a constituting
feature of SRs has two advantages: in the first place it allows to capture the
essential unity of conceptualization of a given situation in spite of the occurrence
of different participants. Furthermore, it avoids multiplying SRs.

The impossibility of strictly defining the various semantic roles and entities used
in the linguistic analyses on which semantic maps are based has led to alternative
views, such as that found in Cysouw’s (2007, 2010) papers. He has introduced the
term “analytical primitives” in order to avoid the problem of determining what
kinds of entities are used for building a semantic map. An analytical primitive is
defined as “any concept that is needed for the analysis of a particular set of data”
(Cysouw 2007: 227). However, even if this approach simplifies the problem posed
by the units of analysis, it has no support in the linguistic data themselves. If the
difference between the concepts used in a semantic map depends on the necessity
felt by the linguist, we run the risk of multiplying the concepts ad infinitum and,
more importantly, making distinctions ad libitum, without a factual basis in the
linguistic data.
It follows, then, that semantic roles, like most linguistic categories, cannot be
kept distinct in a straightforward manner and appear to be organized in a core-
periphery structure with more and less prototypical members inside it. However,
this is not a reason to give up completely the search for some criteria that can be
used to identify them, at least the major prototypes. From this perspective, it seems
more advisable to keep to Haspelmath’s criterion in the analysis of word formation
 Eugenio R. Luján & César Ruiz Abad

patterns, so that semantic roles in word formation patterns will be only identified
as different if they are expressed differently in the languages of the world.
This general principle can have various realizations in the languages:

a. The first, more straightforward possibility is that there exist different word
formation patterns at least in one language for those semantic roles. They will
be consequently identified as different semantic roles.
b. The second possibility is that a given semantic role, even if not having a dedi-
cated word formation pattern of its own in any language, can be identified
as such because it is grouped together with other semantic roles in differ-
ent ways. For instance, the semantic roles Means and Intermediary may not
be linked to exclusive formation patterns of their own, but might be identi-
fied as separate semantic roles if they are grouped together with Instruments
in a certain word formation pattern but they fall with Agents (and not with
Instruments) in another, whether this happens in the same language or in two
­different languages.
c. The third criterion relies on the exclusion of contents. For instance, if a given
word formation pattern includes Agents but Experiencers are excluded
from it, while another pattern allows nouns referring both to Agents and
­Experiencers, they can be considered different semantic roles.

3. Case study: Instrument, Location and related roles in Ancient Greek

Within the general frame presented in § 2, we will develop in this paper the anal-
ysis of word formation patterns for Instrument and Location nouns in Ancient
Greek as a case study. We will begin by defining the semantic roles Instrument
and Location, and then provide an overview of the kind of semantic polysemy or
multifunctionality in which they are involved in word formation. We will analyze
in depth the data provided by word formation patterns in Ancient Greek from this
perspective6 and will conclude with some typological considerations.

3.1 Defining Instrument and Location

In the syntactic domain, Instruments are usually defined in contrast to Agents.
Both are causal semantic roles, that is, they contribute to the carrying out of the

. In this paper the focus will thus be on deverbal derivational suffixes. In some cases, such
as English screwdriver or Spanish abrelatas ‘tin opener’, certain patterns of composition can be
related to these semantic roles, too.
 Semantic roles and word formation 

action.7 The semantic features [control], [intentionality] and [animacy] make the
difference between them. The prototypical Agent is positively marked for each of
these features, while the prototypical Instrument is not (Cruse 1973; Schlesinger
1989); see the following sentences:
(13) John hit the horse with the stick.
(14) The clock was ticking so loudly that it woke up the baby.
(15) The wind blew the tree down.
(16) The key opened the door.

In (13) the prototypical Agent (John) shows all the three features, while the proto-
typical Instrument (the stick) shows none. Agent is a complex category in itself. In
the lowest rank of agentivity stand self-moving devices, such as the clock in (14),
and natural phenomena, such as the wind in (15), usually called ‘Force’ or ‘efficient
cause’ (Crespo 1997: 32). Like Instruments they do not have any of the above men-
tioned features, but “they can act and move independently, and they are not under
the control of another effector” (Van Valin & LaPolla 1997: 121). In Greek, they are
coded by the same markers as prototypical Agents, so there is one single semantic
role Agent-Force (Luraghi 2003: 321; Crespo 1997: 32–33).
Under certain conditions Instruments can be conceived of as Agents. In com-
parison to the stick in (13), the key in (16) is coded as the subject because the Agent
is no longer on the scene. Still it presents the feature [+manipulability] in contrast
to Forces.
In a similar way, Instrument shows interferences with various other semantic
roles, especially Means and Intermediary. This refers to an animate (Intermedi-
ary) or inanimate (Means) entity by means of which the Agent brings about a
state of affairs. Means and Intermediary share the feature [+manipulability] with
Instrument. Luraghi (2003) has identified a separate semantic role Intermediary
and has insisted on the notion of “split control” of the action. The Intermediary,
as a human being, ultimately acts voluntarily; hence, it shares with the Agent the
control of the action (Luraghi 1995: 264).

. See, e.g. Luraghi (2003: 30–38). This is broadly equivalent to Palancar’s (2002: 70–78 and
140–146) semantic macro-role Energizer. Croft (1991: 185) established a difference between
antecedent and consequent roles inside the causal chain of events; the sematic roles under
discussion would belong with the former.
 Eugenio R. Luján & César Ruiz Abad

In the morphological domain, Grossmann (1998: 387) defined Instruments in

this way: “inanimate entities, indirect and non-intentional causes of an event, the
potential function of which is usually serving to bring about an action.”8
In syntax Location is usually defined as the place where the state of affairs
occurs. It does not belong to the causal domain but to the spatial one. However,
since Space is often the source domain in the conceptualisation of the World,
Location keeps relations with several other roles, especially the other spatial
ones (Origin, Direction). Interferences between Location and Instrument are
possible, too.
(17) He paid by credit card.
(18) By land, by sea, by air
(19) We travelled both by bus and by sea.
(20) Ambos viajamos con el autobús y por mar.

In (17) by credit card is an Instrument, while in (18) the same preposition has a
local value. The identification of the role is not straightforward when the item
referred to is a container or a vehicle (by car, by plane…). This fact is confirmed
by (19), since coordination with Location is possible. This ambiguity is still more
evident in other Indo-European languages. In (20), the Spanish translation of (19),
the term referring to a vehicle can select the prototypical Instrument preposition
con ‘with’, while the Location is expressed by por ‘by’.
In morphology, Location nouns have been defined as follows: “the derivatives
whose referents play the semantic role of Location in the state of affairs repre-
sented by the sentence in which they occur.”9 (Grossmann 1998: 388).

3.2 The semantic roles Instrument and Location in word formation

As soon as we deal with the data, one realizes that in spite of the definitions
given above, sometimes it is not easy to attribute a semantic role to a derivative
noun. That should be hardly surprising, since linguistic categories are not clearly
defined, but have a radial structure with prototypes. Even more, when deal-
ing with word formation patterns and, especifically, with derivation, semantic

. “Entità inanimate, cause indirette e non intenzionali di un evento la cui funzione poten-
ziale è quella di servire abitualmente a la realizzazione di un’azione.”
. “I derivati i cui referenti hanno il ruolo semantico di luogo nello stato di cose rappresen-
tato dall’enunciato in cui occorrono.”
 Semantic roles and word formation 

s­ lipperiness is more expected than in other domains such as grammatical mark-

ers in sentence structure.
Agent itself is not a homogeneous semantic role in syntax and this is the case in
word formation, too. For instance, in Spanish the feminine Agent suffix -dora (and
its allomorphs) allows for coining nouns that have an agentive reading but denote
inanimate entities: productora ‘production company’, constructora ‘construction
company’ o planeadora ‘speedboat’, etc. The English suffix -er is even more com-
plex: some -er nouns such as opener, chiller, or stunner can denote an event, but
still have an agentive interpretation (Panther & Thornburg 2002: 283–284).10
The clearest case of interference between Agent and Instrument are the so-
called “impersonal Agents” (Booij 1986; Ryder 1991). In Dutch this is exemplified
by the polysemy of the word zender: (1) ‘person who sends’, (2) ‘radio/TV station’,
and (3) ‘transmitter’. The first meaning refers to an animate, prototypical Agent,
while the third refers, instead, to an entity that can be manipulated like an Instru-
ment. Midway between them, the second meaning has an agentive interpretation,
but it is neither a proper, animate Agent nor is easily controllable. This fuzzy limit
between Agent and Instrument is well represented in modern languages by semi-
automatic devices, such as sprinkler, hairdryer, etc.11 All of them are inanimate
entities but have some control over the action (i.e. they are subject to manipulation
but the Agent is more or less absent). According to Pustejovsky (1995: 99), every
object has a particular telic use, such as cutting for a knife. The telicity is explicitly
reflected in impersonal Agents and Instrument, since they can still be paraphrased
as ‘X for verb-ing’, but this is not possible for animate Agents.
Other languages code Agent and Instrument in a different way. Spanish has
three main deverbal suffixes for these meanings: -dor can denote Agents and pro-
totypical Instruments; -dora is selected for self-moving devices; and -dero/-dera
are used for Instrument and Location. Maa has an Agent suffix a-…-ani and an
instrumental one -ɛt, cf. ɔl-adúŋònì ‘one who divides’ vs. en-duŋét ‘knife’ (Olsen &
Payne 2009). This is also the situation in Classical Nahuatl: compare t­ laxcalchiuani

. Some researchers have proposed to analyze English -er as a nominalizer of the subject
and not of the Agent (Levin & Rappaport 1988; Rappaport & Levin 1992; Bauer 1996 and
others). In fact, nouns such as hearer, rememberer or knower are not Agents. It may be the case
that grammatical roles play a role in these nominalization processes, too.
. In fact, nouns referring to automatic devices have been considered the link between
Agents and Instruments leading to the semantic extension Agent > Instrument (Booij 1986).
However, Rainer (2005) showed that this is wrong, given that most of those nouns referring
to machines were created in the Romance languages when the suffixes were already used for
coining Instrument nouns.
 Eugenio R. Luján & César Ruiz Abad

‘person making tortillas’ vs. tlaxcalchiualoni ‘instrument for making tortillas’

­(Luschützky & Rainer 2011).
We would like to pay attention now to other borderline cases. In some Indo-
European languages animate entities, no matter the degree of control they may
have, cannot display a prototypical Instrument suffix. So in Spanish it is not pos-
sible to coin a noun with the suffix -dero for a person unless used in a metaphori-
cal sense.12 However, in Maa we have ɔl-kípááŕ-ɛt ‘messenger, someone sent by an
elderly or rich person; disciple’, from ɪpaaya ‘send’ (Olsen & Payne 2009). In order
to coin a word denoting a person, the instrumental suffix -ɛt has been selected
instead of the agentive one a-…-ani. The Maa derivative highlights the (partial)
absence of control of the messenger and his dependency on the person who makes
the assignment of sending. Semantically, it is an Intermediary and, as this category
lacks a specific marker in word-formation, Maa codes the derivative noun with the
closest one, Instrument -ɛt.13
In French, we find some marginal values of the prototypical Instrument-­
Location suffix -oir.
(21) Nous devons nous adapter pour ne pas être un
1pl must.1pl 1pl adapt for neg neg be an
trébuchoir mais un attiroir.
obstacle but an attraction
‘We must adapt, so that we are not an obstacle [lit. ‘stumbl-er’], but a center
of attraction [lit. ‘attract-er’].’
(22) La fortune du FN [Front National] vint précisément
The fortune of_the FN (National Front) came exactly
de cette faiblesse : le parti lepéniste était pour
from this weakness the party of_Le_Pen was for
la gauche le repoussoir idéal, fort en apparences,
the left the stimulus ideal strong in appearances
mais inoffensif en réalité.
but harmless in reality
‘FN’s fortune came exactly from this weakness: Le Pen’s party was for the
Left the ideal stimulus [lit. ‘repel-er’], strong in appearances, but harmless
in reality.’

. E.g. coladero means ‘sieve’ or ‘place in which it is easy to slip in’, but metaphorically it can
also refer to a bad goalkeeper.
. Olsen & Payne (2009: 160) consider this nominalization a Theme, intended as “a non-
agentive participant that is in the condition described by the verb stem, i.e. where no change
in state occurs” or as “a participant undergoing literal movement or change in location but no
other change in state” (Olsen & Payne 2009: 160).
 Semantic roles and word formation 

In the foregoing examples only the agentive reading is possible, but the enti-
ties are inanimate. Thus, they have been identified as Forces (Villoing & Namer
2008: 1558).14 We will see below (§ 3.3) some similar cases in Greek.
Another well-known interference between semantic roles is found in nouns
of containers. As Meyer-Lübke (1890: 271) pointed out, it is not clear whether Ital-
ian beccatoio ‘seedtray’ and abbeveratoio ‘drinking trough’ should be classified as
Instruments or Locations. Facing the same problem with French -oir, some authors
have stated that it is not possible to dissociate them (Corbin 1987: 247–248). P
­ lénat
(2005: 250) considers that the differentiation is not useful, since L­ ocation has no
different means of expression. Against this view, Villoing & Namer (2008: 1553)
argue that the distinction Instrument/Location does exist since new dever-
bal nouns in -eur can be Instruments but not Locations (e.g. amortisseur ‘shock
absorber’, from amortir ‘absorb’).15 Note that this is an application of our second
criterion proposed at the end of § 2.2. Villoing and Namer (2008: 1553–1556) have
gone further and they have proposed a test, (23)–(24), in order to classify a noun
in one or the other category:16
(23) NP0 V W avec DET V-oir ↔ NP0 utilise DET V-oir pour Vinf W
(24) A quel endroit est-ce que NP0 V W ? Dans/sur DET V-oir

From the 639 ‘-oir tokens’ of their corpus, 587 (91,8%) were felicitously interpreted
in one of the senses. Unfortunately, we have not found an objective way to apply the
same tests to our corpus, as no one has real linguistic competence in Ancient Greek.
Finally, we have to recall that a well-known problem for semasiological
studies in word-formation is the unpredictability of the semantic evolution of
derivatives. This is also related to the classical difference between Wortbildungs-
bedeutung ‘meaning of the word formation pattern’ and Wortbedeutung ‘meaning
of the word’. For instance, French Instrument-Locative -oir in tiroir ‘drawer’ is
apparently construed as the entity affected by the verb. However, when it was first

. In fact, they chose the term ‘Cause’, but their examples fit with the definition of Force by
Van Valin & LaPolla (1997: 121) that we mentioned above. In order to avoid any confusion
with Cause, which is another differentiated semantic role both in French and in Greek, we
keep the term Force.
. In fact, the same argument is also mentioned by Corbin (1987: 247): “A priori, ces sens
[Instrument et Lieu] sont différents. Par exemple, les noms construits en -on sur une base
verbale peuvent avoir un sens instrumental (bouchon, guidon, lorgnon, pilon, etc.), mais pas
un sens locatif ” (“A priori, these meanings [Instrument and Place] are different. For instance,
nouns in -on from a verbal base can have an instrumental reading, but not a locative one”).
However, potentially any noun in -oir can have both meanings, and so she prefers to set only
one RCM [règle de construction de mots (“word construction rule”)] (Corbin 1987: 248).
. NP0= subject, V=verb, W=object, DET= determinant.
 Eugenio R. Luján & César Ruiz Abad

coined at the beginning of 15th century, the word referred to the ‘door handle’,
so that the later meaning is the result of a metonymic extension. For English
Agent-Instrument -er, it is possible to find similar cases: e.g. keeper may denote
an entity that is subjectively construed as worthy of being kept, as in a sentence
like ‘your boyfriend is a keeper’ (see more examples in Panther & Thornburg
2001: 173–174).
Summing up, in this section, we have discussed some of the major problems
that need to be adressed when trying to identify semantic roles in word forma-
tion. Two main, common fuzzy areas have been referred to, ‘Impersonal Agents’,
which are mid-way between Agents and Instruments, and ‘containers’, which can
be interpreted as Instruments and Locations. Futhermore, word formation pat-
terns serve to coin new words and the lexikon is more likely to undergo unpredict-
able semantic shifts than other domains. Due to specific semantic changes, a given
derivative may not match exactly the definition of the prototypical Semantic Roles
(as given in §3.1). As a result thereof, derivatives can display meanings that are
somewhat unexpected from the prototypical value of the pattern.

3.3 Instrument and Location in word formation in Ancient Greek

In this section, we will focus first on the three main deverbal suffixes which denote
Instrument and/or Location in Ancient Greek:17 -tron, -thron, and -terion.18 We
will then compare their behaviour to that of the suffix -ter.19
A preliminary remark must be made: Greek has a neuter gender and this is the
gender of the nouns in -tron, -thron, and -terion. It is also interesting that there is a
formal relationship between the prototypical masculine Agent suffix -ter, with the
long grade /ē/ characteristic of the Nominative singular of liquid stems, and -tron,
with the neuter ending -on and hence the zero grade suffix (cf. aroter ‘plougher’
vs. árotron ‘plough’). In Table 1 we have listed the meanings of -tron and words for
each meaning occurring in the corpus.

. See Chantraine (1933: 62–64, 330–334, 374–375) and Balles (2008: 191–192, 196–197) on
these suffixes.
. We have obtained the data of the suffixes -tron, -thron, and -terion from a corpus ranging
from the 8th to the 5th century including the Iliad, the Odyssey, Hesiod, the Homeric hymns,
Alcman, Sappho, Alcaeus, Pindar, Hippocrates, Herodotus, Aeschylus, Sophocles, Euripides,
Aristophanes, Thucydides, and Lysias. We only take into account those nouns that were
­synchronically analyzable to a Greek speaker.
. For the suffix -tḗr the whole corpus of Ancient Greek has been taken into account, but
given its high productivity we have not elaborated complete lists.
 Semantic roles and word formation 

Table 1. Meanings of -tron and words in the corpus

Instruments (30 derivatives)
ánkistron ‘fish-hook’ **ankízō *‘fish’
amfíblēstron ‘casting-net’ amfibállō ‘throw round’
árotron ‘plough’ aróō ‘plough’
dértron ‘membrane which contains the bowels’ deírō ‘skin, flay’
díoptron ‘means for seeing through’ dioráō ‘see through’
élytron ‘covering’ eilýō ‘enfold’
énoptron ‘mirror’ enoráō ‘see, look at’
epíklintron ‘couch, arm-chair’ epiklínō ‘lie on’
ésoptron ‘mirror’ eisoráō ‘look into’
zôstra (pl.) ‘belt, girdle’ zonnumi ‘gird’
thýretra (pl.) ‘door-casing, frame’ thýrē ‘door’
kátoptron ‘mirror’ kátopsis ‘sight’
kéntron ‘horse-goad’ kentéō ‘prick, goad’
lístron ‘tool for levelling’ lissós, -á, -ón ‘smooth’
piéstron ‘spatula for children birth’ píezō ‘press’
plástron ‘ear-ring’ plássō ‘mould’
plêktron ‘instrument for striking the lyre’ plessō ‘strike’
podó-psēstron ‘footwiper’ psáō ‘rub, wipe’
róptron ‘the wood in a trap’ répō ‘incline, fall’
semantron ‘seal’ sēmaínō ‘indicate’
stégastron ‘cover, wrapper’ stegázō ‘cover’
táraktron ‘tool for stirring with’ tarássō ‘stir’
téretron ‘borer, gimlet’ teírō ‘bore through’
hypáleiptron ‘spatula for spreading a salve’ hupaleífō ‘spread like salve’
kheiromaktron and kheirómaktron ‘cloth for wiping mássō ‘knead’
the hands, towel’
Non-prototypical instruments
thélgētron ‘charm, spell’ thélgō ‘enchant’
thélktron ‘charm, spell’ thélgō ‘enchant’
loetrá (pl.) in the sense ‘water for the bath’ loéō ‘bath’
podániptra (pl.) ‘water for washing the feet in’ níptrō ‘wash’
phíltron in the sense ‘love-charm’ philéō ‘love’

(Continued)
 Eugenio R. Luján & César Ruiz Abad

Table 1. (Continued)

Prizes (5 derivatives)

thréptra (pl.) ‘reward for rearing, made to nurses by parents’, tréphō ‘nourish’
‘return made by children for their rearing’
kómistron ‘reward for carrying or for saving’ komízō ‘carry’ and ‘save’
lýtron ‘ransom’ lýō ‘unfasten’, ‘save’
menytron ‘reward for information’ mēnýō ‘reveal’
sôstra (pl.) ‘reward for saving’, ‘thank-offering for deliverance so zō ‘save’
from a danger’
Forces (1 derivative)
phóbētra (pl.) ‘terrors’ phobéō ‘terrify’

Containers (3 derivatives)
eksáleiptron ‘unguent-box’ eksaleíphomai ‘anoint’
zýgastron ‘box [which is closed like a yoke]’ zygón ‘a yoke’
phértron ‘bier, litter’ phérō ‘carry’

Location (4 derivatives)
théatron ‘theatre’ theáō ‘contemplate’
léktron ‘couch, bed’ lékhomai ‘lie down’
loetrón in the sense ‘bathing-place’ loéō ‘bath’
pístron ‘drinking-trough for cattle’ pipískō ‘give to drink’

Varia (4 derivatives)
ékmaktron ‘impress’ ekmássō ‘mould’
daitrón ‘portion’ daíō ‘divide’
loetrón in the sense ‘bath’ loéō ‘bath’
phíltron ‘love, affection’ philós, -á, -ón ‘beloved, dear’

The majority of the nouns in -tron are Instruments. We find just a few Loca-
tion nouns and some containers. There is a group of non-prototypical Instruments,
i.e. nouns denoting entities that have some degree of control over the action and
cannot be easily manipulated, such as charms, e.g. thélktron ‘charm, spell’ (from
thélgō ‘enchant’). Another group of nouns, usually found in the plural, is well rep-
resented by polysemous sôstra (Pl.). It denotes both a prize-reward and a sacrifice
for deliverance. This second possibility is not surprising since in Antiquity the
relation god-devotee is one of mutual interest, and so the sacrifice is considered a
kind of payment to the god. However, sôstra as a thanks-offering is not ‘something
 Semantic roles and word formation 

for delivering’ but ‘something paid as exchange of the delivrance’. In this second
acception, sôstra do not represent an entity which precedes the action, that is, the
‘instrumental’ reading of the derivative denoting a sacrifice is no longer so clear.
An interesting case is phóbētra (Pl.) ‘terrors, things that terrify’ – an instrumental
reading is not possible, no Agent has been blurred from the scene and the entity is
inanimate. It is thus a Force, like French trébouchoir ‘obstacle’, attiroir ‘attraction’,
and repoussoir ‘stimulus’ (cf. § 3.2).20
Under the heading “varia” we have classified certain marginal, unpredictable
meanings. Ékmaktron and daitrón ‘portion’ (< daíō ‘divide’) could be classified as
result nouns. Loetrá has two meanings, ‘water for the bath’ and ‘bathing place’,
but it is also attested as an action noun (cf. LSJ). Apart from their unpredictable
meaning, all of them present some particularity. Ékmaktron is an absolute hapax
in the Greek literature, while daitrón and loetrá have an uncommon accentua-
tion. Finally, it could be assumed that unpredictable phíltron ‘love, affection’
could be the result of metonymic semantic change from the instrumental mean-
ing ‘love-charm’ but we have no empirical data since both meanings are early
attested.21
Suffix -thron has the same values as -tron, and, in principle, they are found in
complementary distribution. -thron is productive (rentable in the sense of Corbin
1987: 42) when added to contracted verbs, which are blocked for -tron. In terms of
availability (disponibilité in the sense of Corbin 1987), -thron is quite less produc-
tive for two main reasons: contracted verbs are relatively uncommon and, accord-
ing to Grassmann’s law, -thron becomes -tron if there is a voiceless aspirate stop in
the base.22

. This noun is a neuter plurale tantum because it is a kind of collective. The existence of a
differentiated neuter in Greek may help explain the coining of phóbētra with this suffix.
. See infra stérgēthron, with the same polysemy.
. This is an oversimplication, but we will not go here into the details. We would just like to
mention other phonetic factors that play a role in the distribution of the suffixes and explain
some of the apparent exceptions. Bases ending in a consonantal cluster usually select -ēthron:
mélp-ēthron and stérg-ēthron (and maybe hélkēthron < helkéō or hélkō?) and, according to
Grassmann’s law of dissimilation, thélgēthron (but thélktron with a verbal base ending in a
consonantal cluster). -stron also competes with -thron when the base is monosyballic: podó-
psē-stron, pí-stron. Cases of hesitation can be easily found: kláïstron/kleîthron, the com-
pounds amphí-blēstron/ém-blēthra (from bállō, blē-), kállyntron/kállynthron, enkoímētron/
enkoímēthron, etc. The same distributon is observed in the femenine, homofunctional forms
in -tra, -thra, and -stra.
 Eugenio R. Luján & César Ruiz Abad

Table 2. Meanings of -thron and words in the corpus

Instruments (8 derivatives)
árthron ‘joint’ ararískō ‘join together’
hélkēthron ‘stock of the plough’ helkéō ‘drag about, tear asunder’
kleîthron ‘bar for closing a door’ kleíō ‘close’
kýkēthron ‘ladle for stirring’ kykéō ‘stir’ (used in a metaphorical way, cf. LSJ)
mélpēthra ‘plaything’ mélpō ‘celebrate with song and dance’
skandálēthron ‘stick in a trap for animals on skandálē id.
which the bait is placed and which shuts
the trap’

Non-prototypical instruments
pharýngethron ‘throat’ phárynks id.
psílōthron ‘depilatory’ psilóō ‘strip bare’

Prizes (1 derivative)
epíbathron ‘passenger’s fare’ epibaínō ‘go upon, board’

Location (3 derivatives)
báthron ‘base’ baínō ‘walk, step’
ptolíēthron ‘citadel’ ptólis id.
reîthra (pl.) ‘that which flows, a river, stream’ réō ‘flow, run, stream’

Varia (1 derivative)

stérgēthron ‘love itself ’ < stérgō ‘charm’ loéō ‘bath’

Due to the restrictions of this suffix, the number of nouns in -thron in the
corpus is quite limited. As with -tron, they are for the most part Instruments.23
-terion is a more complex suffix because its derivatives can be the result of
more than one rule of word formation:

1. The “plain” suffix -terion is originally an accumulation of the Agent suffix

-ter plus the denominal Instrument and Location suffix -ion (Chantraine
1933: 62). However, from an early date there are nouns in -terion with the suf-
fix added directly to a verbal base, with no corresponding Agent. The fact that

. The only remarkable case is stérgēthron ‘love itself ’. It is also attested with the instru-
mental meaning ‘love-charm’. However, the former, current meaning is attested since the 5th c.
B.C., while the latter occurs only in the 1st c. Therefore, a semantic extension from the predict-
able, instrumental meaning cannot be argued for.
 Semantic roles and word formation 

the Agent suffix -ter was progressively replaced by -tēs also played a role in the
grammaticalization of -terion.
2. There are reasons24 to think that -terion has been used as a padding suffix to
complete a verse.
3. There are also relational adjectives in -ter-ios (Agent -ter + adj. -ios). They can
provide additional derivative nouns in -terion by elision of the noun that they
originally modified or by substantivization by means of the neuter singular or,
more often, the neuter plural. Elisions are well represented by medical terms:
next to occurrences in which the adjective is substantivized, they frequently
appear in the same texts as adjectives agreeing with neuter nouns of generic
meaning such as phármakon ‘remedy, medicine’ or prosthetón ‘pessary’.25

The number of words occuring in the corpus might give the impression that
Instrument is the most represented category for this suffix, too. However, it should
be noted that 15 nouns are non-prototypical Instruments from medical jargon,
all of them found in Hippocrates, whose work amounts to more than a third of
the corpus. Celebrations, rituals etc. constitute a heterogeneous category and the
number of words must also be taken with caution. As in the case of prizes-rewards
and sacrifices in -tron, offerings, presents and prizes are quite similar. And the
relation between offerings and celebrations/rituals is one of part-whole.26 Out
of the corpus we have found the polysemous word anakalypteria (pl.) ‘festival of
unveiling, when the bride first took off her maiden veil, and received presents from
the bridegroom’ and ‘the presents themselves’ (from anakalýptō ‘unveil onself ’, cf.

. Evidence comes from eklyterion, kēlēterion, klērōterion, thelkterion, thyterion, threpterion,
zeukterion, lyterion, physēterion, and khalinōteria, which only occur in metrical texts. They are
hapax or quasi-hapax in Greek literature and all of them have synonyms which are of common
use. For example, eklyterion, thelkterion, threpterion, and lyterion have a parallel form in -tron.
. One may argue that different rules would require an analysis of different homonymic
suffixes. We have chosen an integrated approach because the rules can be only identified at an
abstract level. The probable (originally) ellipsis in medical jargon of non-prototypical Instru-
ments is only true for some terms since others, such as aleksēterion, pyriāterion, pianterion,
and enkyēterion, have no corresponding adjective. The pattern has hence been “grammatical-
ized”. The same is true for nouns of festivals and sacrifices, which are supposed to have been
derived by means of the neuter plural (but there is also opteria, without the corresponding
adjective). The only formal difference – and a reasonable measure to keep to the idea of
homonym suffixes – lies in the number (singular or plural) of the derivative nouns (but cf.
mysterion and the poetic hapaxes thyterion and lyterion). Another reason is that such an a
priori so heterogeneous suffix is semantically coherent, as we will see.
. Metonymy plays thus an important role in the semantic changes involved in these and
other similar instances. In a cognitive perspective they can be accounted for as the result of
figure/ground effects inside the same frame (Koch 2001).
 Eugenio R. Luján & César Ruiz Abad

Table 3. meanings of -terion and words in the corpus

Instruments (21 derivatives)

ampykterion ‘horse’s headband’ ampykter id.

zeukterion ‘yoke’ zeúgnȳmi ‘yoke’
physēterion ‘wind-instrument, pipe’ physáō ‘blow’
khalinōteria (pl.) ‘mooring-cables’ khalinóō ‘bridle’
perēterion ‘borer’ peráō ‘pierce’,
Non-prototypical instruments
aleksēterion ‘remedy, medicine’ aléksō ‘ward off, defend’
dēlēterion ‘poison’ dēléomai ‘damage’
enkyēterion ‘drug which promotes enkyéomai ‘to be borne in the
conception’ womb’
elaterion ‘squirting cucumber’ and ‘drug prepared therefrom’ elaúnō ‘drive away, expel’
thelkterion ‘charm’ thélgō ‘enchant’
iāterion ‘mode of cure, cure’ iáomai ‘cure’
katharterion ‘purgative’ kathaírō ‘purify’
kēlēterion ‘charm’ kēléō ‘charm’
kyēterion ‘drug which promotes conception’ kyéō ‘be pregnant with’
malthakterion ‘emollient’ malthássō ‘soften, soothe’,
peirāterion ‘test’ peiráō ‘test’
pianterion (pl.) ‘fattening food’ piaínō ‘fatten’
pyriāterion ‘vapour-bath’ pyriáō ‘foment’
skheterion ‘remedy [against hunger]’ ékhō ‘check, keep’
phyteuterion ‘layer’ phyteúō ‘plant’
Containers (5 derivatives)
thymiāterion ‘censer’ thymiáō ‘burn incense’,
klērōterion ‘urn for casting lots or votes’ klēróō ‘vote’
aporranterion ‘a vessel for sprinkling with holy water’ aporraínō ‘sprinkle’,
perirranterion ‘vessel for lustral water’ perirraínō ‘sprinkle’,
poterion ‘drinking-cup’ poter id.
Location (12 derivatives)
akrōterion ‘prominent part’ ákros, -a, -on ‘highest’
bouleuterion ‘council-chamber’ bouleúō ‘deliberate’
desmōterion ‘prison’ desmotēs ‘prisoner’
dikasterion ‘court of justice’ dikázō ‘judge’

(Continued)
 Semantic roles and word formation 

Table 3. (Continued)

enēbēterion ‘place of amusement’ enēbáō ‘spend one’s youth in’

ergasterion ‘workshop, manufactory’ ergázō ‘work’,
eunāterion ‘bed-chamber’ eunáō ‘sleep’
logisterion ‘the place at Athens where the logistaí met’ logízomai ‘calculate’
khresterion in the sense ‘seat of an oracle’ khráō ‘proclaim’
prēterion ‘place for selling, market’ práō ‘sell’
phrontisterion ‘place for meditation’ phrontízō ‘consider’,
phylakterion ‘fort’ phylássō ‘keep watch’
Celebrations, rituals, offerings, presents, prizes (14 derivatives)
diabateria (pl.) ‘offerings before crossing the border’ diabaínō ‘cross’
eklyterion ‘expiatory offering’ eklýō ‘set free’
thoināterion ‘meal, feast’ thoináō ‘feast, sacrifice’
threpterion ‘reward for rearing, made to nurses by parents’, tréphō ‘nourish’,
‘return made by children for their rearing’
thyterion ‘sacrifice’ thýō ‘sacrifice’
thōsterion ‘banquet’ thôsthai ‘feast’
lyterion ‘expiatory offering’ lýō ‘atone for’
meilikteria (pl.) ‘propitiations’ meilíssō ‘appease, propitiate’
mysterion ‘mistery, sacred rite’ myéō ‘initiate into the mysteries’
nikēterion ‘prize of victory’ nikáō ‘win’
nympheuteria (pl.) ‘offerings for the marriage’ nympheúō ‘marry [a woman]’
khresterion in the sense ‘an offering for the oracle’ khráō ‘proclaim’
opteria (pl.) ‘presents upon seeing or for the sight of a person’ horáō ‘see’,
pasteria (pl.) ‘feast on sacrificial meats’ patéomai ‘eat’
Varia (3 derivatives)

sōteria (pl.) ‘things saving, deliverance’ sozō ‘save’

drasterion ‘energy, activity’ drasterios ‘active, efficacious’
khrēsterion in the sense ‘oracular response’ khráō ‘proclaim’

LSJ). Note that on the one hand these meanings form a continuum; on the other,
it is difficult to identify the semantic role at the end of this continuum, especially
when it comes to names of celebrations and rituals.
Under the heading ‘varia’ we have grouped together, again, the marginal,
unpredictable meanings. Sōteria (pl.) ‘things saving, deliverance’ could be a Force
but the scarsity of instances do not allow for a clear interpretation of this derivative
noun or for establishing the differences to the commoner femenine noun sōtēría
 Eugenio R. Luján & César Ruiz Abad

‘deliverance’. Drasterion ‘energy, activity’ is an abstract noun denoting the quality

of the adjective drasterios ‘active, efficacious’, i.e. it is a probable case of substan-
tivization of the neuter singular of an adjective. It might be interpreted as Force,
but this is not so straightforward. Like loutrá/loutrón, khrēsterion shows a complex
polysemy: it can refer to the seat of an oracle, an offering for the oracle and an
oracular response. The polysemy may be the result of metonymic extensions (cf.
‘oracle’ in English) or of analogy to other -ion derivatives such as mateîon in the
sense of ‘oracular response’ (from mántis ‘diviner, seer’).
After analyzing the data of these three suffixes now it is the turn of -ter. In
Greek, like in other Old Indo-European languages, this suffix was primarily used
for deriving Agent nouns. However, we find instances of other meanings, too:27

a. Agents: thyter ‘sacrificer’ (cf. thýō ‘sacrifice’), lȳmanter ‘spoiler, destroyer’ (cf.
lymaínomai ‘outrage’), therapeuter ‘attendant’ (cf. therapeúō ‘do service to’).
b. Forces: prēster ‘hurricane’ (cf. prethō ‘blow out’).
c. Instruments: glypter ‘chisel’ (cf. glýphō ‘engrave’), kalypter ‘covering, lid,
tile(s)’ (cf. kalýptō ‘cover’), katalēpter ‘strap for holding fast’ and ‘clamp’ (cf.
katalambánō ‘seize’), klōster ‘spindle’ (cf. klóthō ‘spin’), kolapter ‘chisel’ (cf.
koláptō ‘carve, engrave’), poter ‘drinking cup’ (cf. pínō ‘drink’).
d. Body parts: mykter ‘nostrils’ (cf. mýssomai ‘blow the nose’).
e. Means: halteres ‘weights [held in the hand to give impetus] in leaping’ (cf. hál-
lomai ‘spring, leap’).
f. Containers: thermanter ‘pot for boiling water’ (cf. thermaínō ‘warm, heat’),
hepsēter ‘dish or pan for boiling’ (cf. hépsō ‘boil’), khōneuter ‘hollow in which
metal was placed for melting’ (cf. khoaneúō/khōneúō ‘cast in a mould’).
g. Locations:28 bater ‘that on which one treads, threshold’ and ‘place from which
one jumps’ (cf. baínō ‘walk, step’), embater ‘threshold’ (cf. embaínō ‘step in’),
klinter ‘couch’ (cf. klínō ‘cause to lean, make to slope’), kampter ‘bend, angle’
and ‘turning-point’ (cf. kámptō ‘bend, curve’).

Even if the suffix was originally used for deriving Agents, when employed for
Instruments and Locations we find a similar polysemy and similar intermediate
and non-prototypical meanings as with the three suffixes previously surveyed.
In fact, a detailed analysis of the attested history of this suffix in Ancient
Greek shows that it has gone through various consecutive phases in which the
main meaning of the suffix has shifted from one semantic role to another. We can
summarize the evidence as follows:

. See Chantraine (1933: 321–329) and Balles (2008: 183 and 190).
. Balles (2008: 196–197) does not list -ter among the suffixes deriving nomina loci.
 Semantic roles and word formation 

i. In the oldest records of the Greek language (i.e. in the Mycenaean dialect of
the Linear B tablets, which dates back to the 2nd millennium BC) -ter occurs
frequently with Agents and is extremely rare for Instruments.29
ii. In the earliest records of 1st millennium Greek, i.e. in the Homeric poems,
there is an increase of the number of Instruments in -ter. Nevertheless, Agents
still outnumber Instruments as the most common reading of nouns in -ter.
This also seems to be the situation in dialects other than Ionian-Attic.
iii. Interestingly, in the Ionian-Attic dialects -ter was given up as a productive
suffix for the formation of Agent nouns and was productive only for deriving
Instruments.30

The findings of our investigation into the meanings of the deverbal suffixes used
for the derivation of Instruments and Location nouns in Ancient Greek allow for
establishing the semantic map of Figure 1.31

Containers

Agent Instrument Place

Force Prizes
-thron
-te¯r
-te¯rion and -tron

Figure 1. Semantic map of the suffixes for instrument and location in Ancient Greek

. See Hernández (2005). The lack of more Instrument nouns in -ter in the extant Mycea-
nean texts cannot be explained away as due to the textual types represented in them, given that
Instrument nouns do occur frequently in them.
. There are few exceptions to this rule, such as lȳmanter ‘spoiler, destroyer’ in Xenophon or
therapeuter ‘attendant’, which is employed by various authors.
. Semantic maps are expected to be of universal validity, but they are implicitly based in
language-specific semantic maps. See François (2008) for a neat example of the application of
the methodology to the domain of lexical semantic associations, in which language-specific
semantic maps are provided as the basis for the general semantic map of the notion ‘breathe’.
 Eugenio R. Luján & César Ruiz Abad

3.4  omparing Instrument and Location in grammatical morphemes

C
and word formation patterns
It is interesting to compare now the findings in the domain of word formation to
grammatical morphemes (cases and prepositions). According to Crespo’s (1997)
data, grammatical morphemes serving to express Instrument and Location are
also used at least for the expression of one of the following:32

a. Accompaniment,
b. Means-Intermediary + Cause,
c. Means-Intermediary + Price.

If we compare this situation to our findings in the domain of word formation we

can gain more insight into the nouns that we have classified as Prizes or non-
prototypical Instruments. In Ancient Greek, as in most languages,33 there are no
specific markers for deriving nouns expressing Accompaniment, Cause, Prize, or
Means-Intermediary. However, as we saw above (§ 3.3), all the suffixes employed
in this language to derive Instrument and Location nouns also provide nouns with
meanings in the area of Means-Intermediaries, Causes, and Prizes. The situation
is, indeed, less clear-cut with word formation patterns than with syntactic mark-
ers and this has to do with the problem of identifying the semantic roles of a word
formation pattern and of the words built by means of it (§§ 2 and 3.2). However,
in the same way as the grammatical morphemes employed for the expression of
Instrument and Location in syntax are used for other semantic roles, too, it is not
surprising that in word formation Instrument and Location behave as adjacent
prototypes inside a fuzzier area. This allows for employing Instrument-Location
suffixes to coin nouns which are not exactly Instruments nor Locations, but are
semantically related to those notions.
As for containers, word formation patterns reflect a similar situation to
grammatical morphemes (Wackernagel 1922: 304; Luraghi 2004: 28). It is well
known that if the lexeme to which a given grammatical morpheme is added
refers to a vehicle or container, the semantic roles Location and Instrument
can be hardly distinguished. This explains instances of coordination like (19)
above for English or (25) for Greek (Thukydides 4.25.7, quoted by Crespo
1997: 30).

. The only possible exception would be the genitive, but it is quite doubtful that it can
be used for the semantic function Instrument (Crespo 1997: 14). Greek adverbs in -êi are
­employed for Instrument, Location, and Means-Intermediary.
. See the remarks in § 3.2, and especially, Comrie & Thompson (1985) and Mel’čuk (1997).
 Semantic roles and word formation 

(25) katà gên kaì taîs nausìn háma estráteuan

by earth.acc and the.dat.pl ship.dat.pl together fought.3pl
‘They fought both by earth and with the navy.’

For Homeric Greek Luraghi (2004) has analyzed in detail the container scheme
and she has shown how in Greek locative prepositions like en ‘in’ are used for the
expression of Containment and they are employed with means of transportation,
too, as in (26) (Il.12.16, quoted by Luraghi 2004: 37).
(26) Argeîoi d’ en nēusì phílēn es
Argive.nom.pl ptc in ship.dat.pl dear.acc.f to
patríd’ ébēsan
homeland.acc.f went.3pl
‘The Argives had gone back in their ships to their dear motherland.’

Exceptionally the plain dative (i.e. with instrumental reading) is found, too, as in
(27) (Od.10.502, quoted by Luraghi 2004: 37).
(27) eis Áidos d’ ou po� tis amphíketo
to Hades.gen ptc neg ptc indf.nom come.aor.m.3sg
nēí melaínēi
ship.dat.f black.dat.f
‘No man ever reached Hades with a black ship.’

In a broader perspective, our results in the domain of word formation – even if

limited to one specific language in this paper – seem to fit with the findings of
general works about the structure of the conceptual space reflected in syntactic
markers. We can compare our findings to the diachronic semantic map proposed
by Narrog (2010: 242) for the Instrument-Companion domain (Figure 2). In this
map, Location is adjacent to Instrument and they are also connected to other
semantic roles, some of which express notions related to those found for the word
formation patterns that we have analyzed in detail in this paper.

Duration
Physical
proximity Co-participant Location Route
Cause/reason
Clausal NP- Comitative Instrument Passive agent
coordination coordination
Ergative agent
Possession Source
Material
Temporal “from”

Figure 2. Semantic map of the instrument domain with directionality (Narrog 2010: 242)
 Eugenio R. Luján & César Ruiz Abad

4. Concluding remarks

In functional-typological approaches to syntax the identification of semantic roles

and the grammatical markers (cases and adpositions) associated with them has
been an important goal. In this paper we have argued that a similar approach to
word formation patterns is possible. Elaborating on the definitions and methodol-
ogy employed in syntactic studies, we have provided criteria for the identification
of the semantic roles in the domain of derivational morphology.
The information provided by typological studies on word formation (e.g. Com-
rie & Thompson 1985 and Mel’čuk 1997) shows that, according to those criteria,
Agent, Instrument, and Location can be considered as different semantic roles in
morphology, since there are languages in which different dedicated suffixes are used
for each of them. However, polysemy between two or more of these semantic roles is
also frequent cross-linguistically (Bauer 2000; Luschützky & Rainer 2011 and 2013).
As a case study we have focused on the deverbal suffixes employed in Ancient
Greek for deriving instrumental and locative nouns. Based on the actual nouns
occurring in a representative corpus of Ancient Greek literature, we have com-
piled exhaustive lists of the nouns derived by means of the three most productive
suffixes in that domain: -tron, -thron, and -terion. The semantic analysis of those
nouns has shown that besides instrumental and locative nouns, the suffixes are
used for deriving other nouns that are semantically related, but cannot be consid-
ered Instruments and Locations proper. Among them we find nouns referring to
non-prototypical instruments, prizes and rewards, and others. Quite interestingly,
the suffixes are also frequently used for deriving nouns of containers, which are
semantically related both to Instruments and Locations. Furthermore, the seman-
tic analysis of one of the most important Agent suffixes in Ancient Greek (-ter)
has revealed that it developed a polysemy similar to that found in the other three
suffixes.
We have compared our findings in morphology to the conceptual organiza-
tion of grammatical morphemes in the same language (Greek) and in other lan-
guages in general. And it appears that both in syntax and morphology a similar
semantic organization is found. This fits well with the fact that the conceptual
semantic organization of morphemes, whether these are grammatical morphemes
(cases and prepositions) or derivational ones, is expected to be the same (Luján
2010: 171–172). This seems to be evidence for arguing that there are certain under-
lying mental structures that would be reflected both at the syntactic and the mor-
phological level. This can be proved, for instance, by the position of containers:
both in syntax and in word formation patterns containers are midway between
Instrument and Location.
The number of semantic roles that can be identified in syntax is higher than
the number of semantic roles reflected in word formation patterns. In this sense,
 Semantic roles and word formation 

semantic maps of word formation patterns are expected to be similar to those

that can be drawn for syntax but more reduced (Luján 2010: 171). For this reason,
when there is no specific suffix for a semantic role such as Means-Intermediary but
a noun expressing it needs to be coined, suffixes for neighbouring semantic roles
(Location and Instrument) will be employed, depending on the available meta-
phor or metonymy that allows for the semantic extension.

References

Balles, Irene. 2008. Latein, Altgriechisch, Vol. 1: Nominale Wortbildung des Indogermanischen in
Grundzügen, Rosemarie Lühr (ed.), Hamburg: Dr. Kovač.
Bauer, Laurie. 1996. English Word-formation, 2nd edn. Cambridge: CUP.
Bauer, Laurie. 2000. What you can do with derivational morphology. In Morphology 2000.
Selected Papers from the 9th Morphology Meeting (Vienna, 25–27 February 2000), Sabrina
Bendjaballah, Wolgang U. Dressler, Oskar Pfeiffer & Maria D. Voeikova (eds.), 37–48.
Amsterdam: John Benjamins.
Booij, Geert. 1986. Form and meaning in morphology: the case of Dutch agent nouns. Linguis-
tics 24: 503–517.
Booij, Geert. 2007. The Grammar of Words, 2nd edn. Oxford: OUP. DOI: 10.1093/acprof:oso/
9780199226245.001.0001
Chantraine, Pierre. 1933. La formation des noms en grec ancien. Paris: Champion.
Comrie, Bernard & Thompson, Sandra. 1985. Lexical nominalization. In Language Typology
and Syntactic Description, Vol. 3: Grammatical Categories and the Lexicon, Timothy Shopen
(ed.), 349–98. Cambridge: CUP.
Corbin, Danielle. 1987. Morphologie dérivationnelle et structuration du lexique. Tübingen:
Niemeyer.
Crespo, Emilio. 1997. Sintaxis de los elementos de relación en griego clásico. In Actas del IX
Congreso Español de Estudios Clásicos, Vol. 2, Francisco R. Adrados & Alfonso Martínez
Díez (eds), 3–42. Madrid: Ediciones Clásicas.
Crespo, William. 1991. Syntactic Categories and Grammatical Relations. The Cognitive Organiza-
tion of Information. Chicago IL: The University of Chicago Press.
Cruse, D., Alan. 1973. Some thoughts on agentivity, Journal of Linguistics 9: 11–23. DOI: 10.1017
/S0022226700003509
Cysouw, Michael. 2007. Building semantic maps: The case of person marking. In New Chal-
lenges in Typology, Bernhard Wälchli & Matti Miestamo (eds), 225–248. Berlin: Mouton
de Gruyter.
Cysouw, Michael. 2010. Semantic maps as metrics of meaning. In Semantic Maps. Methods and
Applications (Linguistic Discovery 8.1), Michael Cysouw, Martin Haspelmath & A ­ ndrej
­Malchukov (eds), 70–95.
Fillmore, Charles J. 1968. The case for case. In Universals in Linguistic Theory, Emmon Bach &
Robert T. Harms (eds), 1–88. New York NY: Holt, Rinehard, and Winston.
François, Alexandre. 2008. Semantic maps and the typology of colexification: intertwining
polysemous networks across languages. In From Polysemy to Semantic Change: Towards
a Typology of Lexical Semantic Associations [Studies in Language Companion Series 106],
Martine Vanhove (ed.), 163–215. Amsterdam: John Benjamins.
 Eugenio R. Luján & César Ruiz Abad

Geeraerts, Dirk. 1997. Diachronic Prototype Semantics. Oxford: Clarendon.

Givón, Talmy. 2001. Syntax: An Introduction, 2 Vols, 2nd edn. Amsterdam: John Benjamins.
Grossmann, Maria. 1998. Formazioni dei nomi di agente, strumento e lugo in italiano. In Atti
del XXI Congresso Internazionale di linguistica e Filologia, Giovanni Rufino (ed.), 383–393.
Tübingen: Max Niemeyer.
Haspelmath, Martin. 1997. From Space to Time: Temporal Adverbials in the World’s Languages.
Munich: Lincom.
Haspelmath, Martin. 2003. The geometry of grammatical meanings: Semantic maps and cross-
linguistic comparison. In The New Psychology of Language, Michael Tomasello (ed.),
211–242. Mahwah: Lawrence Erlbaum Associates.
Hernández, César. 2005. Los nombres de agente en micénico: Aspectos semánticos y morfológi-
cos. MA thesis, Universidad Complutense (Madrid).
Koch, Peter. 2001. Metonymy. Unity in diversity. Journal of Historical Pragmatics 2(2): 201–244.
DOI: 10.1075/jhp.2.2.03koc
Levin, Beth & Rappaport, Malka. 1988. Non-event -er nominals: A probe into argument struc-
ture. Linguistics 26: 1067–1083. DOI: 10.1515/ling.1988.26.6.1067
LSJ = Liddell, Henry George, Robert Scott & Henry Stuart Jones. 1940. A Greek-English Lexicon,
9th edn. Oxford: Clarendon Press.
Luján, Eugenio R. 2010. Semantic maps and word formation: Agents, Instruments, and related
semantic roles. In Semantic Maps. Methods and Applications (Linguistic Discovery 8.1),
Michael Cysouw, Martin Haspelmath & ­Andrej Malchukov, 162–175.
Luraghi, Silvia. 1995. Protypicality and agenthood in Indo-European. In Historical Linguis-
tics 1993. [Current Issues in Linguistic Theory 124], Henning Andersen (ed.), 259–268.
Amsterdam: John Benjamins.
Luraghi, Silvia. 2003. On the Meaning of Prepositions and Cases. The Expression of Semantic
Roles in Ancient Greek [Studies in Language Companion Series 67]. Amsterdam: John
­Benjamins. DOI: 10.1075/slcs.67
Luraghi, Silvia. 2004. The container schema in Homeric Greek. In Linguagem, cultura e cog-
nição: Estudos de Linguística Cognitiva, Augusto Soares de Silva, Amadeu Torres & Miguel
Gonçalves (eds), 25–41. Braga: Almedina. DOI: 10.1524/stuf.2011.0023
Luschützky, Hans C. & Rainer, Franz. 2011. Agent noun polysemy in a cross-linguistic perspec-
tive. Language Typology and Universals 64(4): 287–338. DOI: 10.1515/ling-2013-0051
Luschützky, Hans C. & Rainer, Franz. 2013. Instrument and place nouns. A typological and
diachronic perspective. Linguistics 51(6) : 1301–1359.
Mel’čuk, Igor Aleksandrovič. 1997. Cours de morphologie générale (théorique et descriptive),
Vol. 2 : Deuxième partie. Significations morphologiques. Montréal: Presses de l’Université
de Montréal.
Meyer-lübke, Wilhelm. 1890. Italienische Grammatik. Leipig: Reisland
Mugane, John M. 1997. A Paradigmatic Grammar of Gîkûyû. Stanford CA: CSLI.
Narrog, Heiko. 2010. A diachronic dimension in maps of case functions. In Semantic Maps: Meth-
ods and Applications (=lingusitics Discovery 8.1), Michael Cysouw, Martin H ­ aspelmath &
­Andrej Malchukov (eds), 233–254.
Olsen, Derek & Doris L. Payne. 2009. Maa (Maasai) nominalization: animacy, agentivity and
instrument. In Selected Proceedings of the 38th Annual Conference on African linguistics: lin-
gusitics theory and African Language Documentation, Masangu Matondo, Eric Potsdam &
Fiona Mc laughlin (eds), 151–165. Sommerville, MA: Cascadilla Proceedings Project.
 Semantic roles and word formation 

Palancar, Enrique. 2002. The Origin of Agent Markers. Leipzig: Akademie Verlag.
Panther, Klaus-Uwe & Thornburg, Linda L. 2001. A conceptual analysis of English er n ­ ominals.
In Applied Cognitive Linguistics, Vol. 2: Language Pedagogy, Martin Pütz, Susanne
­Niemeier & René Dirven (eds), 149–200. Berlin: Mouton de Gruyter.
Plénat, Marc. 2005. Brèves remarques sur les déverbaux en -ette. In La syntaxe au coeur de la
grammaire, Recueil offert en hommage au 60ème anniversaire de Claude Muller, Henning
Nølke & Frederic Lambert (eds), 245–258. Rennes: Presses Universitaires de Rennes.
Pustejovsky, James. 1995. The Generative Lexikon. Cambridge, MA: The MIT Press.
Rainer, Franz. 2005. Semantic change in word formation. Linguistics 43(2): 415–441. DOI: 10.1515
/ling.2005.43.2.415
Rappaport, Malka & Levin, Beth. 1992. -er Nominals: Implications for a theory of argument
structure. In Syntax and Semantics 26: Syntax and the Lexicon, Tim Stowell & Eric Wehrli
(eds), 127–153. New York NY: Academic Press.
Ryder, Mary Ellen. 1991. Mixers, mufflers, and mousers: the extending of the -er suffix as a
case of prototype reanalysis. In Proceedings of the Seventh Annual Meeting of the Berkeley
Lingusitics Society, February 15–18, 1991: General session and parasession on the grammar
of Event Structure, Laurel A. Sutton, Christophe Johnson & Ruth Shields (eds), 299–311.
Berkeley: Berkeley Lingusitic Society.
Schlessinger, Izchak. 1989. Instruments as agents: on the nature of semantic relations. Journal of
Linguistics 25: 189–210. DOI: 10.1017/S0022226700012147
Van Valin Jr., Robert D. & LaPolla, Randy J. 1997. Syntax, Structure, Meaning, and Function.
Cambridge: CUP.
Villoing, Florence & Namer, Fiammetta. 2008. Interpréter les noms déverbaux: Quelle relation
avec la structure argumentale du verbe de base? Le cas des noms en -OIR du français. In
Congrès Mondial de Linguistique Française – CMLF’08, Jacques Durand, Bruno Habert &
Bernard Laks (eds), 1551–1569. Paris: Institut de Linguistique Française. 〈http://www.lin-
guistiquefrancaise.org〉 or 〈http://dx.doi.org/10.1051/cmlf08226〉 DOI: 10.1051/cmlf08226
Wackernagel, Jacob. 1922. Vorlesungen über Syntax: mit besonderer Berücksichtigung von
Griechisch, Lateinisch und Deutsch. Basel: Emil Birkhäuser.
 From semantic roles to evaluative markers:
The dative and affected possessors
A constructional comparison between German,
Italian and French

Domenico Niclot
University of Düsseldorf

This paper targets the phenomenon of non valence-governed datives of

affectedness linked to possession. By adopting a constructional approach based
on Goldberg (1995, 2002) and a revision of Raineri and Evola (2008), I work out
a grid that may be adopted for cross-linguistic purposes. By applying this model
to German, French and Italian, I show that the latter language, contrary to the
two former ones has thoroughly grammaticalized this function. I also propose
a tentative account of the grammaticalization paths that may have led to the
heterogeneous behavior of the languages under discussion with respect to this
construction.

Keywords:╇ Construction Grammar; Dative; Beneficiary; Maleficiary;

Affectedness; Grammaticalization

1. Introduction

If any English native speakers heard someone from Germany or Italy say in some-
what odd English my car has broken me, they would probably attempt to make
sense of this utterance by thinking of an imaginary world in which objects take
revenge over humans. Perhaps they would also feel some sympathy towards the
utterer, and indeed a certain degree of sympathy would not be entirely out of
order. In fact, the German and Italian dative, which encodes me in the instance
above, indicates that the referent is affected by the fact that their car has broken
down on them.
This paper targets the phenomenon of datives of affectedness by developing a
constructional model suitable for comparing their cross-linguistic behavior. More
specifically, the focus is set on Italian and German and, to a lesser extent, French.
Most examples are cited from internet websites accessed through the Google
 Domenico Niclot

search engine and available as of 31. 12. 2011.1 This method of data collection has
been adopted on the premise that spontaneous usage events appear particularly
adequate to shed light on the synchronic linguistic inventory. The structure is as
follows.
In Section 2, I discuss Goldberg’s (1995, 2002) analysis of the English Ditran-
sitive Construction within her framework of Construction Grammar. By integrat-
ing it with Kittilä’s (2005) typological observations on semantic roles, I work out
the three core roles encoded by the dative in the languages under discussion, i.e.
the ones of Recipient, Beneficiary and Maleficiary. More specifically, I propose
a distinction between Recipients and Beneficiaries/Maleficiaries. The former are
animate beings intended to receive the transfer of an entity, whereas the latter
are animate beings meant to receive the effects of an event. I also show that the
components of benefaction and adversativity canonically ascribed to Beneficia-
ries and Maleficiaries are not inherent to the dative, but stem from encyclopaedic
knowledge. I posit that the dative encoding of affectedness pertains to a distinct
construction available in all three languages. I regard this construction, which is
determined by the criterion of possession, as the last phase of a path that includes
the previous grammaticalization of Recipients and Beneficiaries/Maleficiaries.
The function of the dative for signaling affectedness is demonstrated by its alterna-
tion with mere possessive marking. I argue that event transfers relating to datives
of affectedness are not necessarily bound to the feature of intentionality, contrary
to event transfers underlying Beneficiaries and Maleficiaries.
In Section 3, I adopt in a slightly modified version of Dabrowska’s (1997) sche-
matic notion of possession as related to the Personal Sphere (PS) of individuals.
I illustrate how the application of this concept to some of the criteria developed
by Raineri and Evola (2008) for analyzing datives of affectedness provides a viable
cross-linguistic grid based on four micro-constructions (i–iv below). Moreover,
I outline that datives of affectedness are evaluative markers that may convey either
the perspective of their referent or the estimation of thirds.
In Section 4, I examine the behavior of the three languages with respect to
micro-construction (i), which encompasses events impacting on entities that
belong to our PS. I highlight three phenomena which can be accounted for in
terms of grammaticalization effects. Firstly, the use of the dative is altogether most
frequent in Italian, followed by French and German. Secondly, German blocks the
dative in conjunction with individuals as Patients. Thirdly, contrary to German, in
Italian and French the dative has occasionally acquired the occasional function of

. In my searches, I excluded all blogs which were clearly recognizable as ill-formed contri-
butions made by non native speakers or as computerized translations of foreign sites.
 From semantic roles to evaluative markers: The dative and affected possessors 

denoting mere possession without implying affectedness proper. Grammaticaliza-

tion also plays a determinant role in Section 5 and 6.
In Section 5, I analyze micro-construction (ii) which relates to changes of
state (henceforth COSs) undergone by entities of someone’s PS due either to
internal or external causation. I work out a critique of the widespread theoretical
claim held for German, most particularly by Hens (1997), according to which the
dative does not profile affectedness, but the component of responsibility attrib-
uted to its referent for not preventing the depicted COS. I also provide arguments
for reversing this tenet, i.e. for regarding the use of the dative as a pragmatic
device that backgrounds responsibility by foregrounding affectedness. I show
that in this micro-construction the dative displays full-fledged status in Italian,
contrary to its lower degree of occurrence in French and to its restriction to the
denotation of adversative affectedness in German. With respect to the latter lan-
guage, I propose two hypotheses that may account for the irreversible grammati-
calization of this sense.
In Section 6, I examine micro-construction (iii) and make some brief con-
siderations about (iv). The former includes events in which entities of our PS
are construed as deliberately acting. Also in this case, I show that the Italian
dative is thoroughly entrenched, whereas in German and French it exhibits the
lower productive status of a still ongoing grammaticalization process. I suggest
that the current development can be plausibly linked to the influence of cog-
nate constructions separately available in these two languages. As for micro-�
construction (iv), I demonstrate, contra Raineri and Evola (2008), that pure
states in which entities of someone’s PS find themselves are incompatible with
the dative in the languages examined here since states infringe the event-trans-
fer requirement.
In Section 7, I summarize the outcome of my analysis and highlight some
issues awaiting further research.

2. A
 constructional delimitation of Recipients, Beneficiaries
and Maleficiaries

The import of a constructional approach to our subject matter can be illustrated by

reviewing the major principles of Goldberg’s Construction Grammar (1995). The
main tenet underlying this model is that constructions are pairings of form and
meaning which cannot thoroughly be compositionally predicted. Let us now see
how this is reflected in Goldberg’s analysis of the English Ditransitive Construc-
tion, which lends itself to our purpose since it displays a high degree of similarity
to the use of the dative for encoding semantic roles in Romance and German.
 Domenico Niclot

This construction exhibits the syntactic structure [Subj [V Obj1 Obj2]] and the
prototypical meaning of ‘X causes Y to receive Z’, whereby both ‘X’ and ‘Y’ refer to
volitional animate entities, as in:
(1) He gave Mary a book.

A further instance cited by Goldberg (1995:â•›141) reads:

(2) Sally baked her sister a cake.

Obj1 of both give and bake is considered to exhibit the argument role of a Recipi-
ent.. However, Goldberg observes that with respect to give the participant role of a
Givee which is already provided by the verb fuses with the argument role provided
by the constructional sense. By contrast, with regard to bake it is the construction
itself that contributes the argument role of a participant role, i.e. a Bakee, which is
not inherent to the verb proper. This is to say that, contrary to give, bake does not
imply an individual to whom an entity is transferred. Thus, we should note that
the same construction may either replicate the argument structure pertinent to a
specific verb or augment it and that the properties of non-compositionality and of
non-predictability posited by Goldberg only apply in the latter case.
A further distinction outlined by Goldberg regards the constructional
sense of bake, which she groups under the class of ‘verbs of creation’ along
with make, build, sew, knit. The meaning attributed to this class is not the one
of ‘successful transfer’ underlying give, but one of ‘intended transfer’, i.e. the
entity may not necessarily reach the Recipient. This seems plausible since we
can, for example, bake someone a cake and lose it on our way to them. Goldberg
views this constructional sense as polysemically related to the prototypical one
underlying give.
A problematic aspect underlying Goldberg’s analysis (1995:â•›38) relates to the
attribution of the prototypical constructional meaning of give to verbs subsumed
under the class of ‘ballistic motion’, which include throw and kick among others.
This stance is hardly tenable since throwing or kicking someone a ball does not
imply that the Recipient will be able to catch it. Thus, if we take the dichotomy
‘successful’ vs ‘intended transfer’ as a criterion for establishing constructionally
interrelated subsenses, Goldberg’s classification based on verb classes requires
restructuring. Moreover, Goldberg ascribes to all the referents of Obj1 the con-
structional role of ‘Recipient’. This classification is somewhat too coarse. In fact, if
we throw someone a ball, we carry out a single event, whereas if we bake someone
a cake, we perform two subevents, i.e. the baking event and the attempted or suc-
cessful event by which we transfer the cake to someone. Thus, transfers linked to
single events imply entities, while transfers linked to two distinct subevents include
both the transfer of the first subevent and of the entity effected or affected through
 From semantic roles to evaluative markers: The dative and affected possessors 

it. I will call individuals targeted by these two kinds of transfer Â�‘Recipients’ and
‘Beneficiaries’ respectively. This differentiation partially overlaps with to the crite-
rion that Kittilä (2005:â•›276) identifies as the distinctive feature of Beneficiaries: the
availability of two subevents the first of which needs to be completed before the
second can occur. Further corroboration for considering Beneficiaries as related to
event transfers is provided by instances such as:
(3) I lit him a candle. (McIntyre 2011:â•›9)

In (3), the referent of Obj1 does not receive an entity proper, i.e. a candle, but the
beneficial effect of the lighting subevent. Moreover, the candle is not created, but
acted on and, as pointed out by McIntyre, it need not even be transferred since it
may already be located in the Beneficiary’s possession.
Although Recipients and Beneficiaries appear to substantially differ as for the
nature of transfers, they also share some properties. All Beneficiaries are construc-
tionally contributed, i.e. their omission leads to a different propositional content.
This also holds for some Recipients: throwing a ball in the air for the mere sake of
playing cannot semantically be equated with throwing someone a ball. Moreover,
with all Beneficiaries and with some Recipients, as discussed above, transfer is
always intended, but not necessarily achieved. The reason why Goldberg does not
operate with the distinction Recipients vs Beneficiaries is to be sought in her claim
that the same syntactic phenomenon, in this case the frame of the Ditransitive
Construction, should encompass the possibly widest range of semantic roles by
maintaining meaning at a highly schematic level. However, this top-down orienta-
tion, which is based on surface generalizations, cannot dispense with specifica-
tions. In fact, Goldberg identifies some classes of verbs the semantics of which
partially differs from the prototypical constructional meaning and she operates
with the criterion of polysemy.
To do justice to Goldberg, she does posit a Benefactive Adjunct Construction
prepositionally introduced by for, as in:
(4) Mina bought a book for Mel.(Goldberg 2002:â•›332)

She remarks that (4) may also appear in the syntactic frame of the Ditransitive
Construction. However, she motivates the necessity to regard these two con-
structions as heterogeneous along two arguments. Firstly, she argues that poly-
semically related instances subsumed under the same construction display more
commonalities than single instances that can participate in either construction.
Adhering to this stance is somewhat problematic if we consider that Goldberg
(2002:â•›334) also subsumes under the Benefactive Adjunct Construction tokens
such as Mina sent a book for her mother’s sake, in which the preposition appears
to denote a circumstantial eventuality and not a participant proper. Secondly,
 Domenico Niclot

she observes that (4) is ambiguous. Besides denoting the individual who was
intended to receive the book, Mel may also refer to the person on whose behalf
the book was bought, so that she did not have to carry out the event herself. This
interpretation corresponds to the subtype of benefaction that Kittilä (2005:â•›273)
labels ‘substitutive benefaction’. Thus, it appears that the Beneficial Adjunct Con-
struction displays a vaster semantic range since it covers both benefactive event
transfers as defined above and substitutive benefaction, contrary to the Ditransi-
tive Construction, which only encompasses the former type. This is consonant
with the ‘Principle of No Synonymy’ posited by Goldberg (1995:â•›67), according to
which “if two constructions are syntactically distinct, they must be semantically
or pragmatically distinct”.
Opting for a micro-constructional distinction of the role of Recipient from
the one of Beneficiary as it may occur in different constructions runs counter to
Goldberg’s principle of generalization in its strong version, but appears profitable
for three reasons. Firstly, it provides a differentiation between these two roles and
other peripheral instances of the Ditransitive Construction and of the Benefac-
tive Adjunct Construction. Just to mention an example, Obj1 of cost, which par-
ticipates in the Ditransitive Construction, does not indicate a Recipient since this
mono-eventive verb lexically lacks the feature of a volitional transfer of an entity,
i.e. the price. Secondly, the spectrum of the English Ditransitive Construction does
not lend itself to cross-linguistic comparative purposes unless we break down its
schematic sense. Finally, evidence for the availability of a distinct representation
of the two roles of Recipient and Beneficiary in the conceptual system of speakers
comes from their partially separate cross-linguistic encoding outlined in Kittilä’s
(2005) typological investigation.
We are now in a condition to formulate a micro-constructional sense of
Recipients and Beneficiaries which can be plausibly adopted cross-linguistically.
Recipients correspond to ‘Y’ in: ‘X intends to transfer/transfers entity Z to Y’. By
contrast, Beneficiaries correspond to ‘Y’ in: ‘X creates/acts on entity Z and subse-
quently transfers/intends to transfer Z or the effect of the act to Y’.
Although Goldberg does not explicitly target Maleficiaries, this role can be
regarded as overlapping with the constructional meaning assigned above to Ben-
eficiaries, the only difference lying in the semantically adversative character of the
depicted event transfer. An instance of Maleficiary is constituted by:
(5) He set her a trap.

Isomorphically to what we have remarked for Beneficiaries, (5) includes a first

subevent, the preparation of a trap, and the intention to damage someone by
its effect. A thoroughly detrimental propositional content of the denoted event
appears to be the essential condition for the instantiation of Maleficiaries. Some
 From semantic roles to evaluative markers: The dative and affected possessors 

hints regarding this issue are provided by Kay (2005:â•›6), who points out the ill-
formedness of:
(6) *Claudine is mixing the neighbor a potion to murder him.

Kay remarks that (6) would only be acceptable in case the event transfer were
�carried out in order to cure the neighbor. Expanding on this observation, we can
claim that if an event bears semantically unspecified or positive associations, it
cannot license a Maleficiary. Interestingly, though, events with detrimental seman-
tics are compatible with Beneficiaries:
(7) I want you to kill me a rabbit.2

In the next subsection, I discuss how the constructional definition proposed for
Recipient, Beneficiary and Maleficiary relates to the use of the dative in German,
Italian and French.

2.1 D
 ative Beneficiaries and Maleficiaries vs dative of affectedness
in German, French and Italian
In these three languages, the dative encodes Recipients (G8-I10), Beneficiaries
(G11-I13) and Maleficiaries (G14-I16):3
(G8) Er gab der Maria ein Buch.4
(F9) Il donna un livre à Maria.
(I10) Diede un libro a Maria.
‘He gave a book to Mary.’

(G11) Sue backte ihrer Schwester einen Kuchen.

(F12) Sue fit un gâteau à sa soeur.
(I13) Sue fece un dolce a sua sorella.
‘Sue baked a cake to her sister.’

. 〈http://www.wilderness-survival.net/forums/archive/index.php/t-8105.html?s=568cf616
10d7460ecdb7121914dee7c0〉
. Shortenings used for the examples quoted in this paper: G (German), F (French), I
(Italian).
. Due to the specific focus of this paper on the dative and to economical reasons, no
interlinear gloss is provided. Instead, the cited examples are directly translated into English
and dative marking is indicated in bold. The gloss for dative is introduced by to since this
preposition also encodes Recipients. In explicit adversative contexts, to is replaced by on,
�according to the occasional use of the preposition for denoting detrimental events in
English. Besides, incorrect spelling in terms of omitted capitals is left unchanged in the
quoted examples.
 Domenico Niclot

(G14) Er stellte ihr eine Falle.

(F15) Il lui tendit un piège.
(I16) Lui le tese una trappola.
‘He set a trap on her.’

Furthermore, in all three languages ‘substitutive benefaction’ cannot be expressed

via the dative and requires prepositional encoding through the counterparts of for,
as in English. English, though, crucially differs from German, Italian and French
in that the Ditransitive Construction does not tolerate any event transfers that
impact on the Beneficiary’s possession:

(17) *He repaired/destroyed me my car.

To be more precise, event transfers relating to possessa are not entirely excluded
in English, as already mentioned with regard to (3). Some analogous instances are
reported by Colleman (2010), who also remarks that the acceptability of this con-
struction is subject to a high degree of language-internal variation. By contrast, the
co-occurrence of the dative with counterparts of (17) is thoroughly well-formed in
German (G18), French (F19) and Italian (I20):5

(G18) Er hat mir den/meinen Wagen repariert/zerstört.

(F19) Il m’a réparé/détruit la/ma voiture.
(I20) Mi ha riparato/distrutto la/la mia macchina.
‘He repaired/destroyed the/my car to me/on me.’

If we observe the examples above, we notice that the dative shows up either with
an additional possessive marker (double marking) or with a sheer definite article
preceding the possessum NP. In the latter case, which is also known as ‘possessor
omission’ (Tuggy 1980), the dative suffices to warrant the recoverability of a pos-
sessive relation. The selection of this marking device in order to avoid the redun-
dancy effects of double marking appears to be a language-specific phenomenon.
This was confirmed by the hits yielded by a Google search that I conducted with
regard to (G18-I20) and which is represented in Table 1.

. For practical reasons, the examples reported throughout this paper are restricted
to dative pronouns. However, the dative also co-occurs with individuals referred to by
proper and common nouns. In this case, German encodes it with the corresponding dative
article preceding the noun, whereas Italian and French do so via the prepositions a and à
�respectively.
 From semantic roles to evaluative markers: The dative and affected possessors 

Table 1.╇ A comparative Google search for the frequency of ‘possessor Â�omission’
in German, French and Italian678
‘Ø Subject repaired (sg+pl) to me the car’6 Ratio Double marking Ratio

German7 ╇21 42% 29 58%

French ╇50 50% 49 50%
Italian 125 96% ╇4 ╇4%
‘Ø Subject destroyed (sg+pl) to me the car’8 Ratio Double marking
German ╇╇3 13% 20 87%
French ╇19 46% 22 54%
Italian 253 94% 15 ╇6%

Altogether, these results reveal that possessor omission is considerably more

widespread in Italian than in German and French. It is also essential to note that
in all three languages the dative need not be instantiated at all and that (G18-I20)
may also solely occur with a plain possessive marker, i.e. the dative alternates with
possessive encoding. This phenomenon has been widely treated within hetero-
geneous frameworks under the dichotomy ‘external possession’ (dative mark-
ing) vs ‘internal possession’ mere (possessive marking).9 An exhaustive sample
of different approaches addressing the alternation from a syntactic, semantic and
syntactico-semantic perspective is provided in Payne and Barshi (1999). The func-
tion consensually attributed to the dative alternant is its emphasis on affected-
ness, which is not yielded by the possessive one. This is consonant with Goldberg’s
‘Principle of No Synonymy’.
A question that arises at this stage is whether the component of affectedness
also underlies the roles of Recipient, Beneficiary and Maleficiary. Kittilä (2005:â•›274)

. The search inputs were the flected forms of the VPs: das Auto reparieren (G), réparer la
voiture (F), riparare la macchina (I).
. These hits, as all the others which will follow in this paper, also include searches for verb-
last position in subordinate clauses for German.
. The search inputs were the flected forms of the VPs: das Auto zerstören (G), détruire la
voiture (F), distruggere la macchina (I).
. These terms are of syntactic nature. ‘Internal possession’ relates to the fact that posses-
sive markers form a single constituent with the possessum, which does not hold for ‘external
Â�possession’ where the dative referring to the possessor is separated from the possessum.
 Domenico Niclot

argues that reception is not sensitive to benefaction or malefaction by citing the

following example:
(21) The dentist sent me a packet with anthrax.

This argument is thoroughly pertinent if we consider that encyclopaedic knowl-

edge should not be confused with encoding devices. Indeed, sending a packet
with toxic substances is likely to affect the Recipient, but this inference stems from
world knowledge and is not yielded by the dative. However, Kittilä (2005:â•›294)
surprisingly suggests that Beneficiaries are affected by stating that they profit from
the denoted event. This is to be discarded on the same grounds that require teas-
ing apart world knowledge from encoding devices with respect to Recipients.
Strong evidence against considering Beneficiaries and Maleficiaries as necessarily
affected is provided by the fact that the evaluation expressed by the dative referent
may defease the effect of the beneficial (22) or adversative intention (23) inherent
to the depicted transfer:
(22) She baked me a cake, but I hate her cakes.
(23) He set me a trap, but his tricks are just ridiculous.

This is also in tune with the constructional meaning of Beneficiaries and Malefi-
ciaries as individuals intended to profit from an event transfer or to be damaged
by it.
The attribution of affectedness to the dative, regardless of the role that it
encodes, is widespread in Cognitive Linguistics. This goes back to Langacker’s
(1991) tenet according to which Recipients represent the role archetype of Expe-
riencers, i.e. of participants less affected by events than Patients, but still unable
to initiate them, contrary to Agents. Langacker ascribes to Experiencers a cer-
tain degree of affectedness which stems from the faculty to apprehend the occur-
rence of events directed towards them. This perspective has been embraced and
developed in numerous analyses focusing on the dative from a cognitive point
of view, most notably in Maldonado (2002) and Dabrowska (1997). However, we
should distinguish an ontogenetic dimension from a phylogenetic one. The former
motivates the origin of the dative as a marker for differentiating Experiencers as
sentient beings from Goals constituted by inanimate entities. The latter relates to
the subsequent developments of its function for specifically encoding affected-
ness, which is demonstrated by heterogeneous cross-linguistic marking of event
transfers involving possessa (e.g. English vs the three languages examined here).
This legitimates positing a distinction between the roles of Recipient, Beneficiary/
Maleficiary and a notion that I will label ‘Affected Person’ for purposes of brevity
bearing in mind that we occasionally also encounter the dative with animals, as in
(F30) to be discussed later.
 From semantic roles to evaluative markers: The dative and affected possessors 

The availability of a dative of affectedness construction leads us to inves-

tigate its diachronic relation to the use of the dative for encoding Recipients,
Beneficiaries and Maleficiaries, i.e. calls into question the phenomenon of
grammaticalization.

2.2 Datives of affectedness as the last stage of grammaticalization paths

A detailed comparison of heterogeneous frameworks dedicated to grammati-
calization would exceed the spectrum of this paper. However, for our purposes
we may summarily refer to two major streams, i.e. the Metaphorical Extension
Approach proposed by Heine et al. (1991) and the Invited Inferencing Theory
developed by Traugott and Dasher (2002).
The former essentially posits that the metaphorically driven extension of
grammatical meaning proceeds along the mapping of a concrete source domain
onto a more abstract target domain, such as spatial markers assuming the func-
tion of encoding possession along the metaphor POSSESSION IS LOCATED IN
SPACE, i.e. by the possessor (Heine 1997).
The latter, by contrast, regards contextually invited inferences as the funda-
mental factor that motivates the functional extension of grammatical markers.
More specifically, invited inferences may be based on metonymization or on gen-
eralization. Metonymization refers to the process by which a semantic component
contextually recurring with a certain marker which does not encode it per se trig-
gers the gradual extension of such marker to separately encode this component as
well. A canonical example is represented by the function acquired by conjunctions
originally denoting simultaneity, such as while, to subsequently indicate contras-
tivity due their occurrence with eventualities bearing contrastive semantics (see
Hopper & Traugott 2003; Chapter 4). The phenomenon of generalization exten-
sively discussed in Bybee et al. (1994), instead, refers to the process by which a
component which is originally inherent to a certain marker along with other
components acquires an independent status. For example, going to, which initially
denoted both a spatial and a temporal trajectory was extended to future events
devoid of locative semantics.
Despite their heterogeneous premises, metaphoric and inferential models are
not mutually exclusive. In fact, some metaphorical pattern, such as for example
the concrete nature of sources vs the abstract one of targets is likely to be retriev-
able in most cases of grammaticalization. However, metaphors should be viewed
as an epistemic criterion suitable for summarizing the outcome of a certain path
rather than the reasoning mechanism which has lead to its gradual development.
In a nutshell, speakers are unlikely to extend meaning by resorting to metaphors,
but rather by drawing inferences on a small scale and along long time spans.
 Domenico Niclot

Regardless of the approach to which one may subscribe, we should also bear in
mind that we are ultimately dependent on highly specific data to corroborate
hypotheses about a certain grammaticalization path and that such diachronic
information is frequently difficult to come by.
If we apply an inference-based approach to the entire dative grammaticaliza-
tion process ranging from Recipients to Affected Persons, we may argue as follows.
Recipients share positive semantic associations with Beneficiaries and negative
ones with Maleficiaries. Therefore, metonymization as the incorporation of a
novel semantic component into the functions of a marker can be plausibly ruled
out. By contrast, generalization nicely accounts for the attenuation of the transfer
of concrete entities and the gradual reinforcement of transferring the effects of
events. The development of the notion of ‘Affected Person’ out of Beneficiaries/
Maleficiaries, instead, can be attributed to metonymization. In fact, with respect
to ‘Affected Person’ the dative conveys an evaluative component inferred from the
semantics of the proposition that underlies event transfers directed towards an
animate Goal. A plausible scenario that allows us to account for this change can
be provided by focusing on the notion of intention, which is pivotal for transfers
towards Beneficiaries and Maleficiaries. In the initial of this grammaticalization
path, the dative is likely to still have encoded this component, as can be illustrated
through a shortened and slightly modified version of (I20) in which the writer is
aware of the detrimental aim pursued by the Agent:
(I24) Il mio rivale mi ha distrutto la mia macchina.
‘My rival destroyed my car on me.’

I will label this pattern ‘targeted event transfer’. In a subsequent phase, the dative
may have been extended to cases in which the event transfer was not deliberately
geared to targeting the specific dative referent (as in: someone destroyed my car)
and was thus reanalyzed as a marker of affectedness via its co-occurrence with
possession. This can be motivated by the fact that events that touch on our belong-
ings are most unlikely to leave us indifferent. However, even from a synchronic
perspective, we should be wary not to consider these two steps as mutually exclu-
sive. This is to say that the feature of targeted event transfer underlying the origi-
nal use of the dative, rather than being thoroughly bleached or desemanticized,10
may still be available in specific contexts. For example, if someone we know had

. For a discussion of bleaching as a phenomenon that does not necessarily imply the
complete deletion of the previous semantic features of the units involved in grammaticaliza-
tion processes, but rather a redistribution and a partial shift of such features see Hopper &
�Traugott 1993 (87 ff.).
 From semantic roles to evaluative markers: The dative and affected possessors 

repaired our car in order to do us a favor, contrary to someone unknown having

done so out of mere professional duties, the use of the dative would both encode
benefaction and affectedness, thus displaying a certain degree of vagueness.
Before concluding this section, it appears helpful providing the reader with
two charts which summarize the notions worked out so far. Table 2 illustrates the
main distinctive features underlying the roles of Recipient, Beneficiary/�Maleficiary
and the notion of ‘Affected Person’.

Table 2.╇ Features of Recipient, Beneficiary, Maleficiary and Affected Person

Role Constructional sense Semantics Function of the dative

Recipient Individual intended to receive transfer beneficial/ Encoding of transfer

of entity detrimental
Beneficiary Individual intended to receive transfer beneficial Encoding of transfer
of event related to entity
Maleficiary Individual intended to receive transfer detrimental Encoding of transfer
of event related to entity
Affected Individual affected by intentional or beneficial/ Encoding of affectedness
Person unintentional events touching on detrimental
his/her possessa

Figure 1 represents the whole grammaticalization path hypothesized here.

Bold marking highlights the respectively profiled component.

INTENTIONAL INTENTIONAL (UN)INTENTIONAL

OBJECT -TRANSFER GENERALIZATION EVENT-TRANSFER METONYMIZATION EVENT-TRANSFER

BENEFICIAL/ BENEFICIAL/ EVALUATION OF

ADVERSATIVE ADVERSATIVE BENEFICIAL/
SEMANTICS SEMANTICS ADVERSATIVE
SEMANTICS
RECIPIENTS BENEFICIARIES/MALEFICIARIES AFFECTED PERSONS

Figure 1.╇ An inference-based grammaticalization path from Recipients to Affected Persons

From a metaphorical angle, we may sum up this grammaticalization path by

positing that Maleficiaries/Beneficiaries originate from Recipients via the metaphor
TRANSFERRED EVENTS ARE TRANSFERRED ENTITIES. The relation between
‘Affected Persons’ and Maleficiaries/Beneficiaries, instead, can be traced back to a
metonymic process by which an effect has been mapped from its cause: AFFECT-
EDNESS INHERENT TO POSSESSION stands for INTENDED �BENEFICIAL/
DETRIMENTAL EVENT TRANSFERS.
 Domenico Niclot

In the next section, I propose some criteria that allow us to identify differ-
ent types of dative micro-constructions of affectedness suitable for cross-linguistic
comparison.

3. A micro-constructional categorization of datives of affectedness

Raineri and Evola (2008) address the issue of non valence-governed dative mark-
ing for conveying affectedness. Their approach appears promising in a twofold
regard. Firstly, it constitutes, to my best knowledge, one of the few attempts to
consider the issue from a cross-linguistic point of view inasmuch as it compares
French with Italian. Secondly, its constructional orientation, reflected in the termi-
nology (Empathetic Dative Construction, henceforth EDC), enables us to some
develop epistemic criteria which exhibit a certain predictive import, contrary to
most descriptive investigations targeting this phenomenon, as Dabrowska’s (1997)
for Polish and Maldonado’s (2002) for Spanish, to quote but two. However, some
of the criteria worked out for the EDC require modification in order to provide an
applicable framework.
Raineri and Evola (2008:â•›4) attribute to the EDC the constructional meaning
of “metaphoric successful transfer of an event and its implications into the dative
referent’s sphere of affect”. They distinguish this sense from the one underlying
Beneficiaries which they regard as solely conveying the transfer of an object, in
conformity with Goldberg and contrary to what I have claimed in 2. Their grid
foots on a syntactico-semantic analysis which includes parameters such as tran-
sitivity and aspectuality. The major difference claimed to subsist between French
and Italian is fact that the former the dative may only co-occur with transitive
events, whereas in the latter it is also compatible with intransitive ones. Among
others, Raineri and Evola cite the following instances, which also have Italian
counterparts:11
(F25) Eloan a commencé à piocher dans celle de Max et il lui a tout mangé!!
‘Eloan started picking at Max’s [food] and ate everything on him.’
(Raineri & Evola 2008:â•›9)
(F26) Du jour au lendemain elle me buvait de 0–20ml à 250–300ml de thé de
�fenouil chaque jour.
‘From one day to the next she began drinking to me from 0–20ml to
250–300 ml of fennel tea a day.’ (Raineri & Evola 2008:â•›9)

. The original glossing format is adapted to the one used in this paper.
 From semantic roles to evaluative markers: The dative and affected possessors 

The co-occurrence of the dative with intransitive verbs in Italian is illustrated

through:
(I27) Il mio bambino non mi mangia.
‘My child isn’t eating to me.’(Raineri & Evola 2008:â•›10)

The major problem posed by this model is that the syntactic criterion of transitiv-
ity is not linked to any semantic contents, i.e. it does not provide any information
as for which events trigger affectedness. This point illustrates the shortcomings
of generalization effects à la Goldberg by which a syntactic phenomenon is sub-
sumed far too schematically under a semantic feature. A comparison of (F25) with
(F26) and (I27) provides a solution to semantic underspecification since it shows
that the meaning of the former example differs substantially from the sense of the
two latter ones. In fact, in (F25) the dative referent is negatively affected by the fact
that someone is eating his food. Affectedness stems from an event impinging on a
Patient, food, to which the dative referent is linked via a possessive relation. In the
two latter utterances, instead, the dative referents, i.e. two mothers, are affected
because it is the Agents, their children, to whom they are bound via a posses-
sive relation, that carry out the drinking and eating events. Thus, it appears that
(F26) and (I27) are semantically analogous despite their heterogeneous surface
structure with regard to transitivity. Raineri and Evola (2008:â•›6) neglect this aspect
since they view the criterion of possession as compatible with the EDC but not
essential to it. The overlooked pre-condition of a possessive relation for instanti-
ating datives of affectedness also underlies two further event types identified by
the writers as compatible with the dative in Italian, contrary to French. These two
types are constituted by verb classes denoting COSs (I28) and what they label
‘stative processes’ (I29):
Allora, mortomi il samsung, con i punti della vodafone mi sono portato a
(I28) 
casa un motorola.
‘So, ever since the Samsung died on me, I brought home a Motorola with
my Vodafone points.’ (Raineri & Evola 2008:â•›11)
(I29) Poi mi è stato buono buono in braccio per la restante mezz’oretta.
‘Then he [baby] was [remained] to me very good [quiet] in my arms for
the rest of the half hour.’ (Raineri & Evola 2008:â•›11)

In (I28), the dative refers to an adversative COS, i.e. becoming broken, undergone
by the writer’s mobile phone, while in (I29) it signals the writer’s positive evalua-
tion of her baby’s stillness.
Integrating the event types outlined so far provides us with the following
micro-constructional grid which can be plausibly adopted for the cross-linguistic
analysis of dative of affectedness:
 Domenico Niclot

(i) is acted on

(ii) changes state

Possessum

(iii) acts

(iv) finds itself in a certain state

Figure 2.╇ A cross-linguistic grid for the categorization of datives of affectedness

A final aspect which needs refinement is the lack of intentionality that Raineri
and Evola’s (2008:â•›17) attribute to the event transfers underlying the EDC. This
tenet only holds partially for (i), as discussed at the end of the previous section
with regard to targeted event transfers. As for (ii), intentionality is indeed to be
ruled out since entities do not deliberately undergo a change of state. This also
applies to (iii) and (iv). To be precise, in the latter two cases animate entities may
occasionally carry out some act or indulge in a certain state in order to please or
tease someone else. However, this is not the default reading of instances such as
(F26), (I27) and (I29). This also legitimates the hypothesis that (ii)–(iv) have either
developed to a later stage than (i) or independently from it.
Positing that datives of affectedness are motivated by a relation of possession
calls for a more fine-grained definition of this notion.

3.1 Affectedness and possession

Both animate entities (as in F26) and inanimate entities (as in I28) may feature as
possessa in dative- of-affectedness-constructions. A notion that may be adapted for
capturing the criterion of possession is the one of Personal Sphere (henceforth PS)
as proposed by Dabrowska (1997:â•›16). The PS includes “the persons, objects, loca-
tions, and facts sufficiently closely associated with an individual that any changes in
them are likely to affect the individual as well”. This schematic classification allows
us to account for concepts that do not denote possessa proper, as ecosystem in:
(F30) Je pense que les pigeons sont très sympa avec cette humanité qui leur a détruit
leur écosystème.(Raineri & Evola 2008:â•›9)
‘I think that the pigeons are really nice with those humans who have
Â�destroyed their ecosystem on them.’

The PS covers temporary possession as well, as it may underlie (I28) in case

the writer were referring to a borrowed phone or (I29) in case the writer were
Â�looking after a friend’s baby. However, Dabrowska’s definition above should be
 From semantic roles to evaluative markers: The dative and affected possessors 

slightly adjusted by including animate beings (see 30) and by taking into account
that changes can also be initiated and not only undergone by entities belonging
to someone’s PS. to ‘changes in them’. The latter modification is motivated by the
fact that in (iii) and (iv) possessa are construed as displaying agent-like and not
patient-like character, contrary to (i) and (ii).
The availability of (iii) and (iv) also partially invalidates the claim that the
dative referent’s affectedness is bound to the condition that possessa be affected.
This tenet, which is advanced by König and Haspelmath (1998), leads us to briefly
illustrate the drawbacks of a non constructional approach towards the phenom-
enon examined in this paper. König and Haspelmath (1998:â•›538–539) introduce
the constraint of affected possessa in order to account for the ill-formedness of
the French counterparts of my head is cracking and my brain is boiling, which,
according to the writers, can solely co-occur with possessives markers. However,
they remark that this restriction does not hold for isolated instances such as the
following, which are instantiated via dative:
(F31) Les oreilles lui tintent.(König & Haspelmath 1998:â•›539)
‘The [i.e. his] ears hum on him.’
Such tokens are posited to have been lexicalized previously to the grammaticaliza-
tion of the affected possessa restriction. This implies that micro-construction (iii)
(i.e. possessa construed as deliberately acting) diachronically precedes micro-con-
struction (i) , which runs counter to the development of (i) out of Beneficiaries/
Maleficiaries as source roles motivated above. In our framework, the availability
of (iii), which goes unnoticed in their analysis, prompts the assumption to be veri-
fied that in French this micro-construction preferably excludes possessa consti-
tuted by body parts. However, König and Haspelmath (1998:â•›573) state that the
latter, as inalienable possessa, rank cross-linguistically higher than alienable pos-
sessa in terms of their co-occurrence with the dative. We may assume that such
claim stems from their extrapolation of the frequent occurrence of body parts in
conjunction with micro-construction (i), which I discuss in 4. These observations
show that operating with semantically based hierarchies of possessa is not epis-
temically profitable unless we embed them in different constructions.
A further notion that needs being addressed with datives related to the PS is
evaluation.

3.2 The dative as an evaluative marker

As pointed out above, the selection of the dative alternant vs the mere possessive
one conveys the judgment of positive or detrimental affectedness triggered by a
certain event. The concept of affectedness encompasses a wide range of emotive
responses such as happiness, surprise, pride on the positive side and anger, disap-
pointment, sadness on the negative one. The function of the dative as an evalua-
 Domenico Niclot

tive marker has not gone unnoticed in the literature. In fact, Maldonado (2002),
�drawing on Langacker (2000), distinguishes between datives reflecting view-
ers, who observe a certain state of affairs off-stage, i.e. objectively, and datives
attributed to conceptualizers, who evaluate it from a subjective on-stage per-
spective. Maldonado’s categorization of datives does not correspond to the one
of this paper since he does not adopt a constructional framework. However, we
may roughly attribute to the dative an objective function with regard to Patients,
Beneficiaries and Maleficiaries inasmuch as its use does not mirror a subjective
choice, but is governed by a marking device which does not tolerate any alterna-
tions. By contrast, the omissible instantiation of the dative in (i)–(iv) is to be
ascribed to conceptualizers.
Arguing from a non cognitive stance, O’ Connor (2007) reaches similar con-
clusions by observing that datives of affectedness display the status of conven-
tional implicatures, i.e. that they contribute a subjective component which can be
detached from the utterance without altering its propositional truth conditions.
As argued in 2, this does not hold for Beneficiaries and Maleficiaries.
At this stage, some terminological precision is in order. So far, I have referred
to the rise of datives of affectedness as a phenomenon of grammaticalization. For
purposes of economy, I will continue to do so unless specification is essential.
However, it is worth signaling that this label is somewhat coarse. In fact, gram-
maticalization is canonically viewed as the development of lexical units into units
endowed with a grammatical function or of units serving core grammatical func-
tions into units that encode more abstract grammatical functions (see Heine et
al. 1991:â•›155 ff.). This principle holds for the distinction between the dative as
encoding both the roles of Recipient and the ones of Beneficiary/Maleficiary inas-
much as the latter two represent adjuncts, contrary to the former, which encodes a
core semantic role. By contrast, datives of affectedness reflect an evaluative stance
toward the narrated event and occasionally the purpose of rendering such dimen-
sion relevant in interaction with thirds. The development of these two functions
can be more appropriately captured with the notions of subjectification and inter-
subjectification respectively (Traugott 2003), which pertain to discourse and not
to grammar. A more recent approach in tune with this perspective is the frame-
work of Thetical Grammar developed by Kaltenböck et al. (2011), which regards
linguistic units that do not contribute to the propositional content of utterances as
relating to the speaker’s attitude or to speaker-hearer interaction.
When referring to evaluation it is also crucial to point out that the source
of evaluation need not overlap with the dative referent, but may also be pro-
jected onto events by other conceptualizers. Evidence for this claim is provided
by instances such as (F30), in which affectedness cannot possibly stem from a
judgment expressed by pigeons. Admittedly, identifying the source of evaluation
 From semantic roles to evaluative markers: The dative and affected possessors 

frequently proves arduous. However, clear indicators are questions or statements

by which the potentially affected person is addressed in the irrealis mode, which
excludes the occurrence of the depicted event. O’ Connor (2007:â•›593) adduces
such an example, pointing out that the omission of the dative in the Spanish
counterpart of if your cat dies, we’ll bury it, would signal rejection of empathy. Yet
empathy is not the only component inherent to evaluation, as the two following
examples show:
La mia amica, che si lamenta che il bimbo non le mangia niente però è così
(I32) 
tondo che quasi rotola.12
‘My friend complains that her son doesn’t eat anything to her, but he is so
round that he could almost roll.’
(I33) Delfine sono mesi che non gli torna a casa, e manco se ne preoccupa.13
‘Delfine has not come back home to him for months and he doesn’t
even care.’
In (I32), the writer reports her friend’s affectedness triggered by the poor appetite
of her son and ironically dissociates herself from the evaluation by the affected
person expressed through the dative. The opposite applies to (I33), in which the
writer uses the dative in order to signal that its referent, a husband, should feel
affected due to his wife’s long absence, but does not do so in the least. In this case,
the use of the dative does not signal empathy, but a reproach for the lack of it.
These two examples show how the dative in conjunction with further contextual
information may reflect a divergence between the perspective of the individual
directly involved in the described event and the one of the speaker/writer.
In the next section I analyze in more detail the behavior of the languages
under discussion with regard to micro-construction (i).

4. The dative with events impacting on the PS

Following the arguments provided so far, we would expect the alternation dative
vs mere possessive marking to reflect the contrast between affectedness and its
absence. Substantiating this claim requires detecting some further contextual
information in which mere possessive encoding correlates with the component of
indifference towards the depicted event. Such clues are indeed difficult to come by

. 〈http://forum.gravidanzaonline.it/forum/viewtopic.php?p=1278794&sid=3d1a1b33dcf2
2d9a1f82648514ce7eba〉
. 〈http://it-arti.confusenet.com/showthread.php?t=134817&p=1288031〉
 Domenico Niclot

and this is intrinsically motivated by the notion of possession. In fact, it is hard to

envisage under what circumstances any desirable or detrimental act performed on
our possessa could fail to trigger an emotive and evaluative reaction.
An alternative way to frame this matter consists in adopting a diachronic
perspective, as König and Haspelmath (1998) do, by positing that the synchronic
state of the dative of affectedness is motivated by the grammaticalization status
of this function according to an implicational hierarchy determined by the rel-
evance of the possessa for their possessors.14 More precisely, within this approach,
inalienable possessa, and particularly body parts, are regarded as cross-linguis-
tically occupying the highest rank on the scale of dative encoding as compared
with alienable possessa. We can verify this claim for the languages under discus-
sion on a small scale by exploring the co-occurrence of dative instantiations with
negatively connoted verbs that can be applicable both to inalienable and alienable
possessa. Let us consider the Google results that I obtained for the counterparts of
break referring to hand and car (Tables 3 and 4):15

Table 3.╇ A comparative Google search for the alternation dative vs mere
possessive �encoding with detrimental events relating to body parts as possessa
Ø Subject broke (sg+pl) Ratio Mere possessive marker Ratio
to me the/my hand’

G 53 78% 15 22%
F 61 96% ╇3 ╇4%
I 41 92% ╇4 ╇8%

Table 4.╇ A comparative Google search for the alternation dative vs mere
possessive �encoding with detrimental events relating to alienable possessa
‘Ø Subject broke (sg+pl) Ratio Mere possessive marker Ratio
to me my/the car’

G ╇44 13% 293 87%

F ╇93 41% 139 59%
I 131 78% ╇37 22%

. For the increasing frequency of tokens as a key element for the grammaticalization of a
certain type see also Bybee (2007).
. The search input for Table 3 consisted of the flected forms of the VPs: die Hand brechen
(G), casser la main (F), rompere la mano (I). The ones for Table 4: das Auto kaputt machen (G),
casser la voiture (F), rompere la macchina (I).
 From semantic roles to evaluative markers: The dative and affected possessors 

A first glance at the results illustrated in Table 3 reveals that in German, the use
of mere possessive markers is higher than in Italian and French, albeit dative encod-
ing prevails in all three languages. Since the proportion of possessive tokens in Ital-
ian and French approaches a null level, it seems plausible to conclude that in these
two languages the co-occurrence of the dative with body parts has acquired such
a degree of entrenchment due to frequency of usage that mere possessive marking
is scarcely productive any longer, whereas it still is in German. By contrast, Table 4
reveals that dative encoding prevails in Italian and that �German displays a com-
paratively low frequency of dative occurrences. The data in both tables also appears
to corroborate König and Haspelmath’s (1998) tenet regarding the crosslinguistic
priority accorded to the dative in conjunction with body parts. This tendency also
emerged in a Google search that I carried out to explore the use of the dative with
regard to beneficial events directed towards body parts and alienable possessa (get-
ting one’s back massaged and getting one’s bike mended; see Tables 5 and 6).16 Table 6
also shows a much lower occurrence of the dative with alienable possessa displayed
by French vs Italian as compared with the data in Table 4.

Table 5.╇ A comparative Google search for the alternation dative vs mere
possessive �encoding with beneficial events relating to body parts
Ø Subject massaged (sg+pl) Ratio Mere possessive marker Ratio
to me the/my back’
G 105 63% 63 37%

F ╇76 90% ╇9 10%

I ╇46 92% ╇4 ╇8%

Table 6.╇ A comparative Google search for the alternation dative vs mere
possessive encoding with beneficial events relating to alienable possessa
Ø Subject repaired (sg+pl) Ratio Mere possessive marker Ratio
to me the/my bike’
G 20 19% 83 81%
F 36 21% 78 69%
I 36 86% ╇6 14%

. The search input for Table 5 consisted of the flected forms of the VPs: den Rücken mas-
sieren (G), masser le dos (F), massaggiare la schiena (I). The ones for Table 6: das Fahrrad
reparieren (G), réparer le vélo (F), riparare la bicicletta (I).
 Domenico Niclot

A criterion that may be included to account for the comparatively scarce

occurrence of the dative in German in Tables 4 and 6 is the one of ‘targeted event
transfer’ that underlies Beneficiaries/Maleficiares, but not necessarily Affected
Persons, as discussed in Section 2.2. More precisely, we might hypothesize that
German has retained to a higher degree than Italian and French the constructional
semantics of Beneficiaries and Maleficiaries. As for the eventualities underlying
Tables 3 and 4, I was not able to detect any contextual clues pointing to the Agent’s
deliberation of specifically damaging the dative referent. By contrast, 17 out of 20
hits reported in Table 6 contained some explicit indications relating to targeted
event transfers, as in the following example where the reparation was carried out
by the writer’s neighbor as a favor:
(G34) Mein anderer Nachbar hat mir das Fahrrad repariert.17
‘Another neighbor of mine repaired the bike to me.’

However, the criterion of targeted event transfer does not seem determinant inas-
much as it also underlies mere possessive encoding, as in:
(G35) Ich bin heute einem Freund dankbar, dass er mein Fahrrad repariert hat.18
‘Today I am thankful to a friend for repairing my bike.’

It goes without saying that the observed rarer instantiation of the dative in German
cannot be generalized without further empirical research. A vast range of verbs
should be examined in their collocation with an equally wide array of body parts
and alienable possessa. Such sample should encompass both verbs with beneficial
semantics (such as clean, wash, install, paint, decorate) and such with adversative
meaning (such as make dirty, steal, destroy, block, eat someone’s food, drink some-
one’s drinks). Besides, pronominal dative referents other than first person singular
ones should be considered as well.
These tasks, which cannot be pursued here for reasons of space, are also rele-
vant to another category of possessa, which is not directly addressed in König and
Haspelmath (1998), i.e. inalienable abstracta. This class is compatible with the
dative in all three languages under discussion in conjunction with both beneficial
(G 36-I38) and detrimental events (G39-I41):
(G36) Danke für den Tipp, damit hast du mir meinen Urlaub gerettet.19
‘Thank you for your advice, You saved my holiday to me.’

. 〈http://archiv.rabeneltern.biz/print.php?threadid=14972&page=1&sid=0a5f0f267a2648
2ecc20e61e56c56279〉
. 〈http://lawendeltreppe.blog.de/2009/09/02/dankbar-6876819/〉
. 〈http://www.team-andro.com/phpBB3/mkp-im-flugzeug-t210646-15.html〉
 From semantic roles to evaluative markers: The dative and affected possessors 

(F37) Mon mari m’a sauvé mon esprit et ma vie par la même occasion.20
‘My husband saved my mood and my life to me on the same occasion.’
(I38) Quella legge mi ha salvato il posto di lavoro!21
‘That law saved my job to me.’
(G39) Du hast mir meinen Plan zerstört.22
‘You have destroyed my plan on me.’
(F40) Il m’a détruit mes plus beaux jours mes plus beaux moments.23
‘He destroyed my most beautiful days and my nicest moments on me.’
Il mio datore di lavoro mi ha distrutto i piani ordinandomi di andare al locale
(I41) 
stasera.24
‘My employer has destroyed my plans on me by telling me that I will have to
go to my working venture tonight.’

It is important to point out that these tokens do not represent any specific intra-
linguistic collocations. In fact, the possessa above, which pertain to domains of the
PS such as time spans, plans, professional life and emotional states, co-occur with
the dative in all three languages. The process whereby the dative has come to signal
affectedness with these class of possessa can be accounted for in terms of general-
ization along the metaphoric mapping of verbs referring to concrete entities onto
abstract targets. This is in tune with the directionality of grammaticalization paths
proceeding from concrete to abstract dimensions.
Yet this principle does not always hold in a uniform manner. For example,
German does not tolerate the use of the dative with verbs above and cognate
ones such as kill, ruin and save when they refer to events in which inalienable
�possessa constituted by animate beings (kinship) feature as Patients. This con-
straint is �particularly surprising if we consider that individuals are likely to be
strongly linked to close animate beings terms of empathy. Moreover, it also par-
tially runs counter to König and Haspelmath’s (1998) claims regarding the cri-
terion of inalienability. Italian and French do not exhibit this restriction, as the

. 〈https://www.google.fr/#q=%22m’a+sauv%C3%A9+mon+esprit%22〉
. 〈http://www.spritz.it/forum/discussione.asp?pagina=5&forum_id=5&topic_
id=5171&forum_title=RIFLESSIONI&topic_title=RAI…++MEDIASET〉
. 〈http://mobil.animexx.de/fanfiction/serie/1891/305049/txt/〉
. 〈http://forum.psychologies.com/psychologiescom/Vivre-une-separation-divorcer/
trompe-avec-femme-sujet_543_1.htm〉
. 〈http://loveboat.forumcommunity.net/?t=518249&st=1965〉
 Domenico Niclot

following examples, for which I did not encounter any German counterpart in my
Google search, illustrate:

(I42) Gli hanno ucciso il padre e poi l’hanno torturato.25

‘They killed the father on him and then tortured him.’

Il padre in lacrime ha abbracciato i due carabinieri che gli hanno salvato

(I43) 
il figlio.26
‘Crying, the father hugged the two policemen who saved the son to him.’

(F44) Il y a des bâtards qui lui ont tué son fils.27

‘Some bastards killed his son on him.’

(F45) Ce chauffeur lui a sauvé sa fille.28

‘That taxi driver saved his daughter to him.’

The constraint underlying German was also confirmed by all my native speaker
informants to whom I submitted the translation of I42-F45 and who considered
these tokens to be ill-formed or at least ‘odd’. Explaining this phenomenon via
the targeted event transfer hypothesis is not satisfactory for the same reasons
given above. Alternatively, we might postulate that the use of the dative is barred
in order to avoid ambiguous readings, as with kill, where the dative referents
might occasionally be interpreted as denoting Beneficiaries proper in case they
had recruited thirds for getting rid of their kins, as in (7). However, this hypoth-
esis seems hardly tenable since it can only account for a highly restricted range
of instances. Thus, the question as to whether this phenomenon may be moti-
vated at all or should rather be regarded as idiosyncratic is left open for future
research.
A final phenomenon which is worthwhile being mentioned relates to the use
of the dative for events in which the component of affectedness seems hardly avail-
able. Let us consider the following examples:

. 〈http://altoadige.gelocal.it/cronaca/2013/10/05/news/se-potessi-tornare-a-casa-lo-farei-
subito-1.7870553〉
. 〈http://iltirreno.gelocal.it/massa/cronaca/2012/07/14/news/tenta-il-suicidio-con-il-
gas-di-scarico-perche-lasciato-dalla-fidanzata-studente-di-26-anni-salvato-dai-carabin-
ieri-1.5405018〉
. 〈http://jeuxdecuisine3.com/gteau-la-crme-l-orange/〉
. 〈http://forumed.sante-dz.org/f30/lamour-est-ce-quil-existe-t21475/page21.html〉
 From semantic roles to evaluative markers: The dative and affected possessors 

Ho posto la domanda direttamente al poliziotto doganale che mi ha

(I46) 
�controllato il passaporto.29
‘I directly asked the customs officer who checked the passport to me.’

(F47) Le contrôleur m’a poinçonné le billet.30

‘The train inspector punched the ticket to me.’

In both cases, it is difficult to figure out how having their pass checked or their
ticket punched could possible affect the dative referents, be it positively or nega-
tively. Moreover, the semantics of the event is also devoid of any beneficial or det-
rimental intention on the part of the Agent.
Interestingly, this pattern can also be observed with respect to verbs that
relate to body parts and that that do not specifically bear a positive or adver-
sative connotation, Let us consider, for example, the counterparts of touch.
Admittedly, touching somebody’s body may lead to positive or negative
responses. However, we also encounter cases in which this component is not
active, as when touching someone’s arm is plainly meant as a signal to attract
their attention:

(I48) Per salutarmi mi ha toccato il braccio.31

‘To greet me, she touched the arm to me.’

(F49) Une fille m’a touché le bras pour me dire que j’ai fait tomber un papier.32
‘A girl touched the arm to me to show me that I had dropped a piece
of Â�paper.’

My Google search did not yield any hits at all for ‘X touched my arm’ in Â�German
(X hat mir meinen Arm berührt), which corroborates the hypothesis that verbs
with a rather neutral connotation are incompatible with the dative in this lan-
guage. For this limited class of verbs (which includes check, punch, control, look
at and cognates) that co-occur with the dative in Italian and French, but not in
�German according to my Google search and to the judgment of native speakers,
we might hypothesize that the grammaticalization path leading from Beneficia-
ries/Maleficiaries to ‘Affected Persons’ has developed to a still further stage, i.e.

. 〈http://autotutela.net/newsite/infonews/espatrio-attenzione/〉
. 〈http://www.wikebec.org/puncher/definition/〉
. 〈http://it.answers.yahoo.com/question/index?qid=20130725120005AARH5KR〉
. 〈http://www.jeuxvideo.com/forums/1-51-22020714-1-0-1-0-il-s-est-passe-beaucoup-
chose.htm〉
 Domenico Niclot

has undergone a semantic bleaching of the component of affectedness eventually

leading to the use of the dative for solely indicating possession.
To summarize this section, the data provided so far suggest that the use of
the dative in this construction is generally available in all three languages despite
heterogeneous levels of frequency of usage which require further detailed verifica-
tion. Put differently, we can hypothesize that in this micro-construction the choice
of either the dative or the possessive alternant is determined by the availability of
entrenched units stored in a specific language rather than by a strategy deliberately
adopted by speakers. This also implies that the process of entrenchment of further
units is not barred. Two exceptions to this conclusion are constituted by the dative
with body parts in Italian and French, which appears to form a rather closed pat-
tern compared with the low productivity of mere possessive marking, and by the
non acceptability of the dative in German with regard to events touching on ani-
mate beings.
The main phenomena outlined for this construction are recapitulated in the
chart below.

Table 7.╇ The dative with transitive events touching on the PS in German, French
and �Italian
Dative with Beneficial Adversative Restriction on Indication of
micro-construction (i) reading reading possessa mere possession

German ✓ ✓ ✓ -
French ✓ ✓ - ✓
Italian ✓ ✓ - ✓

German also behaves differently from Italian and French with respect to
micro-construction (ii), as I will show in the next section.

5. The dative with change-of-state verbs

5.1 Hens’s Dative-of-Inaction Construction and cognate approaches

According to a stance widespread in the German literature, the use of the dative in
conjunction with COSs signals its referent’s responsibility or neglect rather than
denoting affectedness. This claim is particularly put forward by Hens (1997), who
postulates the availability of a Dative-of-Inaction Construction (DI). The latter
exhibits the semantics of dative-encoded participants who have failed to prevent
 From semantic roles to evaluative markers: The dative and affected possessors 

an undesirable COS undergone by an entity belonging to their or to someone else’s

PS, in which case temporary possession is involved. Within our theoretical frame,
we could label the dative referent an ‘unintentional Self-Maleficiary’. The require-
ment for the dative referents to participate in the DI is that they be not the direct
causers of the denoted event. According to Hens (1997:â•›204), this is warranted
by the fact that the verbs participating in the DI belong to the semantic class of
Â�‘mutatives’, which relate to non-agentive transformations due to inherent proper-
ties of the involved entity, such as for example, ‘decay’ (G50-G51) and ‘fragility of
the material’ (G52):
(G50) Ihnen ist die Suppe übergegangen.
‘The soup turned sour on them.’(Hens 1997:â•›204)

(G51) Der Oma ist die Suppe verbrannt.

‘The soup burned on grandma.’(Hens 1997:â•›209)
(G52) Der Oma ist die Vase zerbrochen.
‘The vase broke on grandma.’(Hens 1997:â•›204)

Italian also displays dative counterparts for (G50-G52). Interestingly, the phenom-
enon that Hens subsumes under the DI has been highlighted in other approaches
investigating the dative of affectedness, albeit within a mainly syntactic framework,
particularly by Kallulli (2004) and Schäfer (2007). The latter differ from Hens’s
analysis in that they view the equivalents of break as denoting COSs potentially
triggered by thirds, among others by animate beings. Clarifying the heterogeneous
categorization of the same verb in different approaches firstly requires zeroing in
on the phenomenon of causation.

5.2 Internally vs externally caused COSs revised

COSs may be brought about (un)intentionally by human participants or mechani-
cally by natural forces. A notion that seems adequate to capture this spectrum is
the one of Effector, coined by Van Valin and Wilkins (1996), which is unspecific
with regard to the feature of deliberation.
A canonical criterion for the characterization of causation is the distinction
between internal and external causation. This dichotomy goes back to Smith
(1970) and has further been refined within different frameworks ranging from
syntactico-semantic approaches (McKoon & Macfarland 2000; Levin & Rappaport
Hovav, henceforth L and RH, 1995) to mainly syntactical approaches (Alexiadou
et al. 2006). According to L and RH’s (1995:â•›91) analysis, verbs denoting externally
caused COSs imply an Effector, whereas with regard to internally caused COSs
 Domenico Niclot

“some property inherent to the argument of the verb is responsible for bringing
about the eventuality”. L and RH argue that evidence for the ontological Â�distinction
between internal and external causation is provided by phenomena showing up at
the interface between morpho-syntax and semantics. In fact, contrary to internally
caused COSs, externally caused COSs display the so-called causative-inchoative
alternation.33 This notion refers to the fact that externally caused COSs may be
realized either via an intransitive or a transitive variant depending on the con-
ceptual components which is meant to be profiled. A prototypical example that
they adduce (1985:â•›83ff.) relates to break which in its inchoative instantiation only
profiles the central subevent denoting the COS undergone by the Patient, while in
its transitive realization it profiles the causing subevent triggered by the Effector.
These two constructions can be represented by means of Figure 3 (adapted from
Croft 1990:â•›49).

Causative : The rock (x) broke the windows (y).

x y (y) (y)
∙ ∙ (∙) ∙
cause become broken

Inchoative : The window broke.

y (y) (y)

∙ (∙) (∙)
become broken

Figure 3.╇ The event chain in the causative-inchoative alternation

A viable perspective that allows us to reconcile the discordant classifica-

tion of the status of break as put forward by Hens (1997) vs L and RH (1995),
Kallulli (2004) and Schäfer (2007) can be found in Alexiadou et al. (2006). The
latter suggest that certain verbs participating in the alternation, such as break
may be cause unspecified in that they are compatible with the by itself phrase

. From a formal point of view, zerbrechen, which mostly refers to fragile material, is clas-
sified as a ‘labile alternation’ in that the same lexeme encodes both the transitive and the
intransitive alternant. However, alternants need not be identical. Inchoative kaputt gehen,
‘go broken’, which can be predicated of all breakable items, is classified as an ‘equipollent
alternation’ since the same adjectival stem kaputt, (broken) co-occurs with a different verbal
lexeme, i.e. kaputt machen, (literally: make broken) in the transitive alternant. An exhaustive
crosslinguistic overview of further lexical and morphosyntactic patterns of this alternation
is provided by �Haspelmath (1993).
 From semantic roles to evaluative markers: The dative and affected possessors 

which indicates lack of causation. Expanding on their observation, it seems

�appropriate to posit that the occurrence of by itself with COSs prototypically
regarded as externally caused does not demonstrate lack of causation proper,
but the fact that we conceptualize such COSs as lacking causation. This may
be determined by the fact that these COSs may also be brought about by fac-
tors which rank lower than Effectors represented by animate entities or natural
forces, as in:

(53) The television broke by itself.

(54) The door opened by itself.

These instances are to be considered in a context in which some failure of an inter-

nal technical device, such as a transistor or a lock respectively, has triggered the
denoted COS. This also shows that categorizing COSs on the sole basis of the lexi-
cal availability of alternants turns out to be somewhat reductive. In fact, even with
respect to single verbs we need to also take into account our encyclopedic knowl-
edge about the likelihood that the COS of the involved Patient may be brought
about by Effectors. Applied to (G52), this means that vases are far less likely to
break because of the deterioration of their material than because of external causa-
tion, albeit the former eventuality cannot be entirely dismissed either. By contrast,
the reverse can be posited with regard to break predicated of televisions. This leads
us to view external causation as a radial category with more and less prototypical
members depending on the nature of the Patient.
Considering causation as a continuum enables us to account for a further
aspect addressed by Alexiadou et al. (2006), i.e. the postulation of the component
of causation with respect to some verbs that are canonically classified as internally
caused. As Alexiadou et al. (2006:â•›205) remark, modifying PPs or periphrastic
causative constructions are indicators of this phenomenon. They quote the fol-
lowing example:
(55) The flowers wilted from the heat.

In order to distinguish this notion from the one of Effectors proper, I will call indi-
rect causers ‘Accelerators’. Alexiadou et al.’s claim, which is syntactically founded,
also appears plausible on the basis of world knowledge. For example, flowers have
a limited life span after which they wilt out of biological reasons (internal cau-
sation). However heat may cause them to wilt long before their natural end is
reached, as can individuals by neglecting to water them.
To summarize, we may distinguish between COSs triggered by Effectors, be
they animate or inanimate, COSs conceptualized as determined by the nature of the
 Domenico Niclot

Patients themselves and COSs facilitated by Accelerators. In the next �subsection,

I consider the behavior of the dative with the latter category, which exhibits the
closest affinity to the DI.

5.3 The dative with COSs involving Accelerators

In order to verify whether the correlation of the dative with the verbs subsumed
under the DI only profiles neglect, I conducted a Google search for the verbs listed
by Hens (1997). The search was limited to German and Italian given that French
displays strong restrictions on this construction, as I will show later. Both lan-
guages yielded various hits contextually indicating lack of human responsibility.
For example, in (G56) the writer emphasizes that she used specific equipment and
measures apt to prevent the depicted COS and that she was not able to detect any
Accelerators:
(G56) Die Brownies sind mir angebrannt trotz richtiger Backtemperatur und
�Backzeit.34
‘The brownies burned on me despite the right baking temperature and the
right baking time.’

By contrast, in (I57) the COS is attributed to the non controllable impact of a tech-
nical failure as an Accelerator:
La salsa di soia mi è andata a male complice il frigo che è morto l’altro
(I57) 
�giorno.35
‘The soy sauce went bad on me and its accomplice [i.e. cause] was the fridge
which died [i.e. broke] the other day.’

Thus, the function of the dative in tokens such as the ones above should be viewed
as solely denoting affectedness. This meaning is even more explicit in (G58),
where the Accelerator, a housewife, admits having let her husband’s meal burn.
In this case, the dative exclusively refers to the writer’s empathy for her potentially
affected husband:
(G58) Sein Essen ist ihm verbrannt, weil ich die Platte wieder raufgestellt hatte.36
‘His meal burnt on him because I had turned up the cooker.’

. http://familienforum.liliput.ch/topic.asp?topic_id=923
. http://chezbabs.blogspot.com/2010/04/vegan-noodles.html
. 〈http://www.buchhandlung-weltbuehne.de/metzger-75-untermieter.html〉
 From semantic roles to evaluative markers: The dative and affected possessors 

However, in my Google search, also I encountered further instances in which

the dative correlated with contextual clues pointing to inappropriate handling by
human Accelerators:
(G59) So manche Kartoffel ist mir angebrannt, weil ich es versäumte, auf die kleinste
Energiestufe zurückzustellen.37
‘Some potatoes have burned on me because I neglected to turn the
Â�temperature down to the minimum.’
(I60) Le prime 2 mi sono morte perche le annaffiavo troppo.38
‘The first two [plants] died on me because I watered them too much.’

Justifying the availability of the DI on the sole basis of the contextually pro-
vided component of neglect appears problematic. In fact, this step would require
�splitting the latter case from cases in which the dative only profiles affectedness
due to the non preventability of the depicted COS or from cases in which neither
of the two senses can be determined because no further clues are available. This
would lead to particularization, which runs counter to the principle of generaliza-
tion epistemically inherent to a constructional approach. A further phenomenon
provides evidence for dismissing the tenability of the DI. Both languages exhibit
periphrastic constructions that serve the function of signaling neglect by overtly
indicating Accelerators:
(I61) Mentre scrivevo questo pezzetto ho lasciato bruciare le mie zucchine.39
‘While I was writing this piece [of blog], I let my courgettes burn.’
(G62) Obwohl die nix dafür konnten, da DU ja die Milch hast schlecht werden
�lassen.40

‘However, it’s not their fault because it’s YOU who let that milk turn sour.’

These two tokens conform to the Principle of No Synonymy inasmuch as the peri-
phrastic construction only indicates neglect, whereas the dative one only profiles
affectedness regardless of whether neglect is involved or not. As we will see in the
next subsection with regard to COSs in which individuals may feature as Effectors,
the Principle of No Synonymy is not only semantically, but also pragmatically
motivated.

. 〈http://www.dooyoo.de/toepfe-pfannen/yadel-jasmina-topfset/589742/〉
. 〈http://it.answers.yahoo.com/question/index?qid=20091101090017AAoSwc0〉
. 〈http://www.ibs.it/code/9788806594060/sereni-clara/casalinghitudine.html〉
. 〈http://spickmich.blogspot.com/2007/11/sorry-spickmich-ist-langsam.html〉
 Domenico Niclot

5.4 The dative with COSs brought about by human Effectors

As pointed out in 5.2, the inchoative variant of verbs participating in the causative-
inchoative alternation is cause unspecified. The dative also co-occurs in German
and Italian with COSs caused by human Effectors:
(G63) Sie hat noch einen kopfhörer von mir, den ich ihr fürs krankenhaus geliehen
habe. Er ist ihr aus versehen kaputtgegangen.41
‘She still has a headset of mine which I lent her for hospital. It has
Â�accidentally gone broken on her.’
(I64) Per sbaglio mi si è rotto un pin.42
‘Mistakenly, a pin broke on me.’

Some approaches dealing with the dative in conjunction with inchoative alter-
nants, such as Kallulli’s (2004) and Schäfer’s (2007) do consider affectedness as
one of the constructional senses. Schäfer (2007:â•›49 ff.), for example, discusses the
link between inchoative kaputt gehen, ‘break’, in German as predicated of vases
and states that the dative contributes either an ‘unintentional-causer’ reading or
an affectedness reading and refers to the availability of further contextual clues
for purposes of disambiguation. He argues that the occurrence of adverbials such
as aus Versehen, ‘mistakenly’, triggers the former interpretation, whereas the con-
textual availability of an Effector other than the dative referent enforces the latter
one. Similarly, Kallulli (2004) adduces adverbials represented by the equivalents
of accidentally as a diagnostic tool. In other words, these approaches imply that
Effectors cannot be affected. However, the same adverbials are also attested with
causative alternants:
(G65) Er hat eine getragen die hab ich ihm aus versehen kaputt gemacht.43
‘He used to wear one [chain] which I mistakenly broke on him.’

Una volta gli ho rotto per sbaglio una cosa stupidissima ed ha avuto una
(I66) 
reazione spaventosamente esagerata.44
‘Once, by mistake, I broke on him a trivial item and he really overreacted.’

. 〈http://www.liebeskummer.ch/liebeskummer-herzschmerzen/mal-wieder-schluss_
t3855p9.html〉
. 〈http://www.hwupgrade.it/forum/showthread.php?t=2206982&page=2〉
. 〈http://www.portablegaming.de/fun-forum/50716-ultimative-laber-thread-735.html〉
. 〈http://www.sclerosi.org/forum/viewtopic.php?p=549947〉
 From semantic roles to evaluative markers: The dative and affected possessors 

Bearing in mind the Principle of No Synonymy, we should ask in what regard

transitive and inchoative alternants differ. As for the former, the use of adverbi-
als denoting accidental causation is motivated by the need to emphasize lack of
intentionality. In fact, one may as well break things, or more generally bring about
a certain COS, with a specific purpose. With respect to the inchoative alternant,
instead, I posit that the adverbials fulfill the function of indicating non deliberate
human causation, whereby this aspect is backgrounded and the component of
affectedness is foregrounded. For example, in (G63) the writer, who is the pos-
sessor of the damaged headset, backgrounds the responsibility of the Effector
by depicting her as affected, presumably in order not to expose her to potential
reproaches.
It seems plausible to regard instances such as (G63) and (I64) as originating
from a blending in the sense of Fauconnier and Turner (2002). The two input
spaces are the ones of overt, unintentional human causation, as expressed by
(G65) and (I66), and of metaphorical self-causation which is inherent to (G63)
and (I64) if we omit the adverbials. The component of self-causation is determined
by the fact that kaputt gehen (literally ‘go broken’) displays a high degree of volition
by virtue of the verb go and should be viewed as an instance of middle voice (see
Kemmer 1993, Nava & Maldonado 2005), in which a COS is depicted as though it
were brought about by the Patient acting on itself as an Agent. The same applies to
Italian reflexive si (itself), as shown in Figure 4.

Figure 4.╇ Middles (from Nava & Maldonado 2005:╛475)

As Nava and Maldonado point out, middles profile a high degree of energy
displayed by the entity involved in the event. Thus, the inchoative Patient is lexi-
cally or morphologically constructed as transmitting energy to itself in order
to bring about the COS. This input space of self-causation contains the addi-
tional feature of affectedness directed by the Patient towards its possessor and
expressed by the dative. In metaphorical terms, we may say that the Patient is
profiled as annihilating itself thereby triggering its possessor’s affectedness. The
blending deletes the apparent self-causation of the Patient. It retains the adver-
bial feature of unintentional human causation (first input space) and the feature
of affectedness (second input space). The removal of overt causation (first input
space) is geared towards foregrounding affectedness. Figure 5 illustrates this
blending.
 Domenico Niclot

GENERIC SPACE

CAUSATION

OVERT CAUSATION SELF-CAUSATION

NON INTENTIONALITY TRANSFERRED

AFFECTEDNESS

INPUT 1 INPUT 2

NON INTENTIONALITY

TRANSFERRED
AFFECTEDNESS

BLEND

Figure 5.╇ The dative and the transitive-inchoative blending

Rejecting the independent status of the dative constructions posited by

Kallulli (2004) and Schäfer (2007) is motivated by the same argument that was
advanced for the DI, i.e. by the availability of two alternants which differ from a
semantic point of view. Moreover, instances such as (G64) and (I65) can also be
realized without any adverbials signaling lack of intentionality. In this case, human
Effectors are propositionally deleted but retained in our conceptual representa-
tion (or ‘base’ in the sense of Langacker 1991). Our encyclopedic knowledge tells
us that certain COSs referring to specific Patients are most likely to ensue as a
result of human causation. We may hypothesize a correlation between the use of
the inchoative-dative alternant and the purpose of obscuring one’s or someone
else’s responsibility. Some evidence for this assumption is provided by an infor-
mal �survey which I conducted with 20 German and 20 Italian native speakers. I
�submitted to them the transitive and the inchoative version of the counterpart of
having inadvertently broken one’s glasses and the dative-inchoative alternant both
 From semantic roles to evaluative markers: The dative and affected possessors 

with and without adverbials referring to lack of accidental causation.45 My infor-

mants were asked which option they would have selected for describing this even-
tuality to their opticians if their guarantee had expired and if they had wanted to
persuade them to mend their spectacles free of charge. Interestingly, none of them
opted for the transitive realization. Also, in both languages the choice of the incho-
ative variant without adverbial prevailed over the adverbial one (15/20 for German
and 17/20 for Italian).
We can attribute to the dative the status of a construal (Langacker 1991) as an
information-packaging device which prompts a specific way of viewing a certain
state of affairs, and more specifically in this case as a pragmatic strategy. This is a
case in which the intersubjective dimension mentioned in Section 3.2. comes into
play as a strategy used by speakers to appeal to their hearers.
At this stage, we may ask whether with regard to the tokens that Hens subsumes
under the DI the inchoative alternant is also preferably instantiated in comparison
with the periphrastic one. In my Google search, the former did not co-occur with
any adverbials denoting accidental causation. This suggests that the component of
responsibility may conceptually be anchored to a lesser extent with regard to COSs
brought about by Accelerators. However, this hypothesis awaits further empirical
verification.
A final argument for dismissing the function of the dative for indicating
responsibility with respect to all COSs is provided by its prototypical function
of denoting Experiencers, who display a low degree of activity, contrary to Effec-
tors (see 2.1.). The claim that the dative solely profiles affectedness in conjunction
with COSs leads us to ask from where this sense stems. An appealing hypothesis
consists in postulating an extension, arisen along a grammaticalization path, of
micro-construction (i). In other words, after acquiring the function of denoting
affectedness in (i), the dative might have been applied to other constructions as a
general grammaticalized marker for encoding this component. While this hypoth-
esis may hold for Italian, it appears hardly viable for German, which is governed by
crucial restrictions, as I show in the following subsection.

5.5 Semantic constraints with datives denoting COSs

Legitimating the availability of the DI would also imply positing its counterpart,
i.e. a dative construction which denotes the beneficial occurrence of a certain
COS as facilitated by a human Accelerator. However, Hens (1997:â•›211) observes

. The transitive alternants were: ich habe (aus Versehen) meine Brille kaputt gemacht (G),
(per sbaglio) ho rotto gli occhiali (I). The inchoative variants were: mir ist (aus) Versehen meine
Brille kaputt gegangen (G), (per sbaglio) mi si sono rotti gli occhiali (I).
 Domenico Niclot

that the dative in the DI only yields an adversative reading and cites the following
example:

(G67) Ihm sind die Tomaten gereift.

‘The tomatoes ripened on him.’

The scenario provided for the acceptability of this sentence is the one of an
�experiment in which tomatoes that were due to ripen to a later point in time did
not do so because of some mistake made by the scientist in charge. Contrary to
German, in Italian the dative also yields a beneficial reading of the COS depicted
in (G67):
(I68) Ho coltivato una pianta di zucca, e mi sono maturate 2 zucche.46
‘I have grown a pumpkin plant and two pumpkins have ripened to me.’

In this instance, the writer views the ripening of the pumpkin as a reward for his
agricultural efforts. Hens’s tenet, which is indeed warranted in German, requires
further elaboration. The COS underlying (G67) refers to an event that �conceptually
bears positive associations. Accordingly, the adversative scenario provided by
Hens is conceivable, but rather unlikely to occur. This pattern needs to be differ-
entiated from COSs that display adversative semantics (as in G50) and such that
yield either an adversative or a positive interpretation depending on the context.
With respect to the dichotomy latter, the Italian dative is compatible with either a
beneficial reading (I69) or a detrimental one (I70):
(I69) Sai che mi è dimagrito? Sta ridiventando normale!47
‘You know, it [dog] has got thinner on me. He might be going back to
Â�normal.’
(I70) Lui due mesi fa mi è dimagrito in modo spaventoso.48
‘Two months ago, it [cat] lost weight in an awful way on me.’

In (I69), the writer is relieved, i.e. positively affected, because of the weight loss
of her dog whose overweight had previously caused health problems. In (I70),
instead, the dramatic weight loss undergone by the writer’s pet is viewed as utterly
concerning. My Google search for analogous German counterparts only yielded
hits with adversative semantics, as in:

. 〈http://it.narkive.com/2007/7/16/5260909-raccolta-zucca.html〉
. 〈http://harmel.splinder.com/post/21181032〉
. 〈http://www.micimiao.net/forum/archive/index.php/t-74695.html〉
 From semantic roles to evaluative markers: The dative and affected possessors 

(G71) Er hat mir abgenommen.49

‘He [son] has got thin on me.’

This sentence is uttered by a mother who is giving vent to her worries for her
baby’s lack of appetite. Further evidence for the fact that the dative referring to
COSs only yields an adversative reading in German was provided by an experi-
ment that I carried out by submitting an instance containing a dummy verb to
twenty German and Italian native speakers:
(G72) Die Blumen sind mir aufgekluchst.
‘The flowers have DUMMY me:DAT.’

(I73) I fiori mi sono allodati.

‘The flowers have DUMMY me:DAT.’

All my German informants attributed a negative meaning to this sentence.

By contrast, all my Italian informants judged it as potentially yielding either
a �positive or a detrimental interpretation. In the next subsection I propose
two alternative grammaticalization paths that may account for the behavior of
German.

5.6 “Heading for a fall”: Unusual grammaticalization paths

One of the verbs that Hens (1997) subsumes under the DI is fallen ‘fall’. In terms
of causation, this verb occasionally also tolerates an internal-causation reading as
in case of ripe fruits or dry leaves falling from trees. However, it seems appropri-
ate to classify it as indicating a change of location prototypically brought about
by Effectors. Interestingly, however, in Italian and �German, human Effectors are
instantiated periphrastically as Accelerators:
(G74) Ich habe den Laptop eines Freundes aus Versehen fallen lassen.50
‘By mistake, I let a friend of mine’s laptop fall.’
(I75) Ho fatto cadere per sbaglio una bottiglietta di olio.51
‘I mistakenly let fall a small bottle of oil.’

. http://forum.gofeminin.de/forum/bebeestla/__f77563_bebeestla-Brauche-dringend-
Rat-Spei-Kind-extrem-shock-cry.html
. 〈http://www.gutefrage.net/frage/ich-habe-den-laptop-eines-freundes-aus-versehen-
fallen-lassen-wie-sag-ich-s-meiner-haftpflicht〉
. 〈http://it.answers.yahoo.com/question/index?qid=20111126100237AAuB963〉
 Domenico Niclot

Individuals as Effectors also appear with the dative in the inchoative variant:
(G76) Er ist mir aus Versehen runtergefallen.52
‘It [i.e. laptop] mistakenly fell to me.’

The Italian counterpart of (G76) is equally well-formed. However, contrary to the

instances analyzed in the previous subsections of this section, the dative can be
used to indicate affectedness only in Italian. Clear evidence for this comes from
the fact that the dative also co-occurs with Effectors other than human ones:
(I77) Ieri col vento, mi e’ caduto il vaso con la yucca.53
‘Yesterday, because of the wind the vase with my Yucca fell down to me.’

By contrast, in German we solely encounter dativeless possessive tokens in con-

junction with this pattern of causation:
(G78) Zum letzten mal sind meine Töpfe auf dem Balkon bei Sturm umgefallen.54
‘The last time, my vases on the balcony fell down during a storm.’

This constraint was corroborated by the judgment of my German informants who

remarked that the translation of (I77) was unacceptable since it was not an indi-
vidual who caused the vases to fall. Thus, it appears that fallen thoroughly corre-
sponds to the semantics of neglect attributed by Hens (1997) to the DI. In order
to account for this phenomenon, we need to bear in mind that human Effectors
of fall refer to the Source from which the process leading to the change of location
commences. Sources are encoded via PPs both in German and Italian:
(G79) Er fiel runter vom Balkon.
‘He fell down from the balcony.’
(I80) Cadde dal balcone.
‘He fell down from the balcony.’

An exception to this marking device is constituted by Sources denoting animate

entities, which require dative marking in both languages:
(G81) Mir ist eine Vase hingefallen.
(I82) Mi è caduto un vaso.
‘To me a vase fell down.’

. 〈http://www.gutefrage.net/frage/ich-habe-den-laptop-eines-freundes-aus-versehen-
fallen-lassen-wie-sag-ich-s-meiner-haftpflicht〉
. 〈http://www.i-h-g.it/archivio/y/yucca.htm〉
. 〈http://chiliforum.hot-pain.de/thread-1564.html〉
 From semantic roles to evaluative markers: The dative and affected possessors 

(G81) and (I82) constitute tokens of differential marking, in which the lexically
governed dative encoding of human Sources fulfills the purpose of signaling an
affected Experiencer (see 2.1). Further cases of this phenomenon are represented
in both languages by the equivalents of escape (G: weglaufen, I: scappare) and slip
down (G: herunterrutschen, I: scivolare). All these verbs display negative seman-
tics. Thus, it is plausible to assume that in a subsequent phase of language develop-
ment differential marking may have led to a reanalysis of the dative as a marker
of neglect via metonymization. In fact, if individuals let things fall/slip down or
individuals escape, they may be regarded as incautious. In a still later phase, the
dative is likely to have been extended via generalization to COSs brought about by
Accelerators other than the dative referent (as in G58) or due to the nature of the
Patient (as in break predicated of technical devices). Through this extension the
initial �component of affectedness inherent to Experiencers was reintroduced due to
the lack of the component of neglect. However, the entrenchment of the negative
semantics underlying verbs with differential marking was also retained, thus block-
ing the use of the dative with beneficial COSs (see also the survey conducted with
regard to G72).
Figure 6 illustrates this hypothesized grammaticalization path.

DIFFERENTI AL MARKING METONYMIZATION METONYMIZATION

ADVERSATIVE NEGLECT ADVERSATIVE

AFFECTEDNESS AFFECTEDNESS

Figure 6.╇ A possible grammaticalization path of the German dative with COSs

At this stage, we should note that differential marking also applies to some
verbs with beneficial semantics such as gefallen (appeal in the sense of this
appeals to me) or with neutral semantics such as passieren (happen). However,
the Experiencers of such verbs display a zero degree of activity in terms of par-
ticipation and accordingly do not yield a positive counterpart to the inference
of neglect.
An alternative account of the restrictions underlying the use of the dative with
COSs in German may be worked out by linking the grammaticalization process to
the transitive-inchoative alternation. As previously argued, in the inchoative alter-
nant the notions of Effector and of event-transfer are preserved in our conceptual
representation. This pattern may have constituted the starting point leading to
 Domenico Niclot

the subsequent extension via generalization of the component of affectedness to

COSs that do not imply any Effectors, but only Accelerators, if at all. This path is
represented in Figure 7.

EVENT TRANSFER BY EFFECTOR → GENERALIZATION → LACK OF EVENT TRANSFER

â•…â•… ↓ ↓

╅╇ COS COS

â•…â•… ↓ ↓

â•…ADVERSATIVE AFFECTEDNESS â•…â•…ADVERSATIVE AFFECTEDNESS

Figure 7.╇ An alternative grammaticalization path for the German dative with COSs

The exclusively adversative reading yielded by the dative may be traced back
to the fact that COSs involving Effectors display detrimental semantics. This fea-
ture is likely to have blocked the realization of the dative in conjunction with the
beneficial meaning inherent to some COSs brought about by Accelerators or to
COSs that are conceptualized as internally caused. From a theoretical perspec-
tive, this account differs from the previous one inasmuch as it is not bound to a
remarkably specific phenomenon such as differential marking, but is based on a
widespread construction such as the causative one. However, positing the latter as
the trigger of the grammaticalization path is somewhat speculative and requires
corroboration through diachronic data.
With respect to the use of the dative with COSs in French, neither of these
two grammaticalization paths can be corroborated. In fact, on the one hand, con-
trary to Raineri and Evola’s claim (2008), French does display this construction,
whereby the dative is compatible both with an adversative (F83) and a beneficial
sense (F84):
(F83) Ça a du être la même chose pour tous les autres lapins qui me sont morts.55
‘The same thing must have happened to all the other rabbits that have died
on me.’
(F84) Les cambrias, elles m’ont fleuri au printemps dernier.56
‘The cambrias, they blossomed to me last spring.’

. 〈http://patapons.rabbitforum.org/t729-une-question-pour-lizou〉
. 〈http://www.orchidees.fr/forums/index.php?showtopic=330〉
 From semantic roles to evaluative markers: The dative and affected possessors 

Moreover, French also exhibits differential marking:

(F85) La bouteille m’est tombée par terre.57
‘The bottle fell to me on the floor.’

On the other hand, this construction displays a highly limited quantity of tokens
since most COSs, (se sécher: wilt, s’épanouir: blossom, se casser: break, se plier:
bend), are expressed by reflexive verbs (se: oneself). The latter are constrained to
co-occur with the preposition à (to), which primarily denotes inanimate Goals
and are not compatible with the dative. This also applies to semantic roles which
are even more basic than that of Evaluators. In fact, Recipients of reflexive verbs,
which mostly imply the tranfer of gestures rather than entities, are conceptualized
as Goals:
(F86) Il s’est présenté à moi. *Il me s’est présenté.
‘He introduced himself to me. *He introduced himself to me (DAT).’

Partial restrictions regarding the compatibility of reflexive verbs with the dative
are also exhibited by German. Differently from French, Recipients of reflexives are
encoded via dative, as the counterpart of (F86) shows:
(G87) Er stellte sich mir vor.
‘He introduced himself to me.’

Yet COSs underlying reflexive verbs are not compatible with the dative in German
either, contrary to what is stated by Schäfer (2007 passim). This can be demon-
strated by replacing zerbrechen with sich verformen ‘change shape’ in (G52):
(G88) *Der Oma hat sich die Vase verformt.
‘The vase changed:REFL its shape on grandma.’

We may postulate that this restriction is motivated by the requirement of the fea-
ture of deliberation, which is inherent to event-transfers directed towards Recipi-
ents, but which is not available with reflexive COSs despite their being encoded as
volitional (see 5.4.).
As for the categories of possessa that may enter (ii), we encounter both alien-
able entities (G65, I66, F84) and animate ones (I70, G71, F83). With respect to
body parts, the use of the dative is well-formed for the few German verbs that do
not exhibit any reflexive morphology:

. 〈http://www.google.fr/#q=%22m’est+tombee+par+terre%22&hl=fr&prmd=imvns&ei=
5g3uTpqOFYjk4QTPt5mACQ&start=10&sa=N&bav=on.2,or.r_gc.r_pw.,cf.osb&fp=79f5663
e5c60692e&biw=1280&bih=635〉
 Domenico Niclot

(G89) Die linke Backe ist mir angeschwollen, und die Nase auch ein bißchen.58
‘My left cheek has swollen on me and so has my nose a bit.’

With regard to French, I was not able to detect any non reflexive verbs relating
to body parts. Finally, Italian does not display any restrictions with regard to the
compatibility of the dative with reflexives, as the following example shows, where
the same verb as in (G89), is thoroughly compatible with the dative:
(I90) Stanotte dopo due notti insonni mi si è gonfiata la guancia sinistra.59
‘Last night, after two sleepless nights my left cheek swelled itself on me.’

The phenomena outlined so far point again to the importance of considering

possessa within the constructions in which they show up, as previously argued
contra König and Haspelmath (1998). This is even more relevant if we consider,
for example, that in Italian, the dative can also be instantiated in its inchoative
alternant with body parts, as in:
(I91) In veranda sono caduto e mi si è rotto il braccio!60
‘On the veranda I fell and the arm broke itself on me.’

In this case, the writer is also the Effector of the COS. My Google search only
yielded 4 hits for this token. To account for such low frequency, we need to bear in
mind that the use of the dative in this construction mainly serves the function of
profiling affectedness, which pertains to speaker-hearer interaction. Since break-
ing one’s arm does not usually require concealing responsibility in order to win the
empathy of thirds, contrary to breaking one’s glasses, the pragmatic import of the
dative does not appear to be particularly relevant.
In guise of conclusion, the spectrum of (ii) in the three languages is repre-
sented in Table 8.

Table 8.╇ Major features of the dative with COSs in German, French and Italian
The dative with Beneficial Adversative Compatibility with reflexively
micro-construction (ii) reading reading encoded COSs

German - √ X
French √ √ X
Italian √ √ √

. 〈http://gutenberg.spiegel.de/buch/5098/2〉
. 〈http://www.medico-odontoiatra.it/forum/viewtopic.php?id=425&p=3〉
. 〈http://www.video.mediaset.it/video/camera_cafe/puntate/213255/la-paura.html〉
 From semantic roles to evaluative markers: The dative and affected possessors 

In the next section I illustrate the heterogeneous behavior of the three lan-
guages with respect to micro-construction (iii).

6. The Agentive Dative Construction (ADC)

6. 1 Semantics and spectrum of the ADC

A preliminary remark is in order with regard to label ‘Agentive’. The notion of
agentivity adopted here is not based on the syntactic requirement of a subject and
of a direct object, but on the semantic one proposed by Dowty (1991), according
to which the roles of Proto-Agent and Proto-Patient encompass a cluster of fea-
tures. The latter can, but need not, be all available at the same time with respect to a
specific event or process. Among these properties, Dowty outlines volition, which
does not necessarily imply an Agent acting on another participant. This notion of
agentivity allows us to also consider syntactically intransitive eventualities such as
(F92–I95) as being carried out by Proto-Agents, provided we bear in mind that the
attribution of volition to inanimate entities (F92, G93) constitutes a metaphorical
construal:
(F92) [Le] disque dur me fonctionne pas.
‘The hard disk does not work to me.’
(G93) Der Rechner ist mir abgestürzt.61
‘The PC crashed on me.’
(F94) S’il ne veut pas manger, il me mange pas.62
‘If he does not want to eat, he won’t eat to me.’
(I95) Ragazzi, mi ha pianto il bambino tutta la notte.63
‘Folks, my kid cried on me all night.’

The ADC includes inanimate possessa (F92, G93), animate possessa as kinship
(F94, I95) and body parts:
(I96) Mi tremano le mani ma mi sento pronta.64
‘The hands are shaking on me, but I feel ready.’

. 〈http://www.theofel.de/archives/2012/07/lied-abgestuerzt.html〉
. 〈http://mallia.monbebeblog.com/2216076/Miam/?nextyear=1&month=12&year=2013〉
. 〈http://www.piazzadonegani.it/index.php?pid=2〉
. 〈http://ricerca.repubblica.it/repubblica/archivio/repubblica/2013/04/28/mi-tremano-le-
mani-ma-mi-sento.html〉
 Domenico Niclot

(G97) Mir brummt der Kopf vor lauter Alpträumen.65

‘The head is throbbing on me after so many nightmares.’
(F98) Mon ami, le coeur me bat bien fort.66
‘My friend, my heart is beating to me really fast!’

The co-occurrence of body parts with the dative differs from the tendencies out-
lined with respect to (i). Table 9 reports the outcome of my Google search for the
equivalents of my hands are shaking,67 as in (I96), and my ears are ringing68 in their
mere instantiation possessive and their dative instantiation with possessor/pos-
sessor omission.

Table 9.╇ A comparative Google search for the alternation dative vs mere possessive
�encoding with body parts construed as acting
‘My/the hands are shaking to me’ Ratio ‘My hands are shaking’ Ratio

G 223 64% 129 36%

F ╇74 21% 267 79%
I 195 48% 211 52%
‘My/the ears are ringing to me’ Ratio ‘My ears are ringing’ Ratio

G ╇46 33% ╇93 67%

F ╇53 40% ╇77 60%
I 230 80% ╇58 20%

Although a fine-grained comparison of this data with the outcome of Tables

3 and 5 cannot be carried out for reasons of space, it is worth noting that all three
languages do not exhibit the same constructional fixedness as in (i) since they
display a conspicuous rate of dativeless tokens which in most cases surprisingly
even outnumber the dative ones. Also in this case postulating that the dativeless
instantiations reflect lack of affectedness is hardly justifiable due to the detrimen-
tal semantics of the events. Thus, we can conclude that (i) and (iii) exhibit het-
erogeneous degrees of grammaticalization. Also, this strongly corroborates the

. 〈http://www.gripmastertrails.com/trailnews/bei-blitzschlag-trail/〉
. 〈http://www.dumaspere.com/pages/bibliotheque/chapitre.php?lid=r7&cid=77〉
. The input was as follows: meine Hände zittern/mir zittern die Hände (G), mes mains
tremblent/les mains me tremblent (F).
. The input was as follows: meine Ohren pfeifen/mir pfeifen die/meine Ohren (G), le mie orec-
chie fischiano/mi fischiano/le mie orecchie (I), mes oreilles tintent/les/mes oreilles me tintent (F).
 From semantic roles to evaluative markers: The dative and affected possessors 

hypothesis that (iii) has developed later than (i). The rentability of a differentiated
constructional approach is once again evident if we consider that the data above
invalidates König and Â�Haspelmath’s claim (1998) that only affected body parts may
co-occur with the dative in French. A further claim defeased by the data in Table
9 as well as by (F92) and (F94) is Raineri
Â� and Evola’s (2008) tenet that French only
tolerates the realization of the dative with transitive instances.
Another crucial issue which requires discussion is the observation of the latter
authors that the dative is also compatible with states in Italian (micro-construction
iv). One of the examples that they propose for corroborating this stance reads:
(I99) Mi ha gli occhi verde [sic: i.e. verdi].
‘He has green eyes to me.’(Raineri & Evola 2008:â•›13)

This utterance is attributed to a mother who is surprised, i.e. in our terms positively
or negatively affected, about having given birth to a green-eyed baby. �However, as
a native speaker of Italian, I do not find this token acceptable. All other native
speakers whom I consulted also judged it to be ill-formed, as they did with respect
to the state underlying:
(I100) *Mi conosce l’inglese.
‘He knows English to me.’

The same judgment was expressed by my German and French informants about
the equivalents of (I99) and (I100). By contrast, replacing know in (I100) with an
accomplishment such as learn English leads to an acceptable outcome, albeit only
in Italian:
(I101) La faccio giocare con il dizionario alla mano cosi’ almeno mi impara
l’inglese.69
‘I let her play with a dictionary in her hands so at least she learns English
to me.’

An explanation for this phenomenon can be provided by recalling the event-

transfer metaphor and by arguing that, at least in the languages examined in this
paper, states cannot be transferred. Coining the category of ‘stative processes’ for
instances such as (I29), as Raineri and Evola (2008) do, appears problematic. In
fact, in this example be in the sense of remain pertains to the category of activi-
ties in terms of lexical aspect and can be bounded by grammatical aspect, as
demonstrated by the use of the perfective and of the preposition per, ‘for’, which
indicates the duration boundary. These two criteria do not hold for states proper
as in (I99-I100). We may conclude that French, German and Italian differ in that

. 〈http://forumuniversotombraider.forumcommunity.net/?t=18515419〉
 Domenico Niclot

only the latter tolerates the dative with ‘verbs of posture’, as in (I29), which rep-
resent a subset of the class of activities within (iii). However, positing micro-con-
struction (iv) is legitimated by the compatibility of states proper with the dative
in some Slavic languages such as Slovenian (see Šarić 2002) and Macedonian (see
Mitkovska 2011), where the criterion of affectedness seems to overrule the event-
transfer metaphor. Further cross-linguistic research is required to investigate this
issue.
If we focus on the semantics of the ADC, we observe that it obeys no
restrictions, contrary to micro-construction (ii). In fact, in all three languages
the dative is compatible with VPs that bear per se either beneficial (F92) or
adversative associations (I95). In instances containing a negation, such as (F92)
and (F94), the reading of adversative affectedness is determined by the failed
transmission of the beneficial event inherent to the semantics of the VP. With
respect to VPs that yield either a beneficial or an adversative reading, German
tolerates either interpretation, which also applies to Italian and French. For
example, in (G102) the integration of encyclopedic knowledge (eating sweets
damages children’s health) with the quantifier zu viel, (too much) licenses a
negative interpretation:
(G102) Der isst mir zu viel Schokolade und Zuckerwatte.70
‘He eats too much chocolate and candy floss to me.’

By contrast, in (G103), the not necessarily positive associations inherent to eating

finger food are contextually invalidated by the occurrence of the adverb wenigstens
(at least). Accordingly, the sentence yields the inference that the writer is positively
affected since she considers the depicted circumstances preferable to the eventual-
ity that her son may not eat anything at all:
(G103) Im Moment ist er in einer guten Phase und isst mir wenigstens Fingerfood.71
‘At the moment, he [son] is in a quite good phase and at least is eating finger
food to me.’

As the examples above show, the dative mainly reflects the affectedness of their
writers in terms of expectation (I101), concern (G102), relief (G103). Thus,
appears that the ADC relates primarily to the dimension of subjectivity rather
than to the one of intersubjectivity considered as a strategy for gaining the

. 〈http://www.mamily.de/groups/328/group_posts/16930〉
. 〈http://www.urbia.de/archiv/forum/th-1845691/Ich-kann-nicht-mehr-isst-NICHTS-
freiwillig.html〉
 From semantic roles to evaluative markers: The dative and affected possessors 

sympathy or the approval of thirds. However, the latter eventuality is not to be

excluded either.
As for the relation of possession between the dative referents and the Agents,
we observe that the ADC also encompasses entities of the PS that may be regarded
as peripheral as compared with more prototypical ones. For example, the notion
of PS underlying the following sentences relates to a professional domain in which
individuals interact:
(I104) Se non mi gioca un centrocampista ovviamente sarà sostituito da un altro
centrocampista.72
‘If a midfielder doesn’t play to me, he’ll obviously be replaced by another
midfielder.’
(I105) Non mi studia la materia, cara signora!73
‘She does not learn this subject to me, dear madam.’
In the former example, a football trainer is discussing the logistics of a match
due to take place in a short time and expresses the adversative affectedness that
would ensue for him in case one of his players should not be available for the
match. In the latter instance, a teacher communicates her worries to a mother
whose daughter is not sufficiently performing in a certain subject. Interest-
ingly, Italian even displays a type in which no possessive relation whatsoever is
implied and in which the dative profiles affectedness due to external circum-
stances, such as weather conditions in:
(I106) Se mi piove 2 gg a Fakarava e/o Rangioa, mi metto a piangere come un
�bambino.74
‘If it rains to me two days in Fakarava and/or Rangioa, I will start crying
like a child.’

(I104-I106), for which I did not encounter any French or German counterparts in
my Google search, show that the ADC in Italian exhibits a vaster spectrum than
the other two languages, in which the use of the dative is limited to the domains
of technical devices (F92, G93) and parental issues (F94, G102, G103). These
instances stem from internet blogs in which the writers were mostly young peo-
ple discussing technical issues and young mothers exchanging their experiences.
This constitutes an important indicator for the novelty of this construction in the

. 〈http://forum.fantagazzetta.com/blogs/fantaciccio82/102-news-da-lega-tappetino-
page13.html〉
. 〈http://www.itcdefazio.it/News.asp?ID=301〉
. 〈http://www.polinesia.it/forum/index.php?topic=2666.15〉
 Domenico Niclot

two languages, which may be motivated by the inclination of younger speakers to

�deviate from prescriptive language usage. In fact, my older German and French
informants whom I consulted with regard to the acceptability of these examples
judged them to be ‘odd’.
At this stage, we should attempt to account for the difference between Ital-
ian vs German and French from the point of view of grammaticalization. As for
the behavior of Italian, we can observe that it reflects the mechanism previously
observed with regard to (ii), where the dative is entirely entrenched as a marker
of affectedness. By contrast, in German it is plausible to assume that the spec-
trum of the ADC is not diachronically related to (ii) inasmuch as the former
also tolerates a beneficial reading, contrary to the latter. Finally, with respect
to French it appears difficult to establish a link in terms of motivation between
the ADC and (ii) since both do not display any semantic constraints, but only a
limited occurrence In the next section, I propose a tentative explanation for the
non-thoroughly grammaticalized status of the ADC in German and French by
positing a relation between (iii) and different cognate constructions that display
full-fledged status.

6.2 Th
 e German and French ADC and cognate constructions:
A potential grammaticalization path
German exhibits two entrenched dative constructions, i.e. the so-called Ethical
Dative (Dativus Ethicus) and the Dative of Judgment (Dativus Judicantis). The
former is represented by instances such as:
(G107) Mach mir schön deine Hausaufgaben!
‘Do your homework well to me!’

(G108) Komm mir nicht spät nach Hause!

‘Don’t come late home on me!’

From a syntactic point of view, the Ethical Dative only appears in the imperative
mood and is co-referential with the utterer. By adopting this construction, speak-
ers signal that they would (not) profit from the occurrence of the denoted pro-
cess (Gutzmann 2007). I suggest that this definition should be slightly modified
by arguing that speakers would be positively or negatively affected if the denoted
eventuality took place. The Ethical Dative displays the same function of evaluating
affectedness depending on the positive/negative content semantically or contex-
tually yielded by a sentence, as in the other micro-constructions analzyed here,
i.e. it is a conventional implicature. On the basis of these properties, it may be
assumed that in German the dative is undergoing a grammaticalization shift from
the irrealis of the imperative (Ethical Dative) to the realis of the indicative (ADC),
 From semantic roles to evaluative markers: The dative and affected possessors 

i.e. from potential affectedness to concrete affectedness. Such extension should be

regarded as a case of generalization (see Figure 8).

ETHICAL DATIVE GENERALIZATION AGENTIVE DATIVE CONSTRUCTION

POTENTIAL AFFECTEDNESS FACTUAL AFFECTEDNESS

Figure 8.╇ Grammaticalization path from the Ethical Dative to the ADC

Let us now turn to the Dative of Judgment, which partially also underlies
tokens such as (G102). This construction always co-occurs with zu, ‘too much’,
or zu wenig, ‘too little’ and the dative is attributable to an evaluator who esti-
mates that the extent to which someone acts lies beyond acceptable limits (cf.
English to me, he works too much). The vagueness between an evaluative read-
ing and an affectedness interpretation of (G102) also nicely illustrates how the
latter may be inferred on the basis of the former. In fact, a son eating a quantity
of sweets that his mother generally considers to be excessive, also triggers her
concern in terms of potential health problems. This path can best be accounted
for in terms of metonymization since a judgment is mapped onto affectedness
(see Figure 9).

DATIVE OF JUDGMENT METONYMIZATION AGENTIVE DATIVE CONSTRUCTION

NEGATIVE JUDGMENT OF ACTS CARRIED AFFECTEDNESS MOTIVATED

OUT BY INDIVIDUALS BELONGING BY THE CLOSENESS OF EVALUATOR

TO THE EVALUATOR’S PS AND AGENT

Figure 9.╇ Grammaticalization path from the Dative of Judgment to the ADC

Despite being restricted to adversative semantics, the Dative of Judgment may

also have impacted on the development of the German ADC in conjunction with
the Ethical Dative or separately from it.
With respect to French, we may at first glance posit that the grammati-
calization of the ADC has originated from the spontaneous replication of the
formal syntactic pattern inherent to micro-construction (i), whereby the com-
ponent of affectedness inherent to the latter has been preserved despite the het-
erogeneous constructional sense of (i) vs (iii) (see also critique to Raineri Evola
2008 in section 3). In a subsequent step, the feature of transitivity may have
been dropped. As appealing as it might appear, such hypothesis, which could
 Domenico Niclot

also be advanced for the ADC in German, does not sufficiently explain why
in both languages this construction displays a limited range of tokens. Con-
sidering a widespread French construction, which has been dubbed Narrative
Empathetic Dative Construction by Raineri and Evola (2008), may help us to
work out a supplementary account. This construction is represented by tokens
such as:
(F109) Il est fada celui là, il te boit la bouteille en 3 verres.75
‘That guy is crazy. He drinks to you a bottle in 3 glasses [i.e. very rapidly].’
(F110) Elle te bosse tout l’hiver.76
‘It [horse] works to you for the whole winter.’

As Raineri and Evola (2008:â•›13) point out, “the dative referent is not affected
by the consequences of the event. It is used instead to put the addressee in the
position of a virtual onlooker summoned to take part in the object of discourse”.
In fact, in the examples above, the dative referents do not witness the depicted
events and the Agents do not pertain to their PS. This discourse strategy may
be compared to what Kimps and Davidse (2008) label in another context the
‘conduciveness of illocutionary force’. This is to say that the dative represents
a hearer-oriented device by which speakers attempt to covertly ‘conduct’ their
interlocutors into �feeling affected in an imaginary scenario in order to increase
their interest towards the narration. An indication of how the grammaticaliza-
tion process of the ADC may have developed can be provided by contrasting
(F109) with:
(F111) Elle te boit des bons bibs 180ml.77
‘She [baby] drinks to you whole feed-bottles with 180 ml [of milk].’

In (F109), the dative referent is projected into a scenario in which someone known
to the writer drinks with remarkable rapidity. The dative conveys the sense of sur-
prise which the writer intends to share with their interlocutor. By contrast, in
(F111) the writer evaluates positively the fact that her interlocutor’s baby regu-
larly drinks abundant quantities of milk. Accordingly, the inference governing the
development of the ADC may be postulated as a mapping of positive/negative

. 〈http://www.tmax-mania.com/index.php? s=68515fdc6b0f5563fec4583a115495ce&sh

owtopic=14218&pid=286643&mode=threaded&start=〉
. 〈http://www.forum-equitation.com/index.php?showtopic=9694&st=15〉
. 〈http://www.etreenceinte.com/forum/viewtopic.php?p=2375590&sid=2f5dffd4b174494
7e47f437577cafb70〉
 From semantic roles to evaluative markers: The dative and affected possessors 

surprise onto positive/negative affectedness along a metonymization process (see

Figure 10).

NARRATIVE EMPATHETIC → METONYMIZATION → AGENTIVE DATIVE

CONSTRUCTION CONSTRUCTION
↓ ↓
INVITED NEGATIVE/POSITIVE INVOLVEMENT FACTUAL INVOLVEMENT

Figure 10.╇ Grammaticalization path from the Narrative Empathetic Construction to the ADC

The hypothesis that the Ethical Dative in German and the Narrative Empa-
thetic Dative Construction in French may have functioned as sources for the
ADC awaits corroboration from diachronic research. Should it be confirmed, we
would gain evidence for a counterexample to the canonical tenet of directionality,
according to which grammaticalization processes proceed from concrete domains
to abstract ones.
To conclude, the main features of the ADC in the three languages are sum-
marized in Table 10.

Table 10.╇ Major features of the ADC

Datives with Beneficial/Adversative Unlimited Restrictions
micro-construction (iii) reading domains on possessa

German √ - -
French √ - -
Italian √ √ -

7. Final remarks and proposals for further research

This paper focused on constructions relating to the dative of affectedness in Ger-

man, Italian and French. I showed that three out of the four postulated cross-
linguistic micro-contructions are available in all the languages under discussion.
This seems to provide some evidence for an account à la Croft (2003) based on
conceptual spaces. According to Croft, multifunctional markers that are cross-
linguistically available indicate a relation of similarity between various conceptual
situations anchored in the conceptual representation of speakers. However, this
model should be kept at the highly schematical level of the single micro-contruc-
tional senses. In fact, apart from Italian, which appears to have grammaticalized
 Domenico Niclot

the dative to a general marker of affectedness, German and French exhibit remark-
able constraints of different kinds:

–â•fi Semantic restrictions on the overall constructional sense (only negative

affectedness in German within ii).
–â•fi Semantic restriction on possessa (no animate Patients in German within i).
–â•fi Syntactic restrictions (incompatibility of the dative with reflexive verbs in
German and French within ii).
–â•fi Limited domains of occurrence (German and French within iii).

These restrictions help provide a more particularized account than König and
Haspelmath’s (1998) criterion according to which inalienable possessa are con-
ceptualized as affecting us more than others, which, as we saw, is not generally
tenable.
Future research should include both retrospective and prospective tasks. The
former relates to searching corpora for diachronic tokens that may either con-
firm or invalidate the grammaticalization paths proposed here. As for the latter,
three aspects appear particularly promising. Firstly, investigation on a vast cross-
linguistic sample could shed some light on the availability of the micro-construc-
tions posited here and on their behavior. This particularly holds for (iv), which
seems to be the least widespread one. Secondly, the use of the dative within (iii),
which is potentially open to expansion in German and French, should be moni-
tored. A regularly updated collexeme analysis à la Gries and Stefanowitsch (2004)
geared towards quantifying the co-occurrence of dative vs dativeless tokens
within specific domains may prove a profitable heuristic tool. Finally, recording
spontaneous utterances could provide valuable insights into the pragmatic rel-
evance of the dative for pragmatic purposes.
No matter how daring these objectives may turn out to be, the dative will
certainly reward us by remaining faithful to its Latin etymon, which means give.

Appendix-Abbreviations used in this paper

adc Agentive Dative Construction

cos Change of State
dat Dative
di Dative of Inaction
edc Empathetic Dative Construction
ps Personal Sphere
 From semantic roles to evaluative markers: The dative and affected possessors 

References78

Alexiadou, Artemis, Anagnostopoulou, Elena & Schäfer, Florian. 2006. The properties of anti-
causatives crosslinguistically. In Phases of Interpretation, Mara Frascarelli (ed.), 187–211.
Berlin: Mouton de Gruyter. DOI: 10.1515/9783110197723.4.187
Bybee, Joan, Perkins, Revere & Pagliuca, William. 1994. The Evolution of Grammar. Chicago IL:
The University of Chicago Press.
Bybee, Joan. 2007. Frequency of Usage and the Organization of Language. Oxford: OUP.
Colleman, Timothy. 2010. The benefactive semantic potential of ‘caused reception’constructions.
A case study of English, German, French, and Dutch. In Benefactives and Malefactives.
Case Studies and Typological Perspectives [Typological Studies in Language 92], Fernando
Zúňiga & Seppo Kittilä (eds), 219–243. Amsterdam: John Benjamins.
Croft, William. 1990. Possible verbs and event structure. In Meanings and Prototypes: Studies on
Linguistic Categorization, Savas L. Tsohatzidis (ed.), 48–73. London: Routledge.
Croft, William. 2003. Typology and Universals, 2nd edn. Cambridge: CUP.
Dabrowska, Ewa. 1997. Cognitive Semantics and the Polish Dative. Berlin: Mouton de Gruyter.
DOI: 10.1515/9783110814781
Dowty, David R. 1991. Thematic proto-roles and argument selection. Language 67(3): 547–619.
Fauconnier Gilles & Turner, Mark. 2002. The Way We Think. Conceptual Blending and the Mind’s
Hidden Complexities. New York NY: Basic Books.
Goldberg, Adele E. 1995. Constructions: A Construction Grammar Approach to Argument Struc-
ture. Berlin: Mouton de Gruyter.
Goldberg, Adele E. 2002. Surface generalizations: An alternative to alternations. Cognitive Lin-
guistics 13(4): 327–356. DOI: 10.1515/cogl.2002.022
Gries, Stefan T. & Stefanowitsch, Anatol. 2004. Extending collostructional analysis: A cor-
pus-based perspectives on ‘alternations’. International Journal of Corpus Linguistics 9(1):
97–129. DOI: 10.1075/ijcl.9.1.06gri
Gutzmann, Daniel. 2007. Eine Implikatur konventioneller Art: Der Dativus Ethicus. Linguist-
ische Berichte 211: 277–308.
Haspelmath, Martin. 1993. More on the typology of inchoative/causative verb alternations. In
Causatives and Transitivity [Studies in Language Companion Series 23], Bernard �Comrie &
Maria Polinsky (eds) 87–120. Amsterdam: John Benjamins.
Heine, Bernd, Claudi, Ulrike & Hünnemeyer, Friederike. 1991. Grammaticalization. Chicago IL:
University of Chicago Press.
Heine, Bernd. 1997. Possession: Cognitive Forces, Sources, and Grammaticalization. Cambridge:
CUP. DOI: 10.1017/CBO9780511581908
Hens, Gregor. 1997. Constructional semantics in German: The dative of inaction. American Jour-
nal of Germanic Linguistics and Literatures 9(2): 191–219. DOI: 10.1017/S1040820700002869
Hopper, Paul J. & Traugott, Elizabeth Closs. 2003. Grammaticalization, 2nd edn. Cambridge:
CUP. DOI: 10.1017/CBO9781139165525
Kallulli, Dalina. 2004. De-agentivised causers or non-active causative predications’ predications.
In The Pre-conference Proceedings of the Workshop “Demoting the Agent: Passive and Other
Voice-related Phenomena”. University of Oslo, 59–66.

. All the web-links cited here were available as of 31 December 2011.
 Domenico Niclot

Kaltenböck, Gunther, Heine, Bernd & Kuteva, Tania. 2011. On thetical grammar. Studies in
Language 35(4): 848–893. DOI: 10.1075/sl.35.4.03kal
Kay, Paul. 2005. Argument structure constructions and the argument-adjunct distinction. In
Grammatical Constructions: Back to the Roots [Constructional Approaches to Language
4], Mirjam Fried & Hans Boas (eds), 71–98. Amsterdam: John Benjamins. Cited after the
online version available at: 〈http://www.icsi.berkeley.edu/~kay/ASCs.pdf〉
Kemmer, Suzanne. 1993. Middle Voice [Typological Studies in Language 22]. Amsterdam: John
Benjamins. DOI: 10.1075/tsl.23
Kimps, Ditte & Davidse, Kristin. 2008. Illocutionary force and conduciveness in imperative
Â�constant polarity tag questions: A typology. Text & Talk 28(6): 699–722. DOI: 10.1515/

TEXT.2008.036
Kittilä, Seppo. 2005. Recipient-prominence vs beneficiary-prominence. Linguistic Typology 9(2):
269–97. DOI: 10.1515/lity.2005.9.2.269
König, Ekkehard & Haspelmath, Martin. 1998. Les constructions à possesseur externe dans les
langues d’Europe. In Actance et Valence dans les Langues d’Europe [Empirical Approaches
to Language Typology, EUROTYP 20–2], Jacque Feuillet (ed.), 525–606. Berlin: De
Gruyter. Cited after the online version available at: 〈http://www.eva.mpg.de/lingua/staff/
haspelmath/pdf/Possext1998.pdf〉
Langacker, Ronald W. 1991. Foundations of Cognitive Grammar, Vol. 2: Descriptive Application.
Stanford CA: Stanford University Press.
Langacker, Ronald. 2000. Grammar and Conceptualization. [Cognitive Linguistics Research 14].
Berlin: Mouton de Gruyter.
Levin, Beth & Rappaport-Hovav, Malka. 1995. Unaccusativity. At the Syntax-Lexical Semantics
Interface. Cambridge MA: The MIT Press.
Maldonado, Ricardo. 2002. Objective and subjective datives. Cognitive Linguistics 13(1): 1–65.
DOI: 10.1515/cogl.2002.010
McIntyre, Andrew. 2011. Silent possessive PPs in English Double Object (+Particle) Constructions.
Quoted after the online version available at: 〈http://www2.unine.ch/andrew.mcintyre/
page-300.htm〉
McKoon, Gail & Macfarland, Talke. 2000. Externally and internally caused change of state verbs.
Language 76(4): 833–858. DOI: 10.2307/417201
Mitkovska, Liljana. 2011. Competition between nominal possessive constructions and the pos-
sessive Dative in Macedonian. In The Grammar of Possessivity in South Slavic Languages:
Synchronic and Diachronic Perspectives, Nomachi Motoki (ed.), 83–109. Sapporo: Slavic
Research Center, Hokkaido University.
Nava, Fernando & Maldonado, Ricardo. 2005. Basic voice patterns in Tarascan (P’orhepecha). In
Language, Culture, and Mind, Michel Achard & Suzanne Kemmer (eds), 461–478. Stanford
CA: CSLI. Cited after the online version available at: 〈http://ricardomaldonado.weebly.
com/uploads/2/7/6/3/2763410/nava_maldonado.pdf〉
O’ Connor, Mary Catherine. 2007. External possession and utterance interpretation: A crosslin-
guistic exploration. Linguistics 45(3): 577–613.
Payne Doris L. & Barshi, Immanuel (eds). 1999. External Possession [Typological Studies in
Language 39]. Amsterdam: John Benjamins. DOI: 10.1075/tsl.39
Raineri, Sophie & Evola, Vito. 2008. A Construction Grammar Analysis of the Empathetic
Dative Construction in French and Italian. Ms.
 From semantic roles to evaluative markers: The dative and affected possessors 

Šarić, Ljiljana. 2002. On the semantics of the dative of possession in the Slavic languages: An
analysis on the basis of Russian, Polish, Croatian/Serbian and Slovenian examples. �Glossos 3.
Slavic and East European Language Resource Center, Duke University. Cited after the online
version available at: 〈http://folk.uio.no/kjetilrh/slav4114/2011-02-11/saric_2002_marked-
up.pdf〉
Schäfer, Florian. 2007. On the Nature of Anticausative Morphology: External Arguments
in Change-of State Contexts. Ph.D. dissertation, University of Stuttgart. Cited after the
online version at: 〈http://elib.unistuttgart.de/opus/volltexte/2007/3307/pdf/diss_schaefer_
version_3.pdf〉
Smith, Carlota S. 1970. ‘Jespersen’s ‘move and change’ class and causative verbs in English. In
Linguistic and Literary Studies in Honor of Archibald A. Hill, Mohammad Ali Jazayery,
Edgar Polomé & Werner Winter (eds), 101–110. The Hague: Mouton.
Traugott, Elizabeth C. & Dasher, Richard B. 2002. Regularity in Semantic Change. Cambridge:
CUP.
Tuggy, David. 1980. Ethical datives and possessor omission sí, possessor ascension no! Workpa-
pers of the SIL 24: 97–141. University of North Dakota.
Van Valin Jr., Robert D. & Wilkins, David P. 1996. The case for ‘effector’: Case roles, agents,
and agency revisited. In Grammatical Constructions, Masayoshi Shibatani & Sandra A.
Â�Thompson (eds), 289–322. Oxford: Clarendon Press.
 Author index

A D Handschuh, Corinna 32
Alexiadou, Artemis 297–299 Dąbrowska, Ewa 209 Haspelmath, Martin 3–4,
Amritavalli, R. 113 Dahl, Eystein 5, 14–15, 17, 91, 9–10, 16, 27, 76, 83, 86, 88,
Anderson, Lloyd B. 9, 80–81, 181, 184, 190, 202 106, 120, 140, 152, 159, 171,
100 Dahl, Östen 5, 14–15, 17, 91, 176, 191, 216–217, 223, 242,
Arkhipov, Aleksandr 216, 219, 181, 184, 190, 202 246–247, 287, 290–293, 298,
228 Daniel, Michael A. 5, 14–15, 312, 315, 322
Authier, Gilles 205, 220–221, 17, 137–138, 188, 205, 211, Havers, Wilhelm 120
223, 231 216–217, 221, 225–226 Heine, Bernd 10, 13, 69–79, 81,
de Swart, Peter 144 85–94, 99, 101–102, 106–109,
B Delbrück, Berthold 121, 127, 113–115, 127–130, 233, 281, 288
Bandelt, Hans-Jürgen, 30 135, 195 Hens, Gregor 273, 296–298,
Bennett, David C. 106 Dik, Simon C. 102, 187 300, 305–308
Bentley, John R. 135 Dixon, R.M.W 81, 87, 127, 177 Hettrich, Heinrich 131–132, 134
Benvenuto, Maria Carmela 132 Dowty, David R. 2, 23, 182, Hock, Hans H. 190, 195, 197
Bickel, Balthasar 113, 188, 191 206, 313 Hofmann, J. B. 105, 109
Blake, Barry J. 2, 23, 75, 81–82, Dress, Andreas 30 Hopper, Paul J. 281–282
104, 127 Huumo, Tuomas 99, 112–113,
Blake, Frank R. 2, 23, 75, E 116, 120, 122, 125, 135, 157, 167
81–82, 104, 127 Endruschat, Annette 74, 76
Blansitt, Edward Jr. 107, 169 Erelt, Mati 112–113 I
Bloomfield, Leonard 102 Evans, Vyvyan 103, 106 Itkonen, Erkki 116
Bokarev, Anatolij 219, 222 Evola, Vito 16, 271–273, Ito, Shinya 9, 74, 77, 88
Bornkessel, Ina 2 284–286, 310, 315, 319–320
Bossong, Georg 188, 190 J
Bryant, David 30 F Janda, Laura 4, 120, 209, 233
Bybee, Joan 134, 281, 290 Fauconnier, Gilles 303 Johnson, Mark 99, 101, 124, 127
Fillmore, Charles J. 1, 14, 23, Joki, Aulis J. 116
C 206–207, 234, 246 Jung, Hakyung 80–81
Charachidze, Georges 229 Foley, William A. 2
Comrie Bernard 9, 15–16, 27, Forker, Diana 33, 205, 218, K
137, 152, 154, 156, 171, 175–177, 221–224 Kabata, Kaori 107, 118, 176
206–208, 231, 242–244, 264, Fried, Miriam 186 Kailuweit, Rolf 2
266 Kaufman, Leonard 30
Corbin Danielle 253, 257 G Kenesei, István 111
Coseriu, Eugenio 125 Ganenkov, Dmitry S. 137–138, Khalilova, Zaira 205, 215, 218,
Crazzolara, J P. 27 216–217, 221, 223, 225, 227, 230 222, 224
Creissel, Denis 9, 110 Goldberg, Adele E. 118, 209, Kibrik, Aleksandr 212, 214,
Crespo, Emilio 242, 246, 249, 271–276, 279, 284–285 217–218, 221–223, 227, 230–231
264 Grimm, Scott 182 Kittilä, Seppo 1, 5–7, 9, 14–16,
Croft, William 10–11, 70, Gruber, Jeffrey 1 18, 115–116, 118, 151–152,
100–102, 111, 123, 126, 128, Grünthal, Riho 75, 125 154–155, 158, 169, 272,
130, 249, 298, 321 Gutzmann, Daniel 318 275–276, 279–280
Cysouw, Michael 4–6, 11–12, Klein, Jared 191
16–17, 23–24, 29, 33, 221, H König, Ekkehard 16, 287,
223, 247 Haig, Geoffrey 132, 187, 191 290–293, 312, 315, 322
 Author index

Kornfilt, Jaklin 107, 127, 129 Musaev, Magomed-Salam 213, Rodriguez Aristar, Anthony
Kuryłowicz, Jerzy 5, 119 223 119
Kuteva, Tania 74–78, 85, 87, Mutalov, Rasul 215, 218, 223, Rousseeuw, Peter J 30
106–107, 109, 114, 128, 130, 228
233 S
N Saitou, Naruya 30
L Næss, Åshild 186 Schäfer, Florian 297–298, 302,
Lakoff, George 99, 101, 113, 119, Namer, Fiammetta 253 304, 311
124, 127–128 Narrog, Heiko 1, 5, 9–11, 13–14, Schlesinger I. M. 1, 249
Lander, Yuri 211 17–18, 69, 72, 74, 77–78, Schmidtke-Bode, Karsten 117
Langacker, Ronald W. 101, 125, 87–88, 99, 101–102, 124, 135, Schwyzer, Eduard 74–75, 134
280, 288, 304–305 181, 265 Shibatani, Masayoshi 82
LaPolla, Randy J. 2, 111, 206, Nei, Masatoshi 30 Shin, Yong-Min 2, 17, 124–125
208, 225, 231–232, 234, 249, Newmeyer, Frederick J 1 Siewierska, Anna 81–82
253 Nichols, Johanna 205, 222, 230 Smith, Michael B. 186, 297
Lauerma, Petri 157, 175–176 Nikiforidou, Kiki 4, 114, 128, Spano, Marianna 106
Lehman, Christian 79, 81, 87, 131 Stolz, Thomas 76, 84, 87, 100,
124, 132, 186 Nikitina, Tatiana 106 103, 109, 135–137
Lestrade, Sander 10 Nishimura, Yoshiki 1 Stroh, Cornelia 76
Leumann, Manu 105, 109 Noonan, Michael 13, 75, 104 Sumbatova, Nina 215, 218,
Levin, Beth 2, 154, 182, 206, 223, 228
234, 251, 297 O Szantyr, Anton 105, 109
Luraghi, Silvia 1, 4–5, O’ Connor, Mary Catherine
9–10, 12–13, 17, 71, 74–77, 288–289 T
81, 83, 85–86, 91–92, 94, Ojutkangas, Krista 157, 167 Talibov, Bukar 223
99, 101, 105–107, 109, Tatevosov, Sergej 205
111–112, 115, 117–120, 122–123, P Taylor, John R. 107
125–127, 129, 131, 134–135, Palancar, Enrico 74–77, 79, Thompson, Sandra 242–244,
144, 171, 181, 247, 249, 81–83, 85, 249 264, 266
264–265 Pantcheva, Marina 104–105 Thornburg, Linda L. 118, 251,
Luschütsky, Hans Ch. 242 Panther, Klaus-Uwe 118, 251, 254
Lyons, John 106 254 Toyota, Junichi 83–84, 135
Parodi, Claudia 4 Traugott, Elizabeth C. 281–282,
M Pasquereau, Jeremy 222 288
Magometov, Aleksandr 217, Pennacchietti, Fabrizio 99, Tuggy, David 125, 278
223 122 Turner, Mark 128, 303
Maisak, Timur 205 Plénat, Marc 253 Tyler, Andrea 103
Makhmudova, Svetlana 223 Pompeo, Flavia 132
Malchukov, Andrej 9, 88, Primus, Beatrice 2, 182 U
207, 234 Pustejovsky, Jack 251 Urdze, Aina 76
Maldonado, Ricardo 280, 284, Pütz, Martin 101
288, 303 V
Matisoff, James A. 113 R van den Berg, Helma. 15, 137,
McIntyre, Andrew 275 Radden, Günter 1, 122 219, 222
Mel’čuk, Igor Aleksandrovič Rainer, Franz 242, 251–252, van der Auwera, Johan 11
244, 264, 266 266 Van Valin, Robert D Jr. 2, 23,
Metslang, Helle 112–113 Raineri, Sophie 16, 271–273, 111, 182, 185–187, 206, 208,
Molochieva, Zarina 205, 218, 284–286, 310, 315, 319–320 225, 231–232, 234, 249, 253,
222, 224, 226 Rappaport Hovav, Malka 2, 297
Moulton, Vincent 30 154, 182, 206, 234, 298 Vandeloise, Claude 103
Mounole, Céline 9, 110, 152, Reddy, Michael 138 Västi, Katja 1, 6, 9
156, 171, 175 Rice, Sally 107, 118, 176 Verbeke, Saartje 84, 156
 Author index 

Verhoeven, Elisabeth 112, 188, Wälchli, Bernhard 11, 24 Ylikoski, Jussi 1, 6, 9, 152,
193–194, 197 Wenzel, Heinrich 135 154–155, 169
Villoing, Florence 253 Wiemer, Björn 74, 81, 86
Vovin, Alexander 135 Z
Y Žirkov, Lev 214–215, 220, 223
W Yamaguchi, Kazuyuki 73–74, Zlatev, Jordan 101
Wackernagel, Jacob 264 76–77 Zúñiga, Fernando 115
 Language index

A Drehu 27–28, 33, 41, 68 Germanic 28, 75, 107, 109–110,

Afro-Asiatic 28, 111 Dutch 75, 251 116–117
Agul 215–217, 223, 227, 228, 230 Gikuyu, 244
Akha 27–28, 33, 41, 67 E Godoberi 218, 222, 229
Akhvakh 76, 87, 217, 222 East Caucasian 15, 205, 207, Greek 107, 116, 119
Akusha 223 210–218, 221–235 Ancient 12, 14, 75–77, 81,
Albanian 27–28, 32, 41, 67, Eastern Cushitic 28 83, 86, 107, 109, 110,
74, 76 English 3, 9, 16, 26–28, 31, 37, 115–117, 119, 123, 129, 134,
Altai 28, 34, 41, 67 74–77, 81, 83–84, 91–92, 144, 241–242, 246–249,
Altaic 28, 212, 215, 217–218, 103, 109–111, 114, 116, 118, 253–254, 257, 262–266
220, 222, 226, 228, 229 120–122, 128, 140, 209–210, Homeric 115, 202, 203, 263,
Andi 218, 222, 229 223–224, 233–234, 242, 265
Andic 137 245, 248, 251, 254, 262, Classical 116, 119
Arabic 83, 107 264, 271–273, 276–278, Byzantine 116
Archi 212, 216–218, 223, 228, 280, 315, 319 Modern 109, 116, 129
230–231 Eskimo-Aleut 28 Greenlandic 27–28, 34, 41, 67,
Austronesian 28, 84, 105 Estonian 27–28, 41, 67, 109, 83, 86
Avar 211, 214, 220, 222, 112–113, 151–152, 175
228–230 H
Aymara 27–28, 34, 41, 67 F Hebrew
Azerbaijani 27–28, 34, 41, 68, Faroese 27–28, 32–33, 41, 67 Israeli 105
220 Fijian 105 Biblical 115, 122
Finnic 28, 109, 112, 119, 125, Hinuq 213–214, 217, 222, 224
B 129, 213 Hungarian 82, 111
Bagvalal 214–218, 221–222, Finnish 7, 15–16, 18, 81, 109, Hunzib 213, 218–220, 222
227, 231 112–113, 115–120, 122, 131,
Bats (Tsova-Tush) 222 134–135, 138, 140, 151–153, I
Bezhta 213, 218, 222 155, 157–159, 174–177 Ingush 212, 222, 230
Bolivian Quechua 108 French 6, 10–11, 16, 74–75, Irish 27–28, 41, 67
Botlikh 220, 222, 229 82, 105–106, 108, 110, 113, Italian 6–7, 16, 105, 109, 120,
Breton 108 118, 123, 129, 242, 252–253, 253, 271–273, 277–279,
Budukh 217, 220–221, 223 257, 271–273, 277–279, 284–285, 291–293,
Burmese-Lolo 28 284–285, 287, 291–293, 295–297, 300, 302–308,
295–296, 300, 310–312, 312, 315–318, 321
C 315–322 Itsari 215, 217–218, 223, 228
Celtic 28, 108
Central Khoisan 28 G J
Chamalal 218–220, 222, 229 Georgian 129 Japanese 18, 78, 80, 82–83, 86,
Chechen 212, 217–218, 222, German 6–7, 16, 27, 31, 37, 94, 135
224, 226 40–41, 75–77, 85, 109–110,
Czech 82, 120, 209 113–114, 116, 121, 125, K
132, 209–210, 271–273, Kabardian, 211
D 277–279, 291–296, 300, Kannada 74, 113
Daghestani 188 302, 304–312, 315–322 Karata 217, 222
Daghestanian 137–138 Germanic 28, 75, 107, 109–110, Khinalug 211, 213–216,
Diola 242 116–117 218–220, 223, 226–228, 230
 Language index

Khoekhoe 27–28, 41, 68 N Slavic 82, 121, 136, 207, 316

Khoisan 28 Nahuatl 251 Sonsorol-Tobi 105
Khwarshi 213, 215, 217–218, Nias 27–28, 33, 41, 68 Spanish 77, 105, 109, 241–242,
220, 222, 224 Nilo-Saharan 28 248, 250–252, 284, 289
Korean 27–28, 41, 68 Sundic 28
Kryz 217, 220–221, 223, 231 O Swahili 136
Oceanic 28
L Oromo 27–28, 34, 41, 68 T
Lak 132, 211, 214–216, 220, 223, Tabasaran 223, 230
226, 228–230 P Thai 113
Latin 8, 9, 75, 76, 104, 109, Persian 132 Tibeto-Burman 111, 113
110, 116, 117, 119, 121, 123, Polish 75, 209, 284 Tindi 218, 222, 229
129, 131, 134, 143, 144, Polish 75, 209, 284 Tsakhur 137–138, 212, 217, 221,
188 Portuguese 136 223
Early 117, 202 Tsez 213, 217, 222
Late 104 R Turkic 28, 86, 230
Lezgian 76, 216–217, 223, Romance 75–76, 104, 107–108, Turkish 74, 107, 117, 127, 129, 244
227–228, 230 129, 144, 233, 251, 273
Lezgic 137, 211–212, 215–218, Russian 75, 80–82, 128, 133, U
220, 221, 225–226, 230, 136, 205, 208–210 Udi 212, 217, 223, 225, 227–228,
232, 233 Rutul 217, 221, 223 230
Uralic 28, 111, 169
M S
Ma’di 27–28, 33 Sanskrit 121 Y
Maa 251–252 Vedic 5, 15, 17, 127, 135, 181, Yucatec Maya 188
Mehweb 223, 226 202
Middle English 74 Classical 136 Z
Moru-Ma’di 28 Sino-Tibetan 28 Zulu 243
 Subject index

A Agent, passive 19, 80–1, 83, B

ablative 5, 8, 75–76, 82, 91, 87–9, 91–3, 95–6, 99–100, Benchmark of comparison 215
102, 105–106, 114, 120, 131, 113, 129, 138, 141–2, 149, Beneficiary 7, 15, 75–77, 100,
143–144, 191, 208, 223, 235 271 106–107, 115–120, 123,
ablative-locative transfer 105 Agent, unintentional 215 128–135, 137–140, 142–144,
absolutive 5, 12, 28, 32, 80, 82, agentivity 182, 249, 313 158, 185–186, 194, 271–272,
212 alignment 275–280, 283, 288
accompaniment 244, 264 ergative(-absolutive) 28
accusative 5, 7, 12, 15, 27–28, active-inactive 28 C
31, 34, 80, 82, 84, 86–88, indirective 234 case 2–9, 12–18, 23–27,
90, 93, 105, 121, 127, 129, marked nominative 32–34, 40, 69–94, 99–100,
131–132, 150, 159, 178, (-accusative) 28, 31 103–110, 112–123, 125–128,
181–184, 188, 190–192, neutral 28, 32 131–138, 140–143, 151–153,
194–197, 201–202 nominative (-accusative) 157–161, 163–178, 181–186,
Addressee 5, 15, 18, 106, 138, 184, 194 188–192, 194–197, 201–202,
194, 205, 207, 209–211, secundative 234–235 205–212, 216, 220–221,
217–222, 224–235, 320 allative 15, 102, 105–106, 108, 223–224, 229–233, 235,
Addressee of speech verbs 113, 115–121, 123, 129–130, 241–243, 245–246, 248,
209, 211, 220, 229 138, 140, 150–151, 153, 251, 257–259, 262, 266,
addressive 220–221, 223, 235 157–158, 161, 163–167, 273–275, 277–278, 286,
adpositions 3, 5–9, 15, 18, 170–178, 205, 213, 222–224, 289, 294, 297, 300–301,
26–27, 84, 99, 103, 109, 226, 235 304–305, 307, 312, 314,
114–115, 119, 121, 123, alternation, causative- 317, 319
140–141, 151–152, 157–158, inchoative 298 case roles 69[S4]
160–163, 168–177, 243, 266 animacy 12, 16, 113, 119, 151, causal chain model 70
affectedness 5, 16, 186, 271–273, 154–155, 159, 171, 176, 249 causally affected 182, 184, 186
277, 279–290, 293–294, animate 15, 71, 83, 87, 91, 121, causation 82, 101, 126, 128,
296–297, 300–303, 305, 128, 152, 154–155, 158–163, 130, 182, 273, 297–299,
308–310, 312, 314, 316–319, 165–169, 171, 175–177, 303–305, 307–308
321–322 194–195, 205–206, 211, 215, external 273, 297–299
affective 137–138, 212, 221, 232–235, 249, 251–252, 272, internal 300, 307
223, 235 274, 282, 286–287, 293, causative 81–82, 86, 102, 184,
Agent 1, 3, 9, 13–15, 17, 23, 27, 296–297, 299, 308, 311, 195–197, 201, 235, 298–299,
31–33, 72, 74–75, 77, 79–87, 313, 322 302, 310
90–94, 100, 106–107, 111, anthropocentric concepts 69, causative verb 184, 195–196,
121–123, 127–128, 131–137, 71 201
139–143, 161, 169, 181–184, apud 109[S3] causative-inchoative,
186–187, 191, 194, 198, argument 1–2, 7, 15, 81, 84, alternation 298, 302
201–202, 208–209, 215, 90, 119, 155–156, 161, Cause 8, 70, 76–78, 82–86,
232, 241–242, 244–245, 181–187, 189–192, 194–198, 116–117, 127–130, 140–142,
249, 251, 254, 257–259, 206–207, 211, 218, 226–227, 162, 187, 189, 201, 249–250,
262–263, 266, 282, 287, 231, 235, 253, 274, 280, 298, 253, 262, 264, 283,
292, 295, 303, 313 304–305 298–300, 302
Agent, ergative 74, 79–82, aspect 27, 79, 84, 156, 274, Causee 81–82, 86, 90, 135, 215,
84–7, 90, 92, 95–6, 271 285–286, 299, 303, 315 220
 Subject index

change of state 1, 103, 111, 141, 139–141, 151–164, 166–177, 157–158, 169, 205, 210–211,
182–184, 286, 322 210, 230, 250 225–226, 232–235, 241,
change-of-state verbs 296 directional 87, 90–91, 100, 266, 282
clustering, flat 12, 30 102–103, 108, 117, 119, 127, grammaticalization 2, 7, 13,
clustering, hierarchical 30 130, 138, 141, 143, 151, 158, 16–18, 69–70, 72, 78–79,
cognition verb 188, 192, 199 171, 174, 176, 222 83, 85, 87–91, 93–94,
comitative 13, 74, 76–77, 87, directionality of change 71, 102, 124, 127, 243, 259,
99–101, 106, 109–110, 86, 89 271–273, 281–283, 287–288,
121–122, 124, 126–128, ditransitive 185, 210, 215, 290, 293, 295, 305, 307,
135–137, 139–140, 142–143, 233–234, 272–273, 275–276, 309–310, 314, 318–322
212 278 grammaticalization chain 13,
companion 13, 74, 76–77, 87, ditransitive construction 185, 69, 72, 79, 88, 90
99–100, 108, 121, 124–126, 233, 272–273, 275–276, 278 grammatically relevant
135, 144, 265 semantic role
condition 70, 78, 88, 252, 276, E
285, 287 Effector 249, 297–298, I
consonant mutation 27 302–303, 309–310, 312 illative 120, 151–153, 157–158,
construal 112, 208, 210, emotion verbs 186, 188, 161, 163–167, 170, 172–174,
223, 225, 227, 229–230, 194–197, 199, 201–202 176–178
233–234, 305, 313 empathetic dative construction impersonal verb 196
Construction Grammar 16, 18, 284, 320–322 implicature, conventional 318
271–273 empathy 289, 293, 300, 312 inanimate 12–14, 69, 71, 76–77,
cont 212–215, 218, 221–224, 235 entrenchment 291, 296, 309 85, 87, 91, 102, 110, 116,
containers 14, 103, 111–112, essive 213–216, 218–220, 121–122, 128, 137, 139, 141,
124–125, 138, 142, 144–145, 222–225, 229–230, 232, 235 152, 154–155, 159–162, 164,
165, 176, 253–254, 256, 260, evaluation 280, 285, 287–289 171, 206, 215, 249–251, 253,
262, 264, 266 exemplar semantics 24 257, 280, 286, 299, 311, 313
contextual role 25, 28, 31, 42 Experiencer 3, 5, 14–15, 17, inanimate relations 14, 71, 137
111–113, 133, 137–139, Instrument 1, 8–9, 12–14,
D 141–143, 181–182, 186–198, 75–77, 81–84, 91–92, 94,
dative 5–7, 10, 12, 15–16, 33, 69, 201–202, 212, 215, 309 99–100, 121–128, 135–137,
75, 77, 82–86, 89–90, 94, experiential predicate 112, 139, 141–144, 208, 232,
106, 108, 111–113, 115–116, 197, 201 241–242, 244–245,
119–121, 132–135, 138, 140, 248–255, 258–260, 263–267
143, 150, 178, 181–182, F instrumental 13–14, 71, 74–75,
185–186, 188, 190–191, focus marking 27 77, 79, 81–84, 87, 91–93,
193–194, 202, 205, 207, Force 85, 208, 232, 249, 253, 121–122, 124–125, 127–128,
209, 211–212, 214–218, 257, 261–262, 320 135–136, 141, 150, 184, 191,
220–221, 223–230, 232–235, foregrounding 273, 303 208, 212, 241, 251–253,
245, 265, 271–273, 277–297, formative 25 257–258, 265–266
300–322 intentionality 208, 249, 272,
dative of inaction 322 G 286, 303–304
dativus ethicus 318 generalization 71, 73, 90, 125, Intermediary 121–126, 144, 242,
dativus judicantis 263, 266, 318 141, 143, 158, 174, 231–232, 248–249, 252, 264, 267
deverbal suffix, 241, 248, 251, 276, 281–282, 285, 293, 301, inverted construction 201
253–254 309–310, 319 Invited Inferencing Theory 281
differential marking 9, 315–317 generalized construction 202 isomorphism hypothesis 24,
dimensionality 10, 154, 164, genitive 77, 79–82, 113–114, 123, 28
176 131–132, 134, 143, 150, 178,
Direction 13, 15–16, 18, 69, 75, 188, 191, 198, 212, 235, 264 K
88, 90, 102–107, 109–110, Goal 2, 5, 15, 26, 71, 74–76, 83, knowledge, encyclopaedic
113, 115–118, 120–121, 86, 88, 102–104, 108–109, 299, 304
123–127, 132–133, 137, 130, 143, 151–152, 154, kriging 32
 Subject index 

L N 139–142, 164, 226, 243, 273,

language typology 24 neighbornet 30, 33–34 288, 303–304, 309
lative 15, 205, 213–230, 232–235 nominative 3, 5, 7, 12, 15,
Lexical entailments 182, 27–28, 31, 34–35, 79–80, R
186–187 82, 85–87, 92–93, 127, Reason 15–17, 34, 76–77, 88,
localization 212–215, 217, 220, 131–132, 150, 181–185, 129–133, 138–139, 141,
224, 229, 235–236 188–198, 201–202, 207, 155–156, 163, 205, 229,
Location 7–8, 13, 27, 74–76, 212–213, 235, 254 243–244, 247, 259, 267,
83, 101–114, 120–127, 130, non-argument 186 275
133, 137, 140–141, 143–144, Recipient 3, 6–7, 12, 15, 77, 81,
154–157, 161–162, 168–170, O 83, 85, 89–90, 94, 100,
173–174, 176–177, 212, orientation 212–213, 229, 275, 106–107, 110–113, 115–120,
234, 241–242, 245, 248, 284 123, 128–130, 132–135,
250–254, 256, 258, 260, 137–143, 155, 158–161, 163,
263–267, 307–308 P 168, 181–182, 185–186,
parallel text 23, 25, 35 191–192, 194, 198, 205,
M partitioning around medoids 209–211, 215, 218, 224–228,
macroroles 23, 231 (PAM) 230–235, 272, 274, 276–277,
Maleficiary 119–120, 194, Path 13, 75–76, 87, 102–103, 279–280, 283, 288
271–272, 276–277, 121–127, 130, 138, 140, 272, temporary 232–234, 236,
279–280, 283, 288, 297 281–283, 295, 305, 309–310, 238
Manner 9, 16, 18, 70, 77, 122, 318–319, 321 role(s), context-specific 12,
126, 128, 242–244, 247, 293 Patient 1, 3, 15, 17, 32–34, 23, 29
markedness 151–153, 157, 169, 37, 88–89, 102, 111, 137,
171, 174, 176–177 181–184, 186–187, 191, S
material 36, 77, 127, 144, 198, 202, 235, 244–245, SAE 205, 211, 217–218, 223–226,
297–299 285, 287, 298–299, 303, 229–230, 232–233
meaning extension 99 309, 313 salience 167, 173, 183
Means 14, 248–249, 262 Pearson correlation 33 semantic change 70, 72, 91,
metaphor 13–14, 17–18, 69–70, perception verb 194, 200, 205 101, 257
72, 77, 90–91, 93–94, perlative 86, 103, 118, 123, 130, semantic function 4, 89, 193,
99–100, 102, 106–107, 140, 245 248, 252, 270
110–114, 116, 118–119, 121, personal sphere 16, 272, 286, semantic map 10, 12, 14,
124–127, 130, 132, 134–135, 322 23–24, 29, 88, 99, 182, 187,
138–139, 141–142, 205, place names 156, 162, 171–174, 198–201, 246–247, 263,
208–211, 224, 227–230, 176–177, 221 265
233–235, 241, 267, 281, 283, possession semantic role 2–5, 9, 13–15,
315–316 external 137, 188, 194, 199 17–18, 23, 74, 100, 102–103,
Metaphorical Extension internal 285 109, 111, 113, 117, 122–124,
Approach 234, 281 Possessor 75–76, 79–81, 83, 127, 130, 138–142, 151–152,
metonymization 281–282, 309, 106–109, 113–114, 118–120, 154, 181–182, 184, 187,
319, 321 122–123, 131–143, 215, 205–208, 211, 231–233, 241,
metonymy 72, 91, 99–100, 278–279, 281, 303, 314 244–251, 253, 261–262, 267,
118, 127–128, 141–142, 241, omission 278–279, 314 288[S7]
259, 267 postposition(s) 107, 116, 152, sentience 113, 182, 184
motion 8, 74, 103, 109, 117, 119, 153, 157–159, 162–163, 165 source 5, 8–9, 13, 31, 73–77, 79,
127, 130, 138, 141, 143, 152, Price 264, 276 82–86, 89, 93, 100–110,
154–158, 161–163, 167, 169, proto-role(s) 8, 25, 29–31, 42, 112–114, 118, 120–121,
176–177, 209–210, 225, 187, 209, 242, 329 123–128, 130–133, 137,
227, 274 prototypicality 151–152, 159, 139–144, 174, 176–177,
multidimensional scaling 177, 247 207, 209, 211, 222–223,
(MDS) 29 purpose 15, 26, 30, 76, 78, 227, 231, 250, 281,
multifunctionality, 242, 248 85–86, 115–117, 128–130, 287–288, 308
 Subject index

spatial relations 8–9, 12–13, syncretism 75, 84, 101, 104, V

15, 71, 100–104, 106, 211, 219 verb of bodily sensation 112,
109–110, 114, 118, 121, thematic relation(s) 1, 23 138, 143, 188–189
123–124, 126, 130, 137–144, voice 141, 145, 190, 240, 309
153 T volition 112, 182, 188–189, 192,
split decomposition 12, 30 Theme 3, 154–155, 159–163, 195, 199, 303, 313
stimulus 84, 181, 188, 190–198, 165–167, 169, 205, 234, volition verb 188, 199
201–202, 252, 257 252
sub 102, 137, 212, 217, 219, 236, W
247 U word formation 14, 17, 125,
super 212–216, 218, 222, undergoer 23, 27, 31, 208, 234 241–248, 250–251, 253–254,
227–228, 236 unidirectionality 69, 101, 129 258, 264–267