The evolution of c-structure: prepositions

and PPs from Indo-European to Romance*

NIGEL VINCENT

Abstract

Within X-bar theory prepositions are standardly taken to constitute one of

the core lexical categories along with verbs, nouns, and adjectives definable
by the features [±V, ±N]. Synchronically, however they share properties
with both lexical and functional categories, while diachronically they are
usually the outcome of processes of grammaticalization affecting true lexical
categories such as verbs, nouns, and adjectives. The present paper analyzes
in detail the stages whereby prepositions in Latin and Romance evolve from
earlier adverbial particles and concludes that this diachronic trajectory is
best modeled within a framework that separates the relational structure of
a PP from its configurational representation. It is argued that lexical-
functional grammar, with its systematic distinction between f-structure and
c-structure, provides a better account of these changes than conventional
X-bar theory. In conclusion a possible role for optimality theory within a
formal account of syntactic change is briefly sketched.

1. Introduction

Research in historical syntax sometimes seems in danger of polarization

into two camps, each making diametrically opposed assumptions on a
number of issues including those pertinent to the present paper, namely
word order and constituency. On the one hand, generative historical
syntax (cf. recent conference collections such as Battye and Roberts 1995;
van Kemenade and Vincent 1997) follows synchronic work in that frame-
work in assuming configurational structure as a property of UG.
Languages may differ in overt linear order in virtue of different movement
possibilities and/or due to different settings of the directionality parameter
(cf. Kayne 1994 for a proposal to eliminate the latter in favor of the
former and Roberts 1997a for an exploration of the diachronic

Linguistics 37–6 (1999), 1111–1153 0024–3949/99/0037–1111

© Walter de Gruyter
1112 N. Vincent

consequences thereof ), but all languages have some underlying order

(indeed on Kayne’s proposal they all have the same underlying order), and
that order is in turn a reflection of configurational structure (cf. Kayne’s
‘‘linear correspondence axiom’’). By contrast, most work on grammaticali-
zation (Hopper and Traugott 1993: 50ff.) seems to treat word order as just
that — the linear arrangement of minimal syntactic elements — with no
particular regard to the principles or mechanisms (i.e. the ‘‘grammar’’) that
determine such order beyond an assumption of simple concatenation driven
by either syntactic or pragmatic considerations or both. To put it (perhaps
too) simply, generative grammar seems to err in the direction of assuming
too much in the way of linear order/constituency, while proponents of
grammaticalization assume too little.
A third possibility lies in what Bresnan (forthcoming a) calls ‘‘parallel
correspondence theories.’’ Such approaches treat grammars as formal
objects for exactly the same good reasons as inspired the original
Chomskyan enterprise (Chomsky 1957: 5) but do not require that the
representation of grammatical relations (subject, object, complement, speci-
fier, etc.) be couched in the same terms as the representation of overt
linguistic structure (case marking, work order, etc). One such theory is
LFG, and we will argue that it has much to offer in reconciling the goals
for diachronic syntax of the two approaches sketched in the previous para-
graph. On the one hand, it provides a formalism within which issues can
be formulated and addressed with the precision and explicitness that we
have rightly come to expect in morphosyntactic research. On the other
hand, the formalism does not inherently privilege constituency and linear
order, which are instead seen as one particular type of linguistic realization
(cf. Matthews 1981: chapter 12) of abstract grammatical relations rather
than as the universal metalinguistic vehicle of those relations. The syntacti-
cally relevant properties of a given sentence — henceforth f(unctional )-
structure — are instead expressed in a feature-based notation subject to the
basic principle of unification. This principle determines how smaller
f-structures are combined into larger ones while remaining properly neutral
on the question of how they are expressed. Linguistic expression is deter-
mined instead by the correspondences between f-structures and c-structures,
where the latter may involve configurations, morphological patterns of case
and agreement, or any combination thereof.
A key feature then of the architecture of LFG is the separation of
f-structure and c-structure ( Kaplan and Bresnan 1982).1 For any given
language or construction, the presence of constituency2 and its frequent
concomitant linear order can be seen for what it is: one of the ways —
along with case/prepositions (dependent marking) and agreement (head
marking) — that languages use to realize their underlying grammatical
 The evolution of c-structure 1113

relations. Languages, and individual constructions within languages, may

then differ in the extent of the evidence they offer for constituency just
as they differ in the richness of their case and agreement systems. The
diachronic consequence of this is clear: just as the histories of languages
provide many examples of the rise and fall of case and agreement systems,
we should also look for the possibility that constituency may be the
product and/or victim of the processes of grammatical change. To the
extent that such situations arise, and can be adequately modeled, LFG
can be seen to retain the benefits of a formal approach to syntax while
still characterizing the types of change that were the original impulse
behind work in grammaticalization. In the present paper, we shall exam-
ine one such instance in detail — the changing relations between cases
and prepositions in a historical cross-section from Indo-European to
modern Romance — before reflecting on the theoretical consequences of
an LFG-based approach to the study of morphosyntactic change. Of
particular interest in this connection are the recent moves to inter-
pret LFG in the light of optimality theory (Bresnan forthcoming a,
forthcoming b, forthcoming c).

2. The problem of prepositions

Since the classic article by Jackendoff (1973), it has been standard practice
within generative grammar to accord prepositions the status of a major
category alongside nouns, verbs, and adjectives. Yet the question is by
no means clearcut. On the one hand the core prepositions of a language
seem to form a small enumerable class of items much like determiners
or complementizers; on the other hand if we include so-called complex
prepositions like in light of, by virtue of, in common with, etc., then the
border with NPs is no longer clear (cf. also Lehmann 1998). The concept
of grammaticalization offers a ready answer here. Original NPs may with
the passage of time become compacted down into relation-marking ele-
ments. English because of is a classic instance, where the presence of the
linking of betrays the earlier nominal source. Moreover, prepositions
cross-linguistically provide good evidence for the principle of grammati-
calization sometimes called ‘‘Gabelentz’s law’’ ( Vincent 1993: 145), viz.
(1) Alle Afformativen waren ursprünglich selbständige Wörter
[All grammatical markers were in origin independent words]
Obedience to this ‘‘law’’ is one respect in which prepositions behave like
other functional categories, whose realizations are also not historically
stable. In fact prepositions show a wide range of etymological sources,
1114 N. Vincent

including verbs, nouns, participles, adjectives, adverbs, and various types

of particle ( Kortmann and König 1992: 672; and cf. Vincent 1997a: 212
for a compilation of representative examples from Italo-Romance).
Consultation of the historical grammars of any language with sufficient
recorded time depth will reveal that the modern prepositions are (almost)
always the product of various grammaticalization processes. In this
respect prepositions differ from nouns, verbs, and adjectives, which typi-
cally retain their categorical status over the full length of their recorded
history even though they may undergo quite complex and many-layered
semantic shifts. A second property of (some) prepositions, which allies
them to the class of functional categories, is that they are used to express
purely syntactic relations rather than expressing their own semantic
content. Thus, English of is often argued to be a case marker licensing a
dependent nominal inside a NP/DP, and the demoted subject in the
French causative construction is introduced by à, conventionally glossed
as ‘to’ but in fact serving to express a particular grammatical relation.
Prepositions then are somewhat equivocal members of the class of major
lexical categories. Although they often do have independent lexical
content — such as English up/down, on/off, under/over, before/after, and
equivalent items in many other languages — they just as often do not.
Two possible solutions to this ambivalence suggest themselves. The
first involves abandoning the idea that there is a unified category corre-
sponding to the traditional class of prepositions and distinguishing
between ‘‘pure’’ prepositions on the one hand and case markers on the
other. This approach has found many advocates within generative
grammar. Thus, Rooryck (1996: 226–227) writes,
The property which sets prepositions apart from Case markers such as of and to
is the fact that they are associated with specific thematic roles (Comitative/
Instrument for with, Theme for about). ... This distinction between lexical preposi-
tions and functional preposition-like Case markers is one that is found in many
languages.

I will not discuss here the technical details of Rooryck’s analysis, which
are designed to unify the assignment of case (in a generative sense) under
minimalist assumptions very different from those that will be adopted
here. Rather we may note a number of significant weaknesses in this
general strategy of splitting the erstwhile class of prepositions into two.
First, we have no explanation for the fact that the items that serve as
case markers are homophonous with prepositions. It will be a matter of
lexical arbitrariness that for instance the French causative subject marker
has the same phonological shape as the preposition expressing the the-
matic role goal. Yet we know that cross-linguistically it is a robust
 The evolution of c-structure 1115

generalization that on the one hand goal prepositions are often used to
express the dative, and that on the other the item expressing the demoted
causee subject is often the same as that which marks the dative case (cf.
Ackerman and Moore 1999 for recent discussion and full references).
Moreover, it is not always easy to draw a distinction between the two.
Rooryck (1996: 226) cites English to in Karl gave a book to Fred as an
instance of dative case, but one might just as easily call it the marker of
the thematic role recipient or beneficiary (or even a more abstract kind
of goal ). The range of nuances in this construction is notorious (Green
1974) and indeed there is some evidence that the more the construction
expresses a nonprototypcial relation between giver and recipient, the
more likely it is that the alleged case marker to will be given up in favor
of the double-object construction. Thus, where give behaves as a ‘‘light
verb’’ taking a nominalized verb as object, the preferred form is for
instance Karl gave Fred a slap rather than ?Karl gave a slap to Fred
(Green 1974; Gisborne 1999).
A further objection to treating case-marking uses of apparent preposi-
tions as nonprepositional lies in the fact that, in those languages that
have overt morphological case, both types assign their own case. Thus
consider the following near-minimal pair from Latin cited by Molinelli
(1996: 83):
(2) a. (Pl, Capt 400)
numquid aliud vis
INTERR other.NEUTSG.ACC want.2SG.PRES
patri nuntiari?
father.DATSG tell.PASS.INF
‘Is there anything else you wish to be told to your father?’
b. (Pl, Capt 360)
quae ad patrem vis
which.NEUTPL.ACC to father.ACCSG want.2SG.PRES
nuntiari
tell.PASS.INF
‘which things you wish to be told to your father’
In (2a) the recipient of the information is expressed by the dative case
patri, but in (2b) it is expressed by the construction ad patrem ‘to the
father’. The point here is that the morphological case of patrem in the
second alternative is accusative as required by the preposition ad ‘to’ in
all its uses, whether as here when it substitutes for an independent dative
case or when it is used as a purely prepositional marker of destination
as in venit ad me ‘he came to me’. If we have to distinguish case-marking
ad from prepositional ad in Latin — and the example could be multiplied
1116 N. Vincent

a thousandfold for Latin, Greek, Russian, Lithuanian, and kindred lan-

guages — then it becomes a matter of coincidence that all the different
uses of ad require to be followed by the accusative rather than any other
case. And indeed the mere property of assigning a case formally identifies
such items as prepositional even when in terms of their function they
may be labelled ‘‘case markers.’’
This last example highlights a final problem with the analysis that
separates case markers from prepositions, namely that it will anyway be
necessary to make a similar distinction for simple case forms. Just as
uses of prepositions fall into case-marking and semantic uses, so too
cases may be separated into ‘‘concrete’’ (also called ‘‘local’’) and ‘‘gram-
matical’’ ( Kuryłowicz 1964). On this view the locative in Sanskrit or
Russian for instance is a semantically definable case with a specific
thematic role while the accusative is principally a marker of the grammati-
cal relation object.3 Interestingly, within this tradition too the dative has
been problematic. Szemerényi (1996: 158, note 1) recalls Hübschmann’s
(1875: 131) uncertainty as to how to class the dative and nearly a century
later Kuryłowicz (1964: chapter 8) is still in doubt. The problem is the
familiar one of where to draw the line between arguments and adjuncts
and has nothing to do with whether these are realized through a system
of cases or a system of prepositions or a mixture of the two. Rather than
distinguishing between ‘‘pure’’ prepositions and case markers, therefore,
which leads to the absurdity of a parallel distinction between pure cases
and case-marking cases, we need to have a system that recognizes a class
of prepositions, where these are required by the syntax of the language
in question, but allows an internal distinction to be drawn between (at
least) two subclasses. Moreover, whatever mechanism is utilized to draw
the distinction within the prepositional class should naturally extend to
the parallel distinction between semantic and grammatical uses of cases.
One such mechanism is available within HPSG. In this framework, all
prepositions are characterized as the partition (=value) PREP of the sort
(=feature) SUBSTANTIVE, a treatment that is very close to the tradi-
tional view of prepositions as a small but autonomous word class. The
case-marking uses of prepositions — called ‘‘nonpredicative’’ by Pollard
and Sag (1994: 255–256; 347ff.) — are handled by setting the value for
the CONTENT feature of the PP as equal to that of the argument of
the preposition. This has the same effect as an earlier proposal within
GPSG that such uses were syntactically PPS but corresponded to NP
types in the intensional semantics. The problem from the perspective of
the present study is that this eminently natural account of internal distinc-
tions within the class of prepositions does not easily generalize to an
analogous treatment of the two types of case. We will therefore adopt
 The evolution of c-structure 1117

the LFG solution, in which all semantically contentful items have a value
for a feature PRED. Items that do not have such a value are interpreted
as having a purely grammatical function (see section 5 for the details of
this proposal ).
The fact that prepositions in many languages typically govern cases
just as verbs do also poses a problem for the standard X-bar account of
prepositions even when they are behaving as a major category. It seems
perverse to characterize them as [−N, −V ] since it makes the proto-
typical verbal properties of taking an object and of determining the
object’s case depend on the [−N ] feature rather than the [+V ] one, but
this is the only way to achieve the requisite generalization within the
canonical X-bar feature framework. An alternative feature characteriza-
tion might therefore be the one proposed in Jackendoff (1977) where the
defining properties are not categorical [±N, ±V ] but relational
[±subj, ±obj ]. Verbs and prepositions would then share the property of
being [+obj ], but be distinguished in that verbs are also [+subj ] while
prepositions are [−subj ].4 Yet even this is not a completely satisfactory
solution, and the evidence of history suggests a more complex state of
affairs. Whereas verbs can for the most part be assumed to govern cases
as far back in history as we can go,5 there is clear evidence that the nexus
‘‘preposition+case form,’’ which is so much a part of the traditional
grammar of Latin, German, Russian, etc., is itself the product of a
process of grammaticalization. This suggests the desirability of a more
scalar kind of feature base such as that proposed in Anderson (1997),
though we will not explore the matter further in the present context.
Another, and logically separate, property of prepositions in modern
Romance languages is that they head prepositional phrases, characterized
by strict linear order between the members of the PP and special
co-occurrence restrictions between the prepositions and their specifiers.
PPs, then, are of interest because they allow us to track the separate
historical development of both government and constituency. The histori-
cal facts (to be examined in more detail in the next section) present a
challenge to a universalist X-bar theoretic understanding of phrase struc-
ture. On this view, preposition and PP are part of the major categorical
core defined in terms of the features (whether [±N, ±V ] or [±subj,
±obj ]) and the bar levels. The language-particular choice is whether
there is a P or not. If there is, then PP, Spec of P, Complement of P
come at once and ‘‘for free’’ in virtue of the instantiation of the universal
X-bar schema for the category P. The data we examine here suggest, on
the contrary, that the classic ingredients of a PP — the P itself, the case-
governed complement of P, and the category Spec of P — are the product
of separate historical processes, and at given historical stages some of
1118 N. Vincent

these ingredients may exist independently of others. Moreover, the change

described here is unusual in that it involves the ‘‘capture’’ of an argument
by an item that previously had no argument structure. This is in contrast
to the more commonly studied types of syntactic change that involve loss
of argument or theta structure (Roberts 1985, 1997b; Barron 1997). It
is also unusual — at least from the perspective of X-bar theory — in
that the prepositional head is the grammaticalized result of something
that was in origin a specifier. This last point is, however, consistent with
the idea that when an item grammaticalizes, in generative terms it moves
‘‘down’’ the tree (Roberts 1997b). All in all, prepositions are a complex
and difficult category both synchronically and diachronically, and a closer
look at the historical scenario is clearly called for. I will start from Latin
and work back to Indo-European and forward to Romance.

3. Case and prepositions from Indo-European to Romance

In this section we will document the stages whereby the full-fledged

prepositional syntax of the modern Romance languages emerges. ( The
discussion inevitably involves a certain amount of detailed supportive
argument to establish the historical data. Readers whose interest is princi-
pally in the theoretical conclusions that we argue to follow from these
patterns may prefer to skip ahead to the summary of this section con-
tained in section 3.8, and thence to the more directly theoretical sections
4–6. See also the discussion in Lehmann [1995: ch. 3.4.1.4–6 ].)

3.1. Particles, preverbs, and prepositions

In Latin we find a range of items that are both prepositions, exemplified

in (3a), and also components of compound verb stems (henceforth
‘‘preverbs’’), as in (3b). Thus,
(3) a. sub ‘under’, trans ‘across’, in ‘in, on’, ab ‘from’, ob ‘against’,
cum ‘with’, ex ‘out of ’, pro ‘for’, etc.;
b. submittere ‘to put underneath’, permittere ‘to let through’,
transmittere ‘to send across’, praemittere ‘to send ahead’, etc.;
sufferre ‘to undergo’, transferre ‘to carry across’, perferre ‘to
carry through’, offerre ‘to present’, conferre ‘to gather’,
etc.; subligare ‘to tie below’, obligare ‘to bind’, colligare ‘to
assemble’, deligare ‘to select’, etc.
A link between them can be established through two frequently cited
 The evolution of c-structure 1119

remarks by the grammarian Festus on the language of early prayers

(cf. most recently Cuzzolin 1995):
(4) a. Sub vos placo, in precibus fere cum dicitur, significat id, quod
supplico, ut in legibus transque dato et endoque plorato
‘When people say, mostly in prayers, sub vos placo, it means
the same as supplico and is like the expressions transque dato
and endoque plorato in the laws’
b. ob vos sacro, in quibusdam precationibus est, pro vos obsecro,
ut sub vos placo, pro supplico
‘ob vos sacro in certain prayers stands for vos obsecro, just as
sub vos placo stands for supplico’
Here the classical forms supplico, obsecro ‘I implore, beseech’ are decom-
posed into their constituent parts sub placo and ob sacro and the object
is placed between them. In the other cited examples the preverbs endo
(an earlier form of in) and trans are separated from their verb stems (dato
‘give’, plorato ‘beg’) and act as hosts for the clitic conjunction -que ‘and’.
Other early examples also imply the independence of items that subse-
quently occur only in construction with a verb or a noun. Festus again
quotes a passage from the famous legal text of the fourth–fifth century
BCE, the Twelve Tables:
(5) si calvitur pedemve struit, manum endo iacito
‘if he plays tricks or runs off, lay hands on (him)’
Here endo occurs in isolation but in immediate preverbal position in a
way that is not attested in other texts even in the Old Latin period, but
which has evident congeners elsewhere in Indo-European.6 Thus, in
Ancient Greek and Sanskrit we find examples such as the following (cited
after Coleman 1991: 327; and Hock 1996: 221, note 22):
(6) (Il 12.234)
eks ára de:́ toi épeita theoı̀ phrénas o:́ lesan autoı́
‘so you see the gods themselves thereupon destroyed away his senses’
(7) (RV 3.1.12c)
úd usrı́ya: jánita: yó jaja:́ na
out cow.ACCPL creator.NOMSG REL create.3SG.PERF
‘who as creator created forth the cows’
In (6) the adverb eks ‘away’ serves to reinforce the completive sense of
the verb (one may compare the effect of English up in Eat your food up!).
A similar function is served in (7) by the adverb úd ‘out, forth’.
Items like sub, endo, and ob were not in origin, therefore, prepositions
but rather sentence particles or adverbs that could occupy a variety of
1120 N. Vincent

positions in the sentence.7 The three positions that will be of particular

interest in the present context are
i. sentence-initial and separated from the verb (as in [4], [6 ], and [7])
ii. unstressed and adjacent to the verb
iii. before a local case form
For Latin, the alternation between the first two of these is only attested
residually in the archaic examples adduced in (4) but is entirely systematic
in Vedic (Renou 1933; Pinault 1995) and is also found in a number of
other early Indo-European languages (the traditional name for this con-
struction is tmesis).8 A rather similar kind of alternation between the
second and third possibilities is found with a word like only in English:
(8) a. Fred has only published two articles
b. Fred has published only two articles
Here too the interpretation of (8a) when pronounced with neutral intona-
tion involves only having scope over two articles despite its adjacency to
the verb and its separation from the NP over which it takes scope.
The last property of these ancient items that will be of relevance here
is their prosodic behavior. There is general consensus that when separated
and placed in clause-initial position the particle bore its own stress; when
placed before the verb it was unstressed. Thus, Renou (1933: 50) writes,
‘‘le préverbe tend à être proclitique en subordonnée, ailleurs tonique et
autonome’’ [the preverb tends to be proclitic in a subordinate clause and
elsewhere stressed and free]. From this it follows that the majority of
instances of preverbs in main clauses are separate from the verb and
hence bear stress. Indeed Renou (1933: 52) goes on to observe that ‘‘la
séparation peut être considérée comme courante, sinon tout à fait comme
normale’’ [separation may be considered as common, if not indeed as
completely normal ].9
To sum up, then, in the ancestor language a deictic or orientational
sentence particle could bear focal stress in its own right in absolute initial
position or it could attach to a sentence constituent over which it had
particular scope or it could default to the unstressed preverbal position
(we assume an unmarked clause-final position for the stressed verb).10
Given the foregoing, it would be wrong to see sub vos placo in (4a) as
involving a preposition at all (Cuzzolin 1995: 134). Vos is accusative as
object of placo and sub is a sentence particle; as Coleman (1991: 324)
notes, these particles could occur with intransitive verbs as well as transi-
tive ones, and in the case of the transitive verb it is the particle that is
omissible, not the noun, a distributional property not consistent with
these items being analyzed as prepositions.11 In the subsequent history
of Latin and Romance we find a number of changes that come together
 The evolution of c-structure 1121

to create first the class of prepositions and in due course the rigidly
configurational PPs with which we are familiar from the modern
languages.

3.2. The acquisition of argument structure

A clear property that distinguishes, cross-linguistically, adpositions from

particles is that the former have their own argument structure, both in
the semantic sense of selecting particular theta roles and in the syntactic
sense of determining the case form of that argument. The development
from Indo-European into the daughter languages is no exception and the
former particles come to take a dependent argument in a fixed case. The
source of this change lies in the use of the original particles as specifiers
of the local cases (instrumental, ablative, and locative).12 Thus a particle
such as sub ‘under’ is not inherently directional or locational; it simply
expresses an orientation. If it modifies a directional use of the accusative,
the meaning will be ‘(move) under and to(wards)’; if it modifies a locative
the meaning will be ‘(be) under and in/at’. It is only at a later stage
that the particle and the case form are inseparable, and that we have the
situation in which the preposition may be said to ‘‘take’’ or govern a
particular case form. Thus what was in origin a specifier comes to be a
head. Lehmann (1995: 88) expresses the change somewhat differently,
suggesting that the spatial concepts lexicalized by the original particles
have an inherent semantic argument that expresses the deictic point of
reference. The change then involves the syntacticization of this earlier
semantic relation. Whichever perspective one prefers to adopt, it is worth
noting that such an origin fairly straightforwardly explains what many
have taken to be a puzzle (Baldi 1979; Joseph 1991; Bauer 1995: chap-
ter 5), namely the linear order whereby Latin, usually regarded as an
OV or head-final language, has prepositions rather than postpositions.13
All the available evidence points to specifiers in Indo-European being on
the left. There is no conflict then with a putative I-E OV order on this
interpretation, a point that rather surprisingly seems to have been missed
in the literature (cf. further in section 3.4 below).

3.3. Divergence of preverb and preposition

As long as the adverbial particles are free to move around and occupy
various positions in the sentence it is reasonable to think of them as
single items that have multiple positional possibilities (think again of
1122 N. Vincent

English items such as all, only, and even, and their counterparts in other
languages). Once, however, the sequence particle plus case form is reana-
lyzed in the way described above, the range of distribution of the particle
is reduced and we have the beginning of a split into particles and preposi-
tions. At the same time, and for independent reasons, the particle, when
adjacent to the verb, comes to form a lexical unit with it and to undergo
semantic shifts that often obscure the etymological function of the pre-
verb. Thus obligare is rare in its etymological sense of ‘to bind around,
against’ (<ob+ligare) and the more usual and surviving sense of ‘oblige,
make liable’ does not have a separable component of meaning that could
be attributed to the ob-prefix. Similarly, permittere originally meant ‘to
let go through, to let loose’ (<per ‘through’+mittere ‘to send, cause to
go’) but already by the time of Caesar and Cicero (first century BCE )
had acquired the meaning ‘to allow’, which is all that survives in French
permettre, Italian permettere, etc. This kind of lexicalization is a familiar
process in such circumstances and is described for the Brazilian language
Nadëb by Weir (1986); see also the studies collected in Rousseau (1995).
For more detailed studies of the Latin situation, see Lehmann (1983)
and Rosén (1992).

3.4. Word order I: preposition vs. postposition

Perhaps the most striking property of the newly emergent prepositional

constructions is the rigidity of linear order they exhibit: the order is
always P NP and the two items are very rarely separated by intervening
material. In the larger Latin context, this has been seen as surprising
because (a) Latin word order otherwise shows a fair degree of freedom,
and (b) if there is a dominant tendency, it is for the verb to follow the
object so we might have expected a similar linear order to hold between
an adposition and its object (Hawkins 1983). How are we to explain this
state of affairs?
Baldi (1979) suggests the following sequence of events:
I. attraction of unstressed particles to the preverbal slot, which would
be consistent with the general verb-final typology reconstructable
for I-E;
II. the univerbation of preverb and verb to create compounds of the
permittere, circumire type (cf. section 3.3 above);
III. the refinement of the case semantics of the argument of such verbs
by the duplication of the preverb in a parallel prenominal position:
trans flumen transire/transducere lit. ‘to (cause to) cross across the
river’;
 The evolution of c-structure 1123

IV. the extension of the prenoun+noun groups thus formed to other

contexts where the prenominal element is not supported by a
preverbal element;
V. the eventual loss of the transparent preverbal element once
lexicalization of preverb+verb sets in: e.g. inducere comes to mean
‘introduce, persuade’ and the sense of ‘lead in’ is now expressed by
in+NP ducere.
The problem with this proposal is that the key sentence type — trans
flumen transire — has only a limited attestation, in particular in writers
like Caesar, Cicero, Livy, etc. (i.e. first century BCE and after; cf. also
Nocentini 1992: 228; Coleman 1991: 332). It is not found in early prose
writers such as Cato (234–149 BCE). Even in those writers in which it
is found, it is not always used and the verb is often treated as a simple
transitive (even to the extent of allowing a passive: e.g. Taurus transiri
non potest ‘The Taurus mountains cannot be crossed’ [Cic Att 5,21]). It
is better therefore to regard this construction as an artefact of literary
prose of the classical period, and thus something that came into existence
long after prepositions had been established in the language, and perhaps
even in virtue of the establishment of prepositions, rather than as the
missing link between preverbal and prepositional syntax.
Joseph (1991: 188) also suggests that the apparent anomaly of both
the contiguity of preposition and noun and their relative order in the
broader Latin context calls for some explanation. He too looks to the
preverbal pattern that had established itself with the verb as a model,
reinforced by what he claims is a conspiracy for Latin to have morphosyn-
tactic clusters in which the inflected word is final. Thus, given the choice
between per fugam ‘through flight’ and *fugam per, he suggests only the
former is consistent with the morpholgoical ‘‘target’’ of Latin. There is
an inherent problem of teleology in this mode of explanation, but in any
case there are empirical arguments against Joseph’s position. Not only
are there many instances of elements cliticizing in Latin after the inflec-
tion — the already cited enclitic conjunction -que is a case in point as
are the deictic and emphatic particles contained in items like idem ‘same’,
hic ‘this’, etc. (originally *is-dem, hi-ce) — but in Oscan we find precisely
the development of a new inflectional subsystem out of agglutinated
postpositions (Nocentini 1992).14
Nocentini (1992: 227–230) in fact suggests that the origins of the
matter lie earlier in time. The most common inherited patterns are
NP-P-V and P-NP-V. The former yields the preverb construction, NP
[P-V ], and the latter the prepositional one, [P-NP] V. However, while
this scenario narrows down the range of options, the only pattern it rules
1124 N. Vincent

out absolutely (and correctly) is the possibility of a ‘‘postverb,’’ which

would require as a starting point the unattested sequence *NP-V-P. Since
NP-P-V does occur, we still need an explanation for why that doesn’t
yield a bracketing [NP-P] V, thus creating postpositions. Interestingly
Craig and Hale (1998) propose just such a string as the precondition for
the development of relational preverbs out of postpositions in the
Chibchan language Rama and the Maku language Nadëb ( Weir 1986):
that is, [NP-P] V is reanalyzed as NP [P-V ]. The fact that the change
that must be ruled out in Latin has occurred elsewhere in the world’s
languages means that we cannot give up on the search for a special
property of Latin that allowed P-NP-V to produce prepositions and yet
blocked postpositions from developing out of NP-P-V.
The clue to the answer, I would suggest, lies in the fact, already noted,
that in the ancestor language the preverbal/prepositional elements were
inherently tonic when separated from the verb, a situation that as we
have seen is still fully documentable by the time of the Rigveda (approxi-
mately 1000 BCE). It is the unaccented form that migrates to and
cliticizes on the verb:15 the verb is the unmarked sentence head and a
natural attractor of such sentence-level particles in the unmarked —
hence unstressed — case. Joseph’s discussion of the relative accent of P
and NP (1991: 188–189) starts out on the right track but then makes
the wrong assumption, viz. that the P is proclitic on the NP. This may
have been the later situation (cf. the remark by Quintilian that he
adduces), but we would suggest that the formation of prepositions
could only come about if P and V were split by the NP and that in
such a circumstance P bears its own accent (cf. the different stresses in
English only and even in adnominal and adverbal position) just as
in Vedic.
The stressed particle naturally occurs in initial position, so that
unstressed elements — very commonly pronouns — will follow the par-
ticle by the normal Wackernagel effect. We thus get a very common
pattern (italic means stressed and=marks a clitic boundary):
(9) P=pronoun ... V: sub=vos placo; ob=vos sacro
(cf.endo=que plorato)
It is this pattern that provides the context for the reanalysis in which the
(in origin pronominal ) element attached to the particle becomes interpre-
ted as an argument of the particle, thereby transforming the particle into
a preposition.
A further attraction of this account is that it explains the virtually
exceptionless nature of the process. Contrast a putative application to
Indo-European of the reanalysis model postulated for Rama and Nadëb
 The evolution of c-structure 1125

by Craig and Hale (1988) in which there is a postpositional phase prior

to the reanalysis of the postposition as a preverb, something that they
show happened to different degrees for different members of the class.
This would predict a proportion of postpositions, with perhaps a gradual
dominance of prepositions over time as the other word-order changes
take place, but this is not what we find. The striking thing about the
P-NP order is, as Joseph (1991: 188) rightly noted and Bauer (1995:
128ff.) endorses, that it is early and almost absolute and runs counter to
the prevailing OV typology.
Nonetheless, there are signs of apparent postpositional behavior in
early Latin texts and these need to be explained. It is commonly assumed
that such cases — exemplified in (10) — require the postulation of an
archaic Latin stage in which postpositions were normal (cf. Bauer 1995:
130ff. for a recent restatement of this view):
(10) mecum ‘me+with’
quocum ‘which+with’
ted endo ‘you+in’
quibus ex ‘which+out of ’
quemadmodum ‘which+to+way’=how’
quamobrem ‘which+on account of+thing’=‘why’
In fact they fit in rather well with our story and do not require the
postulation of a separate postpositional phase of the language. To explain
mecum and the like we only have to assume that there could be competi-
tion for the initial stressed position between the particle and a stressed
pronoun. If the latter wins, the particle is forced into second position
and as a relatively unstressed element itself cliticizes: me=cum venit,
typically written as a single word.
In the other type of circumstance — exemplified by quocum, quibus
ex — we have a similar kind of competition for first position though
here the motivation is different. The qu-words are subject to a grammati-
cal rule by which words of this case occupy first position, and if this is
given priority over the initial-particle rule, then the latter will be relegated
to second position. Broadwell (1999) describes a strikingly parallel
pattern in the Otomanguean language, San Dionicio Ocotepec Zapotec
(SDZ ). In declarative clauses this language has prepositions, (11a), but
in interrogatives the prepositions are blocked and appear instead as
postpositions, (11b).
(11) a. ù-diny juààny bèh’cw re’ cùn yàg
COM-hit John dog that with stick
‘John hit that dog with a stick’
1126 N. Vincent

b. xhı́ı́ cùn ù-diny juààny bèh’cw re’

what with COM-hit John dog that
‘What did John hit that dog with?’
Broadwell also interprets this reversal of the expected order in terms of
competition. More precisely, he views order within constituents as due
not to fixed phrase-structure rules but to the interaction of optimality
theory constraints (for further discussions of this approach see section 6).
Although the existence of postposed forms in Sanskrit is widely cited in
favor of an earlier postpositional phase, the detailed study of Renou
(1933: esp. 52ff.) suggests that the determining principle here too is accen-
tual and that in general postpositions are rare in Sanskrit. Nocentini
(1992) similarly adduces a prosodic explanation for the origin of postposi-
tions in Umbrian.16 Broadwell’s data confirm the typological possibility
that Latin particles may in special circumstances be inverted in clause-
initial position without there being any consequent need to postulate a
period in which postpositions held sway over prepositions. Genuine
postpositions are in a minority and arise from another source, namely
nouns (cf. note 12).
To sum up, then, we have two independent but converging patterns
that generate prepositions in a language like Latin. On the other hand,
since prepositions originate as specifiers and since specifiers precede the
case form they modify, we predict the order ‘‘P+caseform’’ rather than
the reverse. Second, separated preverbs would normally occur at the left
edge of the clause and therefore necessarily to the left of any pronominal
items that might later, by reanalysis, come to be interpreted as their
arguments. Apparent instances of postposition (except perhaps Hittite —
see note 14) would arise as the byproduct of competing prosodic require-
ments and would not reflect an earlier stage of genuine postpositional
syntax.

3.5. Word order II: PP-internal fronting

In the previous section we have seen how competition for front position
in the clause leads question words to precede the prepositions that govern
them. The pattern thus generated — quem ad locum et quam ob rem
‘where and why?’ (Plaut Men 823), qua de re ‘what about?’ ( Ter And
184) — bears some resemblance to another pattern found within pre-
positional phrases, what we might call the magna cum laude effect. Here
the adjective that modifies the complement of the preposition occurs
in first position within the PP. This pattern most typically affects
 The evolution of c-structure 1127

quantificational, intensifying, and deictic adjectives such as multus ‘much,

many’, magnus ‘great’, maximus ‘greatest’, summus ‘top’, nimius ‘too
great’, mirus ‘wonderful’, hic ‘this’, etc. Such adjectives seem to be inher-
ently emphatic and are natural candidates for the leftmost position in
the constituent.17 If they occupy this position, they too will trigger this
kind of constituent-internal (or perhaps colon-internal ) Wackernagel
effect18 and force the preposition to be enclitic.
The data reviewed so far suggest a certain freedom in the word order
of Latin prepositional phrases, though this freedom is not absolute. We
have seen that a particular class of adjectives commonly precedes the
preposition. In poetic language, a wider range of adjectives and also
genitives dependent on the argument of the preposition may precede (the
following examples are taken from Penney 1999):
(12) a. ( Ennius Satur 41V )
Nestoris ad patriam
Nestor.GEN at homeland.ACC
‘in Nestor’s homeland’
b. ( Vergil Aen 3.18)
Libyae vertuntur ad oras
Libya.GENSG turn.3PSG to shore.ACCPL
‘they turn toward the shores of Libya’
As (12b) shows, such fronting may even move the genitive out of the
prepositional phrase altogether, and these constructions seem to be part
of poetic style often driven by metrical considerations. Penney (1999) for
example notes that the only occurrence of this pattern to be found in the
plain narrative prose text of Caesar’s De Bello Gallico is quorum per finis
‘through whose territory’, where the fronted genitive quorum is a relative
pronoun, which would, as we have seen, have undergone this movement
in any case. I have checked for comparison another and earlier prose
text, one moreover noted for its lack of rhetorical elaboration — Cato’s
agricultural handbook De Agri Cultura of the second century BCE —
and Penney’s generalization holds true: instances such as quo in agro ‘in
which field’ ( VI.4) are found but otherwise both adjective or genitive
and noun follow the preposition. Similarly poetic is the construction in
which the noun precedes the preposition and its modifier follows (cf. also
Bauer 1995: 135–136):
(13) a. (Lucr 5.1382)
cava per calamorum
hole.ACCPL through reedpipe.GENPL
‘through the holes of the reedpipes’
1128 N. Vincent

b. ( Vergil Aen 4.257)

litus harenosum ad Libyae
shore.ACCSG sandy.ACCSG to Libya.GENSG
‘to the sandy shore of Libya’
c. ( Ennius Ann 187)
arbusta per alta
timber-trees.ACC through tall.ACC
‘through tall timber trees’
In this respect I agree entirely with the distinction Bauer (1995) draws
between those patterns that are attributable to stylistic factors and those
that properly feed into a diachronic account. I differ from her, however,
in that I see the changes in question as the gradual fixing of syntactic
configurationality rather than as a shift from a head-final (‘‘left-
branching’’ in her terms) to a head-initial (‘‘right-branching’’) PP. I see
no difficulty in accepting the usual X-bar account of category structure
for Modern French, Italian, etc., but I would hesitate to project it
backward onto Latin and Proto-Indo-European. One crucial area, there-
fore, in which our approaches differ concerns the role of specifier of PP,
to which I now turn.

3.6. Specifier of PP

The only variant on P-NP order then that needs to be accounted for
within the core syntax of (nonliterary) Latin is that in which a quantifica-
tional or intensifying adjective immediately precedes the preposition. The
fact of immediate linear precedence suggests that this ‘‘displaced’’ adjec-
tive is still within the PP, a property that falls out naturally from the
prosodic competition principle established above.19 Moreover, note that
there is often no semantic difference in such examples between interpre-
ting the adjective as attached to the noun and as taking scope over the
whole PP; in either case summa cum laude, for example, means ‘at the
highest point on the scale of praise’. This in turn suggests that we should
consider the relation of this construction to other forms of prepositional
modification.
Adverbial modification of a prepositional construction is exemplified
in the following:
(14) a. (Cie Or 2.64.259)
si ad te bene ante lucem venisset
if to you well before light come.PLUPF.SUBJ.3SG
‘if he had come to you well before daybreak’
 The evolution of c-structure 1129

b. (Liv I,9)
longe ante alias specie insignem
long before other.FEM.ACCPL beauty famed.ACCSG
‘(a woman) famed for her beauty greatly above others’
c. (Caes BG 6.9)
paulo supra eum locum
a little above that place
‘not far beyond that place’
d. ( Tac Ann 5.3)
haud multum post mortem eius
not much after death her
‘shortly after her death’
Such degree expressions were commonly but not necessarily placed before
the preposition, in other words in the same position as the adjective in
summa cum laude. However, it seems clear that in earlier texts in particular
the PP and its degree modifier could be separated. The following examples
relate to the item usque, glossed in Lewis and Short (1879) as ‘‘all the
way to or from any limit of space or time, etc.’’ This item is of particular
interest both because it later evolves from a fixed pre-prepositional posi-
tion into part of an etymologically complex preposition in French jusqu’à
(<de usque ad) ‘until’, and because it is the nearest we can come in Latin
to an item that only modifies prepositional and adverbial constructions
and thus has the same diagnostic function that an item like right has in
English:20
(15) a. (Cic Rosc Com 7,20)
ab imis unguibus usque ad verticem
from bottom.ABLPL nail.ABLPL right to top.ACCSG
summum
highest.ACCSG
‘from the top of his head to the tips of his toes’
b. (Nep. Hann 2.1)
usque a rubro mari
right from red.ABLSG sea.ABLSG
‘right from the red sea’
c. (Cic. Quint 3,12)
trans Alpes usque transfertur
across Alps.ACC right carry.PRES.PASS.3SG
‘he is carried right across the Alps’
d. (Plin 4,12,21)
ab Attica ad Thessaliam usque
from Attica.ABL to Thessaly.ACC right
‘from Attica right to Thessaly’
1130 N. Vincent

e. ( Ter Ad 90)
mulcavit usque ad mortem
beat.PERF.3SG right to death.ACC
‘he beat [them] to death’
The examples in (15) show clearly that usque can both precede, as in (15a),
(15b), and (15e), and follow, as in (15c) and (15d), the PP it modifies. It
is only in later usage, presaging that which is found in Romance, that
the specifier of a prepositional phrase has a fixed linear position. In sum,
the fixing of the position of the specifier, whether adjectival as in the case
of the summus class of items or adverbial as with usque, longe, etc., postdates
by several centuries the fixing of the position of the preposition. As examples
of this later usage with an item, totum ‘whole’, which did not have such
a function in the classical language, consider the following from the
fourth-century CE text Peregrinatio Aegeriae:21
(16) a. (I,2)
totum per valle illa
whole through valley that
‘right through the valley’
b. (II,3)
totum per mediam vallem ipsam
whole through middle valley the
‘right through the middle of the valley’

3.7. Morphosyntactic equivalence of PPs and case

It is a commonplace of Romance historical syntax that the grammatical

functions of the dative case — the marker of the indirect object — and
the genitive case — the marker of a noun dependent on another noun —
came to be replaced by grammaticalized uses of the prepositions ad ‘to,
toward’ and de ‘down, about’ respectively. What is less commonly
emphasized but is of particular importance in the context of the present
discussion is that this development took place over many hundreds of
years. Thus, consider again the pair of virtually equivalent examples from
Plautus (254–184 BCE) in (2), repeated here as (17):
(17) a. (Capt 400)
numquid aliud vis
INTERR other.NEUTSG.ACC want.2SG.PRES
patri nuntiari?
father.DATSG tell.PASS.INF
‘Is there anything else you wish to be told to your father?’
 The evolution of c-structure 1131

b. (Capt 360)
quae ad patrem vis
which.NEUTPL.ACC to father.ACCSG want.2SG.PRES
nuntiari
tell.PASS.INF
‘You wish these things to be told to your father’
The choice here may well have been dictated by metrical considerations
since the dative case patri and the prepositional phrase ad patrem have
different numbers of syllables, but even so it is hard to believe that the
option would have been available in the first place if the expressions had
not been semantically equivalent or almost so. This in turn requires that
the grammar at this stage of the language be able to capture the fact that
both a bare case form and a prepositional phrase can satisfy the indirect
object requirement of the verb nuntiare ‘to tell, report’.
There are similarly early examples of de+NP being used where classi-
cal norms would have dictated a genitive (again from Molinelli 1996,
who provides ample documentation and valuable discussion):
(18) a. (Plautus Ps 1164)
memento ergo dimidium istinc mihi de
remember.IMP therefore half.ACC thence me.DAT of
praeda dare
booty.ABLSG give.INF
‘Remember then to give me half of the booty from there’
b. (Cato Agr 96.1)
faecem de vino bono
dregs.ACCSG of wine.ABLSG good.ABLSG
‘dregs of good wine’
At the same time there are genitives and datives being used in similar
contexts by writers over half a millennium later. Compare the example
in (19) where there is both a genitive case (loci ipsius) and then a
prepositional phrase de pomis ‘of the fruit’. Note too the dative case nobis
‘us’ expressing the indirect object. The text is the fourth-centry CE
Peregrinatio Aetheriae, which is notorious for its conversational and
colloquial style.
(19) (Per Aeth III,6)
dederunt nobis presbyteri loci ipsius
give.PERF.3PL us.DAT priest.NOMPL place.GEN that.GEN
eulogias, id est de pomis ...
praise.ACCPL that is of fruit.ABLPL
‘the priests sang to us the praises of that place, that is of the fruit ...’
1132 N. Vincent

There are also many other contexts in the language in which de and ad
will have independently definable semantic values. As already noted in
section 2, an adequate grammar will thus have to capture on the one
hand the grammatical and the semantic uses of the same prepositions
and on the other the morphosyntactic equivalence of a bare case form
and a prepositional phrase.22 We will develop an LFG account of this in
section 5 below. Before that, however, let us briefly review a couple of
other developments in the syntax of prepositions in modern Romance.

3.8. Later Romance developments

3.8.1. As is well known, the eventual result of the breakdown and

replacement of the Latin case system, part of which has been described
in the last section, is the complete loss of morphological case as a
grammatical mechanism. As the historical record shows, and as logic
would dictate (Harris 1978), the rise of prepositions precedes the fall of
case. Just as the fixed order of a noun adjacent to its governing preposition
permits the case marking to be lost without loss of content, so the fixed
order of subject, verb, and object allows for loss of grammatical case
marking. The detail of these changes is complex and would take us too
far afield. It is sufficient to note here that between the original, grammati-
cally autonomous semantic cases and the complete loss of case morphol-
ogy is a long-lasting stage in which grammatical case survives beside an
originally semantic use of prepositions, while the latter in turn become
grammaticalized for the expression of non–core arguments as case is
eliminated from the system.23

3.8.2. Just as some prepositions pass from semantic to syntactic func-

tions so a wide range of new prepositions are coined from many different
sources (cf. Vincent 1997a, 1997b; Longobardi 1995). Thus,
(20) Ital senza ‘without’ < Lat absentia(m) ‘absence’
Fren chez ‘with, at the house of, in’ < Lat casa(m) ‘house’
Span bajo ‘underneath’ < Lat bassu(m) ‘low’
The mechanisms here are various, depending on the verbal, nominal, or
adjectival origin of the prepositions in question, but the changes are
facilitated by the fact that the outcome is a class of items, namely
prepositions, that already exist in the language. The grammaticalization
patterns here therefore do not create a new category, as is often the case
elsewhere, but expand and take on the properties of an already existing
class ( Vincent 1995).
 The evolution of c-structure 1133

3.8.3. One particularly interesting development in this latter connection

is the emergence of prepositions whose etymology requires us to assume a
combination of prepositions. Thus, for instance, Italian dopo ‘after’ can be
shown to be the fusion of po<Lat post ‘after’ prefixed by de, the genitive-
marking preposition whose origin we reviewed in section 3.7. Similarly
French avant ‘before’ is from Latin ab ‘from’+ante ‘before’. The most
likely origin of such preposition+preposition sequences is from a stage in
which a preposition could take another PP as complement ( Vincent 1997a:
212–213). Thus the etymology of Italian dopo cena ‘after dinner’ would be
[de [ post cenam]], literally ‘from after dinner’. This is important evidence
for another piece in our puzzle. Classical Latin did not in general allow PPs
to be the complements of other Ps,24 a pattern with which we are of course
very familiar in English (Jackendoff 1973), and we can take their historical
emergence in this function to be yet another sign of the increasing internal
structural complexity of the category PP. It is perhaps not surprising that
the most common prepositions to take PP complements were the two most
grammatical ones de and ad, though others are found: French avant
‘before’<ab+ante, It innanzi ‘before’<in+ante.

3.9. The changes summed up

At this point it will be useful to summarize the sequence of changes that

we have argued for in the preceding sections. There was undoubtedly a
considerable degree of overlap in the instantiation of these changes
(fuelled by the kinds of sociolinguistic factors discussed in Molinelli 1996,
1998), but by separating them out into discrete stages here we can better
highlight the logical connections:

Stage I. Adverbial particles exist that do not have grammatically

determined arguments and can occur in a variety of clausal positions.

Stage II. One particular usage of these particles causes them to precede
and modify semantically independent local case forms of nouns. Another
(independent) usage has them in stressed clause-initial position where
they may host second-position clitic pronouns. (Failing either of these
options, the particles default to an unstressed preverbal position from
which they become attached to the verb to create a new new class of
prefixed verbs.)

Stage III. Both the patterns identified in stage II admit of a reanalysis in

which the nominal or pronominal element may be interpreted as an
1134 N. Vincent

argument of the adverbial particle, thereby converting the particle into a

preposition of which the case form is now a dependent, and creating the
category complement of P. This, as Lehmann (1995: 88) emphasizes, also
involves the syntacticization of the implicit semantic argument that was the
point of reference for the interpretation of the earlier adverbial meaning.

Stage IV. The prepositional phrases thus created come in due course to
be themselves subject to modification by items such as usque, longe, totum.

Stave V. Such secondary modifiers come to occupy a fixed pre-preposi-

tional slot and thus to create a category Spec(ifier) of PP.

Stave VI. As prepositional syntax extends its range and case forms
recede, both changes that are implemented over a number of centuries,
two further changes occur:
i. PPs come to be complements of P;
ii. many new prepositions are created through the grammaticalization
of a wide range of other categories.
Stage VII. Structures of the form [P [P NP] ] that emerge in stage VI
(i) are reanalyzed as [P-P [NP] ] and eventually the compound preposi-
tions so formed relexicalize as single Ps.

Stage VIII. The culmination of all these stages, which are crucially
not only logically but also chronologically distinct, consists in the
rigid, hierarchical, X-bar style, case-free PPs of the modern Romance
languages.

What this sequence of events indicates is that, although the modern

situation may be well described by X-bar theory, the stages that lead up
to it do not necessarily require the postulation of the full projection of
P at the moment the category P comes to exist. In the remainder of this
paper we will try to draw out the theoretical consequences, in both the
diachronic and the synchronic arenas, of such a state of affairs.

4. Grammaticalization and grammatical theory

The case history (in both senses!) we have just sketched is exactly the
kind of evidence that has been at the heart of the recent revival of interest
in grammaticalization (cf. Cuzzolin 1995). Items that in origin had inde-
pendent lexical content gradually extend their semantic range and at the
same time acquire a fixed subcategorization ( Vincent 1980: 56–57). This
 The evolution of c-structure 1135

is perhaps most dramatically evident with the pan-Romance preposition

de/di, which once had a sense of ‘down, away from’ (cf. Latin de-scendo
‘I come down’, de-sum ‘I am absent’) but now marks the purely grammati-
cal relation of one noun on another (la voiture de Paul ‘Paul’s car’, la
chûte du gouvernement ‘the fall of the government’) or of one verb on
another (il a décidé de s’en alleṙ ‘he decided to leave’).
What the grammaticalization approach does not have — and indeed
what some if not all proponents of grammaticalization would argue
against in principle — is a theory whereby one can represent and compute
grammatical structure. Thus, although they make fundamental use of the
concept of ‘‘reanalysis,’’ as indeed all theories of syntactic change cannot
fail to do (Harris and Campbell 1995: chapter 4, though cf. now
Haspelmath 1998), they have no more than the traditional armory of
informal and intuitive grammatical categories with which to define the
input to and output from diachronic reanalyses. The weakness of this
position is perhaps not so obvious when the matter under discussion is
semantic bleaching and the concomitant extension of the class of NPs
associated with a given predicate. However, when we have a discrete shift
from absence to presence of an argument, which is the key factor in the
birth of a class of prepositions, then the inability to fall back on a theory
of argument structure constitutes a serious difficulty.25
In this respect, the acquisition of argument structure, which is, as we
have seen, inherent in the process of preposition formation, is not simply
the opposite of the loss of argument structure, which is commonly, and
reasonably, held to characterize auxiliarization (Roberts 1985, 1993).
More precisely, Roberts (1985) argues that the properties of Modern
English modals derive from the fact that they no longer assign theta roles
when compared to their etymological sources. Loss of theta structure is
a clear formal analogue to the usual accounts of grammaticalization as
involving ‘‘semantic bleaching.’’ By the same logic one might expect
acquisition of theta structure to be a kind of ‘‘degrammaticalization’’ (on
which see Ramat 1992), yet the intuitive parallels break down. Rather,
the fact of adverbs becoming prepositions, as already noted, is usually
thought of as a kind of grammaticalization (Cuzzolin 1995) for the
obvious reason that such items start to contract more than semantic
relations of modification within the clauses to which they belong. What
then is it precisely that unites the loss of argument structure in the case
of the emergence of English modals and the acquisition of argument
structure in the items studied here, and makes both instances of grammati-
calization? The answer is to be sought in the fact that individual lexical
items have (morpho)syntactic consequences to the extent that they deter-
mine or select properties of other items with which they combine. Any
1136 N. Vincent

verb inherently has such an effect through its subcategorization frame.

As the semantic content of this frame is reduced, so that, say, an item
selects a subject without selecting a thematic role to associate with the
subject, we may legitimately talk of grammaticalization (or in Lehmann’s
perhaps more perspicacious term ‘‘syntacticization’’). By the same token,
an adverb has no inherent selection properties, and is thus a more purely
lexical item with greater selectional autonomy. Adding it or removing it
from a sentence may have semantic consequences but will not normally
have syntactic reverberations. The acquisition of a subcategorization
frame, even a semantically determined one, by which an adverb shifts to
prepositional status, is thus a step in the direction of greater grammatical
interaction with other items in the clause, and thus of grammaticalization
in the intended sense. It is then reasonable to say that, as a preposition
moves from having its own semantic content to marking a purely syntactic
relation such as demoted subject in a passive or causative construction,
it acquires an even greater degree of grammaticalization.
It must be emphasized, however, that each such change is a separate
step, which may or may not take place according to the pressures that
are operative in the language and at the historical moment in question.
There is an apparent directionality because of an overriding constraint
that an item cannot acquire increased semantic content, or reacquire lost
semantic content, without being relexicalized. This in turn follows because
of interaction with other items. If, for instance, a verb ceases to determine
the semantic role of its own subject — as happened in the evolution of
raising verbs like seem and tend (Barron 1997) — then that value will be
supplied by the argument of another verb. Since there is no reason for
one verb, having once acquired the privilege of assigning an argument
value to another verb’s empty slot, to give that privilege up, there is no
possibility of the other verb’s reacquiring its earlier semantic potential.
If on the other hand we create a new verb, even from a conjunction or
a preposition as in the much cited to down a glass of wine or the
Shakespearian but me no buts, we endow these items as new members of
the lexical class in question with the properties appropriate to that class.
It is unhelpful, and potentially confusing, to call such changes ‘‘degram-
maticalizations’’ when what they really involve is (re)lexicalization.26
On this view there may be directionality in grammaticalization (Hopper
and Traugott 1993: chapter 5), but there is no teleology (pace Lightfoot
1999). The directionality arises from the nature of grammatical systems
and thus from the nature of possible changes once a system is in a given
state. Whether any further change does or does not take place at a given
time does not, however, depend on what changes have gone before or on a
knowledge of what might be the future endpoint of changes developing
 The evolution of c-structure 1137

out of the current state. Either of these possibilities would indeed be

teleological since they would require a given state of a language to be
endowed with a knowledge of its own history and/or future.
Another criticism that is frequently levelled at research within the
framework of so-called ‘‘grammaticalization theory’’ is its informal nature
(cf. most recently Newmeyer 1988: chapter 5). A range of views are
possible here. At one extreme stand those who regard language as the
kind of phenomenon that by its very essence lies beyond the reach of the
formal, mathematical techniques that characterize the natural sciences.
Grammar, on this view is not a computationally fixable system, but a set
of emergent patterns that inhere in the cognitive and communicative
needs of speakers, and thus the whole formalist, and indeed structuralist,
enterprise is misconceived (Bybee et al. 1994: 1). Such a position seems
to me to be unduly pessimistic. At the other extreme lie those who
consider the only acceptable accounts of natural language to be formal
ones, and what has not been formalized to all intents and purposes does
not exist. In our present state of knowledge this view errs in the other
direction and is ungenerous to the many scholars who have contributed
genuine, though unformalized, insights into the workings of natural
language and natural languages. To take a parallel from the world of
science and engineering, there is a whole continuum of stages from having
the intuition that the laws of physics would permit a vehicle with rotating
blades to fly, through working out in detail the relevant equations and
designing possible configurations of blades, to developing the right tech-
nology and materials to actually build and fly a helicopter. At any point
until the last (and maybe even then given the ever present possibility of
refinement), individuals working on such a project contribute a mix of
more or less formal(ized) and implemented insights.
Transposed back to linguistics, this view means that we should be
prepared to accept insights from any source but that we should not give
up our twentieth-century commitment to aim at formal models of lan-
guage structure. Since one of the conclusions that we argue for in the
present paper is the stepwise acquisition of full configurational structure
in the modern Romance languages, we need to show — at least in
principle — how these changes can be modeled in a framework that does
not build configurationality into its basic architecture, as the various
versions of Chomskyan syntax (GB, principles and parameters, minimal-
ism) all do. This will be the task of the final sections of the paper.

5. Towards a lexical-functional model of change

Canonical X-bar theory can handle most if not all the patterns we have
just observed, but only at the expense of postulating the existence of
1138 N. Vincent

levels of structure prior to the emergence of the forms that would provide
evidence for the structure. Thus, since the presence of a category
P(reposition) in X-bar automatically entails both a complement and a
specifier of P, the slot Spec of P will have to be assumed to exist as soon
as the case forms come to be governed by the former independent
adverbial particle. Indeed, on the widespread view that such adverbial
particles are simply intransitive prepositions (Jackendoff 1973; Emonds
1976: 172ff.), the strongest form of X-bar theory might even be interpreted
to include (unrealized ) Spec and Comp of P at the Proto-Indo-European
stage before case-governing prepositions in the traditional sense have
emerged. Now there is indeed much to be said for the view that assimilates
adverbial particles of this kind to the class of prepositions (cf. already
Jespersen 1992 [1924]: 87ff. for a similar idea), as long as we are not
forced to associate with them unrealized and at some stages unrealizable
arguments and modifiers. In other words, a theory is preferable in which
categories may exist but in which projections of those categories are
associated with them only at a historical stage at which there is overt
evidence for such further layers of structure. On this view, and somewhat
schematically, we could say that PIE has P, early Latin has P and
complement of P, and later Latin/Romance has P, complement of P, and
specifier of P. Projection is then historically delayed until triggered by
the changing distributional frequencies of certain items. The accretion of
each new layer of structure can then be thought as corresponding to, and
making precise, part of the intuitive idea of increased gramaticalization.
Compare now the usual X-theory treatment of morphological case
(cf. Bittner and Hale 1996). On this view, since the Latin ablative, say,
corresponds to an English with-phrase, and since the latter is headed by
the preposition with, then for Latin and other such languages a category
KP (for case phrase) is postulated headed by a case morpheme K, which
will then contract the same structural relation to its associated nominal
as the P does to its nominal.27 (For an instance of this form of argumenta-
tion applied in the historical domain, see Weerman 1997.) Putting this
latter idea together with the standard understanding of projection means
that the postulated functional head K in turn automatically projects Spec
of KP and complement of K. Once again there is a problem of unwanted
structure. The essential objection to this way of proceeding is that it
creates a ‘‘syntax’’ for the morphological form that is never realized in
the phonology (Börjars et al. 1997; Vincent and Börjars 1996). This
‘‘syntactic’’ view of morphology rests on the fallacious representational
assumption that the only way to state morphosyntactic generalizations
is in terms of X-bar trees. What is required rather is a means of represent-
ing morphological information on its own terms but ensuring that it can
 The evolution of c-structure 1139

communicate within the grammar with syntactic information, also

represented on its own terms. The way to achieve this is to extract the
functional equivalences that the two patterns — inflectional case and
prepositions — have, and to state this in a single common representation,
an f(unctional )-structure. This f-structure can then be mapped onto the
independently available and autonomously structured realizations in mor-
phology and syntax. We will flesh out such an account for two types of
situation, one involving a semantic use of case/prepositions and the other
a grammatical use ( Vincent and Börjars 1997).
Let us begin with the representation of the content of a semantic case
such as an original Indo-Eurpean locative must have been ( Kuryłowicz
1964: chapter 8). This can be expressed, following Simpson (1991) and
Andrews (1995), by interpreting the semantic case as having its own
argument. Intuitively, a locative case marker expresses a predicate of
location, the argument of which is the (referent of the) noun to which
the case marker is morphologically attached. Formally, this can be stated
as in (21) for the form urbe ‘in/at the city’:

C D
(21) PRED ‘LOC <(OBJ )>’
h
OBJ urb-
h
In (21) OBJ means a theta-object or an object that has an inherent
h
semantic role, in the present instance locative. In the stage in which
in/endo ‘in, inside’ has not yet become a preposition but is rather a
modifier of this locative case, the representation of endo urbe ‘inside the
city’ will be as in (22), where the modifier is modeled as the ADJ(unct)
function in LFG. (NB: In what follows I adopt the fiction of representing
the item in/endo in this modifier function by its earlier shape endo, and
in its prepositional function by its later shape in. I hope this will clarify
the exposition, though it should be understood that the phonetic evolution
from endo>in was independent of and parallel with the grammatical
shift being analyzed here.)

C D
(22) ADJ PRED ‘endo’
PRED ‘LOC <(OBJ )>’
h
OBJ urb-
h
After the grammaticalization of the preposition, we will have

C D
(23) PRED ‘in <(OBJ )>’
h
OBJ urb-
h
A version of this last with its own modifier would be longe in urbe ‘far
1140 N. Vincent

inside the city’ to which we can assign the following representation:

C D
(24) ADJ PRED ‘longe’
PRED ‘in <(OBJ )>’
h
OBJ urb-
h
What these representations do not capture is the morphological forms
of the nouns in question. We need to ensure that in (21) and (22) the
LOC predicate is realized as the locative case, while in (23) and (24) the
object of in is in the ablative case. In order to achieve this we first
consider how to represent an item that has a grammatical case such as
nominative or accusative (see once more Vincent and Börjars 1997 for
discussion of the two types of case). In such circumstances the case will
be a simple feature of the argument in question but crucially will not
have its own argument structure. Consider the sentence in (25a) and its
associated representation in (25b):
(25) a. Exercitus urbem defendit
army.NOMSG city.ACCSG defend.3SG.PRES
‘The army defends the city’
b. tPRED ‘defendere <(SUBJ ), (OBJ )>’u
NTENSE PRES N

C D
N N
N PRED exercitu- N
NSUBJ CASE NOM N
N NUM SG
N
N N

C D
N PRED urb- N
N N
NOBJ CASE ACC N
v NUM SG w
The association between the functions subject and object and the case
markings nominative and accusative will be achieved through lexical
entries for the cases of the following form:
(26) NOM: (SUBJ ()
(( CASE )=NOM
ACC: (OBJ ()
(( CASE )=ACC
Such entries contain the IO (inside-out) operator, which is placed to the
right of the relevant function. They are to be read as ‘‘the item in question
 The evolution of c-structure 1141

has nominative/accusative case and contributes the value subject/object to

the larger f-structure to which it belongs’’ (for more details on this so-called
‘‘constructive case’’ notation, see Nordlinger 1998: chapter 3). In other
words, adding for example the feature ACC to a noun triggers a morpholog-
ical process that finds the appropriate case form for the lexeme in question
and at the same time says that the item thus formed will count as the object
of the sentence. The complex patterns of morphological realization of these
case/number forms according to the declensional class of the noun in ques-
tion will be the responsibility of a separate morphological component,
whose internal structure we will not further articulate here (cf. Blake [119:
20–23] for the arguments against mapping function straight into morpho-
logical form in a language like Latin). Note in particular that in a system
of this kind there is no need to allocate the subject and object to configura-
tionally determined slots in order for them to receive their case marking.
The kind of direct function–case mapping exploited in (26) fits very well
with the fact that the word order in Latin is relatively free, and that therefore
any linear rearrangement of the sentence in (25a) will retain the same
f-structure values as long as the case forms are not changed. Conversely, if
the case forms are changed, for instance by making exercitus ‘army’ accusa-
tive and urbs ‘city’ nominative, then the meaning and the f-structure of the
sentence change even if word order is held constant.
Let us return now to the question of assigning case forms in the context
of the grammatical change from modifier to preposition itemized in (21)
thorough (24) above. At the original stage represented in (21) the locative
value of the locative case can be expressed as follows:
(27) LOC: (ADJ ()
(( CASE )=LOC
PRED ‘LOC <OBJ >’
h
(27) says that the item in the locative case bears the function of adjunct
in its clause. The prepositional case instead requires the following state-
ment, which must be construed as part of the lexical entry for in:28
(28) in: PRED ‘in <OBJ >’
h
OBJ CASE=ABL
h
In other words the shift from modifier to preposition involves in/endo
acquiring both an OBJ argument in its PRED feature and a case-assigning
feature that determines the morphological shape of that argument.
Inspection of the above f-structures reveals that all the necessary equiva-
lences and differences are represented. When the original locative case
is serving a semantic function than it is functionally equivalent to a
1142 N. Vincent

preposition, and this is expressed in the similarity of their respective PRED

features, viz.
(29) a. loc ‘LOC <OBJ >’ (as in [21] and [22])
h
b. in ‘in <OBJ >’ (as in [23] and [24])
h
When the erstwhile modifier acquires prepositional force, then it takes over
the PRED feature and the case is automatically downgraded to grammatical
status and is assigned in virtue of the f-structure value of the item to which
it attaches, just as happens for clausal subjects and objects.
Finally in this section we see how an extension of the apparatus just
established will provide us with a natural account of the equivalences
between case forms and PPs reviewed in section 3.6 above. I will simplify
the discussion somewhat by extracting the core propositional structure
out of the Plautine examples cited in (17), and repeating it as (30):
(30) a. ea patri nuntias
that.NEUT.ACCPL father.DATSG tell.PRES.2SG
‘you tell those things to your father’
b. ea ad patrem nuntias
that.NEUT.ACCPL to father.DATSG tell.PRES.2SG
‘you tell those things to your father’
We can assume a function–case equivalence for dative inherited from
Indo-European as follows:29
(31) DAT: (IND OBJ ()
(( CASE )=DAT
This will allow us to state the f-structure of (30a) as
(32) tPRED ‘nuntiare <(SUBJ ), (OBJ ), (IND OBJ )>’u
NTENSE PRES N

C D
N N
PRED ‘addressee’
N N
NSUBJ CASE NOM N
N NUM SG N

C D
N N
N PRED is N
NOBJ CASE ACC N
N N
NUM PL
N N

C D
N PRED patr-
N
N N
NIND OBJ CASE DAT N
v NUM SG w
 The evolution of c-structure 1143

To complete the story we now need it to be possible for the f-structure

of ad patrem in (30b) to substitute for and be equivalent to the f-structure
of patri in (30a). In other words, we need to treat the PP here not as
headed by P with an independent PRED feature as in our treatment of
in above, but as constituting a periphrastic case form. To achieve this, I
will exploit the type of analysis developed in Börjars et al. (1997). The
function–case statement required is
(33) AD: (( IND OBJ )
(P-OBJ CASE )=ACC
P
(33) says that ad is a preposition, that it contributes the value for the
indirect object function to the larger f-structure of which it is part, that
is, the prepositional phrase, and that the case form of the object within
that prepositional phrase is accusative. The information in (33) consti-
tutes one lexical entry for the preposition ad. There will be a second
entry for ad in its semantically full sense as a marker of the theta role
goal. This will have the form in (34):
(34) ad PRED ‘ad <OBJ >’
h
OBJ CASE=ACC
h
Now there would be a failure of unification if we were to try to plug in
the preposition ad with the lexical entry in (34) in a context where an
indirect object was required. Thus the choice between which entry for ad
to use in a given sentence is a consequence of the f-structure of that
sentence. Note too that even if ad loses its PRED feature it does not give
up its syntactic subcategorization, which is still accusative, and hence the
correct case is assigned to patrem. An exactly parallel account can be
developed for the grammaticalization of de ‘down, about’ into the marker
of nominal dependent, first in competition with and finally as a replace-
ment for genitive case. The consistent generalization across both the
diachronic scenarios discussed in this paper and those in Börjars et al.
(1997) is that grammaticalization can be neatly modeled in LFG terms
as loss of the PRED feature, with the remainder of the changes following
from the independently required architecture of the model.

6. Toward an optimality-theory account of syntactic change

In the previous section I have sought to show that a parallel corre-

spondence model such as LFG provides a more natural account of the
data outlined in section 3. In recent work, Bresnan (forthcoming a,
1144 N. Vincent

forthcoming b, forthcoming c) has argued further that the LFG frame

work is best construed in optimality-theory terms. On this view, the
relation between f-structure and c-structure is determined by the inter-
action of rerankable constraints rather than in absolute terms. Although
we will not work out a full account here, three observations indicate the
attractiveness of the OT approach also in the present instance.
The first concerns the constraint proposed in Bresnan (forthcoming a)
under the name of the principle of economy of expression:
(35) DON’T PROJECT: All non-preterminal c-structure nodes are
optional and not used unless required by independent principles
As she notes, this constraint ‘‘privileges lexical over phrasal expression
wherever possible.’’ We should note too that for these purposes ‘‘lexical’’
includes fully inflected items according to the principle of lexical integrity.
The consequence of this principle applied to a case form of a language
like Latin or Sanskrit (and hence, for the purposes of the present argu-
ment, to the relevant reconstructible stage of the parent language) is that
unless other factors intervene, such as the requirements of other items in
the f-structure (or members of the ‘‘numeration’’ in minimalist terms),
no further structure is required. In particular, there are no empty Ds or
Ps required. Nor, even if a lexical category node is required, will it be
necessary to project higher-level structure from that node. In the context
of the data studied here, this means that at the earliest Indo-European
stage the morphologically complete lexical entries will not need to be
licensed by empty prepositions or the like. They will simply express the
categories of f-structure directly. As each successive historical reanalysis
takes place, it will introduce new levels of structure via a reranking of
DON’T PROJECT vis-à-vis the endocentricity constraints that ensure
overt expression of grammatical content.
The second observation that leads naturally in the OT direction con-
cerns the notation of morphosyntactic competition (Bresnan forthcoming
b, forthcoming c). In these papers Bresnan develops the idea that there
is a hierarchical ranking of different forms of expression for a given
grammatical content. Just as she proposes a scale for the expression of
pronominal content ranging from zero to full syntactically independent
pronouns, we can think of a similar scale for the expression of case and
prepositional type meanings. On the not-unreasonable assumption that
independent prepositions are higher on such a hierarchy than bound case
morphology, just as pronouns are higher than agreement morphology,
we should expect the development of the former to take precedence over
the historically inherited latter. It is difficult to be more precise at this
point in the research since a full case–preposition scale remains to be
 The evolution of c-structure 1145

worked out. However, for a more detailed investigation of the logic of

this type of argumentation in a diachronic projection of Bresnan’s
pronoun hypothesis, see Vincent (forthcoming).
Third, recall the discussion of the relative ordering of the preposition
and the question word in examples such as quem ad modum, etc. This
requires, as indeed Broadwell (1999) proposes for the San Dominicio
Zapotec data cited in (11), that the constraint forcing the interrogative
to align with the left edge of the clause should outrank the constraint
that the head of the constituent occur on the left edge. Subsequent
developments, in which this order is lost, suggest the reranking of these
constraints. Note further that the increased prominence of the constraint
that makes the preposition leftmost will block the kind of fronting that
we saw in the magna cum laude construction, and indeed neither
possibility has survived into the modern Romance languages.

7. Conclusion

To conclude, then, our aim in the present study has been to trace the
historical evolution of Romance prepositions from their earliest origins
in Indo-European sentence-level particles and to show how the various
ingredients of a full-fledged category of prepositional phrase have
emerged at different stages over a long historical timespan. We have
interpreted this evidence as showing not that there is a future-oriented
teleological pattern to change, but rather that a sequence of changes may
be pushed from behind, each successive stage facilitating the next one.
We have also suggested that this particular example favors an architecture
of grammar in which the system of grammatical relations is separated
from the expression of such relations, and we have finally hinted that the
whole model of grammar may benefit from being interpreted in optimal-
ity-theory terms rather than in the more conventional absolutist sense.

Received 30 June 1998 University of Manchester

Revised version received
12 October 1999

Notes

* Parts of the material in this paper were presented at a workshop on grammaticalization

held during LFG97 at UCSD in June, 1997, and at a workshop on functional categories
and syntactic change held during ICHL13 at Düsseldorf in August, 1997. I am grateful
to those who raised questions on those occasions, and to the organizers, respectively
1146 N. Vincent

Farrell Ackerman and Ans van Kemenade for inviting me to participate. I am grateful
to Ans, ‘‘JC’’ Smith, Joan Bresnan, and Kersti Börjars, who read and commented on
early drafts, and particularly to Ans again for her further suggestions in the light of the
referees’ comments. My thanks too to Andrew Spencer for his comments and notes on
the development of prepositions in Slavic and Greek, and to Anna Morpurgo Davies,
Jim Adams, and David Langslow for conversations on Indo-European and Latin. I
would also like to thank the referees — one of whom subsequently revealed himself to
be Christian Lehmann — for their helpful comments. The work reported here is part
of a larger project funded through a Research Readership from the British Academy,
whose support is gratefully acknowledged.
Correspondence address: Department of Linguistics, University of Manchester,
Manchester M13 9PL, UK. E-mail: [email protected].
1. In this sense LFG enshrines the syntactic equivalent of what Aronoff (1994: 8ff.) calls
the ‘‘separationist hypothesis’’ in morphology. For an elaboration of the latter from
the perspective of LFG, see Börjars et al. (1997) and Vincent and Börjars (1996).
2. On a terminological note, I will use constituency to mean the hierarchical, configura-
tional pattern familiar from X-bar theory. C-structure is the representation within
LFG that encodes the same information. The difference is that within LFG c-structure
may also be ‘‘flat’’ and/or ‘‘morphological’’ according to the dictates of the language
under analysis, whereas X-bar theory is standardly conceived as a universal configura-
tional template to which all morphosyntactic structures in all languages must conform.
3. One referee points out that the accusative in many Indo-European languages — and
plausibly therefore also in the parent language — also had an adverbial function. This
is of course true, but a detailed discussion would take us too far from the main track of
the argument. Suffice it to say that I would envisage an analysis in terms of default case
along the lines of Smith (1996), a solution that he in turn attributes to Gaedicke
(1880). The fact remains that case constructions just as much as prepositional ones
require a distinction between grammatical and concrete uses, and this argues against
treating grammatical uses of prepositions as case markers.
4. Jackendoff ’s system is taken over under the feature names [± predicative] and
[± transitive] in Bresnan (forthcoming d).
5. Horrocks (1997) argues for the origins of Indo-European grammatical case in an
earlier system of semantic relations (cf. already Meillet 1937: 357–359). Even if this
(controversial ) view is accepted, however, the fact remains that the acquisition of case-
marked dependents by prepositions is a much later development within the Indo-
European family. This incontrovertible time lapse between the development of verbal
and prepositional case assignment suffices for the purposes of our present argument.
6. It is the absence of this construction in later Latin that argues against an analysis where
endo is treated as a preposition with a pro object. It is not that Latin did not have object
pro; it did but never with prepositions (van der Wurff 1993; Luraghi 1997; Pieroni
1999), hence it is unlikely that an early example such as this is to be understood in
this way.
7. One of my referees called me to order on the question of terminology here, so a word
of clarification is perhaps useful. ‘‘Particle’’ is a conventional label for a very hetero-
geneous class of elements whose one common property is that they do not inflect and
thus belong to a distinct class from substantives (nouns, adjectives) on the one hand
and verbs on the other. Many of these particles express meanings that modify the
sentence as a whole so that in terms of function they may fairly be described, without
stretching traditional terminology, as adverbs. There is no contradiction therefore in
 The evolution of c-structure 1147

describing them, when necessary, as ‘‘adverbial particles,’’ thereby uniting form and
function in a single composite label.
8. As is often noted, the traditional label is something of a misnomer since it suggests a
division of something that was formerly a unit, whereas at the point at which ‘‘tmesis’’
abounds it is precisely because the full effects of so-called ‘‘univerbation’’ have not yet
set in.
9. For extensive discussion of these matters, and for opposing views as to whether the
principal conditioning factors are prosodic or syntactic, see the exchange between Hale
(1996) and Hock (1996). Neither scholar, however, challenges the traditional view that
separated preverbs in Sanskrit are stressed. This prosodic property, which is clearly
derived from an Indo-European stage at which the items in question were not yet
preverbs but independent clausal elements, is important for our discussion below of the
ordering facts in early Latin.
10. Unstressed main verbs migrate to second position, as already noted by Wackernagel:
cf. Anderson (1993).
11. This conclusion, which is no more than what is standardly agreed by all the manuals
and commentators since the last century, is not threatened by evidence that there were
other items that were already preverbs (and perhaps, though this is less clear, adposi-
tions) in the Indo-European parent language (Pinault 1995). Nor are the objections of
Pinkster (1972: chapter 9.2) especially persuasive. He is certainly right that throughout
attested Latin there is a clear relation of government between preposition and case, but
examples such as those in (4) show us the residue of an earlier stage in which no such
relation was required. The existence of such a stage is amply confirmed by the sister
languages whose attestation is more ancient. The important point for our purposes is
that the items that make up the bulk of the prepositions and preverbs in the daughter
languages display signs of syntactic independence in texts such as those cited here for
Latin and Homeric Greek (8th cent BCE ) and Vedic and Hittite (2nd millenium BCE ).
Even the earliest of these is two or more millennia later than the latest datings for I-E
unity. For a similar interpretation of these facts, see also the discussion in Lehmann
(1995: chapter 3.4.1.4).
12. Note that this origin explains the absence of prepositions governing the dative case,
which did not have local values in the proto-language. Where in a modern language we
find a prepositional dative — e.g. Germanic, Greek — this is due to a later syncretism
of the original dative with one or more of the local cases ( Vincent 1997c).
13. As Christian Lehmann reminds me, if pre-Latin adpositions had by contrast been
denominal and not deadverbial, they would in all probability have been postpositions,
as is the case with the much later items causa ‘because of ’ (<causa ‘reason’), gratia ‘on
account of ’ (<gratia ‘favor’).
14. A referee queries the relevance of the Oscan data here. The point is simply that from a
very similar, perhaps even identical, starting point that language developed a pattern
that Joseph argues was structurally impossible for Latin. Since in other morphological
respects the languages have much in common, the outcome in Oscan suggests that
Joseph was looking in the wrong place to find an answer for the Latin facts. Moreover,
although it is true, as the same referee notes, that items such as -que had always been
enclitic and therefore had a different status from particles in the ancestor language, the
fact remains that they could be attached after the inflection, once again challenging the
claim that the morphological inflection comes last in the word.
15. A clue to the different accentual status of the separated and unseparated preverb in
Latin is found in the vowel alternations in sub placo vs supplico, ob sacro vs obsecro. In
the separated form there is a stress on each part and hence the verb stem has the full
1148 N. Vincent

vowel a. However, the preverb+verb cluster receives a single word stress and hence
the stem vowel reduces: cf. cano ‘I sing’, tango ‘I touch’ beside cecini ‘I sang’, tetigi ‘I
touched’ with stress on the reduplicated perfective prefix. Note that the stress here is
on the Latin initial syllable as a consequence of the normal functioning of the Latin
word-stress rule; it is purely coincidental that the stressed portion of the word
corresponds to the preverb in supplico and obsecro.
16. The one branch where postpositions are undeniable is Hittite, but these are an internal
development within that language, as is shown by the fact that the evidence for post-
positions is stronger in Middle Hittite than in Old Hittite (cf. Starke 1977). Here too,
as Christian Lehmann notes, the source of the postpositions lies in a class of original
adverbs. I am grateful to Anna Morpurgo Davies for bringing the Hittite facts to my
attention and for explaining their significance in the broader context of reconstructing
the syntax of I-E preverbs.
17. Similar emphasis must be attributed to the fronted numeral in duabus de causis ‘for two
reasons’ (Caes BG 6.9). One referee accuses me of circularity at this point: I claim the
fronted adjective is emphatic; how do I know it’s emphatic? because it’s fronted! It is
true that it is often a matter of choice whether to read a passage as emphatic, and in
the absence of native speakers we cannot check our semantic intuitions. In addition,
with some of these adjectives the pattern must have become virtually fixed as in the
summa cum laude formula applied to university degrees in many countries. Nonetheless,
the fact that not all adjectives freely occur in this position outside poetry (where
metrical and other considerations can play a decisive role), and that the class of
adjectives that are found here naturally lend themselves to a contrastive or intensifying
reading suggests that a particular pragmatic effect of highlighting or emphasis is
intended. The effect is similar to the fronting of a focused element in a clause, but it is
important to note that this is a separate constituent-internal pattern. Displacement of
items over a longer distance — what is traditionally called hyperbaton — is much rarer
(cf. also note 17).
18. Though compare Nocentini (1992: 236) for the claim that this pattern is not to be
attributed to the Wackernagel effect.
19. It remains true that Latin had fairly extensive patterns of topicalization or scrambling
in which an element could ‘‘move’’ to the front of its clause and some distance from its
coconstituents (cf. [9b]). It also seems likely that this was characteristic of all styles of
the language since it leads in due course to the new patterns attested in Romance
(cf. Vincent 1998 for some discussion). I will assume however that these patterns are to
be handled independently, an assumption supported by the fact that the grammatical
distance involved and the classes of categories affected are quite different.
20. The only exceptions here are the use of usque with certain locational nouns where it
can modify the bare directional accusative: terminos usque Libyae ‘to the borders of
Libya’ (Just 1,1,5). This is of course a parallel with the type of case-modificational
structure out of which prepositions emerged in the first place (cf. section 3.2). By
contrast, usque is not found with nontemporal and nonspatial prepositions
(cf. Rooryck 1996: 226 on the distribution of right in English).
21. I am grateful to David Langslow for pointing out to me the significance of these
examples.
22. There is a long tradition in Latin grammar of seeking to show that case forms such as
dative and genitive are semantically, and hence distributionally, distinct from PPs with
ad and de respectively (cf. most recently Pinkster 1990). While there are undoubtedly
some tendencies here, the material collected by Molinelli amply demonstrates the
possibility of genuine morphosyntactic and semantic equivalence of a PP and a case
 The evolution of c-structure 1149

form. It is this circumstance that motivates the theoretical discussion in the concluding
section of the present paper.
23. As I was revising the text of this article, I came upon Rosén (1999), which contains
useful discussion of the historical profile of prepositional constructions replacing case-
based ones with many further examples (cf. especially 1999: 137–149).
24. Attested examples such as de in equis, lit. ‘about on horses’ are different in that here
in equis is a metonymic expression meaning ‘those on horseback, i.e. cavalry’.
25. Note that this is not intended as a plea for a particular theory of argument structure;
there are many available in the current literature and the debates about which to choose
would take us too far afield. The point is simply that the kind of change we are
discussing here is hard to conceive of in the absence of some kind of explicit theoretical
characterization of grammatical structure.
26. On the whole question of degrammaticalization, see now van der Auwera (1999). For
an account of why grammaticalization cannot be reversed, see Haspelmath (this issue).
27. Joan Bresnan reminds me that Emonds (1985: 224) had already proposed a variant of
this analysis according to which a bare case form is treated as a structural PP headed
by an empty preposition.
28. I ignore here the additional complexities of the alternation between ablative and
accusative after certain prepositions in Latin. For discussion see Vincent (1997c) and
Vincent and Börjars (1997).
29. In what follows I have departed from LFG orthodoxy by introducing a function IND
OBJ. I do so partly to clarify the exposition for nonspecialist readers but also because
the Latin situation does not readily lend itself to a treatment in terms of the LFG
distinction between an OBJ function and an OBJ-theta function. The function required
here is equivalent to what in relational grammar is labelled ‘‘3.’’

References

Ackerman, Farrell; and Moore, John (1999). Syntagmatic and paradigmatic dimensions of
causee encodings. Linguistics and Philosophy 22, 1–44.
Anderson, John (1997). A Notional Theory of Syntactic Categories. Cambridge: Cambridge
University Press.
Anderson, Stephen (1993). Wackernagel’s revenge: clitics, morphology and the syntax of
second position. Language 69, 68–98.
—(1996). How to put your clitics in their place, or why the best account of second-position
phenomena may be something like the optimal one. Linguistic Review 13, 165–191.
Andrews, Avery (1995). Semantic case-stacking and inside-out unification. Australian
Journal of Linguistics 15, 127–181.
Aronoff, Mark (1994). Morphology by Itself. Cambridge, MA: MIT Press.
Baldi, Philip (1979). Typology and the Indo-European prepositions. Indogermanische
Forschungen 84, 49–61.
Barron, Julia (1997). LFG and the history of raising verbs. Paper presented at LFG97,
University of California at San Diego, June 19–21.
Battye, Adrian; and Roberts, Ian (eds.) (1995). Clause Structure and Language Change.
Oxford and New York: Oxford University Press.
Bauer, Brigitte (1995). The Emergence and Development of SVO Patterning in Latin and
French. Oxford and New York: Oxford University Press.
Bittner, Maria; and Hale, Ken (1996). The structural determination of case and agreement.
Linguistic Inquiry 27, 1–68.
1150 N. Vincent

Blake, Barry (1994). Case. Cambridge: Cambridge University Press.

Börjars, Kersti; Chisarik, Erika; and Payne, John (1999). On the justification of functional
categories in LFG. Paper presented at LFG99, University of Manchester, July 19–21.
Börjars, Kersti; Vincent, Nigel; and Chapman, Carol (1997). Paradigms, periphrases and
pronominal inflection: a feature based account. Yearbook of Morphology 1996, 155–180.
Bresnan, Joan (forthcoming a). Optimal syntax. In The Pointing Finger. Conceptual Studies
in Optimality Theory, J. Dekkers et al. (eds). Oxford: Oxford University Press.
—(forthcoming b). Explaining morphosyntactic competition. To appear in Handbook of
Contemporary Syntactic Theory, Mark Baltin and Chris Collins (eds). Oxford: Blackwell.
—(forthcoming c). The emergence of the unmarked pronoun. In Optimality Theoretic
Syntax, Jane Grimshaw, Geraldine Legendre, and Sten Vikner (eds.). Cambridge, MA:
MIT Press.
—(forthcoming d). Lexical-Functional Grammar. Oxford: Blackwell.
Broadwell, George Aaron (1999). The interaction of focus and constituent order in San
Dionicio Zapotec. Paper presented at LFG99, University of Manchester, July 19–21.
Bybee, Joan; Perkins, Revere; and Pagliuca, William (1994). The Evolution of Grammar.
Tense, Aspect and Modality in the Languages of the World. Chicago and London:
University of Chicago Press.
Chomsky, Noam (1957). Syntactic Structures. The Hague: Mouton.
—(1995). The Minimalist Program. Cambridge, MA: MIT Press.
Coleman, Robert (1991). Latin prepositional syntax in Indo-European perspective. In New
Studies in Latin Linguistics, R. Coleman (ed.), 323–338. Amsterdam: Benjamins.
Craig, Colette; and Hale, Ken (1988). Relational preverbs in some languages of the
Americas: typological and historical perspectives. Language 64, 312–344.
Cuzzolin, Pierluigi (1995). A proposito di sub vos placo e della grammaticalizzazione delle
adposizioni. Archivio Glottologico Italiano 80, 122–143.
Emonds, Joseph (1976). A Tranformational Approach to English Syntax. New York and
London: Academic Press.
—(1985). A Unified Theory of Syntactic Categories. Dordrecht: Foris.
Gaedicke, Carl (1880). Die Akkusativ im Veda. Breslau: Koebner.
Gisborne, Nikolas (1999). English light verbs and the (non)necessity of argument structure.
Paper presented at LFG99, University of Manchester, July 19–21.
Green, Georgia (1974). Semantics and Syntactic Regularity. Bloomington: Indiana
University Press.
Hale, Mark (1996). Deriving Wackernagel’s Law: prosodic and syntactic factors determin-
ing clitic placement in the language of the Rigveda. In Approaching Second. Second
Position Clitics and Related Phenomena, Aaron L. Halpern and Arnold Zwicky (eds.),
165–197. Stanford: CSLI.
Halpern, Aaron L.; and Zwicky, Arnold (eds.) (1996). Approaching Second. Second Position
Clitics and Related Phenomena. Stanford: CSLI.
Harris, Alice; and Campbell, Lyle (1995). Historical Syntax in Cross-Linguistic Perspective.
Cambridge: Cambridge University Press.
Harris, Martin (1978). The inter-relationship between phonological and grammatical
change. In Recent Developments in Historical Phonology, J. Fisiak (ed.), 159–172. The
Hague: Mouton.
Haspelmath, Martin (1998). Does grammaticalization need reanalysis? Studies in Language
22, 49–85.
Hawkins, John (1983). Word Order Universals. New York: Academic Press.
Hock, Hans Heinrich (1996). Who’s on first? Toward a prosodic account of PS clitics. In
Approaching Second. Second Position Clitics and Related Phenomena, Aaron L. Halpern
and Arnold Zwicky (eds.), 199–270. Stanford: CSLI.
 The evolution of c-structure 1151

Hopper, Paul; and Traugott, Elizabeth (1993). Grammaticalization. Cambridge: Cambridge

University Press.
Horrocks, Geoff (1997). Greek voice, Indo-European echoes. Paper delivered to the
Philological Society, Oxford, June 7.
Hübschmann, Heinrich (1875). Zur Kasuslehre. Munich.
Jackendoff, Ray (1973). The base rules for prepositional phrases. In A Festschrift for Morris
Halle, S. R. Anderson and P. Kiparsky (eds.), 345–356. New York: Holt, Rinehart
and Winston.
—(1977). X-bar Syntax. Cambridge, MA: MIT Press.
Jespersen, Otto (1992 [1924]). The Philosophy of Grammar. Chicago: University of Chicago
Press.
Joseph, John (1991). Latin prepositions and Romance syntax. In New Analyses in Romance
Linguistics, D. Wanner and D. A. Kibbee (eds.). Amsterdam: Benjamins.
Kaplan, Ronald M.; and Bresnan, Joan (1982). Lexical-functional grammar: a formal
system for grammatical representation. In The Mental Representation of Grammatical
Relations. J. Bresnan (ed.), 173–281. Cambridge, Mass: MIT Press. (Reprinted [1995]. In
Formal Issues in Lexical-Functional Grammar, M. Dalrymple et al. [eds.], 29–130.
Stanford: CSLI.)
Kayne, Richard (1994). The Antisymmetry of Syntax. Cambridge, MA: MIT Press.
Kortmann, Bernd; and König, Ekkehard (1992). Categorial reanalysis: the case of deverbal
prepositions. Linguistics 30, 671–697.
Kuryłowicz, Jerzy (1964). The Inflectional Categories of Indo-European. Heidelberg: Carl
Winter.
Lehmann, Christian (1983). Latin preverbs and cases. In Latin Linguistics and Linguistic
Theory, Harm Pinkster (ed.), 145–165. Amsterdam: Benjamins.
—(1995). Thoughts on Grammaticalization. Unterschleissheim: LINCOM Europa.
—(1998). German abstract prepositional phrases as reduced subordinate clauses. In Clause
Combining and Text Structure, Iørn Korzen and Michael Herslund (eds.), 87–106.
Copenhagen Studies in Language 22. Copenhagen: Samfundslitteratur.
Lewis, Charlton; and Short, Charles (1879). A Latin Dictionary. Oxford: Oxford University
Press.
Lightfoot, David (1999). The Development of Language. Oxford: Blackwell.
Longobardi, Giuseppe (1995). A case of construct state in Romance. In Scritti linguistici e
filologici in onore di Tristano Bolelli, R. Ajello and S. Sani (eds.), 293–329. Pisa: Pacini.
Luraghi, Silvia (1997). Omissions of the direct object in Latin. Indogermanische Forschungen
102, 239–257.
Matthews, P.H. (1981). Syntax. Cambridge: Cambridge University Press.
Meillet, Antoine (1937). Introduction à l’étude comparative des langues indo-européennes.
Paris: Hachette.
Molinelli, Piera (1996). Casi e preposizioni in latino: lo sviluppo del genitivo e del dativo.
Una prospettiva (anche) sociolinguistica. Linguistica e Filologia 3, 73–125.
—(1998). Dai casi alle preposizioni in latino: analisi sociolinguistica e spiegazione tipolog-
ica. In Sintassi storica, Paolo Ramat and Elisa Roma (eds.), 147–166. Rome: Bulzoni.
Newmeyer, Frederick J. (1998). Language Form and Language Function. Cambridge, MA:
MIT Press.
Nocentini, Alberto (1992). Preposizioni e posposizioni in oscoumbro. Archivio Glottologico
Italiano 77, 196–242.
Nordlinger, Rachel (1998). Constructive Case. Evidence from Australian Languages.
Stanford: CSLI.
1152 N. Vincent

Penney, John (1999). Archaism and innovation in Latin poetic syntax. In Aspects of the
Language of Latin Poetry, J. N. Adams and R. J. Mayer (eds.), 249–268. Proceedings of
the British Academy 93. Oxford: Oxford University Press.
Pieroni, Silvia (1999). Subject properties and semantic roles in Latin. Paper delivered at the
Xth International Colloquium on Latin Linguistics, Paris-Sèvres, April 19–23.
Pinault, Georges-Jean (1995). Le problème du préverbe en indo-européen. In Les Préverbes
dan les langues d’Europe, André Rousseau (ed.), 35–59. Lille: Presses Universitaires du
Septentrion.
Pinkster, Harm (1972). On Latin Adverbs. Amsterdam: North Holland.
—(1990). The development of cases and adpositions in Latin. In Unity in Diversity. Papers
Presented to Simon C. Dik on his 50th Birthday, H. Pinkster and I. Genee (eds.), 195–209.
Amsterdam.
Pollard, Carl; and Sag, Ivan (1994). Head-Driven Phrase Structure Grammar. Chicago and
Stanford: Chicago University Press and CSLI.
Ramat, Paolo (1992). On degrammaticalization. Linguistics 30, 549–560.
—; and Roma, Elisa (eds.) (1998). Sintassi storica. Rome: Bulzoni.
Renou, Louis (1933). La séparation du préverbe et du verbe en védique. BSL 34, 49–96.
Roberts, Ian (1985). Agreement patterns and the development of English modal auxiliaries.
Natural Language and Linguistic Theory 3, 21–58.
—(1993). A formal account of grammaticalization in the history of Romance futures. Folia
Linguistica Historica 13, 219–258.
—(1997a). Directionality and word order change in the history of English. In Parameters of
Morphosyntactic Change, Ans van Kemenade and Nigel Vincent (eds.), 397–426.
Cambridge: Cambridge University Press.
—(1997b). A formal account of grammaticalization. Paper presented at Workshop on
Functional Categories and Morphosyntactic Change, ICHL13, Düsseldorf, August 12.
Rooryck, Johan (1996). Prepositions and minimalist case marking. In Studies in
Comparative Germanic Syntax II, H. Thráinsson, S. D. Epstein, and S. Peter (eds.),
226–256. Amsterdam: Kluwer.
Rosén, Haim (1992). Die Komposita mit co(n)- in funktioneller und vergleichender Sicht.
In Latein und Indogermanisch, O. Panagl and T. Krisch (eds.), 357–367. Innsbruck:
Innsbrucker Beitrage zur Sprachwissenschaften.
Rosén, Hannah (1999). Latine Loqui. Trends and Directions in the Crystallization of Classical
Latin. Munich: Wilhelm Fink.
Rousseau, André (ed.) (1995). Les Préverbes dans les langues d’Europe: introduction à l’étude
de la préverbation. Lille: Presses Universitaires du Septentrion.
Simpson, Jane (1991). Warlpiri Morphosyntax. Dordrecht: Kluwer.
Smith, Henry (1996). Restrictiveness in Case Theory. Cambridge: Cambridge University
Press.
Starke, Frank (1977). Die Funktionen der dimensionalen Kasus und Adverbien im
Althethitischen. Wiesbaden: Harrassowitz.
Szemerényi, Oswald (1996). Introduction to Indo-European Linguistics. Oxford: Oxford
University Press.
van der Auwera, Johan (1999). Degrammaticalization. Paper delivered at the Potsdam
Grammaticalization Conference, June.
van der Wurff, Wim (1993). Null objects and learnability: the case of Latin. Paper presented
at the Xth International Conference on Historical Linguistics, UCLA, August.
van Kemenade, Ans; and Vincent Nigel (eds.) (1997). Parameters of Morphosyntactic
Change. Cambridge: Cambridge University Press.
 The evolution of c-structure 1153

Vincent, Nigel (1980). Iconic and symbolic aspects of syntax: prospects for reconstruction.
In Linguistic Reconstruction and Indo-European Syntax, Paolo Ramat (ed.), 47–68.
Amsterdam: Benjamins.
—(1993). Head vs. dependent marking: the case of the clause. In Heads in Grammatical
Theory, Greville Corbett et al. (eds.), 140–163. Cambridge: Cambridge University Press.
—(1995). Exaptation and grammaticalization. In Historical Linguistics 1993, Henning
Andersen (ed.), 433–445. Amsterdam: Benjamins.
—(1997a). Prepositions. In The Dialects of Italy, Martin Maiden and Mair Parry (eds.),
208–213. London: Routledge.
—(1997b). Lexical and historical grammar: the case of chez. Paper delivered at Conference
on the Boundaries of Morphology and Syntax, Rome, October 23–25.
—(1997c). On the origin of prepositions. Paper delivered to the Philological Society,
Manchester, November 21.
—(1998). Tra grammatica e grammaticalizzazione: articoli e clitici nelle lingue
(italo-)romanze. In Sintassi storica, Paolo Ramat and Elisa Roma (eds.), 411–440.
Rome: Bulzoni.
—(forthcoming). Competition and correspondence in syntactic change: null arguments in
Latin and Romance. In Diachronic Syntax: Models and Mechanisms, Susan Pintsuk,
George Tsoulas, and Anthony Warner (eds.). Oxford and New York: Oxford University
Press.
—; and Börjars, Kersti (1996). Suppletion and syntactic theory. In Papers from the 1st LFG
Colloguium, Grenoble, August 1996, M. Butt and T. King (eds.), 448–462. Grenoble:
Rank Xerox.
—; and Börjars, Kersti (1997). Stating your case. Paper delivered at the First Mediterranean
Morphology Meeting, Mytilini, September.
Weerman, Fred (1997). On the relation between morphological and syntactic case. In
Parameters of Morphosyntactic Change, Ans van Kemenade and Nigel Vincent (eds.),
420–459. Cambridge: Cambridge University Press.
Weir, Ellen M. H. (1986). Footprints of yesterday’s syntax: diachronic development of
certain verb prefixes in an OSV language (Nadëb). Lingua 68, 291–316.