Erschienen in: Syntax - Theory and Analysis : An International Handbook, Vol. 2 / Kiss, Tibor (Hrsg.).

- Berlin : De Gruyter
Mouton, 2015. - (Handbücher zur Sprach- und Kommunikationswissenschaft ; 42,2). - S. 839-874
https://dx.doi.org/10.1515/9783110363708-002
25. Lexical-Functional Grammar 839

25. Lexical-Functional Grammar

1. Introduction
2. LFG basics: c- and f-structure
3. Syntactic phenomena
4. Argument structure
5. Interfaces/other projections
6. OT-LFG
7. Computational issues and resources
8. Psycholinguistic research
9. Conclusion
10. References (selected)

Abstract
LFG is a constraint-based theory with the goal of combining linguistic insight with
computational tractability and implementability. As LFG separates surface form from
functional dependency, restrictive assumptions about configurationality dominate analy-
ses. LFG has been used to analyze diverse phenomena, including discontinuous constitu-
ents and long-distance dependencies in a large number of typologically diverse lan-
guages. As a result, a broad range of theoretical, descriptive and computational linguists
work within LFG.

1. Introduction
Lexical-Functional Grammar (LFG) took shape in the late 1970s when Joan Bresnan’s
linguistic concerns about the continued viability of Transformational Grammar met up
with Ron Kaplan’s ideas about psycholinguistics and computational modelling. The col-
lection of papers in Bresnan (1982b) sets out the fundamental ideas of LFG. The theory

Konstanzer Online-Publikations-System (KOPS)

URL: http://nbn-resolving.de/urn:nbn:de:bsz:352-2-1pbpz08c5itbi6
840 IV. Syntactic Models

has since been extended to include new ideas and cover more data from a wide array of
languages, but the fundamental ideas put forth in the late 1970s and early 1980s continue
to be valid.
LFG has the goal of combining linguistic sophistication with computational imple-
mentability (Dalrymple et al. 1995). A broad range of theoretical, descriptive and compu-
tational linguists work within LFG, with some work also being done in psycho- and
neurolinguistics. Several current textbooks are available (Bresnan 2001; Dalrymple
2001; Falk 2001), as is a description of a major computational grammar development
effort (Butt et al. 1999).
This chapter is structured as follows. Section 2 discusses the basic LFG architecture.
With just this basic architecture, interesting linguistic phenomena can be dealt with sim-
ply and elegantly. This is demonstrated in section 3, where we provide sample analyses
for some core syntactic phenomena. In section 4, we introduce LFG’s linking theory and
its application to data that have been discussed centrally within LFG. In section 5, we
discuss interfaces to other modules of grammar, such as prosody, information-structure,
semantics and morphology, as well as newer developments with respect to combining
Optimality Theory (OT) with LFG. The chapter closes with a look at computational
issues and resources in section 7.

2. LFG basics: c- and f-structure

LFG posits two distinct syntactic representations: c(onstituent)-structure which encodes
surface precedence and dominance in a tree structure and f(unctional)structure which
encodes grammatical relations and other syntactic features in an attribute-value matrix
that is basically a dependency representation. Simplified c- and f-structures for a simple
transitive sentence are shown in (1). Throughout this chapter f-structures are simplified
for expository purposes. In many instantiations of LFG, all semantically relevant material
such as information about telicity, definiteness or animacy is recorded in the f-structure
to then become part of the semantics. See (63) for how the different levels of representa-
tion feed into the semantic meaning of a sentence.

(1) a. Yassin will watch the movie.

b. c-structure c. f-structure
25. Lexical-Functional Grammar 841

Differences in word order are reflected directly at the c-structure: sentences with different
word orders may therefore correspond to identical, or extremely similar, f-structures,
especially in “free” word order languages. (Languages vary as to ordering possibilities
and as to how word order is connected up to information structural differences. See
section 5.2 for a short discussion of proposals within LFG which integrate a representa-
tion of information-structure.) For example, the German equivalent of (1) can have (at
least) the two forms in (2).

(2) a. Yassin wird den Film ansehen. [German]

Yassin will the.M.SG.ACC film.M.SG watch
‘Yassin will watch the movie.’
b. Den Film wird Yassin ansehen. [German]
the.M.SG.ACC film.M.SG will Yassin watch
‘Yassin will watch the movie.’

These will be analyzed with different c-structures, but identical f-structures except for
the choice of sentence topic, cf. Yassin in (2a); the movie in (2b). Indeed, for applications
involving machine translation, f-structure is considered to be the appropriate level to
work with, because this level of representation abstracts away from information that is
too language particular, but still contains enough syntactic information to be useful (e.g.,
Kaplan et al. 1989; Sadler 1993; Frank 1999; Riezler and Maxwell 2006; Graham et
al. 2009).
While we have used a very simplified c-structure representation in (1), LFG does
assume a version of X′-theory that goes back to Bresnan (1977) and that includes restric-
tions on the correspondence between different types of c-structure positions and the
grammatical functions they can correspond to in the f-structure. For current assumptions
about c-structural representations and constraints on c-structure, see Bresnan (2001);
Asudeh and Toivonen (2009) and references therein.
F-structures are referred to as a projection from the c-structure because they are
related to the c-structure via a formal system of annotations. A typical (simplified) exam-
ple for a fragment of English is shown in (3). The c-structure is built up on the basis of
phrase structure rewrite rules. The rule in (3a), for example, says that a sentence consists
of two major constituents, an NP and a VP. In effect, the S (sentence) is rewritten as an
NP and a VP. Rewrite rules are part of the common heritage of modern generative
syntactic theories (cf. Chomsky 1965) and continue to be a standard part of computa-
tional linguistics. As the mathematical basis of LFG is defined in terms of model theory,
rewrite rules as in (3) are formally realized as a system of constraints that are satisfied
by linguistic structures; see Kaplan (1995) and Pullum and Scholz (2001) for further dis-
cussion.

(3) a. S % NP VP
(↑SUBJ)=↓ ↑=↓
b. VP % V NP
↑=↓ (↑OBJ)=↓

The phrase structure rules in (3) are annotated with functional equations. These equations
provide the basis for computing the f-structure (information about number, person, tense,
842 IV. Syntactic Models

etc. is part of the lexical entries of the nouns and the verb in our example). The up arrow
refers to the f-structure of the mother node, i.e., VP in (3b), the down arrow references
the current node, i.e., V or NP in (3b). The annotations relate the c-structure in (1b) and
the f-structure in (1c) via a formal mathematical projection φ.
Without going into too much detail, the representations in (4)−(6) show how the
information provided by the annotations is projected onto the f-structure. In fact, the up
and down arrows stand for variables, which have been instantiated by integers in (4)−
(6). These arbitrary integers are used to label pieces of the f-structure.

(4) (2 PRED) = P ETER

NP 2: [ PRED P ETER ]

Peter

(5) (4 PRED) = DRINK < SUBJ, OBJ>

VP 4: [PRED DRINK < SUBJ, OBJ> ]

drinks

(6)

The f-structure corresponding to the NP node is labeled 2 in (4), the f-structure corre-
sponding to the S node is labeled 1 in (6). The functional annotations are treated as
equations in a mathematical system, which need to be solved. Solving the set of equa-
tions is equivalent to combining bits of f-structural information into one large f-structure
as in (7).
To return to our simple example, the picture in (7) shows the correspondence between
the c-structure and f-structure via the φ-function. That is, the f-structure annotations on
the c-structure have all been solved. The arrows from the individual c-structure nodes
to parts of the f-structure are an informal illustration of the correspondence function.
Note that several c-structure nodes may correspond to a single part of the f-structure.
LFG is a constraint-based theory. One effect of this is that when information from
different parts of the c-structure contribute to a single part of the f-structure, the different
pieces of information must be unified with one another and must therefore be compatible.
For example, if a verb specifies information about its subject’s person and number, these
values (e.g., third person singular) must be compatible with the values for person and
number provided by the subject noun. For a sentence like The boy hops. the subject f-
structure would be as in (8a) where the number and person information are specified by
both the noun boy and the verb hops. An ungrammatical sentence like The boys hops.
would provide conflicting number values, which is shown informally in (8b).
25. Lexical-Functional Grammar 843

(7)

(8) a. b.

The semantically unique PREDicate values provided by most words cannot unify, even if
they appear identical superficially. This is because they carry a semantic index as part
of their value and this unique index cannot unify with a different index. The index
reflects the semantic reference of predicates, which is considered to be unique.
F-structures must obey three constraints: uniqueness, completeness, and coherence.
Uniqueness states that for each f-structure, there is a single attribute of a given type,
e.g., only one subject, only one number; if more than one attribute of a given type is
required, it must be in a set (see section 3.2 on modifiers and section 3.3 on coordina-
tion). Completeness requires every argument of a predicate to be filled, e.g., if a verb
requires a subject and an object, both must be present in the f-structure (*Yassin de-
voured.). Coherence requires governable grammatical functions, i.e. the ones licensed as
arguments of predicates, to be licensed by a predicate if they appear, e.g., an object
cannot appear in an f-structure unless the predicate requires it (*Yassin slept Nadya.).
The f-structures we have seen so far have included two basic grammatical relations:
SUBJ(ect) and OBJ(ect). In LFG, grammatical relations are assumed as part of the syntac-
tic inventory of every language and are referred to as grammatical functions (GF) to
indicate their functional status, which is the relation of arguments and predicational
elements to one another. Because GFs are assumed to not be subject to crosslinguistic
variation, but are a basic part of the syntactic description language for every language,
they are represented at f-structure (and only at f-structure). However, tests for different
types of grammatical functionhood may play out differently across languages; see the
discussion in Dalrymple (2001) for the LFG perspective (also Kroeger 2005; Croft 2001;
Evans and Levinson 2009 for a different perspective). LFG assumes the GFs in (9).

(9) Grammatical Functions

SUBJ OBJ OBJθ OBL(ique)θ COMP(lement) XCOMP(lement) ADJUNCT

Dalrymple (2001: 11−27) provides a useful discussion of the GFs as well as several
syntactic tests by which they can be identified. COMP and XCOMP represent clausal argu-
844 IV. Syntactic Models

ments. Canonically, the COMP is used for finite clauses (e.g., the English that-clause)
while the XCOMP encodes nonfinite, open embedded clauses. An example is the to win
in John wants to win. Here, the embedded verb win does not have an overt subject,
rather, its subject is controlled by the matrix subject John (see Bresnan 1982a for the
classic discussion on control and complementation). Finite clauses like the English that-
clause are considered to be closed because all of the arguments are realized internal to
the clause (i.e. there is no control from the matrix clause).
The canonical example for OBJθ is the indirect dative object in languages like German
and the second object in the English double object construction (10a). The OBL is used
for prepositional phrases which are arguments of the verb. A classic example is the
English indirect to object (10b). Other instances of OBL occur with verbs of motion as
in (11), where the location is subcategorized for by the verb.

(10) a. Kim gave the dog a bone.

b. Kim gave a bone to the dog.
(11) Chris went to/in/behind the house.

The OBJ and OBL are subscripted with a θ to indicate that these GFs are sensitive to
thematic role information. That is, PP arguments as in (11) generally reflect a specific
spatial semantics. Similarly, indirect objects are generally tied to goals.
In fact, there are regularities between the semantics of arguments and their syntactic
expression in terms of GFs. Linking Theories in general attempt to capture these regular-
ities (section 4).
We close this section by noting that the effect of LFG’s projection architecture is that
the levels of representation constrain each other mutually. That is, an analysis can only
be successful if the f-structure information is complete and consistent, and if the phrase
structure rules license the structure. Because the relation between c-structure and f-struc-
ture is stated in terms of a mathematical projection (φ), its inverse can also be computed.
That is, not only can one see into the f-structure from the c-structure, but the f-structure
can refer back to the c-structure. No level is truly primary and no information is ever lost
via derivations. LFG thus falls under the label of constraint-based declarative theories
of syntax (Head-driven Phrase Structure Grammar also falls under this category; see the
HPSG chapter in this volume). In contrast to the fundamental derivational assumptions
of GB/MP (see the Minimalism chapter in this volume), LFG assumes no derivations
from one structure to another. Indeed, this is one of the characteristics which makes LFG
computationally tractable.
With the basic LFG architecture and constraints on the basic c- and f-structure repre-
sentations in mind, the next section examines how LFG analyzes several well known
syntactic phenomena.

3. Syntactic phenomena
For purposes of concrete illustration, we go through several core syntactic phenomena
in this section and show how they are analyzed in LFG. In many cases, the analyses
will share aspects with other syntactic theories, especially within generative syntax and
25. Lexical-Functional Grammar 845

particularly within constraint-based frameworks. However, the analyses also show

marked differences, which are driven by the unique architecture and assumptions of
LFG. The phenomena discussed include long-distance dependencies, modifiers, coordi-
nation, agreement, and case. Argument structure and other projections are described in
subsequent sections.

3.1. Long-distance dependencies

3.1.1. Functional control: Equi and raising

Functional control is described in detail in Dalrymple (2001), Bresnan (2001) and Falk
(2001); the seminal article is Bresnan (1982a). In functional control, a single part of an
f-structure plays two or more roles in the f-structure. Classic examples are so-called equi
verbs as in (12), where the subject of want is identical to the subject of eat, but is only
expressed once, namely in the matrix clause. (Equi verbs can also be analyzed as ana-
phoric control, see section 3.1.3)

(12) a. Kim wants to eat beans.

b.

This functional sharing is expressed at the f-structure by identifying two parts of an f-

structure with one another. In (12) and also generally in the theoretical literature, this is
graphically illustrated by having a line that connects the two parts of the f-structure with
one another.
Bresnan (1982a) shows that this type of functional control is a lexical property. That
is, the type of matrix verb determines the type of functional control that happens. The
verb want, for example, shows both subject and object control, where the matrix object
controls the embedded subject, as in (13a). The verb persuade shows object control, as
in (13b), but the verb promise is a subject control verb, as in (13c).

(13) a. Kim wanted Sandy to eat beans.

b. Kim persuaded Sandy to eat beans.
c. Kim promised Sandy to eat beans.

The X in the XCOMP signifies that it is an open function which expects its subject to be
controlled. Given that this type of functional control is a lexical phenomenon, the lexical
entries of the verbs have to state what kind of functional control the verb allows. A
846 IV. Syntactic Models

typical equation, found in subject control verbs like promise, is shown in (14). Object
control verbs would contain the equation in (15) and verbs like want contain a disjunc-
tion which allows for both options. The equations basically state the subject (or object)
of the matrix verb is the same as the subject of the embedded clause.

(14) (↑SUBJ) = (↑XCOMP SUBJ)

(15) (↑OBJ) = (↑XCOMP SUBJ)

Early generative syntax argued for a distinction between equi verbs of the type above
and raising verbs as in (16). The difference is that in equi verbs, the relevant arguments
are licensed thematically by both the matrix and the embedded verb. For verbs like seem,
on the other hand, it appears that no semantic selectional restrictions are imposed on the
subject and that therefore an expletive subject as in (16) can be used. Verbs like seem
were therefore identified as raising verbs.

(16) a. There seems to be a unicorn in the garden.

b. It seems to be raining.

In LFG, this difference in semantic content (or thematic content, as it has also been
called) is expressed by writing the non-thematic (semantically empty) argument outside
of the angle brackets of the subcategorization frame. This is illustrated in (17b).

(17) a. David seemed to yawn.

b.

The formal mechanism of functional control, by which two parts of an f-structure are
identified with one another, however, remain the same. So the lexical entry of seem
would contain the functional equation in (14). The main distinction posited in LFG
between equi and raising verbs is a semantic one that is expressed within the subcategori-
zation frame of the verb in question.

3.1.2. Functional uncertainty

So far, we have seen instances of functional control that are local. However, functional
control can occur over several layers of embedding, as exemplified by the topicalization
of beans in (18).

(18) Beans, John asked Sandy to persuade Kim to eat.

25. Lexical-Functional Grammar 847

Other dependencies that can be local as well as long-distance are wh-questions and
relative clauses in languages like English. These dependencies, in which the interroga-
tive, relative, or topicalized phrases do not appear in their canonical c-structure position,
are also analyzed via functional control in LFG. Often, the displaced constituent can
play different roles in different sentences, and these roles may be at different levels of
embedding (19).

(19) a. Who hopped? (↑FOCUS-INT) = (↑SUBJ)

b. Who did John say hopped? (↑FOCUS-INT) = (↑COMP SUBJ)
c. What did he eat? (↑FOCUS-INT) = (↑OBJ)
d. What did John say that he ate? (↑FOCUS-INT) = (↑COMP OBJ)
e. What did he want to eat? (↑FOCUS-INT) = (↑XCOMP OBJ)

The control equations needed to identify the role of the displaced constituent with its
GF can be listed in any individual case. However, these possibilities in theory can be
infinite. (In practice, limits are imposed due to performance restrictions, for more discus-
sion on this topic see Pullum and Scholz 2010.) LFG captures this empirical fact via
functional uncertainty equations (Kaplan and Zaenen 1989). For example, Kaplan and
Zaenen (1989) proposed that the rule for English topicalization is as in (20), which says
that there is an XP or S′ which precedes the main sentence, that this XP or S′ is the
overall topic of the clause and that it may correspond to some GF that may be arbitrarily
deeply embedded in a series of COMPs and/or XCOMPs. The technical definition of func-
tional uncertainty is provided by Kaplan and Zaenen (1989: 26): Functional uncertainty:
(fα) = v holds iff f is an f-structure, α is a set of strings, and for some s in the set of
strings α, (fs) = v. When the string s is longer than one: (fas) ≡ ((fa)s) for a symbol a
and a (possibly empty) string of symbols s; (fε) ≡ f, where ε is the empty string.

(20) S′ % XP or S′ S
(↑ TOPIC) = ↓ ↑=↓
(↑ TOPIC) = (↑ {COMP, XCOMP}* (GF-COMP))

In the analysis of a given sentence, the choice of path will depend on how other argu-
ments satisfy the subcategorization frames of the verbs. For example, in (19d) what
cannot be the subject of say or eat because they already have subjects and coherence
would be violated; however, eat has no object and so what can be equated with its
object position.
Functional uncertainty paths vary cross-linguistically, reflecting island conditions and
other language-particular constraints. In addition, different phenomena within a language
(e.g., interrogatives, topicalization, relative clauses) may have different functional uncer-
tainty paths. See Dalrymple (2001) for detailed discussion of functional uncertainty paths
in a number of languages.
Some versions of LFG theory posit traces for certain constructions in certain lan-
guages, while others never posit traces in the c-structure. Bresnan (2001) and Falk (2001,
2007) both argue for traces in LFG for capturing phenomena such as English weak-
crossover constructions and wanna contraction. (But see Dalrymple et al. 2001 for an
alternative account of weak-crossover in terms of linear prominence constraints that do
848 IV. Syntactic Models

not rely on traces.) Such accounts generally posit traces only in highly configurational
languages like English which use structural positions to encode grammatical functions,
instead of case marking or head marking.

3.1.3. Anaphoric control

Functional control involves a single PRED filling two roles in the larger f-structure. How-
ever, not all cases of control are of this type. Consider (21), which at first sight would
appear to involve an embedded XCOMP, just as in the examples discussed under func-
tional control. However, here the subject (David) does not necessarily control the subject
of the embedded clause. That is, there is arbitrary anaphoric control in that the person
leaving is some person that the syntax cannot determine and that needs to be computed
on the basis of discourse or world knowledge. This computation is thought of as being
on a par with anaphora resolution. (LFG posits cross-linguistic binding domains within
sentences for anaphora. Many accounts define appropriate f-structure domains for the
binding relation in conjunction with restrictions on precedence in the c-structure. These
discussions only apply tangentially to anaphoric control.)

(21) David gestured to leave.

An analysis of (21) is provided in (22). The difference to functional control is that the
embedded subject is not empty, but filled with a null pronominal subject (PRO) and,
since the subject is filled, the embedded clause is analyzed as a COMP (a closed function)
rather than an XCOMP (an open function).

(22) David gestured to leave.

Anaphoric control can be obligatory in certain constructions. Equi constructions have

been analyzed as obligatory anaphoric control. In the case of equi, the overt controller
controls a null pronominal subject of the embedded clause. This controlled subject is
absent at c-structure but present in the f-structure. An example is shown in (23) from
Dalrymple (2001: 324).
25. Lexical-Functional Grammar 849

(23) David tried to leave.

Obligatory anaphoric control has also been posited for English tough-constructions (Dal-
rymple and King 2000), as in (24).

(24) Mothsi are tough to kill proi.

3.2. Modifiers

Modifiers or adjuncts are not subcategorized by predicates and in general multiple modi-
fiers are allowed in a given f-structure. In order to satisfy the uniqueness requirement of
f-structures, modifiers belong to a set, even when there is only one of them. Adjectival
modification is illustrated in (25), clausal modification in (26).

(25) the happy little girl

(26) Monday, Nadya walked quickly.

C-structure analyses of modifiers can be as adjunction but may also appear in much
flatter structures. Which type of c-structure is appropriate depends on the language and
is generally determined by constituency tests such as coordination possibilities. The φ-
850 IV. Syntactic Models

mapping between c-structure and f-structure for modifier sets involves specifying them
as elements of the set, as in the rule for adjectival modifiers in (27), where the Kleene
* represents zero or more instances of the AP.

(27) N′ % AP* N
↓ 2 (↑ ADJUNCT) ↑=↓

There are proposals within LFG for constraining the mapping from c- to f-structure in
such a way to allow modifier annotations only in specific configurations. See, for exam-
ple, Bresnan (2001).

3.3. Coordination

Coordination also involves the formation of sets in f-structure. A set will have an element
for each conjunct in the coordination. The canonical example of this is resolved person
and number features in noun phrase coordination, where the coordinated set may have
a different person and number than the individual elements. The features which can be
features of the set itself are called non-distributive features. Consider the sample coordi-
nated NP in (28).

(28) the dog and me

The difference between distributive and non-distributive features is very important for
phenomena like agreement (section 3.4) and case (section 3.5). If a requirement is placed
on a coordinated f-structure for a distributed feature, then that requirement will distribute
to each element of the set and all the elements must unify with that requirement. For
example, if a verb requires a dative object and the object is coordinated, then each
element in the coordinated object must be dative because case is distributive. In contrast,
number and person is generally not distributive and so a verb that requires a plural
subject will be compatible with an f-structure as in (28) in which the set is plural even
though each conjunct is singular.

3.4. Agreement

Agreement is generally analyzed as an instance of feature unification in LFG. Both

elements in the agreement relation will specify a value for a feature, e.g., singular num-
25. Lexical-Functional Grammar 851

ber, and these values must be compatible, see (8). This can be done for verb-argument
agreement as well as for adjective-noun agreement. Note that the feature can either be
present in only one place, e.g., on the subject noun phrase, or in two places with a con-
straint that the values must be identical, e.g., an adjective may state that its number value
is the same as that of its head noun.
A form can show agreement in one of two ways. Building on work in HPSG (Wech-
sler and Zlatić 2003), LFG posits two types of syntactic agreement features: CONCORD
features are closely related to morphological form, while INDEX features are closely re-
lated to semantics. Different predicates will target specific feature sets. In general, CON-
CORD features are used for noun-phrase internal agreement, while INDEX features are
used outside of the noun phrase. Consider (29).

(29) a. This boy and girl hop.

b.

The verb hop agrees with the plural INDEX value of the coordinated NP, which is a
resolved value from the agreement features on the individual conjuncts, while the deter-
miner this agrees with the singular CONCORD value of each of the coordinated arguments.
Recent LFG research has been interested in asymmetric agreement, especially vari-
ants of closest-conjunct agreement (Sadler 2003; Kuhn and Sadler 2007).

3.5. Case

LFG has its own theory of case, developed in a series of papers by Butt and King and
by Rachel Nordlinger. LFG’s approach to case assignment is also often mediated by the
mapping of a(rgument)-structure to GFs, discussed in section 4, whereby certain cases are
associated with certain thematic roles or features at a-structure, e.g., Alsina’s (1996) work
on Romance. We here briefly present the inventory of case types posited by Butt and
King (2003, 2005) and discuss the related proposal by Nordlinger (1998). See Butt (2006,
2008) for a more complete discussion of case in LFG, including in OT-LFG (section 6).

Positional case: Positional case is associated only with syntactic information. That is,
there is assumed to be a syntactic configuration which requires a particular case marking.
An example of positional case is the adnominal genitive in English (see King 1995 for
examples from Russian). As shown in (30), the prenominal NP position is identified as
genitive as part of the positional information in the syntax (the ↑=↓ notation indicates
that the noun is the head of the phrase).
852 IV. Syntactic Models

(30) English Adnominal Genitives (simplified structure)

NP

NP N
(↓CASE)=GEN ↑=↓
Boris’s hat

Structural and default case: Structural case is often an instance of default case and hence
functions as the Elsewhere Case (Kiparsky 1973). For languages which require that all
NPs have case, this can be stated as in (31a), analogous to the Case Filter in GB (Rou-
veret and Vergnaud 1980). If a given NP is not already associated with case due to some
other part of the grammar, then default case assignment principles as in (31b−c) apply.

(31) a. Wellformedness principle: NP: (↑CASE)

b. Default: (↑SUBJ CASE) = NOM
c. Default: (↑OBJ CASE) = ACC

Default case only applies to the core grammatical relations subject and object. The other
grammatical relations tend to involve specialized semantics and therefore do not involve
defaults. The content of the default assignment principles may vary from language to
language, but the existence of a default case for subjects and objects is expected to hold
crosslinguistically.

Quirky case: The term quirky case is used only for those situations in which there is no
regularity to be captured: the case assignment is truly exceptional to the system and no
syntactic or semantic regularities can be detected. Under the assumption that case mor-
phology plays a large role in the fundamental organizing principles of language, quirky
case is expected to be fairly rare. Instead, case morphology is part of a coherent system,
with only a few exceptions along the way. These exceptions are generally due to histori-
cal reasons and have not been eradicated or reanalysed as part of a regularization of the
case system (Butt 2006).

Semantic case: The defining characteristics of semantic case in the sense of Butt and
King (2003) are semantic predictability and subjection to syntactic restrictions, such as
being confined to certain GFs. Indeed, most cases cannot appear on just any GF, but are
restricted to one or two. Under Butt and King’s (2003) analysis, most instances of case
involve semantic case. This is because the bulk of the crosslinguistic case marking phe-
nomena involve an interaction between syntactic and semantic constraints (including the
quirky case found in Icelandic, which is actually mostly subject to semantic regularities).
Consider the accusative/dative ko in Urdu. On direct objects, it signals specificity. That
is, a combination of syntactic (direct objects only) and semantic factors (specificity) are
involved. The ko can also appear on subjects and on indirect objects, as in (32). In either
case, the dative is associated with a more or less abstract goal.
Within Butt and King’s system, the ko is therefore analysed as a semantic case. Butt
and King furthermore pursue a fundamentally lexical semantic approach to case. That
is, lexical entries are posited for individual case markers and these lexical entries contain
the bulk of the information associated with the presence of the case markers. The lexical
entry for ko, for example, is shown in (32).
25. Lexical-Functional Grammar 853

(32) Accusative ko Dative ko

(↑ CASE) = ACC (↑ CASE) = DAT
(OBJ ↑) (GOAL ↑arg−str)
(↑sem−str SPECIFICITY) = + (SUBJ ↑) n (OBJgo ↑)

The entry for ko specifies that it can be used either as an accusative or a dative. As an
accusative, it can only appear on a direct object and is associated with specificity in the
semantic projection. Note the use of the ↑ in the lexical entry of the case marker: the
second line involves inside-out functional designation (Dalrymple 1993, 2001); the ↑
following the specification of a GF formulates a requirement that the constituent should
be analysed as an object. As a dative, it can only appear on either experiencer subjects
or indirect objects (OBJgo) and requires a goal argument at a-structure. (32) illustrates
that the information associated with case morphology interacts with information at sev-
eral levels of representation, e.g., f-structure, semantic projection (section 5.4), and a-
structure (section 4). (Inside-out functional designation can also be applied over func-
tional uncertainty paths, just like the outside-in functional control discussed in section
3.1.2)

Constructive case: Further evidence for the above type of lexical approach to case comes
from Australian languages. Nordlinger (1998, 2000) analyzes two phenomena found in
Australian languages: discontinuous constituents as in (33) and case stacking as in (34).
In the Wambaya example in (33), the NP big dog is a discontinuous constituent. Gener-
ally, Australian languages are known for their free word order and in Wambaya the only
requirement is that there be a finite verb in second position. (I = masculine gender class;
A = transitive subject; O = object)

(33) galalarrinyi-ni gini-ng-a dawu bugayini-ni. [Wambaya]

dog.I-ERG 3SG.M.A-1.O-NFUT bite big.I-ERG
‘The big dog bit me.’
(Nordlinger 1998: 96)

Now also consider the phenomenon of case stacking found in Martuthunira. In (34) the
word thara ‘pouch’ is marked with three cases: one to show that it is signalling a loca-
tion, one to show that it is part of a possessive or accompanying relation to another word
(the proprietive case), and one to show that it is part of (modifying) an accusative case
marked noun. The word mirtily ‘joey’ (a baby euro − a type of kangaroo) has two cases.
The proprietive shows that it stands in an accompanying relationship with another (it is
with the euro), and the accusative to show that it is part of (modifying) an accusative
case marked noun. Finally, ‘euro’ is accusative as the direct object of the clause, while
the first person pronoun (‘I’) is nominative (unmarked). (PROP = Proprietive)

(34) Ngayu nhawu-lha ngurnu tharnta-a mirtily-marta-a [Martuthunira]

I saw-PST that.ACC euro-ACC joey-PROP-ACC
thara-ngka-marta-a.
pouch-LOC-PROP-ACC
‘I saw the euro with a joey in (its) pouch.’
(Dench 1995: 60)
854 IV. Syntactic Models

These facts prompted Nordlinger (1998) to formulate a new perspective on case. She sees
morphology as constructing the syntax of the clause. For example, under her analysis,
the Wambaya ergative ni carries the information that there be a subject and that it be
ergative. These pieces of information are encoded as part of the lexical entry of the
ergative, as shown in (35).

(35) ni: (↑ CASE) = ERG

(SUBJ ↑)

With this lexical approach to case, the effect of the analysis is that the combination of
information from the lexical entries of big, dog and the ergative case in (35) results in
the two partial f-structures shown in (36) and (37). Both the ergative dog and the big
specify that they are parts of the subject because of the information associated with the
ergative case marker in (35). In addition, the dog serves as the head of the phrase and
the big as an adjunct which modifies it (the details of how the adjunct analysis is accom-
plished are left out here).

(36)

(37)

These two sets of information are unified into the structure shown in (38) as a routine
part of the clausal analysis within LFG. The problem of discontinuous constituents is
thus solved by using the case morphology as a primary source of information about
clausal structure.

(38)

The same approach also serves well for instances of case stacking. Since every case
marker contributes not only a value for the case feature at f-structure, but also imposes
a requirement as to what GF it must appear with, the effects of case stacking can be
easily explained via the lexicalist, constructive approach to case.

4. Argument structure
In addition to the basic c- and f-structural representations, LFG’s architecture potentially
allows for several other projections. One standard additional projection is the a(rgu-
ment)-structure. The a-structure encodes predicate-argument relationships in terms of
thematic roles. These thematic roles are arranged in a thematic hierarchy, shown in (39)
(based on Bresnan and Kanerva 1989).
25. Lexical-Functional Grammar 855

(39) Thematic Role Hierarchy

agent > beneficiary > recipient/experiencer > instrument > theme/patient >
location

The GFs as introduced in (9) are also arranged in a hierarchy. Linking is assumed to
prefer a mapping between highest thematic role and highest GF (SUBJ). The default
mapping is therefore straightforward: agents should map to subjects, themes should map
to objects, etc. However, because languages exhibit many phenomena where this default
mapping does not apply, an explicit linking or Lexical Mapping Theory was formulated
in LFG to account for systematic deviations and argument alternations.
The a-structure can be formally represented as an AVM, like the f-structure (e.g.,
Butt 1998), but in keeping with the bulk of the literature on argument structure, represen-
tations like the following are used here: pound < agent theme >.

4.1. Standard LFG Mapping Theory

LFG’s Lexical Mapping Theory grew out of early work like Zaenen et al.’s (1985)
analysis of Icelandic and German and Levin’s (1987) work on English. The discussion
in this section is based on Bresnan and Zaenen (1990), Alsina and Mchombo (1993),
Bresnan and Kanerva (1989), and Bresnan and Moshi (1990). Further reading and discus-
sion can be found in Bresnan (1990) and Bresnan (1994). As in Levin’s first formulation
for linking, thematic roles and GFs are cross-classified by features in standard LFG,
which posits just two relevant features. The feature [±restricted] is a semantically
grounded feature, which indicates whether a given thematic role or GF is sensitive to
semantic restrictions. The feature [±o(bjective)] marks whether thematic roles are likely
to be linked to objectlike GFs. (Alsina [1996: 19] instead proposes two different features:
[±subj(ect)] and [±obl(ique)].)
The [±r,±o] features classify GFs as shown in (40). The clausal COMP and XCOMP are
not considered in this classification (see Berman 2003 and Dalrymple and Lødrup 2000
on the status of clausal arguments).

(40) [−o] [+o]

[−r] SUBJ OBJ
[+r] OBLθ OBJθ

The thematic roles are classified by these same features, as shown in (41) (Bresnan and
Zaenen 1990).

(41) Classification of Thematic Roles

Patientlike roles: [−r]
Secondary patientlike roles: [+o]
All others: [−o]

The possible correspondences between thematic roles and GFs are regulated by the Map-
ping Principles in (42) and wellformedness conditions, some of which are shown in (43)
856 IV. Syntactic Models

(Bresnan and Zaenen 1990). The θ stands for thematic role and θ̂ refers to the highest
argument on the thematic role hierarchy.

(42) Mapping Principles

a. Subject roles:
(i) θ̂ is mapped onto SUBJ; otherwise:
[−o]
(ii) θ is mapped onto SUBJ
[−r]
b. Other roles are mapped onto the lowest compatible function on the marked-
ness hierarchy, where the subject is the least marked.
SUBJ < OBJ, OBLθ < OBJθ

(43) Wellformedness Conditions

a. Subject Condition: Every (verbal) lexical form must have a subject.
b. Function-argument biuniqueness: Each a-structure role must be associated
with a unique grammatical function, and conversely.

The Function-argument biuniqueness condition in (43b) is reminiscent of GB’s θ-Crite-

rion mapping arguments to theta-roles. It has been challenged within LFG (Mohanan
1994; Alsina 1996) for not allowing the necessary flexibility to account for Argument
Fusion in complex predicates (section 4.2). In addition, the Subject Condition may not
hold universally (Bresnan 2001; Dalrymple 2001; Falk 2001).
The feature classifications together with the mapping and wellformedness principles
constitute the essence of LFG’s linking theory. Consider the following example of a
transitive, a passive, an unaccusative and an unergative (taken from Bresnan and Zaenen
1990: 51−52). In (44), the transitive verb pound has two arguments, an agent and a
theme. These are featurally classified according to (41) and then mapped to SUBJ and
OBJ straightforwardly according to the mapping principles.

(44)

Passivization suppresses the highest thematic role, as shown in (45). The only argument
available for linking into the syntax is the theme. This could potentially be linked either
to a subject or an object, but because of the principle in (42), it is linked to the subject.

(45)

(46)
25. Lexical-Functional Grammar 857

The single argument of unaccusatives is a theme, as in (47). This is classified by the

feature [−r] and is linked to a subject rather than an object because of the mapping
principles in (42). The unaccusative situation is thus parallel to the passive in (46). In
the unergative, the only argument of an unergative verb like bark is an agent. This is
classified as a [−o] thematic role and links to a subject.

(47)

The basics of LFG’s mapping theory are thus very simple and yet make for a very
powerful system that has been used to analyze complex case marking and argument
structure phenomena in Germanic, Bantu, Romance and South Asian languages. Some
of the more complex phenomena are briefly described in the remainder of this section.

4.2. Argument alternations and complex predicates

One reason LFG allows for a flexible mapping between thematic roles and GFs is be-
cause of the crosslinguistic recurrence of argument alternations. One famous alternation
investigated within LFG is locative inversion in Chichewa, shown in (48) (Bresnan and
Kanerva 1989; Bresnan 1994). (REC.PST = recent past)

(48) a. a-lendȏ-wo a-na-bwér-á ku-mu-dzi. [Chichewa]

2-visitor-2.those 2.SBJ-REC.PST-come-IND 17−3-village (visitors = subject)
‘Those visitors came to the village.’
(Bresnan and Kanerva 1989: 2)
b. ku-mu-dzi ku-na-bwér-á a-lendo-wȏ. [Chichewa]
17−3-village 17.SBJ-REC.PST-come-IND 2-visitor-2 those (village=subject)
‘To the village came those visitors.’
(Bresnan and Kanerva 1989: 2)

Chichewa is a Bantu language, which does not mark arguments via case, but instead
uses a complex noun class system. The noun classes have a rough semantic/cognitive
basis, but this is only rough, so the different classes are usually indicated by numbers,
e.g., 2 for visitors and 17 for village in (48).
Bresnan and Kanerva (1989) amass evidence which shows that visitors is the subject
in (48a), but not in (48b). In (48b) the subject is village. One piece of evidence for
subject status is verb agreement: in (48a) the verb agrees with visitors via the class 2
marker, in (48b), the verb agrees with village (class 17). Locative inversion is triggered
by focus, so that the location (village) is focused in (48b).
The possibility for locative inversion follows from the standard linking principles.
The thematic roles are classified as shown in (49) via the general classification principles
in (41). Both the theme and the locative could link to either SUBJ or OBL. In (49a) default
linking occurs according to the mapping principles in (42): the theme is linked to the
858 IV. Syntactic Models

subject because it is higher on the thematic role hierarchy than the locative. In (48b),
the locative argument is linked to the subject due to the special focus context. In this
context, locatives are associated with the [−r] feature, which means they can only be
linked to a subject, preempting the theme. Since there cannot be two subjects in a clause,
the theme is linked to the OBJ.

(49) a.

b.

Another instance of an argument alternation that follows from LFG’s linking theory is a
crosslinguistic pattern of causative formation. In causatives, an event is caused by the
action of a causer. In the Chichewa examples in (50) (Alsina and Joshi 1991; Alsina
1997), there is a cooking event which is caused or instigated by a causer external to the
cooking event. There are three syntactic arguments in (50): a causer/agent (the porcu-
pine); another agent (the owl), who is also the causee; a theme/patient of the caused
event (the pumpkins).

(50) a. Nǔngu i-na-phík-ítsa kadzīdzi maûngu. [Chichewa]

porcupine SBJ-PST-cook-CAUS owl pumpkins
‘The porcupine made the owl cook the pumpkins.’
(Alsina and Joshi 1991: 8)
b. Nǔngu i-na-phík-ítsa maûngu kwá kádzīdzi. [Chichewa]
porcupine SBJ-PST-cook-CAUS pumpkins by owl
‘The porcupine had the pumpkins cooked by the owl.’
(Alsina and Joshi 1991: 8)

Causatives also show an argument alternation. As seen in (50) the causee alternates
between a direct argument or an oblique PP in Chichewa. The argument alternation
coincides with a semantic difference. When the causee is realized as a direct argument,
it is interpreted as affected by the action. That is, in (50a) the focus is on the owl having
to cook the pumpkins and how it might feel about that. In (50b), the focus is on the
pumpkins and that they be cooked. It is not important who cooks them, or how they
might feel about it, just that they become cooked.
This semantic difference holds for Urdu (Butt 1998) and Romance (Alsina 1996) as
well. Alsina and Joshi (1991) model this semantic difference via a difference in Argu-
ment Fusion. They examine causatives in Chichewa and Marathi, and propose an analy-
sis by which two argument structures are combined and one of the arguments of each
25. Lexical-Functional Grammar 859

argument structure is identified with an argument in the other. Causative morphemes and
verbs are taken to have three arguments: a causer agent, a patient and a caused event.
This is shown in (51). When a causative morpheme or verb is combined with another
verb, it embeds this verb’s argument structure in its own, as shown in (52).

(51) CAUSE < agent patient event >

(52) CAUSE < agent patient ‘cook’ < agent patient >>

There are four semantic arguments in (52). However, only three arguments are expressed
in the syntax. Two of these arguments fuse at argument structure before being mapped
into the syntax. Alsina and Joshi (1991) posit parameters which allow fusion of the
matrix patient argument with either the embedded agent or the embedded patient.
When the causee (the matrix patient) is fused with the embedded agent, the embedded
agent is no longer available for linking, as shown in (53a). In this case the causee is
considered to be the affected argument of the causation and is mapped to the direct
object. The embedded patient is mapped to a secondary object (49a). When the matrix
patient is fused with the embedded patient, then this argument is no longer available for
linking and the agent of the embedded predicate is linked to an oblique (49b).

(53) Object Causee

a.

b.

LFG’s linking theory is thus primarily concerned with the relationship between argument
structure and grammatical relations (f-structure). A final point with respect to causatives
and linking theories in general relates to the domain of linking. As the name already
indicates, Lexical Mapping Theory assumed that the linking from thematic roles to GFs
was based on a single lexical entry, i.e., was taken care of entirely within the lexical
component. However, data from causatives show that argument alternations and complex
argument structures can arise either in the lexicon, or in the syntax − see Alsina (1997)
for an explicit comparison between the morphological causatives in Bantu and periphras-
tic causatives in Romance. Alsina (1996) and Butt (1995) extended LFG’s linking theory
to account for argument linking where the arguments are contributed by two distinct
words in the syntax. The analyses assume a complex interaction between c-structure, a-
structure and f-structure, where one structure cannot be built up without information
present at another structure. The a-structure is thus not primary, nor is the c-structure.
860 IV. Syntactic Models

That is, each of the levels of representation constrain one another within LFG’s projec-
tion architecture.
Further work on complex predicates which addresses architectural and linking issues
is represented by Manning and Andrews (1999) and Wilson (1999).

4.3. Incorporation of Proto-Roles

So far, we have presented the standard version of LFG’s mapping theory, albeit extended
into the syntax in order to capture complex predication. However, many different ver-
sions of this mapping or linking theory have been proposed. One interesting development
has been the incorporation of Proto-Role information (Dowty 1991), as proposed by
Zaenen (1993), for example.
Zaenen (1993) conducts a detailed study of the interaction between syntax and verbal
lexical semantics in Dutch. Dutch auxiliary selection is one syntactic reflex of unaccusa-
tivity. Unaccusative verbs in Dutch select for zijn ‘be’ while unergatives select for heb-
ben ‘have’. Zaenen shows that semantic factors are at the root of the auxiliary selection
patterns. The have auxiliary is associated with control over an action, whereas the be
auxiliary is selected when an argument is affected or changed (change of state).
These properties are included in Dowty’s (1991) system of Proto-Role entailments:
control is a Proto-Agent property and change of state is a Proto-Patient property. Zaenen
therefore proposes to incorporates Dowty’s Proto-Role entailments into linking theory
as shown in (54).

(54) Association of Features with Participants

1. If a participant has more patient properties than agent properties,
it is marked −r.
2. If a participant has more agent properties than patient properties
it is marked −o.
3. If a participant has an equal number of properties, it is marked −r.
4. If a participant has neither agent nor patient properties, it is marked −o.
(Zaenen 1993: 150,152)

The Proto-Role information allows Zaenen to dispense with thematic roles and the the-
matic role hierarchy. Linking is accomplished via the default association of [±o,r] marked
arguments with the GF hierarchy, as shown in (55).

(55) Association of Features with GFs

Order the participants as follows according to their intrinsic markings:
−o < −r < +o < +r
order the GR [grammatical functions] as follows:
SUBJ < OBJ < OBJθ (< OBL)
Starting from the left, associate the leftmost participant with the leftmost GR it is
compatible with.
(Zaenen 1993: 151)
25. Lexical-Functional Grammar 861

Unaccusatives and unergatives can now be analysed as follows: The single argument of
unaccusatives such as fall has more patient properties than agent properties, is thus
classified as a [−r] role, and is therefore linked to the SUBJ GF. In contrast, the single
argument of an unergative such as dance has more agent properties than patient proper-
ties, and is therefore classified as a [−o] role and is also linked to SUBJ. The difference
in auxiliary selection is sensitive to the [−r] feature, as shown in (56).

(56) Auxiliary Selection

When an −r marked participant is realized as a subject, the auxiliary is zijn ‘be’.
(Zaenen 1993: 149)

Zaenen’s linking architecture remains true to the basic spirit of linking theory, but allows
a better integration of relevant semantic factors. Other approaches have also integrated
Proto-Role properties into an analysis of the relationship between a-structure and GFs.
In his treatment of Romance causatives, for example, Alsina revises LFG’s standard
linking theory considerably and also includes Proto-Role information. Ackerman’s
(1992) ideas are similar in spirit to Zaenen’s analysis, but he offers a different way of
integrating Proto-Roles for a treatment of the locative alternation. Finally, Ackerman and
Moore (2001) incorporate Proto-Role properties into the selection of arguments without
making explicit reference to LFG’s standard linking theory, though they assume their
ideas are compatible with it.

4.4. Lexical rules

Lexical rules manipulate the argument structure of lexical items in systematic ways.
Lexical rules were introduced to capture regular alternations in the lexicon before a-
structure was introduced as part of LFG theory (Bresnan 1982b). Some lexical rules are
now replaced by a-structure alternations. However, alternations that affect GFs, e.g.,
SUBJ and OBJ, instead of thematic roles or argument slots must be stated as lexical rules.
The standard examples of a lexical rule is that of the passive, which rewrites the
subject as NULL or as an oblique agent and rewrites the object as the subject. Thus, the
lexical rules in (57) would rewrite the lexical entry in (58a) and (58b).

(57) Passive lexical rule: SUBJ % NULL

OBJ % SUBJ
(58) a. PRED= ‘persuade <SUBJ, OBJ, XCOMP>’
They persuaded the boys to leave.
b. PRED = ‘persuade <NULL, SUBJ, XCOMP>’
The boys were persuaded to leave.

Rewrites like the passive rule in (57), where arguments are renamed or deleted, are
easier to formulate than rules which introduce an argument, e.g., benefactive construc-
tions. When arguments are introduced, it is unclear where in the PRED structure the new
argument should appear. For this reason, such argument-adding constructions are usually
dealt with within a-structure.
862 IV. Syntactic Models

5. Interfaces/other projections

The f-structure is the primary syntactic projection from the c-structure. However, it is
possible to have other projections off of the c-structure and off of the f-structure or other
projections. Here we briefly discuss proposals for a morphosyntactic projection (section
5.1), information structure (section 5.2), prosodic structure (section 5.3), and semantic
structure (section 5.4).

5.1. Morphology-syntax interface

The morphology-syntax interface can be partially factored out into an m-structure that
is distinct from the f-structure. As originally formulated, the m-structure was meant to
hold features that are only relevant for morpho-syntactic well-formedness while the f-
structure contains features needed for syntax analysis and for semantics. The m-structure
of a clause may thus vary significantly from its f-structure. In the original m-structure
proposal, for example, m-structure is used to account for auxiliary stacking restrictions
in English, which allows for a flatter, simpler f-structure (Butt et al. 1996).

(59) a. Nadya will be hopping.

b. m-structure c. f-structure

M-structure has also been used extensively to analyze French auxiliaries and clitic distri-
bution (Frank 1996). Most LFG analyses do not include m-structure, however, some
approaches have taken up the m-structure idea and have invested it with more theoretical
significance than it had originally. See Sadler and Spencer (2004) for such proposals and
a more detailed discussion of the issues.

5.2. Information-structure

I(nformation)-structure encodes the clause-internal discourse information such as topic

and focus. I-structure often incorporates information from the c-structure as well as the
f-structure and possibly the prosodic structure. A simple example of this is shown in (60)
where the non-canonical word order reflects the discourse functions of the arguments.
Annotations on the c-structure rules create both the f-structure in (60b) and the i-structure
in (60c).
25. Lexical-Functional Grammar 863

(60) a. Èto plat’e šila Inna. [Russian]

this dress sewed Inna
Inna-FOC sewed this dress-TOP
‘It was Inna who sewed this dress.’
b. f-structure c. i-structure

This is accomplished technically by having annotations such as those in (61).

(61) IP % NP IP
(↑ { SUBJ|OBJ }) = ↓ ↑=↓
↓i2 (↑i TOPIC) ↑i=↓i

Some types of topic and focus are GFs and hence appear in the f-structure (Bresnan and
Mchombo 1987). However, for many languages, a separate projection is needed in order
to account for mismatches between the constituency of the f-structure and of the i-
structure. For example, focused verbs cannot be encoded in the f-structure because they
are the heads of their clauses and marking them as focus would result, incorrectly, in
the entire clause being focused (King 1997). Note that the topic is represented as a set
in (61c). This is because there could be multiple topics in clause.
There are a number of differing proposals as to how to exactly integrate the i-structure
projection and what exactly should be represented at this level of analysis, see for exam-
ple Choi (1999), King (1995), O’Connor (2004) and Mycock (2006).

5.3. Prosodic-structure
Work on the analysis of clitics and on discourse functions has led to the incorporation
of prosodic information into the LFG architecture, generally by means of a p(rosodic)-
structure projection (Butt and King 1998; O’Connor 2004; Mycock 2006). The projec-
tion is generally represented as an AVM similar to the f-structure, but there is still
discussion of whether an AVM is the best form for this. For example, Bögel et al. (2009,
2010) argue for using finite-state power to create prosodically bracketed strings which
can then guide c-structure formation and hence influence the f-stucture. In related work,
Asudeh (2009) proposes that linear string adjacency be part of the syntax-phonology
interface in LFG, thereby accounting for complementizer-adjacent extraction effects
without reference to traces. Dalrymple and Mycock (2011) propose a modular architec-
ture with lexical entries containing phonological- and syntactic-forms to explain declara-
tive questions and comma intonation in non-restrictive relative clauses.
Many analyses of clitics account for their unusual behavior by exploiting the mor-
phology-syntax interface, instead of or in addition to using prosodic-structure for this
purpose. Wescoat (2005) proposes lexical sharing to account for English auxiliary clitics,
864 IV. Syntactic Models

while Luís and Otoguro (2005) argue that morphology and phrase structure are separate
levels of analysis. In addition, OT-LFG (section 6) can use the interaction of OT con-
straints in order to govern clitic placement (Estigarribia 2005; Lowe 2011).

5.4. Semantic-structure
Halvorsen (1983) was the first to propose that semantic interpretation in LFG be done
by projecting a s(emantic)-structure from the f-structure. In turn, this s-structure could
be mapped to formulae in intensional logic and model-theoretic interpretation. Note that
f-structures have since been shown to be equivalent to quasi-logical forms (QLF; van
Genabith and Crouch 1996), confirming that f-structures provide a very good input for
further semantic analysis.
Halvorsen’s example (1) is shown simplified in (62). Note that the f-structure repre-
sents an analysis whereby English auxiliaries were taken to embed a VCOMP (infinitive
clause), an analysis that is no longer standard in LFG − rather a flat f-structure is now
assumed with the auxiliary only registering tense/aspect information (Butt et al. 1999).

(62) a. John was flattered by Mary.

b. f-structure

c. s-structure

d. formula of intensional logic

flatter*ʹ (m, j)

In Halvorsen’s original example, the complex f-structure is projected into the flat s-
structure in (62c), which can then be further processed into an intensional logic formula,
as shown in (62d).
Halvorsen and Kaplan (1988) also argue for a s-structure but propose a co-description
analysis whereby the s-structure is not projected in its entirety from the f-structure.
Instead it is created simultaneously as a parallel projection to the f-structure, with certain
25. Lexical-Functional Grammar 865

mutually constraining factors. Co-description analyses have enjoyed a steady popularity

in LFG.
In newer approaches, however, semantic analyses based on LFG f-structures have
largely abandoned the s-structure approach and instead use Glue semantics (Dalrymple
1999). Under this approach, the meanings of words are separate from the mechanisms
which combine them. Lexical entries contain meaning constructors that specify how the
word contributes to the meaning of larger syntactic constituents. The meanings of the
larger constituents are derived using linear-logic deduction on the meanings of the parts.
Combining all of the premises according to a resource logic (linear logic) results in the
meaning of an utterance. Because the composition is governed by this logic, it does not
have to follow the rules of phrasal (e.g., c-structure) composition. Glue semantics is
resource sensitive, because in the deduction each premise is consumed as it is used and
all premises must be consumed. Extensive examples of Glue semantic analyses for differ-
ent syntactic and semantic constructions can be found in Dalrymple (2001).
Within the computational community yet another approach to semantic construction
has recently been developed, namely a system called XFR (Crouch et al. 2011) involving
ordered rewrite rules, which are used to efficiently and robustly produce semantic struc-
tures from f-structures (Crouch 2006; Crouch and King 2006). The resulting semantics
gives a flat representation of the sentence’s predicate argument structure and the semantic
contexts in which those predications hold. These semantic structures in turn are input
to additional ordered rewrite rules which produce Abstract Knowledge Representations
(AKRs) (Bobrow et al. 2005; Bobrow et al. 2007).

5.5. The overall LFG projection architecture

We have now briefly introduced a number of different proposals for additional projec-
tions. Within the LFG community, there are a number of proposals as to how these
various different structures relate to one another. A constant is that f-structure is a projec-
tion from the c-structure, as in the original proposal (Bresnan 1982b). A possible archi-
tecture is represented in (63). Here, the c-structure and f-structure are modulated by the
m(orphosyntactic)-structure and the a(rgument)-structure. In addition, the s(emantic)-
structure comprises information from the f-structure and the c-structure, with the p(ro-
sodic)-structure and i(nformation)-structure mapping between the c-structure and s-struc-
ture. There are many minor variants of the architecture shown in (63); see Asudeh
(2006: 369) for discussion.

(63) The Correspondence Architecture (Asudeh 2006)

866 IV. Syntactic Models

6. OT-LFG
An additional architectural component was added to LFG with the advent of Optimality
Theory (OT). OT was originally formulated to solve problems with respect to alignment
constraints in prosodic morphology (McCarthy and Prince 1993). The establishment of
OT as a serious domain of investigation within syntax is due mainly to articles by Jane
Grimshaw (Grimshaw 1997) and Joan Bresnan (Bresnan 1998). In particular, Bresnan
showed how OT could be used in conjunction with standard LFG.
The next section introduces the basic architecture and assumptions of OT, and the
following section shows how LFG theory can be integrated with OT.

6.1. OT basics
Within OT, the goal is to determine an optimal output (surface form) with respect to a
given input. The optimal output is picked from a set of candidates that compete with
one another. The competition between the candidates is resolved by an evaluation of
constraint violations, as shown in (64) (adapted from Vogel 2001).

(64)

The nature of the input is still in need of precise definition in much of OT. Grimshaw’s
(1997) original paper assumed that the input encompasses the basic argument structure
of a predicate, and a specification of tense/aspect. In later work (Grimshaw and Samek-
Lodovici 1998), the focus/topic specifications were included as part of the input. A
typical input might look as in (65), where the argument structure of give is specified,
along with information about which argument is the topic, and which is in focus.

(65) give(x,y,z), x=topic, z=focus, x=Kim, z=dog, y=bone

6.2. Optimality Theory and LFG

In OT-LFG (Bresnan 2000; Kuhn 2003), the input is assumed to be an underspecified f-
structure in the sense that the GFs are not as yet specified. An example for the transitive
verb drive is shown in (66).

(66)
25. Lexical-Functional Grammar 867

The skeletal f-structure inputs are passed on to a function GEN, which generates a set
of possible output candidates that could correspond to the input. Within OT-LFG, GEN
is assumed to be equivalent to a standard LFG grammar. LFG grammars can be used to
both parse and generate; so the idea is that an underspecified input as in (66) is passed
on to an existing grammar for English. This searches through its rule space and produces
all possible pairings of c-structures and surface strings that could correspond to the input
f-structure. Kuhn (2003) shows that this is computationally viable.
Quite a bit of work is done within OT-LFG, some of the early landmark contributions
are collected in Sells (2001).

7. Computational issues and resources

Right from its inception, LFG was designed as a theory and formalism whose mathemati-
cal basis is solid and well-understood. This property means that LFG is particularly
suitable for computational linguistic research and there has been computational linguistic
work based on LFG for over twenty years. Interest in using LFG in computational lin-
guistics, as well as natural language processing applications is increasing steadily (see
the chapter on computational syntax in this volume).
Perhaps the most visible computational effort involving LFG is the Parallel Grammar
(ParGram) group (Butt et al. 1999; Butt et al. 2002), which implements LFG grammars
of different languages on the XLE grammar development platform (Crouch et al. 2011).
The guiding motivation of ParGram is an effort at parallel analyses across languages
using parallel implementation techniques. That is, much effort goes into sorting through
possible alternative analyses and feature spaces for phenomena across languages and
trying to agree on f-structure analyses that are as parallel as possible across a diverse
number of languages. Grammars that have been developed so far include Arabic, Eng-
lish, French, German, Hungarian, Indonesian, Japanese, Malagasy, Murrinh-Patha, Nor-
wegian, Tigrinya, Turkish, Urdu and Welsh. The English and the Japanese grammars
have been used for industrial purposes in terms of developing query systems and the
Norwegian grammar was used in a machine translation project named LOGON.
Since f-structures are already very close to semantic forms (see section 5.4), the idea
in LFG is that if analyses are kept as parallel across languages as possible, then applica-
tions like machine translation should be able to produce good results more easily
(Frank 1999).
The bulk of the computational work done within LFG is (naturally) symbolic and
rule-based. However, statistical methods can be integrated in several ways, i.e. to pick
the most likely parse or sentence to be generated among a forest of possibilities (Riezler
and Maxwell 2006; Cahill et al. 2007; Graham et al. 2009).
All of the projections discussed above are in principle implementable via the grammar
development platform XLE, though this is rarely done in practice, with most computa-
tional grammars confining themselves to just c- and f-structures. However, an OT-projec-
tion is routinely used in the computational grammars (Frank et al. 2001; Crouch et al.
2011). Its purpose is to help constrain the grammar by dispreferring or preferring certain
rules or lexical items over others.
868 IV. Syntactic Models

8. Psycholinguistic research
LFG has also been used as the basis for psycholinguistic research, though not to the
extent that was once thought. The first standard reference for LFG work, namely Bresnan
(1982b), contains papers on psycholinguistics by Marilyn Ford and Steven Pinker. The
basic tenets of LFG continued to inform their psycholinguistic research (e.g., Levelt
1989; Pinker 1989; Gropen et al. 1991). However, most of the psycholinguistic LFG-
related work seems to have taken place in the 1980s and early 1990s. An exception are
recent publications by Ford and Bresnan (Ford and Bresnan 2013; Bresnan and Ford
2010), which take up Bresnan’s recent work on stochastic approaches to syntax (e.g.,
Bresnan 2007).

9. Conclusion
From its inception, LFG has had the goal of combining linguistic insights with solid
mathematical foundations, computational tractability and implementability (Dalrymple
et al. 1995). Much work in LFG has focused on typological diversity, with major efforts
in a large number of language families. This is reflected in some of the core areas
of research such as case and agreement systems, causatives and complex predicates,
coordination, and anaphora. LFG minimally posits two levels of representation: the c-
structure and the f-structure. The c-structure encodes linear order, constituency and hier-
archical relations while the f-structure focuses on the dependency structure of a clause.
The mapping between the two is realized in terms of a mathematical function and need
not be one-to-one. This fact allows for a flexible architecture which can deal with long-
distance dependencies and other complex linguistic phenomena in a straight-forward and
elegant manner.
A broad range of theoretical, descriptive and computational linguists work within
LFG, with some work also being done in neuro- and psycholinguistics. Several current
textbooks are available (Bresnan 2001; Dalrymple 2001; Falk 2001), as is a description
of the ParGram grammar development effort (Butt et al. 1999).

10. References (selected)

Ackerman, Farrell
1992 Complex predicates and morpholexical relatedness: Locative alternation in Hungarian.
In: I. Sag and A. Szabolcsi (eds.), Lexical Matters, 55−83. Stanford: CSLI Publications.
Ackerman, Farrell, and John Moore
2001 Proto-properties and Grammatical Encoding: A Correspondence Theory of Argument
Selection. Stanford: CSLI Publications.
Alsina, Alex
1996 The Role of Argument Structure in Grammar: Evidence from Romance. Stanford:
CSLI Publications.
Alsina, Alex
1997 A theory of complex predicates: evidence from causatives in Bantu and Romance. In:
A. Alsina, J. Bresnan, and P. Sells (eds.), Complex Predicates, 203−246. Stanford:
CSLI Publications.
25. Lexical-Functional Grammar 869

Alsina, Alex, and Smita Joshi

1991 Parameters in causative constructions. In: Papers from the 27 th Regional Meeting of the
Chicago Linguistic Society, 1−15. University of Chicago.
Alsina, Alex, and Sam Mchombo
1993 Object asymmetries and the Chicheŵa applicative construction. In: S. Mchombo (ed.),
Theoretical Aspects of Bantu Grammar, 17−45. Stanford: CSLI Publications.
Asudeh, Ash
2006 Direct compositionality and the architecture of LFG. In: M. Butt, M. Dalrymple, and T.
H. King (eds.), Intelligent Linguistic Architectures: Variations on themes by Ronald M.
Kaplan. Stanford: CSLI Publications.
Asudeh, Ash
2009 Adjacency and locality: A constraint-based analysis of complementizer-adjacent extrac-
tion. In: M. Butt and T. H. King (eds.), Proceedings of the LFG 2009 Conference, 106−
126, Stanford: CSLI Publications.
Asudeh, Ash, and Ida Toivonen
2009 Lexical-Functional Grammar. In: B. Heine and H. Narrog (eds.), The Oxford Handbook
of Linguistic Analysis. Oxford: Oxford University Press.
Berman, Judith
2003 Clausal Syntax of German. Stanford: CSLI Publications.
Bobrow, Danny, Bob Cheslow, Cleo Condoravdi, Lauri Karttunen, Tracy Holloway King, Rowan
Nairn, Valeria de Paiva, Charlotte Price, and Annie Zaenen
2007 PARC’s bridge and question answering system. In: Proceedings of the Grammar Engi-
neering Across Frameworks (GEAF07) Workshop, 46−66. Stanford: CSLI Publications.
Bobrow, Danny, Cleo Condoravdi, Richard Crouch, Ronald Kaplan, Lauri Karttunen, Tracy Hollo-
way King, Valeria de Paiva, and Annie Zaenen
2005 A basic logic for textual inference. In: Proceedings of the AAAI Workshop on Inference
from Textual Question Answering, Pittsburgh, PA.
Bögel, Tina, Miriam Butt, Ronald M. Kaplan, Tracy Holloway King, and John T. Maxwell III
2009 Clitics and prosodic phonology in LFG. In: M. Butt and T. H. King (eds.), Proceedings
of the LFG 2009 Conference. Stanford: CSLI Publications.
Bögel, Tina, Miriam Butt, Ronald M. Kaplan, Tracy Holloway King, and John T. Maxwell III
2010 Second position clitics and the prosody-syntax interface. In: M. Butt and T. H. King
(eds.), Proceedings of the LFG 2010 Conference, 106−126. Stanford: CSLI Publications.
Bresnan, Joan
1977 Transformation and categories in syntax. In: R. E. Butts and J. Hintikka (eds.), Basic
Problems in Methodology and Linguistics. Part Three of the Proceedings of the Fifth
International Congress of Logic, Methodology, and Philosophy of Science, 261−282.
Dordrecht: Reidel.
Bresnan, Joan
1982a Control and complementation. In: J. Bresnan (ed.), The Mental Representation of Gram-
matical Relations. Cambridge, MA: The MIT Press.
Bresnan, Joan (ed.)
1982b The Mental Representation of Grammatical Relations. Cambridge, MA: The MIT Press.
Bresnan, Joan
1990 Monotonicity and the theory of relation-changes in LFG. Language Research 26(4):
637−652.
Bresnan, Joan
1994 Locative inversion and the architecture of universal grammar. Language 70(1): 72−131.
Bresnan, Joan
1998 Morphology competes with syntax: Explaining typological variation in weak crossover
effects. In: P. Barbosa, D. Fox, P. Hagstrom, M. McGinnis, and D. Pesetsky (eds.), Is
870 IV. Syntactic Models

the Best Good Enough? Optimality and Competition in Syntax, 59−92. Cambridge, MA:
The MIT Press and MIT Working Papers in Linguistics.
Bresnan, Joan
2000 Optimal syntax. In: J. Dekkers, F. van der Leeuw, and J. van de Weijer (eds.), Optimality
Theory: Phonology, Syntax, and Acquisition, 334−385. Oxford: Oxford University Press.
Bresnan, Joan
2001 Lexical-Functional Syntax. Oxford: Blackwell.
Bresnan, Joan
2007 Is syntactic knowledge probabilistic? Experiments with the English dative alternation.
In: S. Featherston and W. Sternefeld (eds.), Roots: Linguistics in Search of Its Evidential
Base, 77−96. Berlin: Mouton de Gruyter.
Bresnan, Joan, and Marilyn Ford
2010 Predicting syntax: Processing dative constructions in American and Australian varieties
of English. Language 86(1): 186−213.
Bresnan, Joan, and Jonni Kanerva
1989 Locative inversion in Chichewa: A case study of factorization in grammar. Linguistic
Inquiry 20: 1−50.
Bresnan, Joan, and Sam Mchombo
1987 Topic, pronoun, and agreement in Chichewa. Language 63(4): 181−254.
Bresnan, Joan, and Lioba Moshi
1990 Object asymmetries in comparative Bantu syntax. Linguistic Inquiry 21(2): 147−185.
Bresnan, Joan, and Annie Zaenen
1990 Deep unaccusativity in LFG. In: K. Dziwirek, P. Farrell, and E. Mejías-Bikandi (eds.),
Grammatical Relations: A Cross-Theoretical Perspective, 45−57. Stanford: CSLI Publi-
cations.
Butt, Miriam
1995 The Structure of Complex Predicates in Urdu. Stanford: CSLI Publications.
Butt, Miriam
1998 Constraining argument merger through aspect. In: E. Hinrichs, A. Kathol, and T. Naka-
zawa (eds.), Complex Predicates in Nonderivational Syntax, 73−113. New York: Aca-
demic Press.
Butt, Miriam
2006 Theories of Case. Cambridge: Cambridge University Press.
Butt, Miriam
2008 Case in Lexical-Functional Grammar. In: A. Malchukov and A. Spencer (eds.), The
Handbook of Case, 59−71. Oxford: Oxford University Press.
Butt, Miriam, Helge Dyvik, Tracy Holloway King, Hiroshi Masuichi, and Christian Rohrer
2002 The parallel grammar project. In: J. Carroll, N. Oostdijk, and R. Sutcliffe (eds.), Pro-
ceedings of the Workshop on Grammar Engineering and Evaluation at the 19 th Interna-
tional Conference on Computational Linguistics, 1− 7.
Butt, Miriam, and Tracy Holloway King
1998 Interfacing phonology with LFG. In: M. Butt and T. H. King (eds.), Proceedings of the
LFG 1998 Conference. Stanford: CSLI Publications.
Butt, Miriam, and Tracy Holloway King
2003 Case systems: Beyond structural distinctions. In: E. Brandner and H. Zinsmeister (eds.),
New Perspectives on Case Theory, 53−87. Stanford: CSLI Publications.
Butt, Miriam, and Tracy Holloway King
2005 The status of case. In: V. Dayal and A. Mahajan (eds.), Clause Structure in South Asian
Languages, 153−198. Berlin: Springer Verlag.
Butt, Miriam, Tracy Holloway King, Maria-Eugenia Niño, and Frédérique Segond
1999 A Grammar Writer’s Cookbook. Stanford: CSLI Publications.
25. Lexical-Functional Grammar 871

Butt, Miriam, Marı́a-Eugenia Niño, and Frédérique Segond

1996 Multilingual processing of auxiliaries in LFG. In: D. Gibbon (ed.), Natural Language
Processing and Speech Technology: Results of the 3rd KONVENS Conference, 111−
122, Bielefeld.
Cahill, Aoife, Martin Forst, and Christian Rohrer
2007 Stochastic realisation ranking for a free word order language. In: S. Busemann (ed.),
Proceedings of the European Workshop on Natural Language Generation (ENLG-07),
Dagstuhl.
Choi, Hyewon
1999 Optimizing Structure in Context: Scrambling and Information Structure. Stanford:
CSLI Publications.
Chomsky, Noam
1965 Aspects of the Theory of Syntax. Cambridge, MA: The MIT Press.
Croft, William
2001 Radical Construction Grammar: Syntactic Theory in Typological Perspective. Oxford:
Oxford University Press.
Crouch, Dick, Mary Dalrymple, Ronald M. Kaplan, Tracy King, John Maxwell, and Paula Newman
2011 XLE Documentation. On-line documentation, Palo Alto Research Center (PARC).
Crouch, Richard
2006 Packed rewriting for mapping text to semantics and KR. In: M. Butt, M. Dalrymple,
and T. H. King (eds.), Intelligent Linguistic Architectures: Variations on Themes by
Ronald M Kaplan, 389−416. Stanford: CSLI Publications.
Crouch, Richard, and Tracy Holloway King
2006 Semantics via F-structure rewriting. In: Proceedings of the LFG 2006 Conference, 145−
165. Stanford: CSLI Publications.
Dalrymple, Mary
1993 The Syntax of Anaphoric Binding. Stanford: CSLI Publications.
Dalrymple, Mary (ed.)
1999 Semantics and Syntax in Lexical Functional Grammar. Cambridge, MA: The MIT Press.
Dalrymple, Mary
2001 Lexical Functional Grammar. Vol. 34 of Syntax and Semantics. New York: Academic
Press.
Dalrymple, Mary, Ronald M. Kaplan, John T. Maxwell III, and Annie Zaenen (eds.)
1995 Formal Issues in Lexical-Functional Grammar. Stanford: CSLI Publications.
Dalrymple, Mary, Ronald M. Kaplan, and Tracy Holloway King
2001 Weak crossover and the absence of traces. In: M. Butt and T. H. King (eds.), Proceedings
of the LFG 2001 Conference. Stanford: CSLI Publications.
Dalrymple, Mary, and Tracy Holloway King
2000 Missing-object constructions: Lexical and constructional variation. In: M. Butt and T.
H. King (eds.), Proceedings of the LFG 2000 Conference. Stanford: CSLI Publications.
Dalrymple, Mary, and Helge Lødrup
2000 The grammatical functions of complement clauses. In: M. Butt and T. H. King (eds.),
Proceedings of the LFG 2000 Conference. Stanford: CSLI Publications.
Dalrymple, Mary, and Louise Mycock
2011 The prosody-semantics interface. In: M. Butt and T. H. King (eds.), Proceedings of the
LFG 2011 Conference. Stanford: CSLI Publications.
Dench, Alan
1995 Martuthunira: a Language of the Pilbara Region of Western Australia. Canberra: Pa-
cific Linguistics.
Dowty, David
1991 Thematic proto-roles and argument selection. Language 67(3): 547−619.
872 IV. Syntactic Models

Estigarribia, Bruno
2005 Direct object clitic doubling in OT-LFG: A new look at Rioplantense Spanish. In: M.
Butt and T. H. King (eds.), Proceedings of the LFG 2005 Conference. CSLI Publications.
Evans, Nicholas, and Stephen Levinson
2009 The myth of language universals: Language diversity and its importance for cognitive
science. The Behavioral and Brain Sciences 32: 429−448.
Falk, Yehuda
2001 Lexical-Functional Grammar: An Introduction to Parallel Constraint-based Syntax.
Stanford: CSLI Publications.
Falk, Yehuda
2007 Do we wanna (or hafta) have empty categories? In: M. Butt and T. H. King (eds.),
Proceedings of the LFG 2007 Conference, 184−197. CSLI Publication.
Ford, Marilyn, and Joan Bresnan
2013 Using convergent evidence from psycholinguistics and usage. In: M. Krug and J.
Schlüter (eds.), Research Methods in Language Variation and Change, 295−312: Cam-
bridge: Cambridge University Press.
Frank, Anette
1996 A note on complex predicate formation: Evidence from auxiliary selection, reflexiviza-
tion, and past participle agreement in French and Italian. In: M. Butt and T. H. King
(eds.), Proceedings of LFG 1996 Conference. Stanford: CSLI Publications.
Frank, Anette
1999 From parallel grammar development towards machine translation. In: Proceedings of the
MT Summit VII: MT in the Great Translation Era, 134−142. Kent Ridge Digital Labs.
Frank, Anette, Tracy Holloway King, Jonas Kuhn, and John T. Maxwell III
2001 Optimality Theory-style constraint ranking in large-scale LFG grammars. In: P. Sells
(ed.), Formal and Empirical Issues in Optimality Theoretical Syntax, 367− 398. Stan-
ford: CSLI Publications.
Graham, Yvette, Anton Bryl, and Josef van Genabith
2009 F-structure transfer-based statistical machine translation. In: M. Butt and T. H. King
(eds.), Proceedings of the LFG 2009, 317−337. Stanford: CSLI Publications.
Grimshaw, Jane
1997 Projection, heads, and optimality. Linguistic Inquiry 28(3): 373−422.
Grimshaw, Jane, and Vieri Samek-Lodovici
1998 Optimal subjects and subject universals. In: P. Barbosa, D. Fox, P. Hagstrom, M.
McGinnis, and D. Pesetsky (eds.), Is the Best Good Enough?, 193−219. Cambridge,
MA: The MIT Press.
Gropen, Jess, Steven Pinker, Michelle Hollander, and Richard Goldberg
1991 Affectedness and direct objects: The role of lexical semantics in the acquisition of verb
argument structure. Cognition 41: 153−195.
Halvorsen, Per-Kristian
1983 Semantics in Lexical-Functional Grammar. Linguistic Inquiry 14: 567−615.
Halvorsen, Per-Kristian, and Ronald M. Kaplan
1988 Projections and semantic description in Lexical-Functional Grammar. In: Proceedings
of the International Conference on Fifth Generation Computer Systems. ICOT.
Kaplan, Ronald M
1995 The formal architecture of Lexical-Functional Grammar. In: M. Dalrymple, R. M. Ka-
plan, J. T. M. III, and A. Zaenen (eds.), Formal Issues in Lexical-Functional Grammar,
7−27. Stanford: CSLI Publications.
Kaplan, Ronald M., Klaus Netter, Jürgen Wedekind, and Annie Zaenen
1989 Translation by structural correspondences. In: Proceedings of the 4th Meeting of the
European Association for Computational Linguistics, 272−281, University of Manches-
ter. Reprinted in Mary Dalrymple, Ronald M. Kaplan, John Maxwell, and Annie Zaenen,
25. Lexical-Functional Grammar 873

eds., Formal Issues in Lexical-Functional Grammar, 311−329. Stanford: CSLI Publica-

tions. 1995.
Kaplan, Ronald M., and Annie Zaenen
1989 Long-distance dependencies, constituent structure, and functional uncertainty. In: M.
Baltin and A. Krock (eds.), Alternative Conceptions of Phrase Structure. Chicago: Chi-
cago University Press.
King, Tracy Holloway
1995 Configuring Topic and Focus in Russian. Stanford: CSLI Publications.
King, Tracy Holloway
1997 Focus domains and information structure. In: M. Butt and T. H. King (eds.), Proceedings
of the LFG 1997 Conference. CSLI Publications.
Kiparsky, Paul
1973 ‘Elsewhere’ in phonology. In: S. R. Anderson and P. Kiparsky (eds.), A Festschrift for
Morris Halle, 93−106. New York: Holt, Rinehart and Winston.
Kroeger, Paul R.
2005 Analyzing Grammar: An Introduction. Cambridge: Cambridge University Press.
Kuhn, Jonas
2003 Optimality-Theoretic Syntax − A Declarative Approach. Stanford: CSLI Publications.
Kuhn, Jonas, and Louisa Sadler
2007 Single conjunct agreement and the formal treatment of coordination in LFG. In: M. Butt
and T. H. King (eds.), Proceedings of the LFG 2007 Conference, 302−322. Stanford:
CSLI Publications.
Levelt, W. J. M.
1989 Speaking: From Intention to Articulation. Cambridge, MA: The MIT Press.
Levin, Lori
1987 Toward a linking theory of relation changing rules in LFG. Technical Report CSLI-87−
115, Stanford.
Lowe, John J.
2011 Ṛgvedic clitics and ‘prosodic movement’. In: M. Butt and T. H. King (eds.), Proceedings
of the LFG 2011 Conference. Stanford: CSLI Publications.
Luı́s, Ana R., and Ryo Otoguro
2005 Morphological and syntactic well-formedness: The case of European Portuguese proclit-
ics. In: M. Butt and T. H. King (eds.), Proceedings of the LFG 2005 Conference. Stan-
ford: CSLI Publications.
Manning, Christopher D., and Avery Andrews
1999 Complex Predicates and Information Spreading. Stanford: CSLI Publications.
McCarthy, John, and Alan Prince
1993 Generalized alignment. In: G. Booij and J. van Marle (eds.), Yearbook of Morphology
1993, 79−153. Berlin: Kluwer Academic Publishers.
Mohanan, Tara
1994 Argument Structure in Hindi. Stanford: CSLI Publications.
Mycock, Louise
2006 The typology of constituent questions: A lexical-functional grammar analysis of wh-
questions. PhD thesis, University of Manchester.
Nordlinger, Rachel
1998 Constructive Case: Evidence from Australian Languages. Stanford: CSLI Publications.
Nordlinger, Rachel
2000 Australian case systems: Towards a constructive solution. In: M. Butt and T. H. King
(eds.), Argument Realization, 41−72. Stanford: CSLI Publications.
O’Connor, Rob
2004 Information structure in lexical-functional grammar: The discourse-prosody correspon-
dence in English and Serbo-Croation. PhD thesis, University of Manchester.
874 IV. Syntactic Models

Pinker, Steven
1989 Learnability and Cognition: The Acquisition of Argument Structure. Cambridge, MA:
The MIT Press.
Pullum, Geoffrey K., and Barbara C. Scholz
2001 On the distinction between generative-enumerative and model-theoretic syntactic frame-
works. In: P. de Groote, G. Morrill, and C. Retor (eds.), Logical Aspects of Computa-
tional Linguistics, 17−43. Berlin: Springer Verlag.
Pullum, Geoffrey K., and Barbara C. Scholz
2010 Recursion and the infinitude claim. In: H. van der Hulst (ed.), Recursion in Human
Language, 113−138. Berlin: Mouton de Gruyter.
Riezler, Stefan, and John T. Maxwell
2006 Grammatical machine translation. In: Human Language Technology Conference − North
American Chapter of the Association for Computational Linguistics Annual Meeting
(HLT-NAACL’06).
Rouveret, Alain, and Jean Roger Vergnaud
1980 Specifying reference to the subject: French causatives and conditions on representations.
Linguistic Inquiry 11: 97−202.
Sadler, Louisa
1993 Co-description and translation. In: F. van Eynde (ed.), Linguistic Issues in Machine
Translation, 44−71. London: Pinter Publishers.
Sadler, Louisa
2003 Coordination and asymmetric agreement in Welsh. In: M. Butt and T. H. King (eds.),
Nominals: Inside and Out, 85−118. Stanford: CSLI Publications.
Sadler, Louisa, and Andrew Spencer (eds.)
2004 Projecting Morphology. Stanford: CSLI Publications.
Sells, Peter (ed.)
2001 Formal and Empirical Issues in Optimality Theoretic Syntax. Stanford: CSLI Publica-
tions.
van Genabith, Josef, and Richard Crouch
1996 F-structures, QLFs and UDRSs. In: M. Butt and T. H. King (eds.), Proceedings of the
First International Conference on Lexical-Functional Grammar, 190−205.
Vogel, Ralf
2001 Case conflict in German free relative constructions: An Optimality Theoretic treatment.
In: G. Müller and W. Sternefeld (eds.), Competition in Syntax, 341−375. Berlin: Mouton
de Gruyter.
Wechsler, Stephen, and Larisa Zlatić
2003 The Many Faces of Agreement. Stanford: CSLI Publications.
Wescoat, Michael T
2005 English nonsyllabic auxiliary contractions: An analysis in LFG with lexical sharing. In:
M. Butt and T. H. King (eds.), Proceedings of the LFG 2005 Conference. Stanford:
CSLI Publications.
Wilson, Stephen
1999 Coverbs and Complex Predicates in Wagiman. Stanford: CSLI Publications.
Zaenen, Annie
1993 Unaccusativity in Dutch: Integrating syntax and lexical semantics. In: J. Pustejovsky
(ed.), Semantics and the Lexicon, 129−161. Berlin: Kluwer Academic Publishers.
Zaenen, Annie, Joan Maling, and Höskuldur Thráinsson
1985 Case and grammatical functions: The Icelandic passive. Natural Language and Linguis-
tic Theory 3: 441−483. Reprinted in Joan Maling and Annie Zaenen (Eds.) Syntax and
Semantics 24: Modern Icelandic Syntax, 95−164. New York: Academic Press. 1990.

Miriam Butt, Konstanz (Germany)

Tracy Holloway King, San Jose (USA)