You are on page 1of 11

A Survey of Lexical-Functional Grammar

Joseph Griego
12/13/14
Prof. Malhotra

1. Introduction. Lexical-functional grammar is an alternative theory of syntax, that,


instead of using movements to generate surface structure, uses a system of multiple, parallel
constraints to create correct structures. While more conventional, movement-based theories
(here referred to as transformational grammar) rely on movements to transform a theoretical
deep-structure of a sentence to the pronounced surface structure, lexical-functional grammar
(LFG) uses multiple simultaneous structures. The functional structure, or f-structure, is roughly
equivalent to transformational grammars deep structure. Another layer, the constituent structure
or c-structure is roughly equivalent to surface structure. Different treatments of LFG also
introduce other layers of structure, such as argument structure (a-structure, equivalent to the
mapping of -roles to arguments in transformational grammar.)
The key piece of LFG is the way in which the interfaces between these layers of structure
mutually constrain the overall structure of a sentence. The most important interface for our
discussion is that between f-structure and c-structure. This interface is determined by functional
equations that relate the functional relationships between nodes in the constituent structure of the
sentence. These functional equations are derived from the lexical items in the sentence and the
grammar of the language.
LFG is motivated by difficulties with accounting for certain phenomena within
a transformational theory. Some of these are constructions like passives, ditransitives,
conjunctions, as well as word order in nonconfigurational languages. LFGs separation of
concerns makes many of these arguably simpler to account for than in a transformational theory.
As we will see, passives, for instance, are accomplished with a single change to the argumentstructure of the verb.
LFG has two main advantages over a transformational theory: first, it is (arguably)
easier to formalize, since it relies on a formal specification of the interfaces between structures;
although transformational theory can (and has) been formalized, LFG lends itself to this process
more readily. Second, it presents a simple way to allow computers to parse sentences, since much
of a languages grammar is stored in its lexicon.

2
2. Functional structure. In LFG, the functional structure is used to represent the
features and relationship between parts of a sentence. Well begin with a simple example:

subj

tense

pred

obj

def
+

num sg

pred `linguist'

fut

`eatsubj, obj'

def
num sg

pred `cake'

{[
}
]

pred `chocolate'
adj

Figure 1. F-structure for The linguist will eat a chocolate cake


F-structure is written by convention as an attribute-value matrix (AVM,) where each
attribute is tied to a value, which can itself be another AVM. In this example, the outermost
matrix represents the sentence as a whole. The subj feature is filled by a noun phrase and
represents the subject of the sentence. The tense feature is, unsurprisingly, the tense of the
sentence. The pred feature is common throughout and represents the lexical item that lends
meaning to the overall structure. In a transformational theory, this would be the head of a phrasal
projection. Verbs in a pred feature are annotated with their arguments, here we have eat<subj,
obj> indicating

that the lexical verb is eat and that two arguments are required, on in the

subject position and one in the object position. This corresponds with the -grids we can draw for
lexical verbs in transformational syntax.
What we would equate to noun phrases are the smaller AVMs in the subj and obj
positions. These carry features that mark them for definiteness (def) and number (num) as well
as their own predicates. We also see the feature adj which allows us to embed adjuncts inside
the f-structure. The notation here is a little different: we expect adj to contain a set of AVMs, as
opposed to a single one.
3. Constituent structure. In LFG, constituent structure is used to represent the grouping
of words into constituents, in the same way as transformational grammar. However, in addition
to rejecting movement, LFG also rejects some ideas taken as principles in transformational

3
grammar. One of the most notable of these is endocentricity. LFG qualifies endocentricity
somewhat. Falk (2001) states a principle, that of Economy of Expression as follows: All
syntactic phrase structure nodes are optional and are not used unless required to license
elements required to create a well-formed f-structure or to add semantic content. Thus, in
LFG, the c-structure of a CP like What I ate could not be analyzed with a phonetically null
complementizer as head of the CPit would instead be analyzed as a CP with no head, violating
endocentricity, but preserving economy of expression.
Violating endocentricity also allows LFG to succinctly account for nonconfigurational
languages. This is a motivating example in Bresnan (2001), where the example of Warlpiri is
used and the existence of a nonconfigurational S node in c-structure:
S
NP

Aux

NP

NP

NP

wita-jarra-rlu
small-dual-erg

ka-pala
pres-3du-subj

wajili-pi-nyi
chase-npast

yalumpu
that-abs

kurdu-jarra-rlu
child-dual-erg

maliki
dog.abs

Figure 2. Nonconfigurational S in Warlpiri (from Bresnan 2001)


The two small children are chasing that dog
In more tame situations, c-structure strongly resembles X-bar syntax and borrows
strongly from it. Facing is the c-structure for the1.sentence
used before, The linguist will eat
CP XP C

a chocolate cake. (Figure 3) Note that it is identical to the


transformational
grammar surface

DP
2. IP

structure of the same sentence.

CP
I

PP

In LFG, constituent structure is governed


phrase structure rules,
3. by
DPlanguage-specific
NP D

just as in a more mainstream approach. A selection


4. Cofphrase
C TP structure rules for English, per

5. T
VP structure rule are optional, per the
Falk (2001) is below. Note that all parts generated
bya T
phrase

principle of Economy of Expression:

6. D D NP

7. VP V DP DP PP

1. CP XP C

DP
CP
I
2. IP

PP

8. PP P DP PP TP
9. NP N PP CP
{
}
IP
10. AP A PP
CP
{
}
AP
11. XP
XP
ADVP

3. DP NP D
4. C C TP
5. T T VP
6. D D NP

7. VP V DP DP PP
8. PP P DP PP TP

IP
CP

}12. XP XP PP

IP
CP

4
TP
T

DP
D

VP

will

NP

the

DP

eat

linguist

NP

N
AdjP

Adj

Adj

cake

chocolate

Figure 3. C-structure for The linguist will eat a chocolate cake


4. Argument structure. LFG also posits an additional layer of structure identified by the
subcategorization of the verb, i.e. its -grid. This is usually shown inline with the pred feature
of the f-structure. This extra structure is called argument structure, or a-structure for short.
The principles that govern a-structure are the same as the theta criterion from transformational
grammar. These essentially boil down to two properties: coherence and completeness. Coherence
requires that every argument function in the f-structure is selected by a local pred feature, while
completeness requires that every argument function specified in the a-structure (pred feature) is
present in the f-structure. These correspond to Principle A and Principle B of the theta criterion
from transformational grammar, which state that each DP in the verb phrase must have exactly
one theta role and that no theta role goes unassigned, respectively. More sophisticated analyses
of a-structure are possible but are outside the scope of this paper.
5. Functional equations. The interface between f-structure and c-structure is governed by
formal equality relations called functional equations. Functional equations arise from both lexical

5
entries of the language and the phrase structure rules of that language. Some examples of phrase
structure rules licensing functional structure are in the way that e.g. in English, the first DP in
an IP construction is the subj of the enclosing function. Therefore, we can annotate the phrase
structure rule with a functional equation as follows:

TP

DP

NP
,
CP

PP
( subj) =

T
=

Figure 4. Annotated phrase structure rule for TPs.


The notation ( subj) = is essentially shorthand that allows us to reference the
f-structure corresponding to the above c-structure node and the one below. In this case
( subj) = would be interpreted as saying that the subject feature of the above node is equated
with the f-structure of the below node.
An example of f-structure being licensed by lexical entries is fairly straightforward, an
example of this is the lexical item dinosaur, which (Falk 2001) can be interpreted as carrying the
functional equations:
( pred) = dinosaur ( num) = sg
These convey the information that the word dinosaur provides the meaning to its
f-structure and constrains the num feature of that structure to be singular. This prevents us from
producing a phrase like two dinosaur, since two would require num to be plural and dinosaur
would require num to be singular--it cant be both, so the structures fail to unify.
As an illustrative example, facing is a c-structure tree of the sentence the hamster will
eat a falafel together with its f-structure and annotated with the functional equations arising from
the lexical entries of the words and the phrase structure rules of English. (Figures 5 and 6)
Note that the functional equations do not merely constrain the f-structure, they fully
specify it; that is, there is no piece of f-structure that is not licensed by a functional equation from
another level of structure. In other words, the f-structure is the least solution of the functional
equations, taken simultaneously.
6. Passives in LFG. One significant advantage of LFG over transformational theories

def
+

subj
num sg

pred
`hamster'

tense fut

pred `eatsubj, obj'

def


obj
num sg

pred `falafel'

Figure 5. F-structure of The hamster will eat a falafel


TP
( subj) =

DP

the

NP

( def) = +

hamster

( pred) = `hamster'
( num) = sg

will

( tense) = fut

VP

( subj) =

DP

eat
`eat. . .'

( pred) =

( def) =
( num) = sg

NP
N

falafel

( pred) = `falafel'
( num) = sg

Figure 6. Annotated c-structure of The hamster will eat a falafel


is the simplicity with which it can account for passives in languages like English. Note that a
recurring element of LFG is that no morphological processes happen at the level of syntactic
derivations. Each word of the sentence comes with all of the features that need matching. So, an
analysis of passives in LFG consists, in the simplest case, of an additional lexical entry for the
passivized form of each verb. That is, for each verb, e.g. eat, there will be a separate lexical
entry for eaten. In particular, the argument structure of these entries will differ.
The argument structure of the active form will be eat<subj, obj> and that of the passive

7
form will be simply eat<, subj> Note that it is the positions in the angle-bracket specification
of a-structure that correspond with the idea of thematic roles. That is, in the passivized form
eaten, the second position still corresponds with the patient of the verb, however it is filled
with the subj feature, as opposed to the object feature. Below is a f-structure and an annotated
c-structure for the sentence The falafel was eaten.
TP
( subj) =

DP

def
+

subj num sg

pred `falafel'

tense pst
pred `eat, subj'

=
=

was

VP

the

( tense) = pst

NP

eaten
`eat. . .'

( def) = +

( pred) =

falafel

( pred) = `falafel'
( num) = sg

Figure 7. F- and c-structure for The falafel was eaten


CP
C
C

TP
T
T
[+T ense]

VP
V
Vpass
be + -en

VP
V
V
eat

DP
D
D
the

NP
N
N
falafel

Figure 8. Transformational treatment of The falafel was eaten

8
Contrast this treatment with the usual transformational treatment of passives (as seen in
Figure 8). In transformational theories, the passive is analyzed as a functional verb that removes
the case-assigning ability of the lexical verb. Without a case, the DP in the object position moves
to the subject of the clause to receive nominative case. Note that the LFG analysis does not
require movement or any kind of morphological change to the words involved.
7. Raising in LFG. Another facet of LFG is the way it expresses the equivalent of traces
in a transformational theory. Since, in a transformational theory, each moved constituent leaves
traces of itself that can be viewed as copies distributed throughout the structure. In LFG,
however, no movement occurs, and we encounter a problem when these traces are necessary to
fill required features.
Consider the sentence, John seems to enjoy falafels Below is the transformational
analysis of this sentence.
CP
C
C

TP
T
T
[+T ense]

VP
V
V
seems

CP
C
C

TP
T
VP

T
to

V

DP
D
D

NP
N
N
John

DP

V
enjoy

D
NP

N
N
falafels

Figure 9. Transformational analysis of John seems to enjoy falafels

9
From this perspective, the DP John is left without case inside its CP and so moves to the
outer clause and receives nominative case from the outermost TP.
From an LFG perspective, however, instead of movement and a trace that allows the
inner structure to share a subject with the outer structure, we simply represent that sharing as
a functional equation. Note that in LFG, instead of a tenseless TP being generated by to, we
instead represent the constituent to enjoy falafels as a VP-bar, a projection created for this
purpose. (Carnie 2007; 447)
S

subj

tense

pred

xcomp

def
+

num sg

pred `John'

pres

`seemxcomp' (subj)

subj []
pred `enjoysubj, obj'

def

obj num pl

pred `falafels'

DP

VP

NP

seems

VP
to

VP
V

N
John

DP

enjoy

D
NP
N
N

Figure 10. F- and c-structure of John seems to enjoy falafels

falafels

This analysis can be glossed as saying that relation between the two subj features is that
of identity--they have the same value. Note also that the a-structure makes a note of the extra
subj

feature but also notes that it is not selected (-marked) by the local pred.
Finally, the overall picture is that this analysis is essentially the same as the

transformational analysis. Both place the DP John in the outer structure and inner structure
simultaneously but do not identify it with a thematic role of the verb seem. The LFG analysis is
complicated somewhat with the open complement function (xcomp) and the VP-bar projection in
the c-structure, but LFG exposes the identity relationship between the two subj positions.

10
8. Conclusion. Lexical-functional grammar can be a compelling alternative to
transformational grammar. However, it does not have significantly greater expressive power.
Its main benefit comes in the form of its preservation of lexical forms. LFG was developed
with efficiency of recognition as a key goal (Darymple) and this continues to be a benefit to
the theory today. LFG is still an active research topic and also has an advantage in that it is
more straightforward to begin a semantic analysis of a sentence, since the f-structure is already
separated, and not mingled with movements and surface structure. Overall, however, both
theories have their merits.

11
References
Carnie, Andrew. 2007. Syntax: A Generative Introduction. 2nd. ed. Malden: Blackwell
Falk, Yehuda. 2001. Lexical-Functional Grammar: An Introduction to Parallel Constraint-Based
Syntax. Stanford: CSLI
Dalrymple, Mary. 2001. Syntax and Semantics: Lexical Functional Grammar. New York:
Academic.
Dalrymple, Mary. Lexical Functional Grammar MIT Encyclopedia of Cognitive Science.
<http://ai.ato.ms/MITECS/Entry/dalrymple.html>
Bresnan, Joan. 2001. Lexical-Functional Syntax. Malden: Blackwell.

You might also like