Lexical-functional grammar

The Lexical-Functional Grammar (LFG) is a unification grammar model . It was created in response to research in the field of transformational grammar and relates primarily to syntax , morphology and semantics , but not to phonology . More recently, however, ideas from phonological optimality theory have become popular in the LFG. The LFG was developed by Joan Bresnan and Ronald Kaplan in the 1970s . They wanted to create a grammar model that had enough depth for linguists and at the same time complied with the strict formalism of computational linguistics and could be efficiently processed by a parser .

The model

In contrast to Chomsky's syntax , which contains separate levels of linguistic representation connected by transformations , the LFG is based on two mutually restrictive structures:

• a constituent tree (C structure, English constituent )
• a feature structure (F-structure, English feature )

Constituents are sequences of words that are internally related. They are also known as phrases. A sentence S usually consists of several constituents (e.g. noun phrases NP, verb phrases VP, adjective phrases AdjP).

Many syntactic problems can be explained by the incomplete correspondence between these two structures. They have to be united to form grammatical sentences. In technical terms: The LFG contradicts the projection principle established in more recent work on transformation grammar . Accordingly, syntactic structures are direct representations of certain lexical information. However, the LFG offers more flexible relationships between the syntactic and semantic structure, thus eliminating the need for transformations.

C structure (constituents)

To understand the C structure, let's first consider the following simple formal grammar :

${\displaystyle {\begin{matrix}(1)&{\text{S }}&\rightarrow &{\text{NP}}\quad {\text{VP}}\\(2)&{\text{NP}}&\rightarrow &{\text{nomen}}\\(3)&{\text{NP}}&\rightarrow &{\text{artikel}}\quad {\text{nomen}}\\(4)&{\text{VP}}&\rightarrow &{\text{verb}}\quad {\text{NP}}\end{matrix}}}$

The terms noun, article and verb stand for terminal symbols in grammar. Terminal symbols are mostly capitalized and non-terminal symbols are capitalized. More specifically, nouns , articles and verb Präterminalsymbole. They are only the preliminary stage to the terminal symbol, they stand for a class of terminal symbols. The actual terminal symbols, the words, are not given in the grammar, but are in a lexicon. If a word occurs in two classes, it appears twice in the lexicon.

With this grammar the sentence The children eat porridge can be developed:

${\displaystyle \left.{\begin{aligned}&\left.{\begin{aligned}&{\text{Die}}\quad &{\text{artikel}}\quad \\&{\text{Kinder}}\quad &{\text{nomen}}\quad \end{aligned}}\right\rangle ^{(3)}\quad {\text{NP}}\\&\left.{\begin{aligned}&{\text{essen}}&\ &{\text{verb}}\\&{\text{Brei}}&\left.{\text{nomen}}\right\rangle ^{(2)}&{\text{NP}}\end{aligned}}\right\rangle ^{(3)}{\text{VP}}\end{aligned}}\right\rangle ^{(1)}{\text{S}}}$

This grammar also allows sentences like The child eat porridge or cars eat streets . Additional properties such as case , gender , tense , number and specification must be taken into account.

F structure (features)

Since the C structure alone is not sufficient to generate correct sentences, the LFG also has the F structure. First we create a lexicon with the additional properties. Properties that are not specified have not yet been determined.

word	Preterminal symbol	F structure
Porridge	noun	case accusative genus Masculine number Singular
The	items	case Nominative genus neuter number Plural
eat	verb	Tense Present subject case Nominative number Plural object case accusative number Singular or plural
child	noun	case Nominative genus neuter number Singular
children	noun	case Nominative genus neuter number Plural

The grammar given above is now expanded to an LFG:

${\displaystyle {\begin{matrix}(1)&{\text{S }}&\rightarrow &{\text{NP}}:(\uparrow {\text{Subjekt}})=\downarrow \quad {\text{VP}}:\uparrow =\downarrow \\(2)&{\text{NP}}&\rightarrow &{\text{nomen}}\\(3)&{\text{NP}}&\rightarrow &{\text{artikel}}\quad {\text{nomen}}\\(4)&{\text{VP}}&\rightarrow &{\text{verb}}\quad {\text{NP}}:(\uparrow {\text{Objekt}})=\downarrow \end{matrix}}}$

The equations indicate that unification (assignment) must take place. With the F-structure of the higher-level node of the C-structure is unified with the F-structure of the subordinate node. The equation unifies the subject node of the F-structure of the higher-level node of the C-structure with the F-structure of the subordinate node. ${\displaystyle \uparrow =\downarrow }$ ${\displaystyle (\uparrow {\text{Subjekt}})=\downarrow }$

We look again at the sentence Children eat porridge and the development given above. However, let's look at the process from the perspective of a parser:

The children create an NP node. The and children are an NP node in the C structure. Its F-structure is created by unifying the F-structures of Die and Kinder . The structures are notated in so-called attribute-value matrices (AWMs) :

${\displaystyle {\begin{bmatrix}{\text{Kasus}}&{\text{Nominativ}}\\{\text{Genus}}&{\text{Neutrum}}\\{\text{Numerus}}&{\text{Plural}}\end{bmatrix}}}$

The equation unifies this F-structure with subject from the F-structure of S: ${\displaystyle (\uparrow {\text{Subjekt}})=\downarrow }$

${\displaystyle {\begin{bmatrix}{\text{Subjekt}}&{\begin{bmatrix}{\text{Kasus}}&{\text{Nominativ}}\\{\text{Genus}}&{\text{Neutrum}}\\{\text{Numerus}}&{\text{Plural}}\end{bmatrix}}\end{bmatrix}}}$

This is then unified with the F-structure of the VP node:

${\displaystyle {\begin{bmatrix}{\text{Tempus}}&{\text{Präsens}}\\{\text{Subjekt}}&{\begin{bmatrix}{\text{Kasus}}&{\text{Nominativ}}\\{\text{Genus}}&{\text{Neutrum}}\\{\text{Numerus}}&{\text{Plural}}\end{bmatrix}}\\{\text{Objekt}}&{\begin{bmatrix}{\text{Kasus}}&{\text{Nominativ}}\\{\text{Genus}}&{\text{Akkusativ}}\\{\text{Numerus}}&{\text{Plural}}\end{bmatrix}}\end{bmatrix}}}$

The child eat porridge is not valid for the given LFG, since the F-structure of the NP cannot be unified with the F-structure from essen :

${\displaystyle {\begin{bmatrix}{\text{Tempus}}&{\text{Präsens}}\\{\text{Subjekt}}&{\begin{bmatrix}{\text{Kasus}}&{\text{Nominativ}}\\{\text{Genus}}&{\text{Neutrum}}\\{\text{Numerus}}&{\text{Plural}}\not ={\text{Singular}}\end{bmatrix}}\\{\text{Objekt}}&{\begin{bmatrix}{\text{Kasus}}&{\text{Nominativ}}\\{\text{Genus}}&{\text{Akkusativ}}\\{\text{Numerus}}&{\text{Plural}}\end{bmatrix}}\end{bmatrix}}}$

outlook

In fact, much more data is stored in the lexicon. In the case of children, for example, the specification is saved in an animated manner. The underlying lexemes are also saved. The lexicon does not have to be listed in full. Since words can be broken down into morphemes , it is often sufficient to save basic forms.

literature

• Joan Bresnan , Ash Asudeh, Ida Toivonen, Stephen Wechsler: Lexical-Functional Syntax. 2nd edition, Wiley-Blackwell, Oxford, 2016, ISBN 978-1-4051-8781-7 .

Web links

• Brief description
• What is LFG? (English)