Lexical Functional Grammar (LFG), developed by Joan Bresnan and Ronald Kaplan beginning in the late 1970s, is a constraint-based grammatical framework that represents syntactic information in two parallel structures: constituent structure (c-structure) and functional structure (f-structure). C-structure captures the surface phrase-structure configuration of a sentence, modeled as a tree generated by context-free rules annotated with functional equations. F-structure captures grammatical relations (subject, object, oblique) and feature information (tense, agreement, case), modeled as a set of attribute-value matrices. The mapping between c-structure and f-structure, mediated by a function φ, is the heart of the formalism.
C-Structure and F-Structure
C-structure trees are generated by annotated phrase-structure rules in which functional equations specify how the functional information from each daughter contributes to the f-structure of the mother. The up-arrow (↑) refers to the f-structure of the mother node, and the down-arrow (↓) refers to the f-structure of the current node. For example, the annotation (↑ SUBJ) = ↓ on an NP daughter of S states that the f-structure of the NP is the subject of the sentence.
(↑ SUBJ)=↓ ↑=↓
F-structure: [SUBJ [PRED 'dog', NUM SG, PERS 3]
PRED 'bark⟨SUBJ⟩'
TENSE PAST]
Functional Uniqueness and Completeness
F-structures must satisfy well-formedness conditions that enforce grammatical coherence. The Uniqueness Condition requires that each attribute have at most one value — preventing conflicting feature assignments. The Completeness Condition requires that all grammatical functions subcategorized by the predicate be present in the f-structure. The Coherence Condition requires that all governable grammatical functions in the f-structure be subcategorized by the predicate. Together, these conditions ensure that f-structures represent well-formed grammatical analyses.
The Xerox Linguistic Environment (XLE), developed by Ron Kaplan and colleagues at Xerox PARC, is the primary computational platform for LFG grammar development. XLE includes an efficient chart-based parser that handles large-scale LFG grammars, a stochastic disambiguation component for selecting the best parse from a packed representation, and a transfer-based machine translation component. The ParGram project uses XLE to develop parallel LFG grammars for over a dozen languages, enabling cross-linguistic computational research.
Cross-Linguistic Applications
One of LFG's greatest strengths is its cross-linguistic applicability. The separation of c-structure (which varies across languages due to differences in word order, configurationality, and agreement) from f-structure (which captures more universal grammatical relations) provides a principled framework for typological variation. Non-configurational languages like Warlpiri, where word order is free and discontinuous constituents are common, can be analyzed with flat c-structures while maintaining structured f-structures that capture the underlying grammatical relations.
LFG has been extended in several directions: LFG-DRT integrates discourse representation theory for semantic interpretation, Glue Semantics provides a logic-based approach to compositional semantics within LFG, and OT-LFG combines the constraint satisfaction architecture with Optimality Theory for modeling competition among candidate analyses. These extensions demonstrate the framework's flexibility and its capacity to integrate syntax with semantics, pragmatics, and phonology.