featurestructure

Feature structure is a formal representation used in linguistics and natural language processing to encode linguistic information as a set of attribute-value pairs. In typed feature structures (TFS), each node has a type from a type system and a set of features, which can be constants, variables, or nested feature structures. The fundamental operation is unification, which combines two feature structures into a most-general common instance if they are compatible, or fails otherwise. Unification allows the sharing of structure and property across different constituents in a grammar.

Feature structures are central to unification-based grammars such as HPSG and Lexical-Functional Grammar. In HPSG, features capture syntax, morphology, and semantics within a single attribute-value matrix, enabling long-distance dependencies to be handled by unification rather than movement. In LFG, there is a distinction between constituent structure (c-structure) and functional structure (f-structure); feature structures encode the latter.

A typical feature structure consists of a root type with features such as CAT (category), NUM (number), PERSON, TENSE, SUBCAT (subcategorization), AGR (agreement), and VALENCES. Features may be mandatory or defaulted by a type hierarchy. Variables in feature structures allow the grammar to enforce locality constraints and cross-branch sharing.

Applications include parsing, grammar development, and formal semantics, where feature structures are used to propagate information about grammatical relations, agreement, and selectional requirements. Although powerful, the approach presupposes a well-designed type system and efficient unification algorithms to be practical in real-time processing.