Generalized phrase structure grammar

Generalized phrase structure grammar (GPSG) is a framework for describing the syntax and semantics of natural languages. It is a type of constraint-based phrase structure grammar. Constraint based grammars are based around defining certain syntactic processes as ungrammatical for a given language and assuming everything not thus dismissed is grammatical within that language. Phrase structure grammars base their framework on constituency relationships, seeing the words in a sentence as ranked, with some words dominating the others. For example, in the sentence "The dog runs", "runs" is seen as dominating "dog" since it is the main focus of the sentence. This view stands in contrast to dependency grammars, which base their assumed structure on the relationship between a single word in a sentence (the sentence head) and its dependents. GPSG was initially developed in the late 1970s by Gerald Gazdar. Other contributors include Ewan Klein, Ivan Sag, and Geoffrey Pullum. Their book Generalized Phrase Structure Grammar, published in 1985, is the main monograph on GPSG, especially as it applies to English syntax. GPSG was in part a reaction against transformational theories of syntax. In fact, the notational extensions to context-free grammars (CFGs) developed in GPSG are claimed to make transformations redundant. One of the chief goals of GPSG is to show that the syntax of natural languages can be described by CFGs (written as ID/LP grammars), with some suitable conventions intended to make writing such grammars easier for syntacticians. Among these conventions are a sophisticated feature structure system and so-called "meta-rules", which are rules generating the productions of a context-free grammar. GPSG further augments syntactic descriptions with semantic annotations that can be used to compute the compositional meaning of a sentence from its syntactic derivation tree. However, it has been argued (for example by Robert Berwick) that these extensions require parsing algorithms of a higher order of computational complexity than those used for basic CFGs.