Routledge Dictionary of Language and Linguistics
1 In its broadest sense, an umbrella term for all generative grammar models, especially those generative grammars that use a unification operation in their rule systems.
2 In a narrower sense, a member of a family of newer grammatical models in which feature unification is used (usually in conjunction with other feature operations) to capture the information flow in derivation. Various particular approaches belong to this group: grammatical models like Generalized Phrase Structure Grammar (GPSG) and Lexical-Functional Grammar (LFG), grammatical formalisms capable of producing expressions for implementation on the computer, like Functional Unification Grammar (FUG) and PATR-II; as well as a series of newer models that present forms mixed from existing approaches and theories like Head-driven Phrase Structure Grammar (HPSG) and Categorial Unification Grammar (CUG). Since all these models were developed at Stanford University and at neighboring institutions in the San Francisco Bay Area, they are known as Bay Area Grammars. Other terminology includes Unification-based Grammars, Constraint-based Grammars, and Information-based Grammars. Unification grammar is based on the further development of linguistic features. Every linguistic unit (word or phrase) is characterized by a feature structure, that is, by a number of attribute-value pairs, whose values can be either atomic symbols or feature structures. Attributive values within a feature structure can be coreferential (also co-indexed), that is, they can describe the same linguistic unit. Feature structures for syntactic units are often termed ‘complex categories.’ They are usually represented as feature matrices (Figure 1) or feature graphics (Figure 2). In the following simplified feature structure of a verb, the coreference of the [AGR] features induces the agreement between the verb and the subject.
In a unification grammar, phrase structure rules indicate which parts of the feature structure of a syntactic unit are coreferent with which parts of the feature structure of their immediate constituents and which are co-referent (
coreferentiality) with the feature structure of the immediately dominating constituent. These coreferences between the descriptions of the constituents in a syntactic tree take care of the information flow in syntactic derivation and are used to represent dependencies between constituents (agreement, government, control, and non-local dependencies). Coreference of two feature structures means that their contents are ‘unified.” If the contents do not contradict each other (i.e. assign incompatible values to at least one feature), the result is unification by the addition of the information in the two unified structures. In the case of a contradiction, the unification does not succeed, and a special category is generated which signals the inconsistency. The unification is usually expressed by brackets, which include the feature structures to be unified. Equivalent notations for Figures 1 and 2 are:
A unification grammar was first suggested by Kay (1979). Independent representational formalisms with unification structures were developed in related work in the area of knowledge representation in artificial intelligence (AitKaci 1984; Smolka and Ait-Kaci 1987). The semantics of unification formalisms was developed by Kaspar and Rounds (1986), Johnson (1988), and Smolka (1988). The result of this work is a feature logic with a bundle theory semantics. A special property of unification grammar is its declarativity. This results from the monotonicity of the unification operation. The order of the steps applied is unimportant for the result of a derivation. In this respect, unification grammar is particularly suited to computational linguistics, since the grammar allows for multiple strategies. It is also not bound to a particular direction of processing; so the same grammar can be used for parsing and generation. Models of unification grammar are differentiated by the role which the phrase structure plays in the syntactic description. In most models, a context-free phrase structure tree is constructed by syntactic rules. The feature structures are associated with the phrase structure nodes and bound together by co-references. In other models (like FUG or HPSG), the phrase structure itself is represented inside the feature structure, so the feature structure is adequate for description. The models also differ in the extensions they use. Frequently used extensions of the grammatical formalisms are generalization or disjunction, templates (feature macros, type-names), functional uncertainty and value bundle features. Significant differences are also found in the expansion of the types of description on the grammatical level: while, for example, GPSG describes only syntactic conformities with the help of feature structures, the feature-based descriptions of HPSG also extend to semantics and phonology. While there are only a few investigations in phonology and phonetics, in semantics there are several attempts to integrate situation semantics and discourse representation theory into models of unification grammar (e.g. Johnson and Klein 1986; Fenstad et al. 1987; Pollard and Sag 1988). In addition to the models of Bay Area Grammar, in their broadest sense later developments like Tree Unification Grammar (TUG) are also unification grammars (Popowich 1989). It is also necessary to include the logical grammars from the tradition of logic programming, in which the feature structures are represented by logic terms and term unification plays the role of feature unification. Theoretically, every formal generative grammar model could probably be encoded as a unification grammar. Thus there are already suggestions that existing grammatical models like dependency grammar and Tree-Adjoining Grammar be supplemented by using the tools of unification grammar (Hellwig 1986; Vijay-Shanker and Joshi 1988).
References
Ait-Kaci, H. 1984. A new model of computation based on a calculus of type subsumption. Philadelphia, PA.
Bresnan, J. (ed.) 1982. The mental representation of grammatical relations. Cambridge, MA.
Calder, J., E.Klein, and H.Zeevat. 1988. Unification Categorial Grammar: a concise, extendible grammar for natural language processing. In COLING 88. Budapest. Vol. 1.83–6.
Fenstad, J.E., P.K.Halvorsen. T.Langholm. and J. van Benthem. 1987. Situations, language and logic. Dordrecht.
Gazdar, G. et al. 1985. Generalized Phrase Structure Grammar. Cambridge, MA.
Hellwig, P. 1986. Dependency Unification Grammar. In COLING 86. Bonn. 195–8.
Johnson, M. 1988.
Attribute—value logic and the theory of grammar. Stanford, CA.
Johnson, M. and E.Klein. 1986. Discourse, anaphora and parsing. In COLING 86. Bonn. 669–75.
Kasper, R. and W.Rounds. 1986. A logical semantics for feature structures. In ACL proceedings 24. New York. 257–66.
Kay, M. 1979. Functional Grammar. PBLS 5.142–58.
——1984. Functional Unification Grammar. In COLING 84. Stanford. CA. 75–8.
——1985. Parsing in Functional Unification Grammar. In D.Dowty, L.Karttunen. and A. Zwicky (eds), Natural language parsing. Cambridge. 251–78.
Pollard, C. and I.A.Sag. 1988. An information-based syntax and semantics, vol. 1: Fundamentals. (CSLI Lecture Notes 13.) Stanford, CA.
Popowich, F. 1989. Tree Unification Grammar. In ACL proceedings 27. Vancouver. 228–36.
Shieber, S.M. et al. 1983. The formalism and imple-mentation of PATR-II. In Research on interactive acquisition and use of knowledge. (SRI international.) Menlo Park, CA. 39–79.
——1986. An introduction to unification-based approaches to grammar. (CSLI Lecture Notes 4.) Stanford, CA.
Shieber, S.M., L.Karttunen, and F.Pereira (eds) 1984. Notes from the unification underground. (SRI Technical Note 327.) Menlo Park, CA.
Smolka, G. 1988. A feature logic with subsorts. In LILOG Report 33. Stuttgart.
Smolka, G. and H.Ait-Kaci. 1987. Inheritance hierarchies: semantics and unification. Journal of Symbolic Computation 7.343–70.
Uszkoreit, H. 1986. Categorial Unification Grammars. In COLING 86. Bonn. 187–94.
——1988. From feature bundles to abstract data types: new directions in the representation and processing of linguistic knowledge. In A.Blaser (ed.), Natural language at the computer. Berlin. 31–64.
Vijay-Shanker, K. and A.K.Joshi. 1988. Feature structure based tree adjoining grammars. In COLING 88. Budapest. Vol. 2, 714–19.
This is the complete article, containing 1,219 words
(approx. 4 pages at 300 words per page).
View More Summaries on Regulus Grammar Compiler
