constraint (n.)

A term used in LINGUISTICS, and especially in GENERATIVE GRAMMAR, to refer to a CONDITION which restricts the application of a RULE, to ensure that the sentences generated are WELL FORMED. For example, in generative PHONOLOGY, a distinction can be made between ‘simultaneous’ and ‘sequential’ constraints: the former states the restrictions on the simultaneous occurrence of FEATURES, e.g. a SEGMENT cannot be at once [+high] and [+low]; the latter states the restrictions on sequences of features, e.g. whether a language permits CONSONANT CLUSTERS. In generative SYNTAX there are also several constraints which have to be imposed in order to prevent the DERIVATION of ILL-FORMED PHRASE-MARKERS, e.g. constraints on the ORDERING of rules. For example, ‘surface structure’ constraints (FILTERS, or ‘OUTPUT conditions’) refer to conditions where a characteristic of SURFACE STRUCTURE decides which phrase-markers are well formed; e.g. no phrase-marker containing an internal boundary symbol can qualify as a well-formed surface structure. Other examples include ISLAND constraints and the CO-ORDINATE structure constraint.

Later generative studies aimed to find constraints which apply to large classes of derivations (i.e. the constraints have a greater EXPLANATORY power) – a trend which contrasts with the local application of the constraints proposed in the 1960s. ‘Constraints’, in this work, are distinguished from ‘filters’: the former are conditions affecting two successive phrase-markers in a derivation: the latter are conditions on a single level of structure, which serves as the output of a given set of rules.

The notion of constraints takes a different direction in OPTIMALITY THEORY, where it is the principal EXPLANATORY device (abbreviated as CON). Here, constraints are ways of characterizing language UNIVERSALS. Each language has its own RANKING of constraints (e.g. which determine MORPHEME position or SYLLABLE STRUCTURE), and differences between these rankings result in the variations observed between languages. Constraints are found only in the constraint hierarchy for a language, i.e. the language’s particular ranking of the universal set of constraints; there are no separate constraints operating on INPUTS or OUTPUTS, and no RULES to be constrained. The approach also uses the notion of constraint satisfaction, representing the extent to which a constraint can be violated in grammatical forms. All constraints are violable. The output forms are the OPTIMAL ones (i.e. with the minimum number of constraint VIOLATIONS), selected by the EVALUATOR component of the theory. The term is also used more generally in generative linguistics with reference to theory construction. A linguistic theory needs to be constrained, in order to restrict the class of potential grammars. In this sense, the main aim of linguistics is said to be the provision of an explanatorily ADEQUATE theory which is maximally constrained.