Processing Language with Logical Types and Active Constraints

Processing Language with Logical Types and Active Constraints Patrick SAINT-DIZIER IRIT Universit6 Paul Sabatier 118, route de Narbonne 31062 Toulouse cedex FRANCE e-mail: suiizier@ixit.irit.fr ABSTRACT In this document, we present a language which associates type construction principles to constraint logic programming. We show that it is very appropriate for language processing, providing more uniform, expressive and efficient tools and treatments. We introduce three kinds of constraints, that we exemplify by motivational examples. Finally, we give the procedural semantics of our language, combining type construction with SLD- resolution. Introduction With the development of highly parameterized syntactic theories like Government and Binding theory and Head-Driven phrase structure grammars and with the development of theories where rewriting and unification plays a central role, like Categorial grammars and Unification Grammars, there is an increasing need for more appropriate and more efficient feature systems. Feature systems must be designed to preserve the adequacy, the expressiveness and the explanatory power of the linguistic system that one wants to model. Real parsing as well as generation systems often require the manipulation of large sets of features, these systems must therefore offer a great flexibility in the specification of features in grammar symbols and a significant modularity so that each linguistic aspect (morphological, categorial .... ) can be dealt with independently. Features are often subject to various constraints. These constraints cannot always be evaluated at the level they are formulated (e.g. a feature value is not yet known) bu~t have to be evaluated later and must be true throughout the whole parsing or generation process. The development of principled-based approaches to language processing also require the definition of more abstract formal systems to handle in an adequate way these principles. Principles indeed often apply not at grammar rule level but they involve a large part of a parse tree. They must be expressed by a constraint system which is global to the whole grammar and not local to a rule, as for example, in DCGs. These brief considerations have motivated our approach: syntactic rules are viewed as type constructions on which constraints arc applied. These constraints are themselves part of the type. To give an appropriate expressive power to constraints and an efficient interpretation, they are interpreted within the Constraint Logic Programming framework. In the next sections, we introduce our description language based on types and constraints. We then give motivational examples which clarify its use. Then we give its procedural interpretation and a constraint resolution mechanism system. 1. A typed-based description language Three main types of operations are at the basis of the typed-based language we have designed for language processing, namely: - the expression of type construction to describe phrase structures, - the expression of relations (either local or long- distance) between types, - the expression of weU-formedness constraints on types. The term type refers here to structured data representation. They must not be confused with types defined in linguistics, as in the eategorial system. The starting point of our description language is CIL (Mukai 85), …

Content truncated to 500 words