Logique, linguistique et informatique
Responsable : C. Rétoré.
This group studies the formalisation of natural language, in particular its syntax and its semantics. The theme fits well within the formal methods team, since the methods involved encompass formal language theory, formal grammars and logic.
Since the beginnings of computer science linguistics has always been at the same time a favoured research field (natural languages and reasoning), a privileged application area (automatic translation) and a source of key concepts (formal language theory). More recently, logic has been recognized as a federating tool for natural language processing with its ability to account for both semantic and syntactic phenomena of natural language. We aim at developing computer models and algorithms for different description levels of natural language study and processing, with a special focus on the syntax-semantics interface.
More precisely, we aim at modelling by means of formal methods such as formal language theory and formal grammars, categorial grammars and resource logic, lambda calculus and higher order logic. The aspects of natural language that we study include the structure of words (morphology) and of sentences both from a syntactic and a semantic or logical viewpoint (this later example can be extended to discourse structure). Among our goals are a closer study of the relationship between formal grammars, a tighter integration of morphology and syntax, the automatic computation of semantic representations from syntactic analyses (from categorial, minimalist, dependency grammars), anaphora resolution, scope ambiguities, and, more generally, the study of the relationship between the predicative structure of a sentence and the logic structure of discourse. Our study can be parted into three layers:
- Syntax is described by formal grammars (chains or trees, context-free of mildly context sensitive, with or without unification). Nevertheless in order to automatically obtain semantic representation that can be processed by a machine, one can use deductive system in resource logics, as categorial grammars do. Syntax also has to take into account word structure, usually described by automata and transducers.
- Logical semantics, which yield the predicative structure of a sentence, makes use of higher order logic, encoded in typed lambda calculus as Church and Montague did, implementing the Fregean principle of compositionality: meaning of whole is function of meaning of its parts and of the way they are syntactically composed. These techniques can be extended from sentences to discourse using lambda-DRT.
- Following Pustejovsky, a way to formalise the relationship between word meanings and other word meanings or the real world are feature structures: they describe the various role a word may play, and Montagovian lambda-terms specify which roles are actually used in a sentence.
The theme focuses on the interaction between these three layers, in particular the automation of the transition between a sentence and its meaning, as well as the mathematical study of the formalisms and transformations.