Guy Perrier

Interaction Grammars

Interaction Grammars (IG) are a new linguistic formalism which is based on descriptions of underspecified trees in the framework of intuitionistic linear logic (ILL). Syntactic composition, which is expressed by deduction in linear logic, is controlled by a system of polarized features. In this way, parsing amounts to generating models of trees descriptions and it is implemented as a constraint satisfaction problem.

Polarization and abstraction of grammatical formalisms as methods for lexical disambiguation

In the context of lexicalized grammars, we propose general methods for lexical disambiguation based on polarization and abstraction of grammatical formalisms. Polarization makes their resource sensitivity explicit and abstraction aims at keeping essentially the mechanism of neutralization between polarities. Parsing with the simplified grammar in the abstract formalism can be used efficiently for filtering lexical selections.

Lexical disambiguation with polarities and automata

We propose a method for lexical disambiguation based on polarities for Interaction Grammars (IGs), well suited for coordination.

Intuitionistic Multiplicative Proof Nets as Models of Directed Acyclic Graph Descriptions

Given an intuitionistic proof net of linear logic, we abstract an order between its atomic formulas. From this order, we represent intuitionistic multiplicative proof nets in the more compact form of models of directed acyclic graph descriptions. If we restrict the logical framework to the implicative fragment of intuitionistic linear logic, we show that proof nets reduce to models of tree descriptions.

Dependency Parsing with Graph Rewriting

We propose to use Graph Rewriting for parsing syntactic dependencies. We present a system of rewriting rules dedicated to French and we evaluate it by parsing the SEQUOIA corpus.

Application of Graph Rewriting to Natural Language Processing

The paradigm of Graph Rewriting is used very little in the field of Natural Language Processing. But graphs are a natural way of representing the deep syntax and the semantics of natural languages. Deep syntax is an abstraction of syntactic dependencies towards semantics in the form of graphs and there is a compact way of representing the semantics in an underspecified logical framework also with graphs. Then, Graph Rewriting reconciles efficiency with linguistic readability for producing representations at some linguistic level by transformation of a neighbor level: from raw text to surface syntax, from surface syntax to deep syntax, from deep syntax to underspecified logical semantics and conversely.