Using feature structures as a unifying representation format for corpora exploration

In this paper we report on the use of feature structures to represent the linguistic information of a corpus. This approach has been adopted in TyPTex, a project which aims at providing a generic architecture for corpora profiling. After a brief overview of the Typtex project, we show that corpora exploration requires manipulating linguistic features in order to obtain a required level of linguistic information or changing the set of features to get a new point of view on the data. We show that feature structures formalism can help the building and management of linguistic features with Meta-Rules based on unification. Finally, we provide an example of marking which uses a mixed approach between projection of information from a static lexicon and contextual marking via Meta-Rules. Results tend to show that the use of feature structures can improve the coverage and reliability of the marking.