Mining Parsing Results for Lexical Corrections
Résumé
Successful parsing depends on the quality of the underlying grammar but also on the correctness of the lexicon that feeds the parser. The development of a lexicon both complete and accurate is an intricate and demanding task. A first step towards the improvement of a lexicon consists in identifying potentially erroneous lexical entries, for instance by using error mining techniques on large corpora (Sagot and de La Clergerie, ACL/COLING 2006) This paper explores the next logical step, namely the suggestion of corrections for those entries. This is achieved by running new analysis on the sentences rejected at the previous step, after having modified the information carried by the identified lexical entries. Afterwards, a statistical computation on the parsing results exhibits the most relevant corrections.
Domaines
Informatique et langage [cs.CL]
Origine : Fichiers produits par l'(les) auteur(s)