Extraire et encoder l’information lexicale de Wiktionary : quel boulot pour étrangler le goulot !

Abstract

We present in this article an effort carried out for a decade which consists in using the content of the Wiktionary collaborative dictionary in order to build free lexical resources. Its main result is the design of machine-readable dictionaries and inflectional lexicons for three languages (French, Italian and English). In this paper, we question the usefulness of such lexical resources at a time when mainstream NLP is based on machine learning and readily do without. We compare different methods of producing resources and more specifically of extracting information from Wiktionary. We then discuss the suitability of standard formats for encoding idiosyncratic resources such as Wiktionary and conclude on the need to prioritize, above all, the production and sharing of resources.