Étiquetage d'un corpus hétérogène de français médiéval : enjeux et modalités
Résumé
We have undertaken a morpho-syntactic tagging of the 2.5 millions words of our corpora of medieval texts. The external and internal heterogeneity of the texts make this task a difficult one. As a result, we had to resort to a double strategy. Since there is actually no tool adapted to our corpora, we had first to rely on a programmable tagger in order to categorize a first text. As a second step, and building on the results obtained with the first text, we produced a tagger based on contextal rule learning. Using this latter tool we subsequently tagged a second, quite "similar" (in terms of external criteria) text. This two-step process was then used once again to tag additional texts.
The next phase will be to evaluate the heterogeneity of texts according to internal criteria. The correlation of internal and external heterogeneity will enable us to elaborate a "fine-grained" typology of texts.
The next phase will be to evaluate the heterogeneity of texts according to internal criteria. The correlation of internal and external heterogeneity will enable us to elaborate a "fine-grained" typology of texts.
Domaines
Linguistique
Loading...