Détection et correction automatique d'erreurs d'annotation morpho-syntaxique du French TreeBank (Detecting and Correcting POS Annotation in the French TreeBank) [in French]

Detecting and correcting POS annotation in the French TreeBank The quality of the Part-Of-Speech (POS) annotation in a corpus has a large impact on training and evaluating POS taggers. In this paper, we present a series of experiments that we have conducted on automatically detecting and correcting annotation errors in the French TreeBank. Two methods are used. The first simply relies on identifying tokens with missing tags and correct them by assigning the tag the same token observed in the corpus. The second method uses n-gram variations to detect and correct conflicting annotations. The evaluation of the automatic correction is performed extrinsically by comparing the performance of different POS taggers in relation to the level of correction. Results show a statistically significant improvement in precision and indicate that the POS annotation quality can be noticeably enhanced by using automatic correction methods. MOTS-CLÉS : Étiquetage morpho-syntaxique, correction automatique, qualité d’annotation.