Analyzing the CoNLL--X Shared Task from a Sentence Accuracy Perspective

Miguel Ballesteros, Jesús Herrera, Virginia Francisco, Pablo Gervás


Nowadays, because of the relevance of the CoNLL shared tasks on Dependency Parsing, the most used evaluation measures are the ones computed in them. These measures, which are token--based, are computed globally for a whole big set of texts considering token by token. But a final user of a dependency parser would expect a high and stable accuracy for every parsed piece of text (usually one sentence). In this cases sentence--based measures add some information that could be relevant. This is why we developed the present study, which is addressed to get a richer description of the performance of dependency parsers.

Texto completo: