A multi-level methodology for the automated translation of a coreference resolution dataset: an application to the Italian language

Abstract In the last decade, demand for readily accessible corpora has touched all areas of natural language processing, including coreference resolution. However, it is one least considered sub-fields in recent developments. Moreover, almost existing resources are only available English language. To overcome this lack, work proposes a methodology to create corpus resolution Italian exploiting knowledge annotated other languages. Starting from OntonNotes, translates and refines utterances obtain respecting grammar, dealing with language-specific phenomena preserving mentions. A quantitative qualitative evaluation performed assess well-formedness generated utterances, considering readability, grammaticality, acceptability indexes. The results have confirmed effectiveness generating good dataset starting an one. goodness also assessed by training model based on BERT model, achieving promising results. Even if been tailored languages, general basis easily extendable adapting small number language-dependent rules generalize most linguistic under examination.