Lost in Grammar Translation Lost in Grammar Translation

1http://www.di.unito.it/∼tutreeb/ Italian Treebank2 (VIT), and the ISST3. None of them is comparable in size with the English Penn Treebank. This limits the possibility to have reliable induced grammars for Italian. Initial studies have shown that probabilistic grammars induced on a small corpus have not impressive performances [5]. Building larger corpora is then needed. We have been working on defining general translators that can transform more expressive grammatical annotations in less expressive ones [1]. These translators can be used to merge corpora with different annotation schemes. Such bigger corpus is better suited for learning reliable grammars. The second problem instead is the one that we had to face in the Evalita comparative study. We wanted to assess the performances of a pre-existing parser, CHAOS4 [2], against an annotated corpus based on a completely different grammar, the TUT. We then translated the grammatical interpretation produced by CHAOS in the target grammatical representation. In this paper we describe the parser we used and how we translated its syntactic interpretation for the purpose of the evaluation (Sec. 2). We analyze the results (Sec. 3). Finally, we draw some conclusions (Sec. 4).