Séminaire de Recherche en Linguistique

 

In this talk I will report ongoing work on phrase structure parsing made at Alpage (Université Paris Diderot). The talk will provide several simple, yet accurate and efficient, algorithms suitable for multilingual phrase structure parsing. In particular I will show that given current data sets such as the SPMRL dataset and multi-view treebanks, lexicalized phrase structure parsing can be made multilingual without significant efforts and achieves state of the art results in a large range of typologically varied languages with a significant gain in speed over latent variable models à la (Petrov 2007). 

 

The second part of the talk outlines some of the issues faced by multilingual parsing (feature selection and sparse data issues) and we will provide a view of some of our research lines in this direction (deep neural architectures for lexicalized parsing).