Section: New Results

Techniques for transferring lexical resources from one language to a closely-related one

Participants : Yves Scherrer, Benoît Sagot.

Developing lexical resources is a costly activity, which means that large resources only exist for a small number of languages. In our work, we address this issue by transferring linguistic annotations from a language with large resources to a closely related language which lacks such resources. This research activity, funded by the Labex EFL, has started in October 2012.

First results include the development of a method to create bilingual dictionaries without any parallel data, depending solely on surface form similarities and their regularities. The resulting bilingual dictionaries are used to transfer part-of-speech annotations from one language to the other. At the moment, our methods are being tested with Wikipedia texts from various languages and dialects closely related to German, such as Dutch and Pfälzisch. We plan to extend this work to data from other language groups and to other types of linguistic annotations, for instance syntactic or semantic resources.