Enhancing tagging performance by combining knowledge sources

The topic of this paper is an ongoing effort to exploit combinations of existing natural language processing (NLP) resources in order to reach part-of-speech (POS) tagging performance in excess of that which any single resource is able to provide. The context of the effort is the ETAP project, a parallel translation corpus project funded by the Bank of Sweden Tercentenary Foundation. The aim of the project is to create an annotated and aligned multilingual translation corpus which will be used as the basis for the development of methods and tools for the automatic extraction of translation equivalents for applications such as machine translation systems. To this end, we are investigating to which extent it is possible to reuse existing – meaning either developed in our department in some other context, or freely available on the WWW – NLP resources for the task of tagging the languages of the project. As a general rule, we may say that the amount of such resources is growing quite fast at the present time. On the other hand, their availability is highly dependent on the language, from almost unlimited numbers for English,