Mapping Semantic Information from FrameNet onto VALLEX1

1. Information on syntactic and semantic properties of verbs, which are traditionally considered as a center of sentence, plays a key role in many rule-based NLP tasks as information retrieval, text summarizing, question answering, machine translation, etc. Lexical resources providing such information are designed, however, within different theoretical frameworks whose theoretical assumptions are reflected in annotation schemes. As a result, there are great differences between individual lexical resources. However, the different theoretical background taken in individual lexical resources has another consequence: each lexical resource captures different types of information. Use of information from several lexical resources then represents an effective way of enriching a particular lexical resource. On the other hand, differences in theoretical assumptions taken in lexical resources bring several difficulties with mapping information: the different level of granularity in word sense disambiguation represents a typical example. Moreover, other requirements for harmonizing linguistic information are imposed on interlinking information from different-lingual lexical resources: a fundamental prerequisite for successful mapping represents especially an accurate translation