Multiword Expressions in Syntactic and Semantic Parsing Environments

点击翻译按钮获取中文摘要

Multiword expressions (MWEs) are combinations of multiple lexemes whose overall properties are not readily predictable by those of their components, which makes them a challenge for natural language processing (NLP). Verbal MWEs (VMWEs) stand out as a particularly difficult and interesting subclass because of properties like discontinuity, overlap, varying word order, idiomaticity, and syntactic or semantic ambiguity. The project contributes to treating VMWEs in NLP settings by a) improving MWE identification approaches, b) integrating MWE identification with syntactic parsing, and c) extending the approach to an MWE-aware semantic parser. Concerning b) and c), the project will use architectures based on Tree Adjoining Grammars, Combinatory Categorial Grammar and Discourse Representation Theory, in combination with an extension of the Parallel Meaning Bank (PMB) with semantic annotations of MWEs.