Parsing with Intraclausal Coordination and Clause Detection

Syntactic analysis, i.e., parsing of text is used during various tasks, e.g., machine translation, question answering, etc. The structure of a sentence is represented with a tree. Parsing long sentences is a difficult task. The motivation was to analyze sub-units of the sentence independently, which could improve the overall parsing accuracy. We developed a new parsing algorithm that includes intraclausal coordination and clause detection. Parsing using clause detection was first tried by Abney (1), whose algorithm delimits non-embedded clauses before the complete parse is made. In (2), there is a short description of a rule-based parser where clause identification is included in the parsing process. A detailed description of our new algorithm can be found in (3). To our knowledge, the algorithm is the first to use intraclausal coordination detection in cojunction with clause detection before parsing. The most important contribution is the decrease in the number of parsing errors by 7.1% and 6.4% for Slovene, compared to the Malt (4) and MSTP (5) baseline parsers, respectively.