Joint Natural Language Processing with Global Information

点击翻译按钮获取中文摘要

We investigated the usage of global information for joint processing of natural language processing such as word segmentation, part-of-speech tagging, parsing and predicate argument structure analysis.For English, we collected multi-word expressions that act as functional words and implemented part-of-speech tagging and syntactic analysis using them. We also proposed various syntactic processing methods, one that utilizes both top-down and bottom-up information jointly and another that keeps multiple candidates in the form of parse forest and uses re-ranking algorithm to select the best answer.For predicate argument structure analysis, we proposed a method that utilizes various types of features in a joint framework and another method that used Markov Logic Networks to incorporate various constraints jointly.