Recognizing Arguing Subjectivity and Argument Tags

摘要

In this paper we investigate two distinct tasks. The first task involves detecting arguing subjectivity, a type of linguistic subjectivity on which relatively little work has yet to be done. The second task involves labeling instances of arguing subjectivity with argument tags reflecting the conceptual argument being made. We refer to these two tasks collectively as "recognizing arguments". We develop a new annotation scheme and assemble a new annotated corpus to support our learning efforts. Through our machine learning experiments, we investigate the utility of a sentiment lexicon, discourse parser, and semantic similarity measures with respect to recognizing arguments. By incorporating information gained from these resources, we outperform a unigram baseline by a significant margin. In addition, we explore a two-phase approach to recognizing arguments, with promising results.