Understanding Editorial Text: A Computer Model of Argument Comprehension
by Michael G. Dyer Natural language processing (NLP) is an area of research within Artificial Intelligence (AI) concerned with the comprehension and generation of natural language text. Comprehension involves the dynamic construction of conceptual representations, linked by causal relationships and organized/indexed for subsequent retrieval. Once these conceptual representations have been created, comprehension can be tested by means of such tasks as paraphrasing, question answering, and summarization. Higher-level cognitive tasks are also modeled within the NLP paradigm and include: translation, acquisition of word meanings and concepts through reading, analysis of goals and plans in multi-agent environments (e. g., coalition and counterplanning behavior by narrative characters), invention of novel stories, recognition of abstract themes (such as irony and hypocrisy), extraction of the moral or point of a story, and justification/refutation of beliefs through argumentation. The robustness of conceptually-based text comprehension systems is directly related to the nature and scope of the knowledge constructs applied during conceptual analysis of the text. Until recently, conceptually-based natural language systems were developed for, and applied to, the task of narrative comprehension (Dyer, 1983a; Schank and Abelson, 1977; Wilensky, 1983). These systems worked by recognizing the goals and plans of narrative characters, and. using this knowledge to build a conceptual representation of the narrative, xx UNDERSTANDING EDITORIAL TEXT including actions and intentions which must be inferred to complete the representation. A large portion of text appearing in newspapers and magazines, however, is editorial in nature.