Constraint Solving for Language Processing and Bioinformatics

点击翻译按钮获取中文摘要

In this proposal I address the problem of building executable theories of language intelligence. The grand goal in this area of Artificial intelligence (AI) is to automatically extract knowledge from human language expressions. This would enable knowledge encoded as natural language text to be combined with the power of AI reasoning mechanisms, which would facilitate a variety of AI applications, such as scientific discovery from web-based documents, within various disciplines. In particular, I propose to: (1) develop an appropriate constraint-based model(s) for parsing queries into knowledge and parsing potential "answers" (i.e., the subtexts that existing search engines return as possibly containing the answers) into knowledge bases from which inferences can be made to answer the query, so that a real answer is obtained instead of a URL; (2) incorporate correct ontological reasoning extensions with good complexity properties to remedy the problem that some of the subtexts proposed as answers in fact answer logically flawed interpretations of the query; and (3) demonstrate these methods in a prototype for molecular biology text mining that uses both English and DNA text as test corpora, paving the way to developing executable theories of language intelligence that could be applied to languages of molecular biology, as well as to human languages. My research will develop foundational and innovative interdisciplinary Web search methods and have immensely high practical and commercial potential: language understanding being the major bottleneck for search engine providers, any progress is likely to result in commercial opportunities and/or collaborations. Moreover, the envisaged DNA linguistics application has high social relevance and may be of great value to a large research community, e.g. allowing biologists to command computers directly through English, for tasks such as extracting genetic information of interest encoded in DNA strings, or gleaning relationships between biological entities mentioned in a large number of written documents, or simply obtaining a question's answer.