Research Statement – Shafiq Joty

The Internet is a great source of human knowledge, but most of the information is in the form of unstructured text. In Natural Language Processing (NLP), we focus on adding structure to this text to uncover relevant information, and to use it in developing end-user application programs. To this end, my primary research goal is twofold: (i) developing NLP tools to automatically understand language phenomena that go beyond the individual clauses or sentences of a text, i.e., the discourse structure of the text; and (ii) exploiting these discourse analysis tools effectively in downstream NLP applications including machine translation, summarization, question answering, and sentiment analysis. One methodology emphasized throughout my research is to first identify the inherent semantic structures in a given problem, and then to develop structured machine learning models to exploit such structures effectively. My work has relied on deep learning for better representation of the input text and on probabilistic graphical models for capturing dependencies in the output.