Computational Discourse Analysis

点击翻译按钮获取中文摘要

MOTIVATION "Writing an NSERC DG proposal takes time; Jo wrote hers in two months." In a coherent text, textual units are not understood in isolation but in relation with each other through discourse relations that may or may not be explicitly marked. The fact that Jo wrote her proposal in two months, illustrates that writing an NSERC DG application takes time! Research on discourse analysis tries to identify and model the coherence relations that hold between textual units. These, in turn, allow us to interpret the communicative purpose of a text's units and construct an understanding of the text as a whole. Being able to recognize and model these relations automatically is necessary to develop more natural downstream Natural Language Processing (NLP) applications such as dialogue systems and natural language generation. LONG TERM OBJECTIVE Discourse coherence entails the existence of logical connections (for example, illustration, result, purpose...) between textual units. Although much research work has addressed computational discourse analysis in recent years, the field is still in its infancy and many important open questions still remain. The long term objective of this research program is to explore more advanced computational aspects of discourse modeling and their use in downstream NLP applications. In particular, we will study the modeling of hierarchical discourse structures, cross-lingual and cross-genre discourse phenomena, and the role of discourse representations to improvement coherence in text generation and dialogue systems. SCIENTIFIC APPROACH The proposed research program will explore three inter-related areas of research: Area 1: Modeling and disentangling shallow and hierarchical discourse representations. Area 2: Analyzing discourse phenomena across languages and textual genres. Area 3: Exploring the use of discourse modeling for Natural Language Generation (NLG) and Dialogue Systems (DS). SIGNIFICANCE OF THE WORK Amazing recent progress in the field of NLP has allowed the development of real-life applications such as language-enabled personal assistants, as well as a new generation of research based on Deep Learning. However, much research is still required for these applications to go beyond individual sentences, and understand and produce coherent multi-sentence texts. The proposed research program can make a significant contribution to the field of NLP in these exciting times. The use of discourse information in downstream applications will allow us to model why a sentence was produced and its effect on the preceding or upcoming communication and will allow us to build systems that can converse naturally. Given the current economic importance of AI and NLP in Canada, I am confident that our work can easily be transferred to the Canadian industry.