Can I also have a question, please? On Perspectives of Interactive Clustering-Driven Question-Answering

1 Introduction Question Answering (QA) is an active research area that gained an interest of many researchers, including but not limited only to the TREC community 1. The systems that participate in the TREC QA track are designed to provide answers to factoid questions, from 2 gigabytes of text, collection that includes mostly newspaper or newswire articles[3]. Most of the systems that participate in the TREC evaluation, uses a variety of linguistic resources and employs sophisticated text processing techniques, which help to retrieve a sufficient set of answer candidates by resolving anaphor and synonyms, uncovering complex lexical, syntactic and semantic relations between a question and answer candidates. Since the beginning of TREC evaluation one can observe a constant improvement of the systems performance, maturing of commonly used methods and increasing sophistication of used tools and linguistic resources. Users expectations concerning a naturalness, reliability and intuitiveness of access to information as well as ever-expanding search space of various types of resources , including dynamically growing textual documents influence the environment and working conditions for wide range of tools such as databases querying systems, Internet search engines, QA-Systems etc. The amount of text available online has been rapidly increasing; in 2005, a popular search engine indexed 19.2 billion web pages 2 and their numbers continue to grow. In [1], Brill demonstrated that web redundancy can be effectively used in QA-Systems to retrieve an answer rich set of passages, reducing the overall complexity of the system and its dependency on external tools and linguistic resources. We think that a similar effect can be achieved by redefining Question-Answering by introducing the means for a system-user interaction. As we demonstrate below, such interaction while beneficial for a system by providing the means to adjust the search space and select an answer rich set of documents closely related to a question addresses also some of users expectations concerning naturalness and intu-itiveness of accessing sought-after information. This approach aims to reassemble a part of a conversation between two human beings, when a person who is supposed to answer a given question, asks for more details to eliminate potential ambiguity of a question or to specify the proper answer scope expected by other person. We think that research on interactive QA is also valuable for the development of dialog and chat systems. So far, most of these systems employs various discussion strategies to avoid a direct answer to …