Corpus-based Pattern Induction for a Knowledge-based Question Answering Approach

Abstract

In this paper, we present an approach which, given a knowledge base and an appropriate text corpus, automatically induces patterns which can be used to query the knowledge base. We do not only consider simple questions but text problems consisting of several sentences. Answers to complex text problems are determined by an inference process which computes the answer on the basis of the background knowledge in the knowledge base (KB) as well as the textual description of the problem. The question formulated in natural language thus needs to be translated into appropriate KB structures and queries. Our approach to translating the natural language question uses an underlying corpus and the knowledge base to derive meaningful and relevant patterns which can then be used to process the questions and capture their meaning with respect to the underlying knowledge base. We apply our approach to Advanced Placement (AP) questions in three areas: physics, chemistry and biology. We report results of a first evaluation of the translation procedure for the three domains.

Towards portable natural language interfaces to knowledge bases – The case of the ORAKEL system

“…Finally, an interesting question for future research is whether the mappings between linguistic expressions and relations defined in the knowledge base can be learned automatically from a corpus. This idea was already preliminary explored in [14,19], but further research is needed to clarify its full potential as well as weaknesses.…”

“…This research has been supported by the following projects: the BMBF project SmartWeb, 19 financed by the German Ministry of Education and Research as well as the EU projects Dot.Kom,…”

“…Nevertheless, such approaches strongly depend on the quality and coverage of the lexical resources used. Recent work by ourselves [19], in which an approach based on lexical matching is explored, has in fact shown that one can rely less on lexical matching the more technical the domains get. In fact, we can not expect to have the complete lexical knowledge necessary for very technical domains in general resources such as WordNet.…”

Fuzzy-Based Answer Ranking in Question Answering Communities

“…They considered that two different words are equal if its distance value is greater than 0.6. Cimiano et al (2007) presented an approach which, given a knowledge base and an appropriate text corpus, automatically induces patterns which can be used to query the knowledge base. Given a partial frame and a matching predicate-argument structure, they determined the highest-scoring mapping between attributes of the partial frame and arguments of the predicate-argument structure by calculating the average Levenshtein distance.…”