PUB - Publikationen an der Universität Bielefeld

Question Answering over Linked Data (QALD) has emerged in the last years as an important topic of research. QALD systems provide access to a growing body of linked open data on the Web for casual end users that are empowered to satisfy their information needs intuitively, using natural language. In this paper we focus on analyzing the infamous lexical gap that arises in any information or question answering system. The lexical gap refers to the problem that there can be a vocabulary mismatch between the vocabulary used in a user question or query and the vocabulary used in the data to describe the relevant information. We build on a semantic parsing approach to QALD and adapt the approach of Zettlemoyer and Collins (2005) to the task at hand. We evaluate our approach on the QALD-4 benchmark and show that performance of a semantic parsing approach can be substantially improved if the right lexical knowledge is available. For this, we model a set of lexical entries by hand to quantify the number of entries that would be needed. Further, we analyze if a state-of-the-art tool for inducing ontology lexical from corpora can derive these lexical entries automatically. We conclude that further research and investments are needed to derive the lexical knowledge automatically or semi-automatically to increase performance of QALD systems.