A Novel Approach Towards Incorporating Context Processing Capabilities in NLIDB System

International Joint Conference on Natural Language Processing, pages 1216–1222, Nagoya, Japan, 14-18 October 2013. A Novel Approach Towards Incorporating Context Processing Capabilities in NLIDB System Arjun R. Akula, Rajeev Sangal, Radhika Mamidi Language Technologies Research Center, IIIT Hyderabad, India. arjunreddy.akula@research.iiit.ac.in, {sangal,radhika.mamidi}@iiit.ac.in Abstract This paper presents a novel approach to categorize, model and identify contextual information in natural language interface to database (NLIDB) systems. The inter- actions between user and system are cat- egorized and modeled based on the way in which the contextual information is uti- lized in the interactions. A relationship schema among the responses (user and system responses) is proposed. We present a novel method to identify contextual in- formation in one specific type of user- system interaction. We report on results of experiments with the university related queries. 1 Introduction Natural Language Interface to Database (NLIDB) systems allow the users to query databases in a natural language (Androutsopoulos et al., 1995; Meng and Wang, 2001; Popescu et al., 2003; Stratica et al., 2005; Li et al., 2005; Giordani, 2008; Giordani and Moschitti, 2009; Gupta et al., 2012). Although NLIDB systems are able to answer a wide range of natural language queries (NL queries), they are not used much in com- mercial applications. One of the main reasons for the less acceptance of these systems in real- time applications is that they lack robust context processing capabilities (Bertomeu et al., 2006). Currently there is very little work which explic- itly aims to investigate the role of context pro- cessing capabilities in NLIDB systems. However, the importance of context processing capabilities has been explored extensively in Question An- swering systems (Chai and Jin, 2004; Kato et al., 2004; Kirschner and Bernardi, 2007; Negri and Kouylekov, 2007; Kirschner and Bernardi, 2010). Users often fail to express their intention (in- formation need) in a single NL query (user re- sponse) (Bertomeu et al., 2006). Hence to an- swer a sequence of related NL queries, NLIDB systems should keep track of contextual informa- tion. NLIDB systems which do not use contextual information (non-contextual NLIDB) fail to com- pletely capture the user’s intention. Figure 1: An example of context based user- system interaction For example, let us consider a user-system inter- action shown in Figure 1. User responses are rep- resented as U1, U2, etc. and system responses are represented as S1, S2, etc. In this example, to interpret U2, information present in the preced- ing query U1 is needed. That means information present in U1 is the contextual information for U2. Query U3 does not depend on the informa- tion present in preceding queries. Semester name ‘Monsoon 2011’ present in U1 and the professor name ‘Einstein’ present in S3 are needed to inter- pret U4. 1.1 Background In a semantic template based non-contextual NLIDB system (Gupta et al., 2012), the main stages involved in extracting answers (system’s re- sponse) from the database are shown in Figure 2. At the syntactic analysis stage, the linguistic in- formation is extracted from the NL query. …

Content truncated to 500 words