Resolving Anaphora in a Portable Natural Language Front End to Databases

Resolving Anaphora in a Portable Natural Language Front End to Databases Fhlvia A. Burros and Anne DeRoeck Department of Computer Science University of Essex Colchester - CO4 3SQ - U.K. [barrf - deroe] ~essex.ac.uk Abstract An analysis of the evolution of Natu- ral Language front ends in the last three decades shows that the growth in portabil- ity brought, as a side effect, the narrow- ing of the provided coverage of contextu- ally based linguistic phenomena, such as anaphora and ellipsis. This paper presents the design and state of development of a computational mecha- nism which provides pronominal Anaphora Resolution within the environment of a highly portable Natural Language front end to databases, SQUIRREL3 Simple cases of Ellipsis are also treated by the pro- posed model. An Overview of SQUIRREL is presented, followed by a description of the Discourse Module and the results achieved so far. The prototype is implemented in C-Prolog. 1 Introduction The development of Natural Language (NL) systems for data retrieval has been a central issue in NL Pro- cessing research for the last three decades, motivated by the aim of helping non-expert database users. When we try to draw a line of evolution of such systems, it can be observed that growth in portabil- ity, essential for commercial viability, came at a cost in terms of broader linguistic coverage. 2 Earlier systems, mostly research motivated, were mainly developed for a single application, using domain-dependent information for treating contex- tual phenomena (eg, DEACON (Craig et al., 1966), SHRDLU (Winograd, 1972), LUNAR (Woods, 1The current system forms the base line for a joint SERC/DTI funded collaborative project between the University of Essex and Status IQ Ltd. for construct- ing an integrated platform for the retrieval of structured and textual data through Natural Language queries. 2See (Burros and DeRoeck, 1993) for a comprehensive review on Portable NL front ends. 1973), LADDER (Hendrix et al., 1978)). In con- trast, the subsequent generation of interfaces car- ried a higher emphasis on portability in their de- sign (eg, INTELLECT (Harris, 1984), IRUS (Bates et al., 1986), TEAM (Grosz et al., 1987)). These systems, however, offer a reduced coverage of dis- course phenomena, a central issue when continuity in the database consultation carries some priority. Thus, the ideal NL Front End (NLFE) should carry a broader linguistic coverage, in order to support a user focused query process, combined with a high degree of portability. In this light, we designed a Discourse Module, which is incorporated into a highly portable NLFE, SQUIRREL (DeRoeck et al., 1991). The system was originally conceived with a single-query based mode of consultation. By providing for anaphora and sim- ple cases of ellipsis resolution, the Discourse Mod- ule yields continuous consultations without the use of world models (to maintain the system's general portability). 2 Issues in Anaphora Resolution Our primary goal is the achievement of dialogue-like querying by extending SQUIRREL to a system ca- pable of dealing with basic pronominal anaphora and ellipsis. Information about each query is made …

Content truncated to 500 words