An architecture for anaphora resolution

Abstract

In this paper, we describe the pronominal anaphora resolution module of Lucy, a portable English understanding system. The design of this mo;clule was motivated by the observation that, although there exist many theories of anaphora resolution, no one of these theories is complete. Thus we have implemented a blackboard-like architecture in which individual partial theories can be encoded as separate modules that can interact to propose candidate antecedents and to evaluate each other's proposals.

Use of Domain Knowledge in Resolving Pronominal Anaphora

“…Selectional constraints on verbs are also identified in Lucy (Rich & LuperFoy (1988)), a natural-language front-end for knowledge-based systems. For instance, the object of the verb to eat is required to be of the type 'edible substance'.…”

Corpus-based anaphora resolution towards antecedent preference

“…Whereas knowledge-based systems like (Carbonell and Brown, 1988) and (Rich and LuperFoy, 1988) combining multiple resolution strategies are expensive in the cost of human effort at development time and limited ability to scale to new domains, more recent knowledge-poor approaches like (Kennedy and Boguraev, 1996) and (Mitkov, 1998) address the problem without sophisticated linguistic knowledge. Similarly to them we do not use any sentence parsing or structural analysis, but just rely on morphosyntactic and semantic word information.…”

Automatic Anaphora Resolution for Norwegian (ARN)

“…The systems relied heavily on linguistic knowledge, and tried to limit the amount of pragmatic (world) knowledge used. Many of the systems were never or only partially implemented, such as Rich and LuperFoy (1988). Some achieved good results, but on very limited corpora, for instance Carbonell and Brown (1988) achieved 87% on a sample of 31 anaphora.…”

“…Different anaphora systems have used different score ranges, e.g. from -1 to +2 (MARS by Mitkov 2002), from -5 to +5 (Pronominal anaphora resolution module of Lucy by Rich and LuperFoy 1988), and from 0 to 100 (RAP by Lappin and Leass 1994). ARN uses points in the range of -100 to +100.…”

“…The point scores are often modified by multiplying them with some other factor which could for instance reflect the candidate's proximity to the anaphor (Lappin and Leass 1994) or express the factors confidence (Rich and LuperFoy 1988). In ARN, the final candidate scores are multiplied by a number indicating their proximity to the anaphor.…”