A Rule-Based System for Unrestricted Bridging Resolution: Recognizing Bridging Anaphora and Finding Links to Antecedents

Abstract

Bridging resolution plays an important role in establishing (local) entity coherence. This paper proposes a rule-based approach for the challenging task of unrestricted bridging resolution, where bridging anaphors are not limited to definite NPs and semantic relations between anaphors and their antecedents are not restricted to meronymic relations. The system consists of eight rules which target different relations based on linguistic insights. Our rule-based system significantly outperforms a reimplementation of a previous rule-based system (Vieira and Poesio, 2000). Furthermore, it performs better than a learning-based approach which has access to the same knowledge resources as the rule-based system. Additionally, incorporating the rules and more features into the learning-based system yields a minor improvement over the rule-based system.

Rule- and Learning-based Methods for Bridging Resolution in the ARRAU Corpus

“…We started with a re-implementation of the state-ofthe-art rule-based method by Hou et al (2014), which did not achieve satisfactory performance when being applied to the ARRAU corpus. We found that the reasons for this lie in the different bridging annotations.…”

“…As a starting point, we adopt the approach by Hou et al (2014) and re-implement a rule-based system for full bridging resolution 1 . The system contains eight rules.…”

“…Recently, there has only been few work on these tasks (Hou et al, 2014(Hou et al, , 2013bMarkert et al, 2012;Rahman and Ng, 2012), which is partly due to the lack of annotated data, which makes the application of statistical methods difficult. Most recent work has focused on the corpus ISNotes (Markert et al, 2012), on which Hou et al (2014)'s rule-based system currently achieves state-of-theart results.…”

“…Most recent work has focused on the corpus ISNotes (Markert et al, 2012), on which Hou et al (2014)'s rule-based system currently achieves state-of-theart results.…”

“…This paper presents a rule-based and a learningbased system, as submitted to the shared task. We start with a re-implementation of Hou et al (2014)'s rule-based approach, which we then apply to the ARRAU corpus. Although the approach was designed for the same domain (news), we find that the performance is very poor.…”