ARKref: a rule-based coreference resolution system

ARKref is a tool for noun phrase coreference that is based on the systemdescribed byHaghighi and Klein (2009) (which was never publicly released). It was originally written in 2009. At the time of writing, the last released version was in March 2011. This document describes that version, which is open-source and publicly available at http://www.ark.cs.cmu.edu/ARKref.1 ARKref is a deterministic, rule-based system that uses syntactic information from a constituent parser, and semantic information from an entity recognition component, to constrain the set of possible mention candidates (i.e., noun phrases) that could be antecedents for a given mention. It encodes syntactic constraints such as the fact that the noun phrases in predicative nominative constructions corefer (e.g., John was the teacher.), as well as semantic constraints such as the fact that he cannot corefer with a noun labeled as a location. After filtering candidates with these constraints, it selects as the antecedent the candidate noun phrase with the shortest (cross-sentence) tree distance from the target. Antecedent decisions are aggregated with a transitive closure to create the final entity graph. ARKref belongs to a family of rule-based coreference systems that use rich syntactic and semantic information to make antecedent selection decisions. Besides Haghighi and Klein, current work in this vein includes Lee et al. (2013), which was one of the best performing systems in a recent CoNLL shared task. The following example provides an illustration of ARKref’s output, in which brackets denote the extent of noun phrases and indices denote the entity to which each noun phrase refers. This example emphasizes the syntactic selection criteria: