A Constraint-Based Hypergraph Partitioning Approach to Coreference Resolution

Abstract

This work is focused on research in machine learning for coreference resolution. Coreference resolution is a natural language processing task that consists of determining the expressions in a discourse that refer to the same entity.\ud The main contributions of this article are (i) a new approach to coreference resolution\ud based on constraint satisfaction, using a hypergraph to represent the problem and solving it by relaxation labeling; and (ii) research towards improving coreference resolution performance using world knowledge extracted from Wikipedia.\ud The developed approach is able to use an entity-mention classification model with more\ud expressiveness than the pair-based ones, and overcome the weaknesses of previous approaches in the state of the art such as linking contradictions, classifications without context, and lack of information evaluating pairs. Furthermore, the approach allows the incorporation of new information by adding constraints, and research has been done in order to use world knowledge to improve performances.\ud RelaxCor, the implementation of the approach, achieved results at the state-of-the-art level, and participated in international competitions: SemEval-2010 and CoNLL-2011. RelaxCor achieved second place in CoNLL-2011.Postprint (published version

Coreference resolution: an empirical study based on SemEval-2010 shared Task 1

“…RELAXCOR (Sapena et al, 2010a) is a coreference resolution system based on constraint satisfaction. It represents the problem as a graph connecting any pair of candidate coreferent mentions, and it applies relaxation labeling over a set of constraints to decide the set of most compatible coreference relations.…”

“…For the present study, all constraints were learned automatically using more than a hundred features over the mention pairs in the training sets. The typical attributes were used, like those in Sapena et al (2010b), but binarized for each possible value. In addition, other features that could help, such as whether a mention is an NE of location type or a possessive phrase, were included.…”

“…The RELAXCOR implementation used in the present study is an improved version of the system that participated in the SemEval-2010 Task 1 (Sapena et al, 2010b). The largest differences involve the training and development processes.…”

Latent Trees for Coreference Resolution

“…Cai and Strube (2010), Cai, Mujdricza-Maydt, and Strube (2011) and Sapena, Padró, and Turmo (2013 present systems that implement global decision via hypergraph partitioning. Whereas Cai and Strube (2010) and Cai, Mujdricza-Maydt, and Strube (2011) use a spectral clustering algorithm to perform hypergraph partitioning, Sapena, Padró, and Turmo (2013) use relaxation labeling for the resolution process. There are also approaches that perform global inference using integer linear programming to enforce consistency on the extracted coreference chains (Denis and Baldridge 2007;Klenner 2007;Finkel and Manning 2008).…”

TextServer: Cloud-Based Multilingual Natural Language Processing

“…This services applies a reimplementation of the CoNLL-2010 shared task second ranked system [17], which was, however, the first in the ranking based on machine learning techniques, and thus, more easily adaptable to new languages. Concretely, it is based on graph partitioning via constraint relaxation labeling, where constraints were automatically learned.…”