Understanding the Value of Features for Coreference Resolution

By Eric Bengtson, Eric Bengtson, Dan Roth. 2008

321 citations recorded by the source corpus.

In recent years there has been substantial work on the important problem of coreference resolution, most of which has concentrated on the development of new models and algorithmic techniques. These works often show that complex models improve over a weak pairwise baseline. However, less attention has been given to the importance of selecting strong features to support learning a coreference model. This paper describes a rather simple pairwise classification model for coreference resolution, developed with a well-designed set of features. We show that this produces a state-of-the-art system that outperforms systems built with complex models. We suggest that our system can be used as a baseline for the development of more complex models -- which may have less impact when a more robust set of features is used. The paper also presents an ablation study and discusses the relative contributions of various features.

Subfields: Artificial Intelligence, Control and Systems Engineering, Natural Language Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Control and Systems Engineering

In Natural Language Processing

Citations in the corpus

Reasoning about Quantities in Natural Language (2015, methodology): “We use the Illinois Coreference Resolver (Bengtson and Roth, 2008; Chang et al., 2013) and the Illinois SRL (Punyakanok et al., 2008), for coreference and semantic role labelling, respectively.”
End-to-end Neural Coreference Resolution (2017, background): “They can typically be categorized as (1) mention-pair classiﬁers (Ng and Cardie, 2002; Bengtson and Roth, 2008), (2) entity-level models (Haghighi and Klein, 2010; Clark and Manning, 2015, 2016b; Wiseman et al., 2016), (3) latent-tree models (Fernandes et al., 2012; Bj¨orkelund and Kuhn, 2014;…”
Easy Victories and Uphill Battles in Coreference Resolution (2013, methodology): “Acknowledging this complexity, coreference systems, either learning-based (Bengtson and Roth, 2008; Stoyanov et al., 2010; Haghighi and Klein, 2010; Rahman and Ng, 2011b) or rule-based (Haghighi and Klein, 2009; Lee et al., 2011), draw on diverse information sources and complex heuristics to…”
Solving Hard Coreference Problems (2019, methodology): “Although there is a plethora of works on this task (Soon et al., 2001a; Ng and Cardie, 2002a; Ng, 2004; Bengtson and Roth, 2008; Pradhan et al., 2012; Kummerfeld and Klein, 2013; Chang et al., 2013), it is still deemed an unsolved problem due to intricate and ambiguous nature of natural language”
Using Coreference Links to Improve Spanish-to-English Machine Translation (2017, background): “Interpolation between the best and worse candidate is considered (Wiseman et al., 2015; Bengtson and Roth, 2008).”
Learning-based Multi-Sieve Co-reference Resolution with Knowledge (2012, methodology): “2.”
Coreference Semantics from Web Features (2012, background): “It includes over 80 core features that exploit various automatically generated annotations such as named entity tags, syntactic parses, and WordNet classes, inspired by Soon et al. (2001), Ng and Cardie (2002), and Bengtson and Roth (2008).”
Dynamic Knowledge-Base Alignment for Coreference Resolution (2013, methodology): “The training for our system is similar to Bengston and Roth (2008). The positive training examples are generated from mentions and their immediate preceding antecedent. The negative examples are generated from mentions and all their preceding non-coreferent mentions. If the mention is not a pronoun,”
Learning Global Features for Coreference Resolution (2016, background): “This contrasts with mention-pair models (e.g., Bengtson and Roth (2008)), which score all pairs of mentions in a cluster, as well as with certain cluster-based models (see discussion in Culotta et al. (2007)).”
Learning Anaphoricity and Antecedent Ranking Features for Coreference Resolution (2015, methodology): “the output of an anaphoricity classifier as a feature in a downstream coreference system (Ng, 2004; Bengtson and Roth, 2008).”
Inference Protocols for Coreference Resolution (2011, methodology): “This flexible decoder architecture allows linguistic or knowledge-based constraints to be easily added to the system: constraints may force mentions to be coreferent or non-coreferent and can be optionally used in either of the inference protocols.”
Evaluation Metrics For End-to-End Coreference Resolution Systems (2010, methodology): “We use the ACE2004 (Mitchell et al., 2004) English training data which we split into three sets following Bengtson & Roth (2008): Train (268 docs), Dev (76), and Test (107).”
An Active Learning Approach to Coreference Resolution (2015, background): “, 2013] and UIUC systems [Bengtson and Roth, 2008], [Dakwale et al.”
Removing the Training Wheels: A Coreference Dataset that Entertains Humans and Challenges Computers (2015, methodology): “Our simple classiﬁer yields a binary decision for each mention pair, a method that had been very popular before the last ﬁve years (Soon et al., 2001; Bengtson and Roth, 2008; Stoyanov et al., 2010).”
Coreference Resolution with World Knowledge (2011, unknown): “World knowledge is extracted primarily from three data sources, online encyclo-pedia (e.g., Ponzetto and Strube (2006), Uryupina et al. (2011)), unannotated data (e.g., Daum´e III and Marcu (2005), Ng (2007)), and coreference-annotated data (e.g., Bengtson and Roth (2008)).”
A Multi-Pass Sieve for Coreference Resolution (2010, methodology): “Most coreference resolution approaches perform the task by aggregating local decisions about pairs of mentions (Bengston and Roth, 2008; Finkel and Manning, 2008; Haghighi and Klein, 2009; Stoyanov, 2010).”
Learning Structured Perceptrons for Coreference Resolution with Latent Antecedents and Non-local Features (2014, background): “…used to train coreference re-solvers either by casting the problem as a binary classiﬁcation problem that considers pairs of mentions in isolation (Bengtson and Roth, 2008; Stoy-anov et al., 2009; Chang et al., 2012, inter alia ) or in the structured manner, where a clustering for an entire…”