Bootstrapping Path-Based Pronoun Resolution

By Shane Bergsma, Dekang Lin. 2006

153 citations recorded by the source corpus.

We present an approach to pronoun resolution based on syntactic paths. Through a simple bootstrapping procedure, we learn the likelihood of coreference between a pronoun and a candidate noun based on the path in the parse tree between the two entities. This path information enables us to handle previously challenging resolution instances, and also robustly addresses traditional syntactic coreference constraints. Highly coreferent paths also allow mining of precise probabilistic gender/number information. We combine statistical knowledge with well known features in a Support Vector Machine pronoun resolution classifier. Significant gains in performance are observed on several datasets.

Subfields: Artificial Intelligence, 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 Natural Language Processing

Citations in the corpus

Gender Bias in Coreference Resolution (2018, methodology): “Noun Gender and Number Many coreference resolution systems, including those described here, make use of a common resource released by Bergsma and Lin (2006) 3 (“B&L”): a large list of English nouns and noun phrases with gender and number counts over 85GB of web news.”
Toward Gender-Inclusive Coreference Resolution (2019, methodology): “” And the most frequently used of these is the resource developed by Bergsma and Lin [2006] (henceforth, B+L), in which a large quantity of text was processed with “high precision” anaphora resolution links to associate names with “genders.””
Causal Mediation Analysis for Interpreting Neural NLP: The Case of Gender Bias (2020, background): “Winogender additionally includes gender estimates from text (Bergsma and Lin, 2006), which we also include in our analysis.”
Gender Bias in Coreference Resolution: Evaluation and Debiasing Methods (2018, background): “vious feature rich and rule-based approaches rely on corpus based gender statistics mined from external resources (Bergsma and Lin, 2006).”
Easy Victories and Uphill Battles in Coreference Resolution (2013, methodology): “As with the features of the S UR - FACE system, two conjoined variants of each feature are included: ﬁrst with the type of the current mention ( NOMINAL , PROPER , or the citation form of the pro-noun), then with the types of both mentions in the pair.”
Word Embeddings (Also) Encode Human Personality Stereotypes (2019, result): “This line of work is in stark contrast with earlier work in the field, which treated human stereotypes encoded in text as common sense knowledge that could be helpful in automating tasks such as named entity tagging and coreference resolution (Bergsma and Lin, 2006; Ji and Lin, 2009).”
Findings of the 2016 WMT Shared Task on Cross-lingual Pronoun Prediction (2019, methodology): “Except for the English wordlist with gender distributions by Bergsma and Lin (2006), only the shared task data was used in the CUNI systems.”
RtGender: A Corpus for Studying Differential Responses to Gender (2018, methodology): “As a baseline for convenience we provide masculine/feminine ratios for these ﬁrst names from Bergsma and Lin (2006).”
Visual Coreference Resolution in Visual Dialog using Neural Module Networks (2018, background): “The linguistic community defines coreference resolution as the task of clustering phrases, such as noun phrases and pronouns, which refer to the same entity in the world (see, for example, [10]).”
Chinese Zero Pronoun Resolution: An Unsupervised Probabilistic Model Rivaling Supervised Resolvers (2014, methodology): “Otherwise, we determine its gender by looking up a gender word list constructed by Bergsma and Lin's (2006) approach.”
Why discourse affects speakers’ choice of referring expressions (2015, background): “If r = new, sample that new referent r from the base distribution over entities with probability 1 U· (count U· denotes a total number of unseen entities that is estimated from a named entity list (Bergsma and Lin, 2006)).”
Supervised Noun Phrase Coreference Research: The First Fifteen Years (2010, methodology): “Given an automatically parsed corpus, Bergsma and Lin (2006) extract from each parse tree a dependency path, which is represented as a sequence of nodes and dependency labels connecting a pronoun and a candidate antecedent, and collect statistical information from these paths to determine the…”
CoNLL-2011 Shared Task: Modeling Unrestricted Coreference in OntoNotes (2011, methodology): “…accuracies, so in order to ensure that participants in the closed track were working from the same data, thus allowing clearer algorithmic comparisons, we spec-iﬁed a particular table of number and gender predictions generated by Bergsma and Lin (2006), for use during both training and testing.”
The Swedish Winogender Dataset (2021, methodology): “In Rudinger et al. (2018), the noun gender and number dataset from Bergsma and Lin (2006) is used to this end, cited as a frequently used source of this type of information in actual pro-noun resolution systems.”
Stylometric Analysis of Scientific Articles (2012, methodology): “We therefore use the data of Bergsma and Lin (2006).”
A Multi-Pass Sieve for Coreference Resolution (2010, methodology): “Gender – we assign gender attributes from static lexicons from (Bergsma and Lin, 2006; Ji and Lin, 2009).”
CoNLL-2012 Shared Task: Modeling Multilingual Unrestricted Coreference in OntoNotes (2012, methodology): “Proceedings of the Joint Conference on EMNLP and CoNLL: Shared Task, pages 1–40, Jeju Island, Korea, July 13, 2012. c©2012 Association for Computational Linguistics”
Toward Gender-Inclusive Coreference Resolution: An Analysis of Gender and Bias Throughout the Machine Learning Lifecycle* (2021, methodology): “…B + L), in which a large quantity of text was ), a confusion matrix between the pronoun(s) they use (rows) and the inferred gender of their name (based on (Bergsma and Lin 2006)) in columns (where Masc = “he”, Fem = “she”, Neut = “it” and Plur = “they”; “Unk” means the name was not found).”
RelaxCor Participation in CoNLL Shared Task on Coreference Resolution (2011, methodology): “All the knowledge required by the feature functions is obtained from the annotations of the corpora and no external resources have been used with the exception of WordNet (Miller, 1995), gender and number information (Bergsma and Lin, 2006) and sense inventories.”