Free the Plural: Unrestricted Split-Antecedent Anaphora Resolution

Abstract

Now that the performance of coreference resolvers on the simpler forms of anaphoric reference has greatly improved, more attention is devoted to more complex aspects of anaphora. One limitation of virtually all coreference resolution models is the focus on single-antecedent anaphors. Plural anaphors with multiple antecedents-so-called split-antecedent anaphors (as in John met Mary. They went to the movies)-have not been widely studied, because they are not annotated in ONTONOTES and are relatively infrequent in other corpora. In this paper, we introduce the first model for unrestricted resolution of split-antecedent anaphors. We start with a strong baseline enhanced by BERT embeddings, and show that we can substantially improve its performance by addressing the sparsity issue. To do this, we experiment with auxiliary corpora where split-antecedent anaphors were annotated by the crowd, and with transfer learning models using element-of bridging references and single-antecedent coreference as auxiliary tasks. Evaluation on the gold annotated ARRAU corpus shows that the out best model uses a combination of three auxiliary corpora achieved F1 scores of 70% and 43.6% when evaluated in a lenient and strict setting, respectively, i.e., 11 and 21 percentage points gain when compared with our baseline. 1(1)The Joneses i went to the park. They i had a good time.(2) John and Mary i went to the park. They i had a good time.(3) John i met Mary j in the park. They i,j had a good chat .(4) John likes green i , Mary likes blue j , but Tom likes both colours i,j .

Stay Together: A System for Single and Split-antecedent Anaphora Resolution

“…In this work, we propose an alternative, an extended version of LEA (Moosavi and Strube, 2016) that does joint evaluation of single/split-antecedent anaphors by explicitly representing plural entities. Yu et al (2020a) introduced the first system to resolve all split-antecedent anaphors annotated in the ARRAU corpus. Their work focuses on the data sparsity problem; split-antecedent anaphora resolution is helped using four auxiliary corpora created from a crowdsourced corpus and other anaphoric annotations in the ARRAU corpus.…”

“…Thanks in part to the latest developments in deep neural network architectures and contextual word embeddings (e.g., ELMo (Peters et al, 2018) and BERT (Devlin et al, 2019)), the performance of models for single-antecedent anaphora resolution has greatly improved (Wiseman et al, 2016;Clark and Manning, 2016b;Lee et al, 2017Kantor and Globerson, 2019;Joshi et al, 2020). So recently, the attention has turned to more complex cases of anaphora, such as anaphora requiring some sort of commonsense knowledge as in the Winograd Schema Challenge (Rahman and Ng, 2012;Peng et al, 2015;Liu et al, 2017;Sakaguchi et al, 2020); pronominal anaphors that cannot be resolved purely using gender (Webster et al, 2018), bridging reference (Hou, 2020;Yu and Poesio, 2020), discourse deixis (Kolhatkar and Hirst, 2014;Marasović et al, 2017; and, finally, split-antecedent anaphora (Zhou and Choi, 2018;Yu et al, 2020a) -plural anaphoric reference in which the two antecedents are not part of a single noun phrase.…”

“…The most frequently used extra resource is single-antecedent anaphors. Due to the complexity of these tasks, previous work is usually based on assuming that either gold anaphors (Hou, 2020;Yu et al, 2020a) or gold mentions (Zhou and Choi, 2018;Yu and Poesio, 2020) are provided. By contrast, in this work we introduce a system that resolves both single and split-antecedent anaphors, and is evaluated in a more realistic setting that does not rely on gold anaphors/mentions.…”

“…We further evaluate our best system in the gold setting to compare with the Yu et al (2020a) system. The model achieved better performance when compared to their system that is designed solely for split-antecedent task.…”

The CODI-CRAC 2021 Shared Task on Anaphora, Bridging, and Discourse Deixis in Dialogue

“…In addition, more research has been carried out on aspects of anaphoric interpretation that go beyond identity anaphora but are covered by datasets * Work done when the author was a student at CMU such as ARRAU (Poesio et al, 2018;Uryupina et al, 2020). These include, e.g., bridging reference (Clark, 1977;Hou et al, 2018;Hou, 2020;Yu and Poesio, 2020;Kobayashi and Ng, 2021), discourse deixis (Webber, 1991;Marasović et al, 2017; or split-antecedent anaphora (Eschenbach et al, 1989;Vala et al, 2016;Zhou and Choi, 2018;Yu et al, 2020b.…”