Robust Coreference Resolution and Entity Linking on Dialogues: Character Identification on TV Show Transcripts

Proceedings of the 21st Conference on Computational Natural Language Learning (CoNLL 2017), pages 216–225, Vancouver, Canada, August 3 - August 4, 2017. c⃝2017 Association for Computational Linguistics Robust Coreference Resolution and Entity Linking on Dialogues: Character Identification on TV Show Transcripts Henry Y. Chen, Ethan Zhou, Jinho D. Choi Math and Computer Science Emory University Atlanta, GA 30322, USA {henry.chen, ethan.zhou, jinho.choi}@emory.edu Abstract This paper presents a novel approach to character identification, that is an entity linking task that maps mentions to charac- ters in dialogues from TV show transcripts. We first augment and correct several cases of annotation errors in an existing corpus so the corpus is clearer and cleaner for statisti- cal learning. We also introduce the agglom- erative convolutional neural network that takes groups of features and learns mention and mention-pair embeddings for corefer- ence resolution. We then propose another neural model that employs the embeddings learned and creates cluster embeddings for entity linking. Our coreference resolution model shows comparable results to other state-of-the-art systems. Our entity linking model significantly outperforms the previ- ous work, showing the F1 score of 86.76% and the accuracy of 95.30% for character identification. 1 Introduction Character identification (Chen and Choi, 2016) is a task that identifies each mention as a character in a multiparty dialogue.1 Let a mention be a nominal referring to a human (e.g., she, mom, Judy), and an entity be a character in the dialogue. The objective is to assign each mention to an entity, who may or may not appear as a speaker in the dialogue. For the example in Table 1, the mention comedian is not one of the speakers in the dialogue; nonetheless, it clearly refers to a real person that may appear in some other dialogues. Identifying such mentions as actual characters requires cross-document entity resolution, which makes this task challenging. 1The dialogues are extracted from TV show transcripts by the previous work (Chen and Choi, 2016). Character identification can be viewed as a task of entity linking. Most of the previous work on entity linking focuses on Wikification (Mihalcea and Cso- mai, 2007a; Ratinov et al., 2011a; Guo et al., 2013). Unlike Wikification, entities in this task have no precompiled information from a knowledge base, which is another challenging aspect. This task is similar to coreference resolution in the sense that it groups mentions into entities, but distinct because it requires the identification of mention groups as real entities. Furthermore, even if it can be tackled as a coreference resolution task, only a few coref- erence resolution systems are designed to handle dialogues well (Rocha, 1999; Niraula et al., 2014) although several state-of-the-art systems have been proposed for the general domain (Peng et al., 2015; Clark and Manning, 2016; Wiseman et al., 2016). Due to the nature of multiparty dialogues where speakers take turns to complete a context, charac- ter identification becomes a critical step to adapt higher-level NLP tasks (e.g., question answering, summarization) to this domain. This task can also bring another …

Content truncated to 500 words