Dependency-Guided LSTM-CRF for Named Entity Recognition

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 3862–3872, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 3862 Dependency-Guided LSTM-CRF for Named Entity Recognition Zhanming Jie and Wei Lu StatNLP Research Group Singapore University of Technology and Design zhanming jie@mymail.sutd.edu.sg, luwei@sutd.edu.sg Abstract Dependency tree structures capture long- distance and syntactic relationships between words in a sentence. The syntactic relations (e.g., nominal subject, object) can potentially infer the existence of certain named entities. In addition, the performance of a named en- tity recognizer could benefit from the long- distance dependencies between the words in dependency trees. In this work, we pro- pose a simple yet effective dependency-guided LSTM-CRF model to encode the complete de- pendency trees and capture the above proper- ties for the task of named entity recognition (NER). The data statistics show strong cor- relations between the entity types and depen- dency relations. We conduct extensive experi- ments on several standard datasets and demon- strate the effectiveness of the proposed model in improving NER and achieving state-of-the- art performance. Our analysis reveals that the significant improvements mainly result from the dependency relations and long-distance in- teractions provided by dependency trees. 1 Introduction Named entity recognition (NER) is one of the most important and fundamental tasks in natural language processing (NLP). Named entities cap- ture useful semantic information which was shown helpful for downstream NLP tasks such as corefer- ence resolution (Lee et al., 2017), relation extrac- tion (Miwa and Bansal, 2016) and semantic pars- ing (Dong and Lapata, 2018). On the other hand, dependency trees also capture useful semantic in- formation within natural language sentences. Cur- rently, research efforts have derived useful discrete features from dependency structures (Sasano and Kurohashi, 2008; Cucchiarelli and Velardi, 2001; Ling and Weld, 2012) or structural constraints (Jie et al., 2017) to help the NER task. However, how LOC EVENT GPE They suffer from drug dealing and loitering near their premises nsubj prep prep pobj cc conj prep poss pobj The seminar on the actual practice of tax reform was held in Hong Kong det prep det amod pobj prep nn pobj auxpass prep pobj nn nsubjpass Figure 1: Example sentences annotated with named en- titeis and dependencies in the OntoNotes 5.0 dataset. to make good use of the rich relational informa- tion as well as complex long-distance interactions among words as conveyed by the complete depen- dency structures for improved NER remains a re- search question to be answered. The first example in Figure 1 illustrates the re- lationship between a dependency structure and a named entity. Specifically, the word “premises”, which is a named entity of type LOC (location), is characterized by the incoming arc with label “pobj” (prepositional object). This arc reveals a certain level of the semantic role that the word “premises” plays in the sentence. Similarly, the two words “Hong Kong” in the second example that form an entity of type …

Content truncated to 500 words