The Importance of Automatic Syntactic Features in Vietnamese Named Entity Recognition

31st Pacific Asia Conference on Language, Information and Computation (PACLIC 31), pages 97–103 Cebu City, Philippines, November 16-18, 2017 Copyright c⃝2017 Hoang Pham and Phuong Le-Hong The Importance of Automatic Syntactic Features in Vietnamese Named Entity Recognition Thai-Hoang Pham R&D Department Alt Inc Hanoi, Vietnam phamthaihoang.hn@gmail.com Phuong Le-Hong College of Science Vietname National University in Hanoi Hanoi, Vietnam phuonglh@vnu.edu.vn Abstract This paper presents a state-of-the-art system for Vietnamese Named Entity Recognition (NER). By incorporating automatic syntactic features with word embeddings as input for bidirectional Long Short-Term Memory (Bi- LSTM), our system, although simpler than some deep learning architectures, achieves a much better result for Vietnamese NER. The proposed method achieves an overall F1 score of 92.05% on the test set of an evalua- tion campaign, organized in late 2016 by the Vietnamese Language and Speech Processing (VLSP) community. Our named entity recog- nition system outperforms the best previous systems for Vietnamese NER by a large mar- gin. 1 Introduction Named entity recognition (NER) is an essential task in natural language processing that falls under the domain of information extraction. The function of this task is to identify noun phrases and categorize them into a predefined class. NER is a crucial pre- processing step used in some NLP applications such as question answering, automatic translation, speech processing, and biomedical science. In two shared tasks, CoNLL 20021 and CoNLL 20032, language independent NER systems were evaluated for En- glish, German, Spanish, and Dutch. These systems focus on four named entity types namely person, or- ganization, location, and remaining miscellaneous entities. 1http://www.cnts.ua.ac.be/conll2002/ner/ 2http://www.cnts.ua.ac.be/conll2003/ner/ Lately, an evaluation campaign that systemati- cally compared NER systems for the Vietnamese language has been launched by the Vietnamese Lan- guage and Speech Processing (VLSP)3 community. They collect data from electronic newspapers on the web and annotate named entities in this corpus. Sim- ilar to the CoNLL 2003 share task, there are four named entity types in VLSP dataset: person (PER), organization (ORG), location (LOC), and miscella- neous entity (MISC). In this paper, we present a state-of-the-art NER system for Vietnamese language that uses auto- matic syntactic features with word embedding in Bi- LSTM. Our system outperforms the leading system of the VLSP campaign utilizing a number of syn- tactic and hand-crafted features, and an end-to-end system described in (Pham and Le-Hong, 2017) that is a combination of Bi-LSTM, Convolutional Neu- ral Network (CNN), and Conditional Random Field (CRF) about 3%. To sum up, the overall F1 score of our system is 92.05% as assessed by the standard test set of VLSP. The contributions of this work con- sist of: • We demonstrate a deep learning model reach- ing the state-of-the-art performance for Viet- namese NER task. By incorporating automatic syntactic features, our system (Bi-LSTM), al- though simpler than (Bi-LSTM-CNN-CRF) model described in (Pham and Le-Hong, 2017), achieves a much better result on Vietnamese NER dataset. The simple architecture also contributes to the feasibility of our system in 3http://vlsp.org.vn/ 97 practice because it requires less time for in- ference …

Content truncated to 500 words