VnCoreNLP: A Vietnamese Natural Language Processing Toolkit

Thanh Vu, Dat Quoc Nguyen, Dai Quoc Nguyen, Mark Dras, Mark Johnson

Abstract

We present an easy-to-use and fast toolkit, namely VnCoreNLP—a Java NLP annotation pipeline for Vietnamese. Our VnCoreNLP supports key natural language processing (NLP) tasks including word segmentation, part-of-speech (POS) tagging, named entity recognition (NER) and dependency parsing, and obtains state-of-the-art (SOTA) results for these tasks. We release VnCoreNLP to provide rich linguistic annotations to facilitate research work on Vietnamese NLP. Our VnCoreNLP is open-source and available at: https://github.com/vncorenlp/VnCoreNLP

Introduction

While Natural Language Processing (NLP) has seen massive global growth, many languages—including Vietnamese—lack integrated, production-ready toolkits similar to the Stanford CoreNLP. Prior Vietnamese tools were often fragmented, outdated, or computationally expensive.

Vietnamese presents unique linguistic challenges: unlike English, white space is used to separate syllables rather than words (e.g., "học sinh" is one word meaning "student"). Consequently, word segmentation is a mandatory first step. Existing neural toolkits like NNVLP were too slow for large-scale applications (~300 words/second), creating a need for a toolkit that is "Easy-to-use, Fast, and Accurate."

Methods

VnCoreNLP adopts a pipeline architecture where an Annotation object is passed through several specialized components. Rather than relying solely on heavy deep learning architectures, the authors utilized optimized feature-based models that achieve state-of-the-art (SOTA) results with significantly lower latency.

VnCoreNLP pipeline architecture

The toolkit components include:

  • Word Segmentation: Uses a transformation rule-based learning model to handle the syllable-to-word mapping.
  • POS Tagging: Employs MarMoT, a generic CRF-based framework known for its efficiency in morphological tagging.
  • NER: Utilizes a dynamic feature induction model that automatically optimizes feature combinations, reducing the need for manual feature engineering.
  • Dependency Parsing: Implements a greedy version of a transition-based parsing model with selectional branching, which balances the speed of greedy parsers with the accuracy of search-based ones.

Datasets

The authors evaluated the toolkit using standard Vietnamese benchmark datasets: VLSP 2013 for word segmentation and POS tagging, VLSP 2016 for Named Entity Recognition (NER), and the Vietnamese Dependency Treebank (VnDT) for training and testing the dependency parser.

Key Results

VnCoreNLP established new SOTA benchmarks across all four core tasks. Notably, in the context of Vietnamese NLP, traditional feature-based models outperformed BiLSTM-CRF neural models in NER tasks while being roughly 10 times faster.

NER performance comparison
  • Word Segmentation: 97.90% F1 score at a blistering speed of 62,000 words/second.
  • POS Tagging: 95.88% accuracy at 25,000 words/second.
  • NER: 88.55% F1 score, outperforming BiLSTM-CRF variants and processing 18,000 words/second.
  • Dependency Parsing: Achieved a Labeled Attachment Score (LAS) of 70.23 (with automatic POS tags), surpassing previous joint models like jPTDP while maintaining a speed of 8,000 words/second.

Discussion and Insights

Vietnamese is an isolating language with no word inflection. The authors found that neural models (BiLSTM) don't benefit much from additional POS tag features because the word order itself conveys most of the grammatical information. Furthermore, even in feature-based models, capturing character-level information is vital for NER to identify common syllables used in Vietnamese names. By focusing on speed, the toolkit is specifically designed to be useful for researchers in regions with limited computational resources.

Conclusion

VnCoreNLP provides a high-performance, unified pipeline for Vietnamese NLP. While highly effective, future iterations could improve by incorporating more specific Vietnamese linguistic features. The toolkit is open-source and available on GitHub at https://github.com/vncorenlp/VnCoreNLP.

Summary updated

Abstract

We present an easy-to-use and fast toolkit, namely VnCoreNLP---a Java NLP annotation pipeline for Vietnamese. Our VnCoreNLP supports key natural language processing (NLP) tasks including word segmentation, part-of-speech (POS) tagging, named entity recognition (NER) and dependency parsing, and obtains state-of-the-art (SOTA) results for these tasks. We release VnCoreNLP to provide rich linguistic annotations to facilitate research work on Vietnamese NLP. Our VnCoreNLP is open-source and available at: https://github.com/vncorenlp/VnCoreNLP
BibTeX
@inproceedings{vu2018vncorenlp,
  author = {Thanh Vu and Dat Quoc Nguyen and Dai Quoc Nguyen and Mark Dras and Mark Johnson},
  title = {{VnCoreNLP: A Vietnamese Natural Language Processing Toolkit}},
  booktitle = {Conference of the North American Chapter of the Association for Computational Linguistics},
  year = {2018},
  url = {https://openreview.net/forum?id=SkVWiQ-_br},
}
Figure