[Paper Review] N-LTP: A Open-source Neural Chinese Language Technology Platform with Pretrained Models.

Wanxiang Che, Yunlong Feng+2 more|arXiv (Cornell University)|Sep 24, 2020

Topic Modeling25 references36 citations

TL;DR

N-LTP is an open-source neural Chinese NLP platform that unifies six core tasks—lexical, syntactic, and semantic analysis—using a shared, pre-trained multi-task model with knowledge distillation to improve performance. It outperforms single-task baselines and provides accessible APIs and visualization tools.

ABSTRACT

We introduce N-LTP, an open-source neural language technology platform supporting six fundamental Chinese NLP tasks: lexical analysis (Chinese word segmentation, part-of-speech tagging, and named entity recognition), {syntactic parsing} (dependency parsing), and {semantic parsing} (semantic dependency parsing and semantic role labeling). Unlike the existing state-of-the-art toolkits, such as Stanza, that adopt an independent model for each task, N-LTP adopts the multi-task framework by using a shared pre-trained model, which has the advantage of capturing the shared knowledge across relevant Chinese tasks. In addition, knowledge distillation where the single-task model teaches the multi-task model is further introduced to encourage the multi-task model to surpass its single-task teacher. Finally, we provide a collection of easy-to-use APIs and a visualization tool to make users easier to use and view the processing results directly. To the best of our knowledge, this is the first toolkit to support six Chinese NLP fundamental tasks. Source code, documentation, and pre-trained models are available at \url{this https URL}.

Motivation & Objective

• To develop a unified, open-source platform for six fundamental Chinese NLP tasks.
• To address the limitation of existing toolkits that use isolated models per task by introducing a multi-task learning framework.
• To enhance model performance through knowledge distillation, where single-task models guide the multi-task model.
• To improve usability by providing intuitive APIs and a visualization tool for direct result inspection.

Proposed method

• Adopting a multi-task learning framework with a shared pre-trained backbone to capture cross-task knowledge in Chinese NLP.
• Integrating six core NLP tasks: word segmentation, POS tagging, named entity recognition, dependency parsing, semantic dependency parsing, and semantic role labeling.
• Applying knowledge distillation by training the multi-task model to mimic the predictions of individual single-task teacher models.
• Designing a modular architecture with standardized APIs for easy integration and deployment.
• Implementing a visualization tool to enable real-time inspection of NLP output structures.
• Releasing pre-trained models, source code, and comprehensive documentation at a public URL.

Experimental results

Research questions

• RQ1 Can a shared multi-task learning framework effectively improve performance across six diverse Chinese NLP tasks compared to isolated models?
• RQ2 To what extent does knowledge distillation from single-task teachers enhance the performance of a unified multi-task model?
• RQ3 How does the integration of lexical, syntactic, and semantic parsing in a single platform affect end-to-end NLP pipeline efficiency and accuracy?
• RQ4 Can an open-source, user-friendly toolkit with visualization support effectively support researchers and practitioners in Chinese NLP?

Key findings

• N-LTP is the first open-source toolkit to support all six fundamental Chinese NLP tasks in a single unified framework.
• The multi-task model, enhanced by knowledge distillation, achieves superior performance compared to its single-task teacher models.
• The shared pre-trained architecture enables effective knowledge transfer across related tasks, improving generalization.
• The platform's APIs and visualization tool significantly reduce the barrier to entry for researchers and developers.
• Pre-trained models and full documentation are publicly available, enabling reproducibility and widespread adoption.
• The integration of syntactic and semantic parsing within the same framework enables more coherent and contextually consistent linguistic analysis.