Adversarial Multi-Criteria Learning for Chinese Word Segmentation Xinchi Chen (Fudan University) Advisors: Prof. Xuanjing Huang Prof. Xipeng Qiu Direction: Natural Language Processing
What is Chinese word segmentation (CWS) ? 中国官员应邀到美国开会 中国/官员/应邀/到/美国/开会 大学生活好，还是中学生活好 大学/生活/好，还是/中学/生活/好 如果你饿了，我就下面给你吃 如果/你/饿/了，我/就/下/面/给/你/吃
基于词的方法 基于字的序列标注方法1 基于词和基于字的方法的结合 基于统计的分词 1. N. Xue. 2003. Chinese word segmentation as character tagging. Computational Linguistics and Chinese Language Processing 8(1):29–48.
大学/生活/好/，/还是/中学/生活/好 基于字的序列标注方法 B E B E S S B E B E B E S 大学/生活/好/，/还是/中学/生活/好
Long Short-term Neural Network based CWS [X Chen, X Qiu, C Zhu, P Liu, X Huang; EMNLP 2015]
Adversarial Multi-Criteria Learning for Chinese Word Segmentation [X Chen, Zhan Shi, X Qiu, X Huang; ACL 2017]
Adversarial Multi-Criteria Learning for Chinese Word Segmentation [X Chen, Zhan Shi, X Qiu, X Huang; ACL 2017]
Adversarial Multi-Criteria Learning for Chinese Word Segmentation [X Chen, Zhan Shi, X Qiu, X Huang; ACL 2017]
Objective function
Unsupervised Domain Adaptation by Backpropagation [Yaroslav Ganin, et al.]
Adversarial Multi-Criteria Learning for Chinese Word Segmentation [X Chen, Zhan Shi, X Qiu, X Huang; ACL 2017]
Adversarial loss function The criterion discriminator maximizes the cross-entropy of predicted criterion distribution p( |X) and true criterion. An adversarial loss aims to produce shared features, such that a criterion discriminator cannot reliably predict the criterion by using these shared features. Therefore, we maximize the entropy of predicted criterion distribution when training shared parameters. Unlike (Ganin et al., 2016), we use entropy term instead of negative cross-entropy.
Training
Experiments
Experiments
Experiments
Experiments
Experiments
Error Analysis
Case Study
Simplified Chinese to Traditional Chinese Knowledge Transfer Simplified Chinese to Traditional Chinese Formal Texts to Informal Texts
Simplified Chinese to Traditional Chinese
Formal Texts to Informal Texts
Thank you for your attention! Xinchi Chen (Fudan University) Advisors: Prof. Xuanjing Huang Prof. Xipeng Qiu Direction: Natural Language Processing