That Slepen Al the Nyght with Open Ye! Cross-era Sequence Segmentation with Switch-memory

Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 7830 - 7840 May 22-27, 2022 c⃝2022 Association for Computational Linguistics That Slepen Al the Nyght with Open Ye! Cross-era Sequence Segmentation with Switch-memory Xuemei Tang1,2 and Qi Su2,3,4∗ 1Department of Information Management, Peking University, Bejing, China 2Digital Humanities Center of Peking University, Bejing, China 3School of Foreign Languages, Peking University, Bejing, China 4MOE Key Lab of Computational Linguistics, School of EECS, Peking University, Bejing, China tagxuemei@stu.pku.edu.cn sukia@pku.edu.cn Abstract The evolution of language follows the rule of gradual change. Grammar, vocabulary, and lexical semantic shifts take place over time, re- sulting in a diachronic linguistic gap. As such, a considerable amount of texts are written in languages of different eras, which creates ob- stacles for natural language processing tasks, such as word segmentation and machine trans- lation. Although the Chinese language has a long history, previous Chinese natural language processing research has primarily focused on tasks within a specific era. Therefore, we pro- pose a cross-era learning framework for Chi- nese word segmentation (CWS), CROSSWISE, which uses the Switch-memory (SM) module to incorporate era-specific linguistic knowledge. Experiments on four corpora from different eras show that the performance of each corpus significantly improves. Further analyses also demonstrate that the SM can effectively inte- grate the knowledge of the eras into the neural network. 1 Introduction As a human-learnable communication system, lan- guage does not remain static but instead evolves over time. The rate of change between different aspects of language, such as grammar, vocabulary, and word meaning, vary due to language contact and many other factors, which has led to the di- achronic linguistic gap. An example of this can be seen in, “That slepen al the nyght with open ye (That sleep all the night with open eye),” which is a sentence from The Canterbury Tales, written in Middle English by Geoffrey Chaucer at the end of the 14th century. It is difficult for people without an understanding of Middle English to make sense of this sentence. Furthermore, some discourses ∗ Corresponding author Sample from MSR Golds (wait) (who) (come) (slove) (ne)？ 等待 谁 来 解决 呢 ？ PKUSeg 等待 谁 来 解决 呢 ？ JiaYan 等 待 谁 来 解 决 呢 ？ Sample from AWIKI Golds (Qi)(Cui Shu)(lead army)(attack)(Lv)。 齐 崔杼 帅师 伐 莒 。 PKUSeg 齐崔 杼帅 师伐莒 。 JiaYan 齐 崔杼 帅师 伐 莒 。 Table 1: Illustration of the different segmentation results for a modern Chinese sentence and an ancient Chinese sentence with different segmentation toolkits. contain both modern English and Old English due to citation or rhetorical need. For example, Shake- speare’s fourteen lines of poetry are often quoted in contemporary novels. This kind of era-hybrid text creates barriers to natural language process- ing tasks, such as word segmentation and machine translation. The Chinese language has the honor of being listed as one of the world’s oldest languages and, as such, has seen several changes over its …

Content truncated to 500 words