文句修飾與推薦的輔助寫作系統

Auxiliary writing system based on sentence modification and recommendation

程煒倫(Wei-Lun Cheng)

指導教授 ： 許聞廉

國立清華大學/電機資訊學院/資訊系統與應用研究所/碩士(2021年)

摘要

讓計算機像人類一樣進行中文寫作是相當困難的議題。一來，中文語態和時態上比英文模糊；二來，在詞彙組合上，根據不同語境有其合適的搭配詞組。若直接替換成意義相近的詞彙進行改寫，都容易使句子產生語病謬誤。因此，要能有效傳達完整語意且通順句子的計算機寫作，理解句子結構與詞彙排列都是重要的要素。
本論文提出一「句子修改輔助系統」，利用替換搭配詞組讓語句更加通順。該方法實現以中心詞為主軸的規則式關係抽取，詳細抽取出句子的主語、謂語、定語、狀語與賓語之訊息。在測試階段，將各自同義詞的所有組合，根據對應依存關係頻率較高者作為系統推薦修改的搭配詞。
本論文以1000篇聯合報財經新聞，共拆解出5000個複合句提供訓練。實驗結果顯示，動賓關係（VOB）和定中關係（ATT）之詞組推薦分別得到了89.1%和92.8%的精準度。相信對於處理中文語句的句法結構以及搭配詞組替換有很大的幫助，此舉讓計算機在語句寫作上實現更細緻的語意操作。

關鍵字

語意抽取 ； 搭配詞替換 ； 中文語句改寫

並列摘要

It is an essential research topic for computers to write Chinese articles as human beings. Firstly, Passive voice and tense in Chinese are more vague than English. Secondly, there are suitable collocation phrases according to different contexts. If we replace words with their synonyms directly, it is easy to cause the linguistic fallacy. Therefore, understanding sentence structure and vocabulary arrangement are both essential elements to make computers write semantic complete and fluent sentences.
This thesis proposes a novel "sentence modification auxiliary system" which can supply more suitable collocation to make sentences more fluent. This rule-based relation extraction method is centered on the head word and can extract the information of subject, predicate, attributive, adverbial, and object in the sentence detailedly. In the test phase, all synonym combinations are ranked according to the frequency of the corresponding dependency relationships. These collocation words are the recommended modifications by the system.
We collect modifications from 1,000 financial news articles which were disassembled to 5,000 compound sentences as a training data set. The experimental results show that the phrase recommendation of the verb-object relationship (VOB) and the attribute relationship (ATT) achieved accuracy of 89.1% and 92.7%, respectively. We believe it could be helpful in the Chinese syntactic structure and Chinese collocation substitution. This allows the computer to achieve more detailed semantic operations in sentence writing.

並列關鍵字

Semantic Extraction ； Collocation Word Replacement ； Chinese Sentence Rewriting

參考文獻

[1] J. Allen, Natural language understanding. Benjamin-Cummings Publishing Co., Inc., 1988.

[2] E. Reiter and R. Dale, “Building applied natural language generation systems,” Natural Language Engineering, vol. 3, no. 1, pp. 57–87, 1997.

[3] C. D. Paice, “Constructing literature abstracts by computer: techniques and prospects,” Information Processing & Management, vol. 26, no. 1, pp. 171–186, 1990.

[4] M. Gambhir and V. Gupta, “Recent automatic text summarization techniques: a survey,” Artificial Intelligence Review, vol. 47, no. 1, pp. 1–66, 2017.

[5] A. H. Oh and A. I. Rudnicky, “Stochastic natural language generation for spoken dialog systems,” Computer Speech & Language, vol. 16, no. 3-4, pp. 387–407, 2002.