建構一個中文對聯創作的知識評價架構

Abstract

對聯是一個重要的傳統中華文化之一。在這篇論文裡，我們提出了 Couplet Analysis System (CAS)，目的是為了能擷取並評價對聯的知識。而對聯的創作，有它固定的一些創作上的限制像是音韻、用字、以及語義上的要求，根據這些創作對聯的要求及特性，我們採用知識庫的方式定義出對聯中的知識屬性來擷取對聯的知識，然而這三個限制之中，語義的處理最為困難，所以這篇論文針對對聯語義處理的部分提出了重要的處理方法。為了針對對聯之中的各個詞彙來擷取知識，首先必須先將對聯先作準確的斷詞動作，因此我們提出了HRWS的對聯斷詞方法。接著在對聯語義的處理上，我們利用E-HowNet的架構，提出了HBA的方法來解決詞義歧異性問題，並且提出EH-SSC方法來決定上下聯間語義相似度。最後提出了知識評價機制，利用所擷取出來的對聯知識屬性來評價一幅對聯。為了評估CAS系統是否確實有效，我們利用了東吳大學的「全球徵聯」對聯比賽的2510篇參賽作品來做實驗，結果達到42%的對聯評價準確度。總結來說，CAS確實可以幫助使用者分析及評價他們的對聯。

The Chinese couplet called du□ li□n is an important part of traditional Chinese culture. In this thesis, we propose Couplet Analysis System (CAS) that its goal is to extract and evaluate knowledge of a couplet. To analyze a couplet, the constraints about tone, word, and semantic meaning are concerned as important features in a couplet. We use knowledge–based approach to define the knowledge attributes that to extract knowledge of a couplet. Among these three features, the analysis of semantic meaning is the most difficult process. Therefore the thesis focuses on the semantic meaning analysis. Before processing the constraints, the word segmentation is addressed. Then Heuristic Rule-based Word Segmentation is proposed to solve this problem. In analysis of semantic, E-HowNet is employed to compute the semantic similarity. Following structure of E-HowNet, the thesis proposes Heuristic-based approach to solve the semantic tagging for word problem and E-HowNet based semantic similarity approach to compute semantic similarity value between sentences of a couplet. Finally, the thesis proposes Knowledge Evaluation mechanism by using the knowledge attributes to evaluate the couplet. The evaluation results of the system are compared with that of domain experts. The result shows that our approach yields 42% precision. To sum up, CAS can help couplet writers analyze and evaluate couplets.