利用Ontological Chain解決跨語言資訊檢索系統中的翻譯歧義性問題

Abstract

翻譯檢索問句為本的跨語言資訊檢索系統會遭遇到翻譯歧義性的問題，目前解析歧義性的方法主要有同義詞典為本和語料庫為本的方法，前者的涵蓋範圍不夠，詞鍵關係過少；後者構需要耗費龐大成本來建構語料庫。本論文提出一套知識本體鏈(Ontological Chain)的方法，解決跨語言資訊檢索系統中翻譯歧義性(Transilation Ambiguity)問題。運用知識本體表示專家建構的領域知識(Domain Knowledge)，從知識本體相關的節點延伸出知識本體鏈，替每個中文詞鍵找到最適當的英文翻譯。本論文以英國聖安德魯大學照片資料集(The Eurovision ST Andrews Photographic Collection,簡稱ESTA)兩萬八千篇影像和照片說明為例，實作一個跨語言資訊檢索系統。本系統的平均準確率可達 49%，並且達到單語言資訊檢索系統的81%效能。

Bilingual dictionaries have been commonly used for query translation in cross-language information retrieval(CLIR). However, the problem of translation ambiguity happens in query translation. Recent studies suggest traversing WordNet for selecting appropriate translations. This paper proposes an ontological chain approach to resolve translation ambiguity. First, we find the most smilar ontology nodes for each query. Second, we construct a semantic graph according to the semantic distances between these nodes. And finally we select the connected component with the highest score as our ontological chain. We show that our approach reaches 81% effect of monolingual information retrieval systems. When there are many candidate translations, our system performs better than monolingual information retrieval system.