- 學位論文
物體描述之共通性分析與應用 -以鳥類之口語化描述語料為例
Analysis and Application of the Commonality of Object Descriptions -A Study Based on the Descriptive Corpus of Birds
摘要
物體描述共通性指的是在描述一件事物時所共有的特性、重點，如何處理及應用描述式的語言在資訊檢索、知識查詢等領域研究中均扮演重要的角色。本研究利用句子結構化的方式探討口語化語言的共通性，建立物體之特徵描述及口語化知識庫，並建立一套可用描述式語言查詢之系統。 本研究首先針對台灣一般賞鳥民眾，分析其口語化描述特性，進而應用於鳥類知識語料庫之建立。其工作項目有：(1)語料收集與校正；(2) 語料前置處理；(3) 語料結構化；(4) 相似度比對，最後，透過上述處理流程，建立一套鳥類查詢系統，提供一般使用者以口語化描述方式進行鳥種的查詢。所得到的結構化知識，亦可作為解說教育時做為參考的重點，以傳遞最符合人們需求、易於理解的生態知識。 透過收集了628筆訓練語料及360位測試者的口語化描述查詢實驗，結果發現：一般使用者之口語描述查詢結果在前十名（Top-10）命中率（hit ratio）達到99%；包含率（inclusion ratio）達到72%，顯示本研究的方法適合一般口語化語料之建立及查詢之應用。本研究未來可藉由導入權重之計算及增加口語語意的判斷技術，進而更提升口語化查詢之正確性。
並列摘要
Description commonality refers to the common characteristics and key points while using descriptive language to describe an object. It is essential to process and apply the descriptive language in the topics of information retrieval and information inquiry. This research utilized the structure of language to seek out the commonality of the descriptive language, built up the descriptive characteristics and description database and established a useful inquiry system which dealt with the descriptive language. This research aimed at the analysis of general public’s description commonality on birds. The methodology contains four major tasks: (1) object description corpus collection, (2) linguistic processing of the corpus, (3) analysis of the corpus structure pattern, and (4) building the cognition commonality model. The bird inquiry system allows general users to do birds inquiring based on their descriptions was built through the above procedures. The systemic knowledge gathered from this research can be utilized as the references in interpretation education, in order to deliver the ecology knowledge which can be understood by general public and match their needs. After collecting 628 training and 360 testing descriptions, the experimental results showed that the hit ratio of the general users’ description inquiry reached up to 99%. In addition, the inclusion ratio could be as high as 72%. The result demonstrates that the research methodology is suitable for building the commonality of descriptive language and the inquiry system. The further research recommendation is to increase the inclusion ratio by evaluating different weighting approaches and integrating semantic understanding techniques into the language processing module.