- 學位論文
醫療主訴句型預測及中文詞翻譯之研究
A Study on Sentence Pattern Prediction and Medical Entity Translation for Assisted Editing of Chief Complaint
摘要
病歷摘要是醫護人員用以作為醫病構通橋梁的重要依據，撰寫中文化病歷摘要更是近年來衛生署為了提升病患對本身病情「知的權利」加以推廣之。目前病歷摘要中文化撰寫時所遭遇問題，如：1)中文輸入速度不夠快；2)撰寫花費過多時間；3)醫療詞彙難以明確翻譯成中文表達；4)中英語文法架構差異甚遠，轉譯時容易產生誤差等。 本研究的目的為針對病歷摘要主訴，發展建立一套可供醫護人員於撰寫中文化病歷摘要主訴時，參考之書撰建議系統，其功能包括醫療用語中文化輔助翻譯及主訴句型書寫預測提醒。其特定目標包括：(1) 蒐集並建置醫療專有名詞之中英平行詞庫，包含：病歷主訴資料的文字語料庫、中英平行語料庫、病歷主訴關鍵詞庫、POS-based(part of speech)平行語料庫、Class-based平行語料庫等工作；(2)資料分析與句型預測模型之建立：包含詞彙前後Bi-gram 語言模型、POS-based Bi-gram語言模型、Class-based Bi-gram語言模型以及Partial Sentence Tree 模型；(3) 建置病歷主訴撰寫介面；(4)系統評估。 本研究整合上述方法與模組，開發出一套病歷主訴撰寫系統，包括病歷主訴撰寫介面、句型預測模組副程式與中文翻譯模組副程式等子程式。並進行系統成效評估，主要測試系統的輸入效能及輸出之正確性。 實驗結果顯示，使用者透過POS-based Bigram下一字詞提示功能時，對於英文主訴撰寫之Keystroke數與構句時間能有小幅度改善；而使用完整語句預測模型時，能有效降低輸入之Keystroke數與構句時間，減少主訴撰寫時耗費的時間與人力。
並列摘要
The medical records are important information for the communication of the illness condition. In recent years, to compose medical records in Chinese has increased demand since most patients are not aware of English medical terms. While composing a medical records in Chinese, several problems may arise: 1)input method of Chinese is not quick enough; 2)wasted too much time when writing of chief complaint; 3)not accurate translation of medical vocabulary; 4)the grammatical structure is very different. The goal of this paper is to construct a system for chief complaint composing with sentence pattern prediction. Two major modules are developed in this study: 1) chief complaint collection and construction module: to setup original chief complaint corpus, parallel corpus, medical keywords corpus, POS-based corpus, Class-based corpus; 2) data analysis and chief complaints prediction module: bi-gram language model, POS-based bi-gram language model, class-based bi-gram language model and Partial Sentence Tree Model. By integrating of the above modules, chief complaints composition supporting system with friendly input interface, sentence prediction function and chief complaints translation function was built. Several experiments and assessment of the effectiveness of the system was carried out. The experimental results showed that the Keystrokes and composing time of chief complaint composition are improved by adapting the proposed approach. It encourages us that the proposed system can be utilized on chief complaints composing and will be helpful in reducing the burden of medical staffs.