本篇論文致力於研究學術論文摘要中文步結構的自動分析,藉以協助非以英語為母語的學者進行學術論文摘要寫作。我們的方法,能為每篇摘要中所有的句子都以各種不同語言特徵之文步序列進行自動標記。 此方法利用一組少量的人工標記文步的摘要群及n-grams,自動在大量未標記的摘要中訓練文步與n-grams之間的關係,以擴充產生可用來標示文步的顯著n-grams,並且使用已標記的摘要來訓練文步序列的馬可夫模型。以一組少量由人工標記的摘要做評估。最後,我們提供一個自動分析文步模型,能夠自動快速地把摘要標示上文步。 本篇研究結果顯示,我們的自動分析文步模型對於學術論文的摘要呈現了合理的精確性,並且快速的文步標記工具提供非以英語為母語的學者一個自動化、實用的方式,預期可以縮短了解或寫作學術論文中良好結構摘要的時間。
This paper presents a method for automatically labeling the move structures of academic abstracts to assist non-native speakers of English in writing academic abstracts. In our approach, sentences in a given abstract are automatically labeled with move sequences. The method involves an annotating small set of abstracts and n-grams with moves by hand, and learning the relationships between moves and n-grams in large unlabeled abstracts, and training a Hidden Markov Model of move sequences. We also implement and evaluate an automatic move tagger. The result of this paper shows that our automatic move tagger performs with reasonably high precision, and providing an automatic, practical and fast move tagger for non-native speakers to understand or write well-structured abstracts.