探究語句模型技術應用於摘錄式語音文件摘要

Sentence Modeling Techniques for Extractive Spoken Document Summarization

張皓欽(Hao-Chin Chang)

指導教授 ： 陳柏琳

國立臺灣師範大學/理學院/資訊工程學系/碩士(2013年)

摘要

摘錄式語音摘要是根據事先定義的摘要比例，從語音文件中選取一些重要的語句來產生簡潔的摘要以代表原始文件的主旨或主題，在近幾年已成為一項非常熱門的研究議題。其中，使用語言模型(Language Modeling)架構結合庫爾貝克-萊伯勒差異量(Kullback-Leibler Divergence)來進行重要語句選取的方法，在一些文字與語音文件摘要任務上已展現不錯的效能。本論文延伸此一方法而三個主要貢獻。首先，基於所謂關聯性(Relevance)的概念，我們探索新穎的語句模型技術。透過不同層次(例如詞或音節)索引單位的使用所建立的語句模型能與文件模型進行比對，來估算候選摘要語句與語音文件的關係。再者，我們不僅使用了語音文件中所含有語彙資訊(Lexical Information)，也使用了語音文件中所含隱含的主題資訊(Topical Information)來建立各種語句模型。最後，為了改善關聯模型(Relevance Modeling)需要初次檢索的問題，本論文提出了詞關聯模型(Word Relevance Modeling)。語音摘要實驗是在中文廣播新聞上進行；相較於其它非監督式摘要方法，本論文所提出摘要方法似乎能有一定的效能提升。

關鍵字

語音摘要 ； 語句模型 ； 語言模型 ； 庫爾貝克-萊伯勒差異量

並列摘要

Extractive speech summarization, aiming to select an indicative set of sentences from a spoken document so as to concisely represent the most important aspects of the document, has emerged as an attractive area of research and experimentation. A recent school of thought is to employ the language modeling (LM) framework along with the Kullback-Leibler (KL) divergence measure for important sentence selection, which has shown preliminary promise for extractive speech summarization. Our work in this paper continues this general line of research in three significant aspects. First, we explore a novel sentence modeling approach built on top of the notion of relevance, where the relationship between a candidate summary sentence and the spoken document to be summarized is discovered through various granularities of semantic context for relevance modeling. Second, not only lexical but also topical cues inherent in the spoken document are exploited for sentence modeling. Third, to counteract the shortcoming of the RM approach, need of resorting to a time-consuming retrieval procedure for relevance modeling, we present a word relevance modeling(WRM) approach. Experiments on broadcast news summarization seem to demonstrate the performance merits of our methods when compared to several existing unsupervised methods.

並列關鍵字

Speech summarization ； sentence modeling ； language modeling ； Kullback-Leibler divergence

參考文獻

[Barzilay and Elhadad 1997] R. Barzilay and M. Elhadad, “Using lexical chains for text summarization,” Proceedings of Workshop on Intelligent Scalable Text Summarization, pp. 10-17, 1997.

[Baxendale 1958] P. Baxendale “Machine-made index for technical literature - an experiment,” IBM Journal of Research and Development, Vol. 2, No. 4, pp. 354-361, 1958.

[Brin and Page 1998] S. Brin and L. Page, “The anatomy of a large-scale hypertextual web search engine,” Computer Networks and ISDN System, Vol. 30, No. 1-7, pp. 107-117, 1998.

[Carbonell and Goldstein 1998] J. Carbonell and J. Goldstein, “The use of MMR diversity-based reranking for reordering documents and producing summaries,” Proceedings of the 21th Annual International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR 1998), pp. 335-336, 1998.

[Chen et al. 2009] Y.-T. Chen, B. Chen and H.-M. Wang, “A probabilistic generative framework for extractive broadcast news speech summarization,” IEEE Transactions on Audio, Speech and Language Processing, Vol. 17, No. 1, pp. 95-106, 2009.