Cross-Lingual Korean Speech-to-Text Summarization

Abstract

The development of a cross-lingual text summarization of a language differing from that of the source document has been a challenge in recent years. This paper describes a summarization system built to auto-translate Korean speech into an English summary text. Recent studies have discussed two separate tasks in this area, namely, obtaining the analysis information from one of the two languages by providing early or late translation approaches. The early translation tries to translate the original documents into the target language, and then summarizes the results by considering the information of the translated texts, whereas the late translation approach attempts to summarize the original documents and then translate them into the target language. We propose a method for automatically converting Korean speech into an English summary text. The Korean transcript is segmented and analyzed for sentence clustering. A word-graph is then used to compress and generate a unique, concise, and informative compression. Experiments prove that our method achieves better accuracy in comparison with other methods.