容錯性中文語詞比對架構的設計

Abstract

對於設計中文輸入法而言，容錯或錯誤改正能力是非常需要的。尤其是在以語音辯識技術為主的輸入法而言更是有用，因為錯誤是難以避免的。在一些實際應用上，如自然語言的語音辯識的製作上，我們常需要處理極大的詞庫。我們所面對的問題就是在極大詞庫下如何設計一個有錯誤改正能力而且即時的語句比對系統。 這篇論文試著在一個極大的詞庫下去製作一個索引架構來幫助具錯誤改正能力的詞句比對運算。所採取的方法主要是以下三種概念：1. Cartesian Product File 2. Covering between buckets 3. Gradual expansion of search region。實驗的結果顯示在一個極大的詞庫下製作一個多錯誤改正能力而即時的語句比對系統是可行的。

Error tolerant capability is very desirable in designing a Chinese computer input method. It is especially useful in designing an input method based on speech recognition technology because where errors are inevitable. In practical applications, such as natural language speech recognition, we need to handle very large phrase tables. How to do error tolerant phrase matching with very large phrase tables in a real-time speech recognition environment is the problem we are facing. This thesis developed an index scheme to help the error tolerant phrase matching calculations with very large phrase tables. The approach is based on three concepts. 1. Cartesian Product File. 2. Covering between buckets. 3. Gradual expansion of search region. The results show that doing multiple error tolerant phrase matching with very large phrase tables is feasible.