| Summary: | 博士 === 國立清華大學 === 資訊系統與應用研究所 === 100 === We introduce a learning method for predicting text completion in writing, and grammatical constructions to assist in the translation of a source text. In the proposed approach, predictions are offered on the fly during sentence translation to help the user in making appropriate lexical and grammar choices, thus improving writing quality and productivity. The method involves automatically extracting and evaluating sublexical/constituent translations for out-of-vocabulary (hereafter referred to as OOV) words (i.e., out-of-vocabulary module for text prediction), automatically analyzing target-language sentences to generate general and syntax-based phraseological tendencies (i.e., target-language writing suggestion module for grammar prediction), and automatically learning high-confidence word- or phrase-level translation equivalents (i.e., text prediction). At run-time, the source text and the translation prefix entered by the user are broken down into n-grams to generate grammar and translation predictions, which are further combined and ranked via translation and language models. These ranked prediction candidates are then displayed to the user in a pop-up menu as translation or writing hints. We present a prototype writing assistant, TransAhead, that applies the method to a human-computer collaborative environment for computer-assisted translation and computer-assisted language learning. Experimental results show that the OOV module indeed provides good translations for unknown words, and eases the impact of OOV on translation quality. It was also found that language learners substantially benefit from the writing module’s phraseology information. Overall, our methodology supports inline text and grammar predictions and has great potential for assisting language learners or novice translators in the process of translation, writing or even language learning.
|
|---|
