This paper describes ATR’s hybrid approach to spoken language translation and it’s application to the IWSLT 2005 translation task. Multiple corpus-based translation engines are used to translate the same input, whereby the best translation among the element MT outputs is selected according to statistical models. The evaluation results of the Japanese-to-English and Chinese-to-English translation tasks for different training data conditions showed the potential of the proposed hybrid approach and revealed new directions in how to improve the current system performance. 1.

We intended to demonstrate the effect of using N-best speech recognition hypotheses for improving speech translation performance. A log-linear model, which integrated features from speech recognition and statistical machine translation, was used to rescore the translation candidates. Model parameters were estimated by optimizing an objectively measurable but subjectively relevant translation quality metric. Experimental results have shown that the proposed N-best approach improved translation quality over the conventional single-best approach. The improvements were confirmed consistently by several automatic translation evaluation metrics. 1.

Alignment models have been more prominent in the statistical (SMT) rather than the example-based machine translation (EBMT) research tradition. Word alignment is one of the oldest concepts in machine translation, and it is now readily available thanks to statistical tools. However, since real translations are rarely word-by-word, word alignment usually makes use of two linguistically unreasonable concepts: empty cepts and distortion. In this thesis, we develop a two-phase EBMT approach on the basis of a two-dimensional word alignment model. This approach avoids alignment to the empty cept and does not make use of distortion. In two-phase (or precompiled) EBMT, translation examples are converted into translation rules during the preprocessing phase. However, since the sentence to be translated is not known at this stage, preprocessing must be sensitive, as it entails a great risk of losing valuable information. In order to enable a more informed matching, translation rules must remain representative of the original example translation. Such a translation rule we call translation frame, and its major task is to capture the structural discrepancies in the sentence pair. We propose to generate translation frames on the basis of