Improvements in Statistical Phrase-Based Interactive Machine Translation

State-of-the-art Machine Translation (MT) systems are still far from being perfect. An alternative is the so-called Interactive Machine Translation (IMT). In this paper, we present some novel methods to improve the statistical phrase-based IMT. We utilize dynamic distortion limitation to balance the requirements of long distance reordering and decoding speed. And we introduce the difference function to the translation hypothesis extension as a heuristic function, to make the final translation candidates as diverse as possible. We also use the user validated prefix to direct the word selection of suffix based on a word co-occurrence model. All these methods aim at optimizing the first N-best candidate translations and look forward to reducing the cognitive burden of the users. The experiential results show the effectiveness of our methods.