We present two methods that merge ideas from statistical machine translation (SMT) and translation memories (TM). We use a TM to retrieve matches for source segments, and replace the mismatched parts with instructions to an SMT system to fill in the gap. We show that for fuzzy matches of over 70%, one method outperforms both SMT and TM baselines. 1

We improve translation memory (TM)inspired consistent phrase-based statistical machine translation (PB-SMT) using rich linguistic information including lexical, part-of-speech, dependency, and semantic role features to predict whether a TM-derived sub-segment should constrain PB-SMT translation. Besides better translation consistency, for English-to-Chinese Symantec TMs we report a 1.01 BLEU point improvement over a regular state-of-the-art PB-SMT system, and a 0.45 BLEU point improvement over a TM-constrained PB-SMT system without access to rich linguistic information, both statistically significant (p <0.01). We analyze the system output and summarize the benefits of using linguistic annotations to characterise the nature of translation consistency. 1