Abstract
We introduce a method for learning to extract Synchronous Grammar Patterns (SGP) from a parallel corpus that can be used in assisted writing for English learners, as well as machine translation. In our approach, The method involves automatically recognizing the grammar patterns in source language, automatically aligning the source grammar patterns to target, and automatically selecting good SGPs and examples. We present a prototype system, Translation Pattern Assis- tant (TPA), that applies our method to providing English writing suggestion with Chinese translation. The evaluation on a set of randomly sampled SGP shows that TPA provides satisfactory suggestion for learners. Additionally, we propose a method for using SGP to improve machine translation quality.