Exploring Grammatical Error Correction with Not-So-Crummy Machine Translation

To date, most work in grammatical error correction has focused on targeting specific error types. We present a probe study into whether we can use round-trip translations obtained from Google Translate via 8 different pivot languages for whole-sentence grammatical error correction. We develop a novel alignment algorithm for combining multiple round-trip translations into a lattice using the TERp machine translation metric. We further implement six different methods for extracting whole-sentence corrections from the lattice. Our preliminary experiments yield fairly satisfactory results but leave significant room for improvement. Most importantly, though, they make it clear the methods we propose have strong potential and require further study.