A Tree Transducer Model for Grammatical Error Correction

We present an approach to grammatical error correction for the CoNLL 2013 shared task based on a weighted tree-to-string transducer. Rules for the transducer are extracted from the NUCLE training data. An n-gram language model is used to rerank k-best sentence lists generated by the transducer. Our system obtains a precision, recall and F1 score of 0.27, 0.1333 and 0.1785, respectively, on the official test set. On the revised annotations, the F1 score increases to 0.2505. Our system ranked 6th out of the participating teams on both the original and revised test set annotations.