CoNLL 2014 shared task: Grammatical error correction with a syntactic N-gram language model from a big corpora

Abstract

We describe our approach to grammatical error correction presented in the CoNLL Shared Task 2014. Our work is focused on error detection in sentences with a language model based on syntactic tri-grams and bi-grams extracted from dependency trees generated from 90% of the English Wikipedia. Also, we add a naïve module to error correction that outputs a set of possible answers, those sentences are scored using a syntactic n-gram language model. The sentence with the best score is the final suggestion of the system. The system was ranked 11th, evidently this is a very simple approach, but since the beginning our main goal was to test the syntactic n-gram language model with a big corpus to future comparison.