# The CUED's Grammatical Error Correction Systems for BEA-2019

## TL;DR

This paper presents two grammatical error correction systems from Cambridge University for BEA-2019, one based on finite state transducers and neural models for low-resource scenarios, and a purely neural system for the restricted track.

## Contribution

It introduces a novel combination of finite state transducers with neural language models for low-resource grammatical error correction and a neural-only system for the restricted track.

## Key findings

- Finite state transducer and neural model hybrid system for low-resource correction
- Pure neural system with back-translation and checkpoint averaging for restricted track
- Systems achieved competitive results in BEA-2019 shared task

## Abstract

We describe two entries from the Cambridge University Engineering Department to the BEA 2019 Shared Task on grammatical error correction. Our submission to the low-resource track is based on prior work on using finite state transducers together with strong neural language models. Our system for the restricted track is a purely neural system consisting of neural language models and neural machine translation models trained with back-translation and a combination of checkpoint averaging and fine-tuning -- without the help of any additional tools like spell checkers. The latter system has been used inside a separate system combination entry in cooperation with the Cambridge University Computer Lab.