# Training on Synthetic Noise Improves Robustness to Natural Noise in Machine Translation

## TL;DR

Training machine translation models with synthetic character noise enhances their robustness to real-world spelling errors, maintaining performance on clean text while improving translation quality on noisy inputs.

## Contribution

Demonstrates that adding synthetic noise during training significantly improves robustness to natural noise without needing natural noise data.

## Key findings

- Synthetic noise training improves robustness to real spelling errors.
- Performance on clean text remains unaffected.
- Robustness achieved without access to natural noise data.

## Abstract

We consider the problem of making machine translation more robust to character-level variation at the source side, such as typos. Existing methods achieve greater coverage by applying subword models such as byte-pair encoding (BPE) and character-level encoders, but these methods are highly sensitive to spelling mistakes. We show how training on a mild amount of random synthetic noise can dramatically improve robustness to these variations, without diminishing performance on clean text. We focus on translation performance on natural noise, as captured by frequent corrections in Wikipedia edit logs, and show that robustness to such noise can be achieved using a balanced diet of simple synthetic noises at training time, without access to the natural noise data or distribution.