# CUNI System for the WMT19 Robustness Task

## TL;DR

This paper describes the CUNI Transformer system's robustness in translation tasks, demonstrating significant improvements over LSTM baselines, especially when fine-tuned on noisy data, without sacrificing news domain translation quality.

## Contribution

The paper introduces a robust Transformer-based translation system and shows how fine-tuning on noisy data enhances robustness without degrading performance on clean data.

## Key findings

- Transformer system outperforms LSTM baseline in noisy input scenarios
- Fine-tuning on noisy data improves robustness
- Translation quality remains high on news domain

## Abstract

We present our submission to the WMT19 Robustness Task. Our baseline system is the Charles University (CUNI) Transformer system trained for the WMT18 shared task on News Translation. Quantitative results show that the CUNI Transformer system is already far more robust to noisy input than the LSTM-based baseline provided by the task organizers. We further improved the performance of our model by fine-tuning on the in-domain noisy data without influencing the translation quality on the news domain.