On the Impact of Various Types of Noise on Neural Machine Translation

TL;DR

This paper investigates how different artificial noise types in training data affect neural machine translation quality, revealing neural models are more sensitive to noise and can learn to copy inputs under severe noise conditions.

Contribution

The study introduces five types of artificial noise and compares their impact on neural and statistical machine translation, highlighting neural models' vulnerability.

Findings

Neural models are more affected by noise than statistical models.

Severe noise can cause neural models to copy input sentences.

Different noise types degrade translation quality to varying degrees.

Abstract

We examine how various types of noise in the parallel training data impact the quality of neural machine translation systems. We create five types of artificial noise and analyze how they degrade performance in neural and statistical machine translation. We find that neural models are generally more harmed by noise than statistical models. For one especially egregious type of noise they learn to just copy the input sentence.