# Low-Resource Corpus Filtering using Multilingual Sentence Embeddings

## TL;DR

This paper presents a multilingual sentence embedding approach using LASER for filtering noisy low-resource parallel corpora, achieving top performance in WMT19 shared tasks and demonstrating effectiveness in low-resource scenarios.

## Contribution

The paper introduces a novel filtering method based on LASER embeddings that outperforms existing techniques in low-resource machine translation tasks.

## Key findings

- Achieved the best performance in Nepali-English and Sinhala-English tasks.
- Ensemble of scoring methods improved filtering accuracy.
- Effective in low-resource and zero-resource translation scenarios.

## Abstract

In this paper, we describe our submission to the WMT19 low-resource parallel corpus filtering shared task. Our main approach is based on the LASER toolkit (Language-Agnostic SEntence Representations), which uses an encoder-decoder architecture trained on a parallel corpus to obtain multilingual sentence representations. We then use the representations directly to score and filter the noisy parallel sentences without additionally training a scoring function. We contrast our approach to other promising methods and show that LASER yields strong results. Finally, we produce an ensemble of different scoring methods and obtain additional gains. Our submission achieved the best overall performance for both the Nepali-English and Sinhala-English 1M tasks by a margin of 1.3 and 1.4 BLEU respectively, as compared to the second best systems. Moreover, our experiments show that this technique is promising for low and even no-resource scenarios.