Data Processing Matters: SRPH-Konvergen AI's Machine Translation System for WMT'21

Abstract

In this paper, we describe the submission of the joint Samsung Research Philippines-Konvergen AI team for the WMT'21 Large Scale Multilingual Translation Task -Small Track 2. We submit a standard Seq2Seq Transformer model to the shared task without any training or architecture tricks, relying mainly on the strength of our data preprocessing techniques to boost performance. Our final submission model scored 22.92 average BLEU on the FLORES-101 devtest set, and scored 22.97 average BLEU on the contest's hidden test set, ranking us sixth overall. Despite using only a standard Transformer, our model ranked first in Indonesian → Javanese, showing that data preprocessing matters equally, if not more, than cutting edge model architectures and training techniques.

Samsung R&D Institute Philippines at WMT 2023

“…We employ ratio-based filters on tokenized sentence pairs following Cruz and Sutawika (2022) and Sutawika and Cruz (2021). We first tokenize using SacreMoses 2 then apply the following ratio-based filters: 7,143,725 115,239,312 95,954,020 Synthetic he→en 73,278,018 1,471,827,973 1,056,677,671 Synthetic he→en Filtered 47,372,416 659,409,236 541,376,459 Table 1: Corpus Statistics.…”