The ILSP/ARC submission to the WMT 2018 Parallel Corpus Filtering Shared Task

Abstract

This paper describes the submission of the Institute for Language and Speech Processing/Athena Research and Innovation Center (ILSP/ARC) for the WMT 2018 Parallel Corpus Filtering shared task. We explore several properties of sentences and sentence pairs that our system explored in the context of the task with the purpose of clustering sentence pairs according to their appropriateness in training MT systems. We also discuss alternative methods for ranking the sentence pairs of the most appropriate clusters with the aim of generating the two datasets (of 10 and 100 million words as required in the task) that were evaluated. By summarizing the results of several experiments that were carried out by the organizers during the evaluation phase, our submission achieved an average BLEU score of 26.41, even though it does not make use of any language-specific resources like bilingual lexica, monolingual corpora, or MT output, while the average score of the best participant system was 27.91.

NTT Neural Machine Translation Systems at WAT 2019

“…The ParaCrawl project successfully created a large-scale parallel corpus between English and other European languages by extensively crawling the web (Bañón et al, 2020). Typical bitext-mining projects, including ParaCrawl, took the following steps to identify parallel sentences from the web (Resnik and Smith, 2003): (1) find multilingual websites, which may contain parallel sentences, from the web (Papavassiliou et al, 2018;Bañón et al, 2020); (2) find parallel documents from websites (Thompson and Koehn, 2020;El-Kishky and Guzmán, 2020); (3) extract parallel sentences from parallel web URLs (Thompson and Koehn, 2019;Chousa et al, 2020). Our work focuses on the first step: finding bilingual targetdomain web URLs.…”

BitextEdit: Automatic Bitext Editing for Improved Low-Resource Machine Translation

“…Bitext Quality Improvement The most standardized approach to improving bitext either discards an example or treats it as a perfect training instance . Past submissions to the Parallel Corpus Filtering WMT shared task employ a diverse set of approaches covering simple pre-filtering rules based on language identifiers and sentence features (Rossenbach et al, 2018;Lu et al, 2018;Ash et al, 2018), learning to weight scoring functions based on language models, extracting features from neural translation models and lexical translation probabilities (Sánchez-Cartagena et al, 2018), combining pre-trained embeddings (Papavassiliou et al, 2018), and dualcross entropy . In contrast to prior work, and similar to ours, Briakou and Carpuat (2022) propose to revise imperfect translations in bitext via selectively replace them with synthetic translations generated by NMT of sufficient quality.…”

BitextEdit: Automatic Bitext Editing for Improved Low-Resource Machine Translation

“…The most standardized approach to improving bitext either discards an example or treats it as a perfect training instance . Past submissions to the "Parallel Corpus Filtering" WMT shared task employ a diverse set of approaches covering simple pre-filtering rules based on language identifiers and sentence features (Rossenbach et al, 2018;Lu et al, 2018;Ash et al, 2018), learning to weight scoring functions based on language models, extracting features from neural translation models and lexical translation probabilities (Sánchez-Cartagena et al, 2018), combining pre-trained embeddings (Papavassiliou et al, 2018), and dual-cross entropy . Our work builds on top of prior work and instead of filtering out all the imperfect translation candidates, we selectively edit them and keep them in the pool of training data for NMT.…”