The University of Helsinki Submissions to the WMT19 Similar Language Translation Task

Abstract

This paper describes the University of Helsinki Language Technology group's participation in the WMT 2019 similar language translation task. We trained neural machine translation models for the language pairs Czech ↔ Polish and Spanish ↔ Portuguese. Our experiments focused on different subword segmentation methods, and in particular on the comparison of a cognate-aware segmentation method, Cognate Morfessor, with character segmentation and unsupervised segmentation methods for which the data from different languages were simply concatenated. We did not observe major benefits from cognate-aware segmentation methods, but further research may be needed to explore larger parts of the parameter space. Character-level models proved to be competitive for translation between Spanish and Portuguese, but they are slower in training and decoding.

Dialect-to-Standard Normalization: A Large-Scale Multilingual Evaluation

“…• We voluntarily restrict our dataset to "clean" corpora, i.e., interviews transcribed and normalized by trained experts. This contrasts with other data collections specifically aimed at extracting dialectal content from social media (e.g., Ueberwasser and Stark, 2017;Mubarak, 2018;Barnes et al, 2021;Kuparinen, 2023). Such datasets compound the features and challenges of both dialect-to-standard normalization and UGC normalization.…”

Why don’t people use character-level machine translation?

“…This system, however, makes many suboptimal design choices and ended up as the last one in the manual evaluation. Scherrer et al (2019) experimented with character-level systems for similar language translation and observed that characters outperform other segmentations for Spanish-Portuguese translation, but not for Czech-Polish. Knowles et al (2020) experimented with different subword vocabulary sizes for English-Inuktikut translation and reached the best results using a subword vocabulary of size 1k, which makes it close to the character level.…”

Why don't people use character-level machine translation?

“…This system however makes many unusual and suboptimal design choices and ended up as the last one in the manual evaluation. Scherrer et al (2019) experimented with character-level systems for similar language translation and observed that characters outperform other segmentations for Spanish-Portuguese translation, but not Czech-Polish. Knowles et al (2020) experimented with differ-ent subword vocabulary sizes for English-Inuktikut translation and reached the best results used a subword vocabulary of size 1k, which makes it close to the character level.…”