An Exploratory Analysis of Multilingual Word-Level Quality Estimation with Cross-Lingual Transformers
This reduces the need for labelled data and maintenance costs for QE models, making them more practical for real-world machine translation applications, though it is incremental as it builds on existing transformer methods.
The paper tackled the problem of high cost and data requirements for language-specific word-level Quality Estimation (QE) models by exploring multilingual approaches, showing they perform on par with current models and enable accurate zero-shot and few-shot predictions for new language pairs.
Most studies on word-level Quality Estimation (QE) of machine translation focus on language-specific models. The obvious disadvantages of these approaches are the need for labelled data for each language pair and the high cost required to maintain several language-specific models. To overcome these problems, we explore different approaches to multilingual, word-level QE. We show that these QE models perform on par with the current language-specific models. In the cases of zero-shot and few-shot QE, we demonstrate that it is possible to accurately predict word-level quality for any given new language pair from models trained on other language pairs. Our findings suggest that the word-level QE models based on powerful pre-trained transformers that we propose in this paper generalise well across languages, making them more useful in real-world scenarios.