The paper ‘SurreyAI 2023 Submission for the Quality Estimation Shared Task’ discusses the use of Quality Estimation (QE) systems in machine translation. The SurreyAI team adopted an approach that builds upon the TransQuest framework, exploring various autoencoder pre-trained language models within the MonoTransQuest architecture. The models used are XLMV, InfoXLM-large, and XLMR-large. The study evaluates the relationship between machine-predicted quality scores and human judgments for 5 language pairs. The approach using MonoTQ-InfoXLM-large emerged as a robust strategy, surpassing all other models in the study by significantly improving over the baseline for the majority of the language pairs.
Publication date: 4 Dec 2023
Project Page: surrey.ac.uk
Paper: https://arxiv.org/pdf/2312.00525
