Alibaba-Translate China's Submission for WMT 2022 Metrics Shared Task
This is an incremental improvement for machine translation evaluation researchers, focusing on adapting an existing method to a new competition.
The paper tackles the problem of unified translation evaluation by building a system based on UNITE for the WMT 2022 Metrics Shared Task, using pseudo-labeled data pre-training with data cropping and score normalization, and fine-tuning with DA and MQM data, but no concrete results or numbers are provided.
In this report, we present our submission to the WMT 2022 Metrics Shared Task. We build our system based on the core idea of UNITE (Unified Translation Evaluation), which unifies source-only, reference-only, and source-reference-combined evaluation scenarios into one single model. Specifically, during the model pre-training phase, we first apply the pseudo-labeled data examples to continuously pre-train UNITE. Notably, to reduce the gap between pre-training and fine-tuning, we use data cropping and a ranking-based score normalization strategy. During the fine-tuning phase, we use both Direct Assessment (DA) and Multidimensional Quality Metrics (MQM) data from past years' WMT competitions. Specially, we collect the results from models with different pre-trained language model backbones, and use different ensembling strategies for involved translation directions.