HuQ: An English-Hungarian Corpus for Quality Estimation

Quality estimation for machine translation is an important task. The standard automatic evaluation methods that use reference translations cannot perform the evaluation task well enough. These methods produce low correlation with human evaluation for English-Hungarian. Quality estimation is a new approach to solve this problem. This method is a prediction task estimating the quality of translations for which features are extracted from only the source and translated sentences. Quality estimation systems have not been implemented for Hungarian before, thus there is no such training corpus either. In this study, we created a dataset to build quality estimation models for English-Hungarian. We also did experiments to optimize the quality estimation system for Hungarian. In the optimization task we did research in the field of feature engineering and feature selection. We created optimized feature sets, which produced better results than the baseline feature set.