Quality and Coverage: The AFRL Submission to the WMT19 Parallel Corpus Filtering For Low-Resource Conditions Task

The WMT19 Parallel Corpus Filtering For Low-Resource Conditions Task aims to test various methods of filtering noisy parallel corpora, to make them useful for training machine translation systems. This year the noisy corpora are from the relatively low-resource language pairs of English-Nepali and English-Sinhala. This papers describes the Air Force Research Laboratory (AFRL) submissions, including preprocessing methods and scoring metrics. Numerical results indicate a benefit over baseline and the relative effects of different options.