A Multifaceted Evaluation of Neural versus Phrase-Based Machine Translation for 9 Language Directions
This provides a detailed evaluation of neural vs. phrase-based machine translation for researchers and practitioners, though it is incremental as it builds on existing paradigms.
The study compared neural and phrase-based machine translation across 9 language directions, finding that neural systems produce more fluent, accurate translations with better word order and inflection, but struggle with very long sentences.
We aim to shed light on the strengths and weaknesses of the newly introduced neural machine translation paradigm. To that end, we conduct a multifaceted evaluation in which we compare outputs produced by state-of-the-art neural machine translation and phrase-based machine translation systems for 9 language directions across a number of dimensions. Specifically, we measure the similarity of the outputs, their fluency and amount of reordering, the effect of sentence length and performance across different error categories. We find out that translations produced by neural machine translation systems are considerably different, more fluent and more accurate in terms of word order compared to those produced by phrase-based systems. Neural machine translation systems are also more accurate at producing inflected forms, but they perform poorly when translating very long sentences.