We conduct in this work an evaluation study comparing offline and online neural machine translation architectures. Two sequence-to-sequence models: convolutional Pervasive Attention (Elbayad et al. 2018) and attention-based Transformer (Vaswani et al. 2017) are considered. We investigate, for both architectures, the impact of online decoding constraints on the translation quality through a carefully designed human evaluation on English-German and German-English language pairs, the latter being particularly sensitive to latency constraints. The evaluation results allow us to identify the strengths and shortcomings of each model when we shift to the online setup.
Neural machine translation (NMT) was shown to produce more fluent output than phrase-based statistical (PBMT) and rule-based machine translation (RBMT). However, improved fluency makes it more difficult for post editors to identify and correct adequacy errors, because unlike RBMT and SMT, in NMT adequacy errors are frequently not anticipated by fluency errors. Omissions and additions of content in otherwise flawlessly fluent NMT output are the most prominent types of such adequacy errors, which can only be detected with reference to source texts. This contribution explores the degree of semantic similarity between source texts, NMT output and post edited output. In this way, computational semantic similarity scores (cosine similarity) are related to human quality judgments. The analyses are based on publicly available NMT post editing data annotated for errors in three language pairs (EN-DE, EN-LV, EN-HR) with the Multidimensional Quality Metrics (MQM). Methodologically, this contribution tests whether cross-language aligned word embeddings as the sole source of semantic information mirror human error annotation.
Despite the growing importance of data in translation, there is no data repository that equally meets the requirements of translation industry and academia alike. Therefore, we plan to develop a freely available, multilingual and expandable bank of translations and their source texts aligned at the sentence level. Special emphasis will be placed on the labelling of metadata that precisely describe the relations between translated texts and their originals. This metadata-centric approach gives users the opportunity to compile and download custom corpora on demand. Such a general-purpose data repository may help to bridge the gap between translation theory and the language industry, including translation technology providers and NLP.