Recently, neural machine translation has revolutionised the field of machine translation, and now results in many research tasks keep improving every year. The new neural models have greatly improved translation quality, but have very different sorts of errors than the traditional statistical machine translation technology. An important challenge is to incorporate this technology improvement into commercial products and ensure that machine translation users get the best value while still keeping the product features they rely on for their work. SDL provides machine translation technology in a variety of products and markets. Our customers have expectations related to decoding speed, support for dictionaries and tags, and other functionality, so they can successfully integrate MT in their workflows. When it comes to commercialising MT, ensuring that these expectations are met is as important as improvements in BLEU score. In this talk I will focus on these important practical aspects in the context of the current NMT developments.
展开▼
