How to Move to Neural Machine Translation for Enterprise-Scale Programs - An Early Adoption Case Study

How to Move to Neural Machine Translation for Enterprise-Scale Programs—An Early Adoption Case Study Tanja Schmidt Welocalize, Inc. Frederick, MD, United States tanja.schmidt@welocalize.com Lena Marg Welocalize, Inc. Frederick, MD, United States lena.marg@welocalize.com Abstract While Neural Machine Translation (NMT) technology has been around for a few years now in research and develop- ment, it is still in its infancy when it comes to customization readiness and experience with implementation on an enterprise scale with Language Service Providers (LSPs). For large, multi- language LSPs, it is therefore not only important to stay up-to-date on latest re- search on the technology as such, the best use cases, as well as main advantages and disadvantages. Moreover, due to this in- fancy, the challenges encountered during an early adoption of the technology in an enterprise-scale translation program are of a very practical and concrete nature and range from the quality of the NMT output over availability of language pairs in (customizable) NMT systems to addi- tional translation workflow investments and considerations with regard to involv- ing the supply chain. In an attempt to outline the above chal- lenges and possible approaches to over- come them, this paper describes the mi- gration of an established enterprise-scale machine translation program of 28 lan- guage pairs with post-editing from a Sta- tistical Machine Translation (SMT) setup to NMT. 1 Introduction The idea of using recurrent neural networks for machine translation was first presented by Kalchbrenner and Blunsom (2013), followed soon after by Cho et al. (2014) and Sutskever et al. (2014). In a mere three years from those papers, NMT systems were outperforming SMT systems for several translation tasks at the Association for Computational Linguists’ Conference on Machine Translation (WMT). At the same time, large translation providers such as Systran, Google and Microsoft announced deployments of NMT systems for public consumption. The combination of these factors quickly made both buyers and providers of translation services aware of the new opportunities. The rapid emergence of NMT has necessitated that LSPs focus on many new areas, including: qualitative evaluation of individual NMT systems, comparing translation quality and productivity of NMT and SMT systems, implementation and deployment of NMT systems, and building customized NMT systems for specific domains and/or clients. 2 Contextualization Since the deployment of machine translation technology for commercial use, and especially the breakthrough of Statistical MT solutions, requests for MT as part of regular translation programs have constantly been on the rise. An explosion in the amount of content published as well as increasing pressure to publish content fast and simultaneously in different target markets and languages have caused clients to look into alternative, cheaper options and LSPs to adjust their translation workflows and processes. The continually improving quality of MT systems and new developments such as NMT add to this demand. As a major global LSP, we count a range of big global companies among our end clients, for whom we typically provide ongoing, on-demand translation services into 20+ languages, covering various content types (= enterprise-scale translation …

Content truncated to 500 words