MTUOC: easy and free integration of NMT systems in professional translation environments

MTUOC: easy and free integration of NMT systems in professional translation environments Antoni Oliver Universitat Oberta de Catalunya aoliverg@uoc.edu Abstract In this paper the MTUOC project, aim- ing to provide an easy integration of neu- ral and statistical machine translation sys- tems, is presented. Almost all the required software to train and use neural and sta- tistical MT systems is released under free licences. However, their use is not al- ways easy and intuitive and medium-high specialized skills are required. MTUOC project provides simplified scripts for pre- processing and training MT systems, and a server and client for easy use of the trained systems. The server is compatible with popular CAT tools for a seamless integra- tion. The project also distributes some free engines. 1 Introduction MTUOC is a project from the Arts and Humanities department at the Universitat Oberta de Catalunya (UOC) to facilitate the use and integration of neu- ral and statistical machine translation systems. Most of the software needed for training and using such systems is distributed under free per- missive licences. So this technology is, in princi- ple, freely available for any professional, company or organization. The use of MT toolkits presents some problems: • Technological skills: medium-high techno- logical skills are required. Knowledge of some programming (as Python, for exam- ple) and scripting (as Bash, for example) lan- guages are necessary. On the other hand, the c⃝ 2020 The authors. This article is licensed under a Creative Commons 3.0 licence, no derivative works, attribution, CC- BY-ND. documentation of these toolkits is not always detailed enough and some time in trial and er- ror is spent. • Integration: the resulting systems are not eas- ily integrable in existing workflows. Most of the toolkits provide access through some kind of API, usually using a server-client configu- ration. Some CAT tools offer plugins to ac- cess some existing systems. But not all CAT Tool - MT system combinations are available. • Hardware: relatively high hardware require- ments are present, especially for training the systems. For training SMT systems lots of RAM memory is required. For training NMT systems one or more powerful GPU units are compulsory. MTUOC tries to offer solutions for the first two problems. Regarding the technological skills prob- lem, it provides a series of easy-to-use and easy- to-understand Python and Bash scripts for corpus pre-processing and training. All these scripts are well documented and can be adapted and extended in an easy way. Regarding the integration prob- lem, a fully configurable server and client are pro- vided. The server can mimic the behaviour of sev- eral kinds of servers, so it can be used with a large range of CAT tools. For example, the server can use a Marian engine but behave as a Moses server so it can be directly integrated with OmegaT. The client can deal with several widely used file for- mats (as XLIFF, for example) and generate TMX translation memories that can be used in any CAT tool. Regarding the hardware …

Content truncated to 500 words