Pangeanic travels to Japan to present neural networks research and Cor technologies at the next TAUS Technology event: Tokyo
Pangeanic has been working in the integration of two research projects over the last year and a half. One is the product of the EU’s EXPERT project, which gave birth to ActivaTM, a matrix-like, scalable, infinite language database. Based on Elastic Search, it incorporates many features that traditional translation memories (desktop or server) cannot. The project also added very strong customization and features specifically useful for translation companies in terms of plugin compatibility with CAT tools, artificial intelligence (deep learning) applied to tag handling in order to solve interoperability issues, etc. Cor is the result of national research in Spain, co-funded by EU FEDER funds. The project has now grown into a fully operational platform designed to disrupt traditional LSP business and the way translation is delivered to the industry. Cor is a very advanced software platform that automates translation processes by developing new theorems, algorithms, interfaces and applications providing added value for both professional and end-users. Cor automates order processing, document analysis and also website analysis in one place, safe delivering translated content without interfering with an existing website’s structure. It does not provide messy proxy translation where users loose control over their hosting and content. Website owners can create content whilst Cor takes it up and translates it making use of ActivaTM and statistical or neural machine translation in the background.
TAUS program has published an abridged version of Manuel’s presentation in Tokyo: “Neural research has taken the MT community like a storm and it looks the game changer for the foreseeable future. Although the basis for neural developments have been around for several years, it was only the discovery of GPU power, almost by chance that made it possible. Pangeanic has developed two technologies for the translation community: one is the result of a national R&D project. Cor not only can crawl, detect new or deleted content on a website, it can also pull the content out in a translation-friendly format. Elastic Search-based ActivaTM is the result of a EU project and it interfaces with Cor as a CAT-tool independent TM, resolving some interoperability issues thanks to machine learning for tag and in-line handling. Finally, our own NMT results using TAUS data help translators do the job. This presentation will also compare our experiences between SMT and NMT, with open questions on hybridization for the future.”