Towards a User-Friendly Webservice Architecture for Statistical Machine Translation in the PANACEA project

Towards a User-Friendly Webservice Architecture for Statistical Machine Translation in the PANACEA project∗ Antonio Toral, Pavel Pecina, Andy Way School of Computing Dublin City University Dublin, Ireland {atoral,ppecina,away}@computing.dcu.ie Marc Poch IULA Universitat Pompeu Fabra Barcelona, Spain marc.pochriera@upf.edu Abstract This paper presents a webservice archi- tecture for Statistical Machine Translation aimed at non-technical users. A workflow editor allows a user to combine different webservices using a graphical user inter- face. In the current state of this project, the webservices have been implemented for a range of sentential and sub-sentential aligners. The advantage of a common in- terface and a common data format allows the user to build workflows exchanging different aligners. 1 Introduction A human translator willing to use a state-of-the- -art Statistical Machine Translation (SMT) system has two main options nowadays: to use an on-line translation system (e.g. Google Translate1) or to train his or her own MT system from scratch (e.g. Moses (Koehn et al., 2007)) using any available parallel corpus (e.g. Europarl2). Both options, however, present some important drawbacks. On one hand, the on-line systems are not customisable and might be inadequate if the user wants to trans- late sensitive data. On the other hand, installing an MT system with all its dependencies and train- ing it requires technical expertise which is usually beyond the competences of an average translator. This paper presents an architecture for building SMT systems where each component is deployed ∗We would like to thank the developers of Soaplab and Tav- erna for solving our questions and requests. This research has been partially funded by the EU project PANACEA (7FP- ITC-248064). ∗c⃝ 2011 European Association for Machine Translation. 1http://translate.google.com/ 2http://www.statmt.org/europarl/ as a webservice. This way the user does not have to deal with technical issues regarding the tools, such as their installation, configuration or maintenance. A workflow editor allows the user to combine the different webservices using a graphical user inter- face. In the current state of this project, webser- vices have been created for a range of sentential and sub-sentential aligners. This work is part of the FP7 PANACEA project,3 which addresses the most critical aspect of MT: the language-resource bottleneck. Its ob- jective is to build a factory of language resources that automates the stages involved in the acquisi- tion, production, updating and maintenance of lan- guage resources required by MT systems. This is done by creating a platform, designed as a dedi- cated workflow manager, for the composition of a number of processes for LR production, based on combinations of different web services. The tech- niques developed are language independent, while the use case language-pairs are English–French, English–German and English–Greek. The rest of the document is structured as fol- lows. The following section details the implemen- tation of webservices for each of the aligners con- sidered. In addition, we discuss the procedures de- veloped to convert the format of these aligners to a common format and present a set of workflows that demonstrate the usage of the aligners in pro- cessing …

Content truncated to 500 words