WEBTRAN: a controlled language machine translation system for building multilingual services on Internet.
Aarno Lehtola, Jarno Tenni, Catherine Bounsaythip, Kristiina Jaaranen · 1999
This paper presents how to embed a fully automatic controlled language (CL) translation into online services on internet. CLs are natural sublanguages with specific domains, limited vocabulary, restricted syntax, and minimised ambiguities. The CLs enable practical and costeffective language processing solutions, e.g., for machine translation and text mining. We have designed a generic CL translation software, Webtran, which is intended for building multilingual services on Internet. We describe the Augmented Lexical Entries formalism provided for a controlled language definition. These entries can be simple, single or multilingual, word or idiom descriptions. They may define dependency relations, control semantic admissibility, specify language repair, and even depict high-level document syntax. Webtran includes a language-modelling tool which supports the ALE-formalism and implements human-assisted machine learning to intensify the modelling. We present our experiences of adapting Webtran in translating product articles in an online catalogue from Swedish to Finnish. Regardless of the test languages, the Webtran software is in principle language independent, and tentative tests have been carried out with Estonian, French and English as targets. We also discuss here the embedding of Webtran to the overall catalogue production process. 1