SYSTRAN Translation Stylesheets: Machine Translation driven by XSLT

Track: Case Studies, Publishing, End-User Applications

Audience Level: High Level/Technical view

Time: Thursday, November 17 14:00

Author: Pierre Senellart, SYSTRAN S.A., INRIA Futurs

Author: Jean Senellart, SYSTRAN S.A.

Keywords: XSLT, Stylesheet, Machine Translation, Publishing


We present the SYSTRAN Translation Stylesheets (STS) used by the SYSTRAN machine translation engines that power the most prominent online translation services on the Web today (including Google, Yahoo!, AltaVista's BabelFish, and AOL). SYSTRAN develops and markets the world's most widely used machine translation software which powers products and solutions for the desktop, network infrastructures and the Internet that facilitate communication in 40 language combinations and in 20 specialized domains. The choice of leading global corporations, search engines and governments, SYSTRAN's software is applied across diverse best-practice solutions for intra-company communications, content management, online customer support, eCommerce, email systems, chat, and more. SYSTRAN's expertise spans over three decades of building customized translation solutions through open and robust architectures.

XSL Transformation stylesheets are usually used to transform a document described in an XML formalism into another XML formalism, to modify an XML document, or to publish content stored into an XML document to a publishing format (XSL-FO, (X)HTML…). SYSTRAN Translation Stylesheets (STS) use XSLT to drive and control the machine translation of XML documents (native XML document formats or XML representations — such as XLIFF — of other kinds of document formats).

STS do not only provide a simple way to indicate which part of the document text is to be translated, but also enables the fine-tuning of translation, especially by using the structure of the document to help disambiguate natural language semantics and determine proper context. For instance, the phrase “Access From Front Door” is to be analyzed as “The access from front door” within a title, and as “Do access (something) from front door” in the text body. In that case, the STS would pass a title option to the translation engine. The stylesheet can activate specialized domain dictionaries for some parts of the document and can mark some expressions as not to be translated, in the same manner.

Another key application of STS is to consider machine translation as part of the authoring and publishing process: source documents can be annotated with natural language markup produced by the author, markup which will be processed by STS to improve the quality of translation, the gateway to the automatic publishing of a multilingual website from a monolingual (annotated) source. This composition, inside the publishing process, is a real breakthrough for Web content translation. Traditionally, machine translation is applied after publishing and does not have access to the original structure of the document, but only to its HTML representation.

The mechanism is implemented through XSLT extension functions. In particular, the stylesheet uses a systran:translate function to translate an XML fragment, and systran:getValue/systran:pushValue/systran:popValue functions for consulting and for setting linguistics options in the translation engine. Proper management of character properties is also provided so that, for instance, the translation of a phrase in bold font will appear in bold font, even if the phrase has moved within the translated sentence.

This process is highly customizable by the addition of new templates into the stylesheets. Because the translation is driven by the document structure, it is easier to leverage this structure during the translation process and to keep it in the translated document, than with traditional document filters, which process the entire document linearly.

STS are part of SYSTRAN's commercial version 5.0 product line and are also used for and by specific corporate-customers.