Machine Translation and User-Generated Content

Zia Daniell Wigder | December 20, 2007, 10:58 AM

There have been numerous comparisons of different machine translation (MT) services – today’s assessment in the Wall Street Journal was of particular interest to me as it compared the Arabic capabilities of four MT providers: Google, Applied Language, WorldLingo and Systran. Since much of my free time is devoted to organizing interactive dialogues with the Middle East, I’ve been actively following the evolution of Arabic translation tools, especially as an increasing number of MT providers have tackled this language (by contrast, I’m still waiting for a free Farsi-to-English translator).

The WSJ article comes to the same conclusion as many other assessments: the tools can be helpful, but clearly aren’t ready for prime time. Indeed, in conversations I’ve had with companies about the use of MT for interactive dialogue, they’ve all highlighted this area as one of MT’s weakest: the colloquialisms used in dialogue and disjointed nature of most conversations make it almost impossible to apply MT, which performs best with highly structured text.

MT’s poor performance when it comes to unstructured text means this option may not yet suffice for global companies hoping to use MT to translate large volumes of user-generated content (UGC). Companies are aware that UGC is an increasingly valuable tool: over a third of all online users find ratings and reviews from other consumers helpful during online shopping, for example. Yet to translate a large volume of content through professional translators is prohibitively expensive, particularly when the precise ROI of user-generated content remains unclear.

Companies aiming to expand their UGC offerings into new international markets are therefore facing a variety of different options, none of which provide a silver bullet. They can roll out a framework for international UGC to develop organically, much in the way YouTube did in Europe – such an option relies on having (or building rapidly) an engaged base of users ready to contribute content in every market. Other companies have experimented with translating a handful of consumer entries professionally to seed new international offerings with content. Still others are eyeing community-based translations as an alternative to pricey professional versions.

Our first report of 2008 will address this issue of user-generated content in international markets: the report will help companies determine which strategies make the most sense given the particular nature of their content.

Until then, happy holidays and best wishes for a wonderful new year!