Lost en Translation

I once ran a Wordpress plugin on the Sciencebase site that used the translation engines of AltaVista’s Babelfish and Google to generate an international version of the site in any of more than a dozen languages. It worked very well, from the technical point of view. Indeed, it created a whole stack of new spiderable content and essentially multiplied the reach of the site approximately twelve times as far as the search engines were concerned.

Trouble is, after a few days, I got so tired of the complaints from native speakers of the various languages rendered pointing out how silly were many of the phrases produced by the automatic machine translation that I disabled the plugin pretty soon thereafter. Lesson learned: machine translation does not a professional-looking website make, even if it does increase the number of pages that the search engines can spider on your site.

Machine translation is fine for an on-the-fly snapshot of a general type website where technical language is minimal, of course, and many people who speak a language other than English can access sites across the globe and get an approximate idea of content using such tools.

Not for the first time, researchers in the field of business and computing recently pointed out some of the pitfalls and potentially some of the possible ways around the issues that arise when things get lost in translation.

Writing in the International Journal of Information and Operations Management Education (2007, 2, 194-210), Milam Aiken and grad student José Ablanedo of the School of Business Administration, at the University of Mississippi and Mahesh Vanjani, Associate Professor of Management Information Systems, at Texas Southern University, in Houston, point out that even natural, as opposed to machine, translation can give rise to problems.

When Parker Pen marketed a ball-point pen in Mexico, its advertisements were meant to say “It won’t leak in your pocket and embarrass you”.
However, the company mistranslated the Spanish word ‘embarazar’ (as to embarrass) and what they ended up saying was “It won’t leak in your pocket and make you pregnant”. A T-shirt maker in Miami printed shirts for the Spanish market that promoted the Pope’s visit. The shirts proclaimed “I Saw the Potato” instead of “I Saw the Pope”.

Even in that great seat of learning, Cambridge, England, mistakes are made. In an effort at inclusive multiculturalism the owner’s of a shopping mall pasted signs in a dozen languages to purportedly “welcome” shoppers to the center. Unfortunately, the Chinese characters used actually meant “adjective”, or literally “form containing word” instead of the word “welcome”.

With such embarrassments in mind, Aiken and colleagues explain that, because of the increase in global business, travel, and communication, there is a growing need to accommodate multiple languages. While human translators are still important, automatic natural language translation can increase the productivity of global travelers and office workers reading email, browsing web pages, and reviewing documents.

They compared English to Spanish translation software – Spanish Assistant version 5 (running on a Windows XP computer), and SYSTRAN (Systran, 2007) accessed via the aforementioned Alta Vista Babelfish service. They come to the perhaps rather obvious conclusion, that nevertheless ought to be emphasized, especially so for anyone considering adding a machine translation plugin to their website and/or relies on mission critical translation.

“The results show that the computer translations were much faster but less accurate. However, the trade off in speed versus accuracy might allow computers to be used for a rough, first pass, while more expensive, and time-consuming human translation can be used for more critical text.” This is basically the same conclusion at which I arrived in my international experiment on Sciencebase.