In an OCR + MT experiment, Quest Visual Word Lens says that Grilled Sausages -> A LA PARILLA SALCHICHAS

One of the more interesting developments at the guided tours business over the past few years has been an increase in the proportion of weird walks sold as younger customers have started using decent, cheap handheld-hosted apps to cover the basic "¡Look! ¡Ze catedral!" legacy (OK, zombie) guide territory.

Real-time optical character recognition with machine translation - already available in various guises in military environments - is an obvious complement to electronic guides of this type, and yesterday a brief burst of interest greeted an iPhone app which purports to offer hyperreal functionality of this type.

The simple, contrastive graphics used for Quest Visual's YouTube Word Lens English ⇄ Spanish demo seem to indicate that it can manage only very basic OCR, but the results of its MT do look rather impressive:



Unfortunately, rudimentary investigation suggests that this particular horse may have been nobbled, and that the video does not give a fair picture of the capabilities of the tool. As Hans Klis observes, translation is word-for-word, so that, for example, where Google Translate correctly renders grilled sausages as salchichas a la parrilla, etc etc, Word Lens fails at the most basic of syntax hurdles and serves up a la parrilla salchichas and francés patatas fritas.

Someone - probably Google - will surely do this very well, very soon. But official guides can breathe a sigh of relief this Christmas, and caveat emptor continues to be the first rule of the iPhone Apps jungle.

Labels: