The Perils of Computer Language Recognition

This hilarious video demonstrates the perils of voice recognition. And I have an Irish friend who has “trained” his computer to use voice recognition – undoubtedly a security asset as nobody else will be able to use his system.

But there are also dangers involved in simple language text translation, as the following illustrates.

As Emeritus Professor at the University of Paris, I occasionally lunch at the cafeteria there. Before going, I check the online menu. This particular Friday, when there is usually fish, the French language menu was out of order, but they directed users to the translated, English version. The fish option read “Net of Place”. Well, that sounds right. “Net” is certainly a fishy sort of term for a menu, redolent of a catch at sea; and “place” is certainly a flatfish. But wait a minute; surely the fish is spelt “plaice”; and whatever their faults, computers do NOT make spelling mistakes.

Intrigued, I set out for the restaurant, to check what exactly was on the menu. It turned out that the fish option was “filet de lieu”. Now, the French word “filet” means fillet or net; and the poor computer had chosen the wrong option. Similarly, “lieu” is French for “place” or the fish “coley”, similar to cod, but “fillet of coley” was really not to my taste.

So I departed, a wiser, if hungrier, man.