How Machine Translation Works

A metaphor for the machine translation process

Imagine this: you are sitting in a room. It is sparsely furnished, with nothing but a desk, a chair, and an enormous bookcase. The shelves are packed with books, but every book has exactly the same cover, making them impossible to tell apart. You are good at your job, though, and know exactly where to find everything.

Every page of every book is divided into two columns: squiggles on the left and scribbles on the right. You have no idea what any of them mean. All you know is that there must be some connection between the squiggles and the scribbles.

The room also has two small windows. Every so often, a sheet of paper covered in squiggles drops through the window on the left and lands on your desk. Your job is to search the books in your library for the scribbles that correspond to those squiggles.

You are a professional. One glance at a sequence of squiggles, and you instantly remember which page of which book contained something similar. You take the right book from the shelf and carefully copy the corresponding scribbles onto a separate sheet of paper. Little by little, you fill the page. Once it is complete, you toss it through the other window.

Sometimes you receive squiggles that do not appear anywhere in the books. At other times, the same squiggles correspond to several different scribbles. When that happens, you choose whichever scribbles you like best.

You have no idea what the books are about. You do not understand what the squiggles mean. You do not know whether the scribbles you so carefully copy make any sense—or whether they are of any use to anyone.

But none of that really matters to you.

You have just read a brief metaphorical explanation of how machine translation works.