Why a picture beats a translation

Here is a word you do not know: lluvia. It means rain, in Spanish. You have now read that, and if you are like everybody else you will have lost it before the end of this page.

Now try it the other way. Lluvia sounds a bit like "you via". Picture a road sign that says VIA, and rain hammering down on it so hard the letters run. The sign is soaked. That is all.

Come back to this paragraph in an hour and see which one is still there.

What is actually happening

A translation pair gives your memory nothing to hold on to. Two arbitrary strings, joined by nothing except the fact that you saw them next to each other. Memory is not built for that and never has been. It is built for places, faces, stories and things that happen.

The keyword method exploits that. You take the new word, find something in a language you already speak that sounds like part of it, and then build one concrete image that ties that sound to the meaning. The sound gets you to the keyword, the image gets you from the keyword to the meaning, and the whole thing runs in about a second once it is set.

It was formalised by Atkinson and Raugh in the mid seventies with Russian vocabulary, and the effect has been replicated so many times since that the interesting question stopped being whether it works. The interesting questions now are which images work best, and why a technique this well established is so rarely built into the software people actually use.

What makes an image work

• Concrete beats abstract. A castle, not an idea of shelter. You cannot picture an idea, so an abstract image quietly becomes another word pair.
• One scene, not a story. If you have to remember a plot to get to the word, the plot is now the thing you have to remember.
• Something has to happen. A cat next to a cake is furniture. A cat sitting in the middle of the cake, licking the icing off its paws, is a scene.
• Slightly ridiculous is the point. Not for fun. An ordinary image is competing with every other ordinary image you have ever seen.
• Yours beats ours. This is the one people resist, and it is the best established finding of the lot. A hook you generate yourself is remembered better than a better one somebody hands you, because generating it is the work.

Where it stops working

It stops working for joining words, and this is worth being honest about because a lot of vocabulary apps quietly pretend otherwise. There is no picture for de, or for el, or for the difference between por and para. Those are learned by meeting them in sentences, hundreds of times, and no image will short cut it.

So the product does not try. Function words get a usage note and two examples instead of a forced picture. Pretending a picture holds "the" is the app lying about its own method, and you would find out within a week.

The half that a picture cannot do

A hook makes the first meeting stick. It does nothing about the fifth week. That is the other half of the problem and it needs a different tool, which is the subject of the next piece.