People who think about language talk about how the sounds of words bear no relation to objects in the world. Indeed, this has been a leading assumption in much of modern literary criticism, philosophy, and even linguistics.

Not necessarily so, says Brent Berlin, an anthropologist at the University of Georgia in Athens. Berlin suspected that there was more than an arbitrary connection between word sounds and the physical characteristics of objects being described; he set up an experiment to test his hypothesis.

First, he examined the words for two animals, the tapir and the squirrel, in 19 South American Indian languages. He was searching for similarities in sound-pattern. In 14 of those languages, the tapir--which is a big, slow-moving beast--was given a name with the sound "aah," whereas the small, quick squirrel was given a name based on the sound "ee."

Next, to see if these sound-meanings could be generalized, Berlin read the unfamiliar words to a group of English speaking test subjects, asking them to guess which word meant "squirrel." He reasoned that if word and object are arbitrarily connected by language, the result should be random; sometimes right, sometimes wrong. In fact he found a greater-than-chance-level number of correct guesses.

Something else interesting showed up: when he used words from the five languages that didn't fit the original "aah"-"ee" pattern, the subjects' responses were indeed random--unless the "ee" sound happened to be present, in which case they tended to guess "squirrel."