To what extent do English words sound like what they describe?

This is an ancient question with much attendant scholarship rejoicing in the name of 'phonosemantics', qv http://en.wikipedia.org/wiki/Sound_symbolism

Friendly introdution with bouba and kiki here: http://www.visiblemantra.org/phonosemantics.html

This is a very difficult (if not impossible) question to answer, since testing whether a word "sounds" like a concept is largely subjective. It is further complicated by the fact that humans will tend to "chunk" the memory of an auditory stimulus of a word with its associated visual stimuli. In other words, when one hears the word "book", one is very likely to simultaneously, automatically, recall the memory of the sight of a book. Therefore, one might be biased to think a word is onomatopoeic simply because one subconsciously already has an association between the sound of the word and its associated concept.

The easiest way I can think of to non-subjectively test the theory you propose is to take a group of non-English speakers (preferably those who don't even speak an Indo-European language), present them with English words and pictures of the concepts the words represent, and then ask them to match the English words with their associated concepts. I am not aware of any such experiment that has actually been performed, however.

This is further complicated if we accept that language evolution exists on a continuum: The current meaning of a word in modern English can have little or nothing to do with its earlier meanings. For example, about 1000 years ago, the word "book" (bóc) had a meaning more similar to "writing tablet" or "single sheet of paper", the latter of which doesn't have a "hard quality" at all.

Włodzimierz Sobkowiak did some research on this in the 1990s, studying English onomatopoeia's phonostatistics (the paper is available here). He took the words in Kloe's dictionary on onomotopoeia and statistically compared their phonetics to the rest of the English lexicon. Sobkowiak concluded that the words in the dictionary (that were specifically identified as being onomatopoetic) had statistically significanty different phonetic patterns than the rest of English, suggesting that not all English words are onomatopoetic.

Once you exclude onomatopoeia, the theory makes little sense.

Do the sounds things make influence the words that are used for them? Clearly this happens to some extent, as the cuckoo demonstrates, but this is onomatopoeia and we must discard it from discussion. Adjectives that sound like the sounds they describe? Again, we must discard them.

With what remains, you are asking if the tonal qualities of the words have a tendency to match the nature of the thing the word stands for. That relies heavily on the meaning that you attach to tonal qualities, and I would suggest that those meanings are likely to have formed the other way around; for example, lethargic has a heavy, slow feeling to it because of its meaning, instead of meaning something slothful because of the way it "feels".

Synonyms are another good way of testing the theory past breaking point. "Amanuensis" has a very rounded, flowing sound to it; "Secretary" is full of stops and sharp edges. The emotional quality of the tones are very different, but the words mean the same.

In short, I don't think this theory holds much water.

My friend in college actually did an undergrad linguistics capstone project on this exact topic. First, he found a bunch of proposed sound/meaning correspondences that had been put forth by people ranging from Aristotle (who claimed that a 't' sound symbolizes something standing still) to modern researchers (who have for example proposed that front vowels like the 'i' in "him" correspond to small things, while back vowels like the 'a' in "father" correspond to big things, reflecting the fact that the front vowels resonate in a smaller space in the mouth - and my friend sarcastically said "This is of course why we have the words 'big' and 'small'"). Then he used a program to make up a bunch of words that sounded nothing like any real (English) words, but exemplified some of the proposed sound/meaning correspondences. And then he had (natively English-speaking) people choose from among several possible definitions for the made-up words. One definition for each pseudoword was the target definition based on the sound/meaning correspondence.

His finding? People had no tendency whatsoever to pick the "right" definition for his made-up words. He concluded that ordinary words (as against onomatopoeias) are in fact pretty much completely arbitrary, as far as his data could allow him to say. He further posited that the reason this theory has stuck around for so long is just that it sounds so good to people, probably for the reasons mentioned by user1579. I was convinced, even though before his presentation I would've thought the opposite. (I had heard about kiki and bouba, but as Neil Coffey points out, it's important not to read too much into it.)