Monday, 23 April 2012

Baboons can read!

Or rather, baboons can learn to recognise words and then learn the frequency of particular letters in them to predict if a new item is a word or not.

What they did was let baboons play with a computer in return for treats (so they could choose not to play at all if they didn't want to, or play a lot if they were greedy buggers.). They were presented with a whole load of words, all four letters long, all with a vowel and three consonants (as far as I can tell - I don't have access to the original paper). Some of the words were real, and some were non-words like virt or dran. The non-words were, I think, possible words (that is, they conform to the rules of the phonology of the language (phonotactic constraints). I don't know which language - the study was done in France but the phonotactic constraints of French and English are not all that different). The baboons had to press a button if they thought it was a real word, and another if they thought it wasn't. I don't think any of them did any better than chance at first, but they were given a treat if they guessed right until they had seen the words enough times to remember them and guess right 80% of the time. 

This is Dan the Baboon, class swot. He remembered 300 words:
Dan the Baboon
Note that this isn't reading - that means identifying a word as a word (connecting it with a sound and/or meaning). This is just object recognition, which I think is the authors' point - that reading isn't a linguistic skill, but rather object recognition and then we have other linguistic processes going on to make the next step, which baboons don't have. Fortunately, otherwise they'd start writing us notes and making demands for better treats or more fun computer games.

Once the baboons knew a load of words, the researchers gave them new ones to look at. Now, they could guess when a word was a real word at slightly better than chance, or 60% of the time. Apparently, they were recognising letter combinations like 'th' and assuming that new words with that letter combination were also words. (Note that Language Log has comprehensively dissected and basically trashed this claim, but I don't have access to the paper so I'm going to have to take the conclusions as they say.)

So the baboons seem to be able to break the words down into their letters, or at least smaller parts. I don't know if this is a new finding. The authors say it's like how you can recognise a table you've never seen before, because you know how a table basically looks, with legs and a top and the relations between the parts.

I don't think these baboons are so clever though. I want to know how they'd do with a non-word that had the letter combination 'th' - presumably they'd wrongly think that it was a word.

They would also (I'm fairly sure) do badly if they were asked to identify possible vs impossible non-words. This is what I mentioned before: impossible words violate the phonotactic constraints of the particular language. So blod is a possible (but not real) English word, but bkod is not. The baboons would not know this (of course, as they aren't associating the words with sounds and they don't speak English (they're French baboons) and they don't have any concept of phonotactics anyway). I don't know if they could learn to recognise them either, though - I think it might be too complex and the rules just too arbitrary. Even for Dan. 

No comments:

Post a Comment