Evolving Language Experiment

[Terra]

I recently watched a documentary on language-learning, and was intrigued by an experiment it included: http://www.youtube.com/watch?feature=pl ... j8Y#t=114s (It lasts about 5 minutes.)

To summarize, they create a language for categorizing made-up fruits by type/shape, color, and number. At first, the word assigned to each of the 27 (3x3x3) possible fruits is completely random. Then, a tester is shown the word and its corresponding fruit for only *some* of the fruits. They're then asked to recall the words for certain fruits that they just saw, but also the words for fruits that they didn't see. Then, this process is repeated with a new tester, except this time the word for each fruit is replaced by the new, evolved word that the previous tester gave, whether it's the same as the one before it or not. Within 9 generations (Each generation is a single person, I assume.), a somewhat-regular pattern emerges that describes the fruits' properties.

1) Has anyone here read/seen/performed such an experiment before?
2) Here's the guy's site for publications: http://www.lel.ed.ac.uk/~simon/publications.html . I assume that the paper about this experiment is this one: "Complex Adaptive Systems and the Origins of Adaptive Structure: What Experiments Can Tell Us" (http://onlinelibrary.wiley.com/doi/10.1 ... x/abstract), but have no idea how I could get access to it.

[Kezdő]

Well, that was probably the most intriguing and thought-provoking thing I've seen in a while. We subconsciously structure things, make order from chaos. I really had no idea! Thanks very much for putting that up here.

As for your questions, the closest things I can remember are some experiments on synesthesia (which were quite far-removed anyway). I think you have to create an account with the library to gain access.

[brandrinn]

We should run a similar experiment on the ZBB.

[Terra]

I think you have to create an account with the library to gain access.

Well of course, but it costs money.

Anyways, I found the paper: http://www.lel.ed.ac.uk/~monica/LACAS-final-resub.pdf But, it seems to be about a different instance of this experiment, with differences like motion replacing number as a category, and the colors and shapes involved.

Also, I expect/wonder about 1st-language interference. "r-" seemed to become to mean "red", and "l-" "blue".

We should run a similar experiment on the ZBB.

I don't think that the ZBB would provide an unbiased audience, since they're/we're knowledgeable about linguistics, though I'm really tempted to try anyways.

Doing the experiment over the internet would also require the participant to be trusted more than they would in a controllable setting; They'd have to be trusted to neither review the words presented to them, nor record any notes about them.

[brandrinn]

We should run a similar experiment on the ZBB.

I don't think that the ZBB would provide an unbiased audience, since they're/we're knowledgeable about linguistics, though I'm really tempted to try anyways.

Doing the experiment over the internet would also require the participant to be trusted more than they would in a controllable setting; They'd have to be trusted to neither review the words presented to them, nor record any notes about them.

I think it would work. We know about linguistics, but we don't know what other people will name the fruit. The patterns will still emerge organically and not by conscious planning. And how could we review anything? It's not like the other players' answers are written down in a central place.

In fact, I'm going to give it a try. I'll make a new thread about it if you don't want me to hijack yours.

[Terra]

I think it would work. We know about linguistics, but we don't know what other people will name the fruit. The patterns will still emerge organically and not by conscious planning.

Maybe. The participants in the study were not aware that the experiment was about creating patterns for an evolving language; They thought that it was about simply memorizing the words of a language.

And how could we review anything? It's not like the other players' answers are written down in a central place.

During "training" (showing the participants half of the fruits with their corresponding words), each pair was shown the same amount of time, and presumably only once (The paper doesn't say exactly how many times.).

In fact, I'm going to give it a try. I'll make a new thread about it if you don't want me to hijack yours.

Actually, I think that it'd be neat to do the experiment multiple times, but with the same initial set, to compare how they evolve separately. Perhaps a bouba-kiki effect might become apparent. Do you want to try such a thing?

I'm also interested in what would happen if people didn't perform the experiment alone, but in groups. Would it promote regularity or not?

I've already got some things ready. I've got a bunch of randomly generated words, and images to go with them. Because fruits are hard to draw, I settled for objects with 3 properties: shape, color, and number.

I've only to decide how to actually perform the test. Perhaps, first have a training folder with half of the complete pairs, then a test folder with all the objects, but no words. Still, it relies upon the participant to properly train, and then label.

[brandrinn]

I tell you what. I'll just run a test experiment in another thread, and based on how well that goes, you can decide how to run the proper experiment.

[Terra]

Alright, sounds good.