Even though I presented my poster today at the university's small poster conference, only a few people actually read it through (I shouldn't have put mathematical equations at the beginning of the poster because apparently it scared people away...). Anyway, I make it public here in case it might be of interest to someone.
Let's imagine that there are two talkers A and B. In describing dative events (i.e., events where things go from one person to another), A always uses DO construction (e.g., The boy gave the girl an apple), and B always uses PO construction (e.g., The boy gave an apple to the girl).
Previous studies have repeatedly shown that people tend to syntactically align with the conversation partner. In other words, people tend to use more DOs when talking with A and more POs when talking with B, even without realization. This is called structural priming. This can be regarded as adaptation to the syntactic environment that each talker provides.
Imagine a situation where people are sequantially faced with A and B. They should first align with A and produces more DOs by creating a model that strongly favors DO construction. But what happens if they are successively faced with B who only produces POs? Do they adjust the previous model or do they create a whole new model?
If people adjust the previous model and try to adapt to the new environment, they should mix the statistical information each talker provides. This can be called talker-general adaptation because they do not distinguish between the two talkers in terms of adaptation.
If, on the other hand, people create a whole new model and try to adapt to the new talker they should separately maintain the two kinds of statistical information. This can be called talker-specific adaptation because they distinguish between the two talkers.
My prediction is, the specificity of the adaptation will be affected by the length of exposure to the first talker (e.g., A), with longer exposure being more likely to induce talker-specific adaptaiton.
The reason is that longer exposure to one statistical environment (e.g., a completely DO-biased environment) makes the model more sensitive to deviations (e.g., POs), and in response to the deviation they should be more likely to put aside the previous model and create a new one.
If true, it has significance in bilingualism. If second language acqusition is considered as a kind of adaptation to a new statistical environment, it is similar to the adaptation to B in this context.
If less exposure to the first environment leads to talker-general adaptation (mixing of statistical information) and more exposure leads to talker-specific adaptation (separation of statistical information), that imples that child second language (L2) learners tend to integrate the two languages while adult L2 leaneres tend to separate them.
Integration of representations could be very helpful because you do not have to inhibit one representation to use another. You would not suffer from forgetting caused by this kind of inhibition. You would treat the second language as if it was the first language.
The last part is only speculative but very interesting to me. If you have any comments please don't hesitate to give.
If you are interested in the mathematical detail please look at the original poster's Hypothesis part. Also if you are interested in the methodological scheme please refer to its Experiment part.