On Edge

[Balthasar Bickel]

Dear all

In response to Frans’s worries about the visibility of universals research, I guess it is fair to say that universals in the classical sense are somewhat in a crisis. There are reasons:
- absolute universals can’t be established by samples (see a nice recent paper by Piantadosi & Gibson in Cogn. Science), especially as we can’t be sure about the population from which our samples are drawn (e.g. what survived the human populations bottlenecks 20-60ky ago?) 
- absolute universals can be established by fighting counterexamples (re-analyzing them, explaining them away, adding additional conditions etc), but that has always had a touch of arbitrariness
- absolute universals can be established by considerations of explanatory adequacy à la Chomsky, but that has become ever more problematic given recent demonstrations of learnability of CFGs (e.g. Perfors et al. in Cognition 2011) and given Chomsky's retreat to recursion in the mathematical sense, i.e. something which characterizes computation in far more cognitive domains than language and which allows for a vast array of formulating specific grammars and thereby for an equally vast array of possible proposals for formal universals.
- statistical universals can be established based on samples, but there is a serious risk of formulating spurious correlations (see a nice study here by Roberts & Winters in PloS One 2013). It’s not enough to develop just-so stories and post-hoc explanations of the generalizations one happens to find in a sample. (Perhaps indeed many of us had had enough of this, and this might explain why we now see less proposals of implicational universals and the like than in the last century.) 

What is needed is detailed research on *causal* theories that predict specific systematic biases in how languages change, regardless of time and space, and robust quantitative methods that are able to capture such biases. No easy task, and a task that obviously requires more than linguistics. (I guess I agree on this with Wolfgang Schulze.)

Balthasar.


PS to avoid misunderstandings: I am not saying there can’t be solid universals like ‘languages don’t count’, to pick one of David’s examples. I just think we should do better than proposing descriptive generalizations. (But I am not going to enter the debate about science :-)


-----------------------------------------
http://www.comparativelinguistics.uzh.ch/bickel_en.html