Probabilstic Models of learning

In Barner 2011 the study design showed that even when “only” was added, which forces grammatical strengthening (disambiguating whether the soft or hard alternative is implied), children failed to catch false statements with context independent scales like “some”, but caught false statements made with context dependent scales like “two”. Barner et al argues this evidence is consistent with an inability to generate relevant scalar alternatives for context independent scales. 

In investigating the origins of scalar implicatures Stiller 2011 conducted 3 studies whose findings support a statistical linguistic account of scalar pragmatics. In reference to their third experiment that manipulates informativeness of features by showing figures beforehand and showed how rarity of afeature is correlated with informativeness they argue, "a theory of implicature must integrate finegrained statistical information about such shared context." This is in contrast to the counterfactual and simple linguistic alternatives theory.

The conclusion made by Stiller reminds me of a theory of learning we've discussed in SymSys 1. It says that learning and decision making is all probabilistic, that we're always making new judgements and decisions based on past evidence. It's a convincing argument to me that statistical analysis plays a large, unconscious role in almost everything we do, including the generation of scalar alternatives. And I wonder how many more fields of human behavior related science are beginning to incorporate an account of probability.