All of my cake

Both the articles assigned this week covered the same subject, from different perspectives: children "misunderstanding" the word "some". That is, both studies focused on how children performed with scalar implication. If I'd eaten all of the cake, would children in general agree with a statement that I'd eaten some of it?

Barner et. al. shows that they do, in contrast with adults, and concludes that it's because they have not had enough linguistic experience to access a set of scalar alternatives so quickly and intuitively. While their study doesn't disprove the claim that it's, because their brains, still not fully developed, do not have the processing power to do so, it weakens that claim. Stiller et. al., on the other hand, shows that children do have the processing power in an experiment where a task requiring the same processing (but with different lexical items) is completed satisfactorily. Their findings also support the counterfactual theory that a fact with more information than is given is false - if I say my friend has glasses, both adults and children would guess that it is someone with only glasses rather than someone with both glasses and a top hat. At the same time, they refute the linguistic alternatives theory that Barner et. al. assumes to an extent - that this phenomenon occurs because people refer to a list of scalar alternatives when deciding whether an utterance with scalar implication is true.

This leaves the question open - why are children unable to come to the same conclusion as adults about "some of the cake" as opposed to "all of the cake", if they clearly have the processing power and neither do adults access a list of scalar alternatives? While I have no background in either pragmatics or developmental psychology and am unqualified to answer this conclusively, there seems to still be a lack of experience that prevents the children from interpreting "some" as automatically excluding "all". As Stiller et. al. mentioned in their conclusion, the children performed on an equal level as the adults with scalar implication if by chance different lexical terms were used, and mentioned the role of world knowledge in the decision. Perhaps this could also be applied to Barner et. al.'s results. In a slightly different twist of their own conclusion, the children might simply not have an exclusive definition of "some" and "all" - not because they haven't been explicitly taught that set of scales, but because they have not seen as many circumstances where a differentiation would be necessary, and - as the authors mentioned - they have not been taught the difference.

Interestingly enough, this particular scenario appears to be slightly useless with regards to application to machine learning. Contrary to humans, computers are usually programmed to find all the occurrences of something, and usually will only show some of those occurrences due to limitations in the interface (for example, Google will not show you all of the billions of webpages in a search on one page.) Furthermore, if absolutely necessary, it is very easy to hard-code the differences using logic gates, and it may often be more practical to do so rather than to let the machine learning algorithm acquire the difference the hard way. If for some reason one were to do that, however, it's conceivable that different results will show up, as, according to the conclusion synthesized from the two papers, machine learning algorithms will likely not have more trouble learning from a circumstance where the linguistic alternatives theory is plausible than one where it is not.