LING 641, Spring 2015, Lab 1

Defining the learning problem

Posted Tuesday February 3rd, due Tuesday February 17th.

Read the following:

Each of these works describes a learning problem in a different domain of grammar. The learning problem that Chomsky describes has achieved a huge amount of attention. The learning problem addressed by Pinker has received moderate attention. The learning problem investigated by Goro has received very little attention.

Do these three problems present the same challenges for learners? If they present different learning challenges, explain how they are different. Give a clear statement of what the learning problem is in each case. If you see similarities and differences between the three learning problems, then it would be helpful to explain those.

To what extent could the three challenges be solved by assuming that the child has substantial innate knowledge? In other words, could the challenges be solved by assuming that the child does not, in fact, have to learn about each of the phenomena, and instead has the relevant knowledge ‘built in’? The typical assumption is that if property X is innate, then it should be both universal across languages, and should not need to be learned.

Or could any of the challenges be solved by assuming that learners have a very powerful distributional learning mechanism, i.e., a mechanism that keeps a detailed count of things that do and don’t occur in the input? In that case, what information in the input would the child need to keep track of in order to successfully learn [be specific], and how plausible do you think it is that such experience is reliably available to the learner? Relevant input could consist of individual strings, collections of strings, sentence-situation pairings, or whatever you think might work. 

For purposes of this assignment you are free to assume as much specific, innate knowledge as you wish, and you can also assume a highly powerful distributional learning mechanism. You could give your learner arbitrarily good memory or computational abilities – no limits. But you can’t change the facts of the target language. You should feel free to comment on the plausibility of your assumptions, but it’s more important to consider how the learning problems could in principle be solved. First find some solution, then worry later about its plausibility.

For this assignment, you should pay attention to detail as much as possible. In particular, be explicit about (i) what the learner’s possible hypotheses could be, and (ii) what evidence – linguistic, situational, or some combination – could help the learner to arrive at the appropriate conclusion. Could the learner figure things out based on individual examples in the input? If so, say what those examples are. Or does the learner need to combine information from multiple examples, i.e., tracking frequencies, or tracking the possibility of different types of sentences? Could the problem be solved by just assuming detailed innate knowledge, or is there a minimum that absolutely must be learned, because it is not universal in all languages?