Week 13

05 Dec 2017

This week I focused on trying to build an outline and presentation for the paper I’m hoping to be submitting to CogSci in February. Sidenote: with my past research experiences, the presentation and paper go hand in hand. That is, it’s easier to write a paper or create a presentation first and then do the other. Ideally, you work on them in tandem…

The overall goal of the project is try and use speech recognition to better understand language acquisition. To do this, we want to train a speech recognition system on child-directed speech to try and see if the acoustics of the parent input accounts for any of the variation in how children acquire language:

Here I outline how typical human speech recognition works. When it’s between two adult humans who don’t have any aphasia’s or communication disorders it’s usually very accurate. Two individuals can carry on in a conversation where multiple acoustic strings are exchanged:

In order to better understand how humans got so good, it’s important to look at the field of early (first) language acquisition:

First language acquisition outlines a number of theories on how humans acquire language:

One of the main theories of how children acquire is by reasoning with analogy or imitating their linguistic input (parents):

Perhaps it’s reinforcement learning and correction that explains the ability for children make progress. More clearly, children have some internal policy that is a function of their parents such that when they utter incorrect phrases or construct they are “punished” (there is some correction cost to an incorrect utterance). However, when children say correct utterances the lack of correct reinforces good behavior:

Parental input has become a crucial component to assessing how children acquire language. Of course, parent input (how parents choose to teach their children language) differs across cultures. These cultures differences are often based on geographical location, class, and ethnicity. For example, there are many Native American tribes that engage with their children as true conversational partners as opposed to talking to their children with impoverished speech like some of their midwestern middle-class American counterparts.

In fact, because we notice the variance in parental input and because it’s possible to find counterexamples to reinforcement and correction, and pattern recognition. Perhaps we have to appeal to something much deeper, intrinsic to children. The theory of Universal Grammar allows us to vision language acquisition as much more than reinforcement and pattern recognition, but outlines how all humans utilize their intuitions about the systematicity of language: