How Humans Learn Language – Hi! This is an optional and very brief look at how humans learn language. So how do humans learn to talk or to sign their sign language? There are many theories and this is an area of research called psycholinguistics, and the summary is that there is no consensus. But I thought might be illuminating to look at the different theories that we have and how they compare to what we do with computers.
The main question that we have is whether being exposed to enough, to many sentences of a language is enough to learn the language and this is particularly important for babies: If you talk enough of a language to a baby, what are they gonna do with that?
In reality we see that babies learn language almost miraculously and in two to three years they’re speaking a language. The main question we have in psycholinguistics is: Is being exposed to many sentences enough to learn a language? Some currents of thought say “yes”, that the input, the sentences that you hear, are very rich. Rich enough that you can get cognitive extrapolation essentially statistical extrapolation of patterns and that in time these patterns will emerge as a human language.
There’s many different theories for how this could go, and from the past to the present they have been called behaviorism, structuralism, functionalism but in essence they say that the input is all you need. That somehow your brain is going to pull together the cognitive resources to compute language and to acquire it.
There’s a second current of thought that says “no”. That being exposed to many sentences of a language is not enough. That somehow the input is not rich enough. This is called “poverty of stimulus” and that we have something in our brain that fills in the blanks and that essentially the input is a way for us to fill in those blanks and that those rules or mechanisms that we have in our brain essentially will instantiate language.
This kind of thought is called “generativism“. You might be familiar with Noam Chomsky, for example. He’s one of the proponents of these theories and essentially what would happen is that you could have something… theories that are very aggressive in saying that humans have a lot of principles and parameters, for example, in your brain.
They have parameters that need to be switched on and off depending on the input that we get. There’s some theories like “Minimalism” and “Merge” that say that we have a few operations to merge cognitive objects and doing that we string together structures in a language. But essentially these theories say that the input is not enough. That there’s something about us humans and about our cognition that is dedicated to language and that makes us very good at language.
Again, maybe this evolved. Maybe it’s some sort of separate module in our brain, but that there’s something language specific in us, whereas theories like functionalism think that we learn language using a pooling of all of our cognitive resources. There’s one piece of evidence in favor of generativism which is that children learn languages very fast but adults don’t. If you’ve seen a baby learn… Again, they, in three years they will learn a human language and if they’re doing it at that rate there must be learning about 50 words a day.
They can regularize patterns very easily so for example what we see there is an object called a “wug”. And now there’s another one. There’s two of them. There are two ___, you’re probably gonna say “wugs”, and children are able to successfully do this exercise from a very young age. They’re very good at generalizing and exploiting patterns in the data or at filling in their language acquisition device with the data. However adults are not good at this.
They need instruction, concentration and years to learn another language. Generativists say that we have some sort of language acquisition device in our brain that is on from a very early age and specialized towards language and that then switches off as we have other evolutionary focuses, worries in life, such as socialization, for example. The functionalists would say that when you’re a baby your priority is to learn language, so we focus all of your cognitive resources into this task. Whatever is happening in your brain we do have a clear idea of how babies learn, of what the process is for them to learn how to speak.
They start with very simple expressions: cooing, babbling, they go through single words and then telegraphic stages where they can just have one or two words together, and then suddenly their knowledge explodes and they can put together many longer sentences. The errors that they make are fascinating because they over generalize.
When they make a mistake with a word, it’s usually a word that is a weird exception and they try to say it with the general pattern. So they are very good at learning. They learn very fast and they, whatever errors they make, are very different from those of a computer so we probably don’t want to model our computer learning of language in the human learning. So in summary there are several theories about how humans learn languages. Some theories say that we extrapolate from the data and using general cognition arrive at knowledge of the language.
Some theories say that the data is not enough, that there’s something unique about us as humans that makes us particularly good about languages. And there are models for how children learn how to speak but our computational models not only are not related to them but they don’t need to be. We do not need to imitate how humans learn language. All we need to do is to imitate their capacity at understanding language and producing language. We don’t necessarily need to do it in exactly the same way that humans are. Thanks.
Web enthusiast. Thinker. Evil coffeeaholic. Food specialist. Reader. Twitter fanatic. Music maven. AI and Machine Learning!