Interview

Rod Ellis

Page 1 | Page 2 | Page 3

Over the course of your research, how have your views of language acquisition and research about it changed?
When I was a teacher back in 1967 the prevailing psychological theory that informed teaching then was still really behaviorist, a notion that in some way you could inculcate a habit through extensive, thorough practice. So I guess my starting point was that.

But even before I ever began to research or read about language acquisition, I began to have doubts about that. The things one was practicing, that students were getting right, in the context of the practice, they didn't get right once they left the practice. They went back to trying to use their own language.

So, it seemed to me that there was something essentially wrong with the behaviorist view of language learning. And then of course I began to read people like Pit Corder, and began to understand notions like "interlanguage." I began to see that the process of acquiring competence in a language is an extremely gradual one, and nor was it an incremental one, a bit by bit building-block process. It was an organic process-an evolving a competence in the language.

And I guess the only thing that's likely changed is perhaps what the nature of language representation is. I still used to adhere to the idea that learners learn rules. Maybe the rules that they internalize are not target language rules but rather interlanguage rules. But I still felt that they probably construct rules. A lot of current theorizing about second language acquisition based on the connectionist model of language and language learning suggests that there are no rules, that all we have is a network of enormous, elaborate connections between neurons in the brain. And that language is somehow represented in this complex network.

As a result of the exposure of using the language certain pathways in that network get well trodden and as a result of that learners appear to be performing in accordance with rules but those rules don't actually exist mentally. All that exists is a hugely elaborate network of bits and pieces of information, not necessarily even corresponding to words or morphemes, etc. And that I find a compelling vision of how input comes to be represented in the human mind.

Yes, then behaviorism obviously cannot handle…
No, but neither can a generative Chomskyan model, or even the traditional grammar model because they're all premised on the idea of rules. We can formulate specific rules about grammar but they don't necessarily reflect what actually is represented in the brain.

So we can talk about a rule for regular past tense, we can talk about a rule for making relative causes but it simply doesn't follow that the knowledge that we use in order to produce sentences with regular past tense, or sentences with relative causes actually consists of those rules. Your behavior, if you like, becomes rule-like without you representing knowledge in terms of rules. And that is what current research is saying.

Of course it is still controversial, people still cling to the idea that even implicit knowledge does consist of rules. But computer simulations and theorizing with a connectionist model are going to take us an enormously long way towards actually explaining what our knowledge of language consists of.

So there is no black box.
(Laughs) There is no black box, because the other thing that would follow from a connectionist theory is that the mechanisms by which we develop these networks are essentially the same as the ones involved in any other type of learning, like perception. How it is we come to perceive objects and the way in which we perceive them? And again it is the result of exposure to a variety of objects that enable us to develop patterns in a network and so to very rapidly perceive things in certain ways.

Similarly with language. Perception and language involve the same pattern detector. Then if there is a black box, it is simply a pattern detector. That is to say, we are simply equipped with the ability to see patterns and regularities in input, whether it is visual or oral input.

What about consciousness-raising? How does that fit in with these ideas on rules?
I have for long time thought that perhaps the best way to actually teach language as opposed to simply giving learners experience with using and hearing language is to treat it as an object so that we can teach a bit about it. The only way that we can do this it seems to me that is compatible with what we know about learning is through consciousness-raising. That is to say we don't actually directly try to influence the construction of the complex network that I've been talking about, because really learners can only do it themselves. We cannot implant rules into that network.

Learners extract from the available information around them the regularities that go into their knowledge system. If that is the case, all that we can do is make them aware of some of these patterns and bits and pieces of language and how they work under the assumption that if you have an awareness of them, then ultimately your pattern detector might function a bit more efficiently. For example, if you know that relative causes have a certain structure, a shape, then maybe you are more able to detect them in the input to which you are exposed. And through the process of the detection, you gradually build up the connectionist network that I've been talking about.

Thus to me, consciousness-raising serves to equip people with explicit knowledge, which in someway may facilitate the construction of implicit connectionist systems.

The constructionists think that for people to really learn well, they have to construct a pattern by themselves.
You can't do it for them. You can give them hooks. You can perhaps choose the way in which the pattern detector works by making them aware of the kinds of things they may well come across when they are exposed to language.

Page 1 | Page 2 | Page 3

<<Back Number | Top | Recent Issue>>