Why Rules Don't Matter in Language Learning

Rules don't matter in language learning, at least not the

rules that are taught at school.

I'm going to talk about that because I just finished reading a book written by

Manfred Spitzer, whom I have mentioned before, called "The Mind Within the Net"

where he covers how the brain learns, what happens when we learn, and it's

kind of a very good follow up to a previous book I read by Manfred

Spitzer in German called "Lernen"

I like Manfred Spitzer.

He's very good at explaining what happens in our brains when we learn.

But this latest book is particularly interesting because much of language

instruction is based on teaching rules.

That is largely counterproductive.

Spitzer begins his book with the example of how babies, or young kids,

learn the past tense of English verbs.

Surprisingly, they first learn some of the very common irregular verbs

for the simple reason that those verbs are very common: went, came, et cetera.

And so they learn these and by learning these, and by hearing other

verbs,

they gradually start to develop some rules, some patterns for

how the verbs work in English.

And so after learning a few irregular verbs, because they're the most

frequent, then they become quite good at learning the sort of regular verbs.

That is where the -ed is added to the verb.

And studies have shown that adults learn the same way.

And then as they develop this pattern, because they hear more and more of

these, verbs with the -ed attached them.

So they begin to understand that the past tense is formed by adding -ed.

Then they start to apply the -ed past tense to verbs that don't take the -ed.

In other words, other irregular verbs.

So after that initial period where the very common irregular verbs

are learned, they then go through a period where they're very good with

the regular verbs that take -ed, and then subsequently they learn the not

so frequent verbs that don't take -ed.

That is the pattern.

That is the pattern for kids, and it's the pattern for adult learners.

What does that tell us?

It tells us that it is not a matter of teaching a rule.

It's a matter of exposure to the language, allowing the brain to form its own rules.

He again illustrates this point...

Well, first of all, he talks about the father trying to teach

the child, you know, "This is how the past tense is formed."

That's unnecessary.

If the father is speaking in a loud voice and pointing his finger at the

child, the child will only form the the following pattern: that is, my father

always talks in a loud voice and points his finger. Or he may deduce another

pattern that all fathers talk in a loud voice and point their finger.

The rule that matters is how the language is used, and that is

acquired from interaction with others.

In fact, one of the important things in childhood is the ability to play, because

play for the child is an opportunity, and is very important from an evolutionary

perspective, where the child can

make mistakes, interact, experiment with different things, and start to form

patterns as to, you know, how life works.

Learn to anticipate what's going to happen in his or her life.

And it's done in an environment which is relatively safe.

It's not like you take the child out,

you know, for caveman days, hunting where a mistake can have catastrophic consequences

In play, it's basically a consequence free opportunity to experiment.

And speaking of experiments, just to sort of contrast the

sort of rules-based language

production with the sort of naturally evolved process of correct language production,

Spitzer takes the example of text-to-speech.

So there was a system of text-to-speech.

The original text to speech was called something like DECtalk, D-E-C talk.

Where this was used with English,

there was a problem because, in Spanish, the spelling is very consistent.

For example, Spanish and other languages, spelling is very consistent. In English,

of course, spelling is notoriously not consistent. So the solution was to

develop a formula or an algorithm, which looked at three letters on either side

of the letter that they were trying to develop a sound for in order to

develop the correct sound for that word.

But even with that rule, the process was very slow. But as the technology

improved and people developing text- to-speech were able to access massive

amounts of, you know, corpora, massive amounts of language, content, audio,

text, audio, text, they found that it was much faster and more accurate to

allow the computer, without any rule, without any instruction, to develop its

own patterns, its own rules in order to produce accurately text-to-speech from

written English.

And that's a bit like what happens in our brains.

We don't need specific instruction, we just need to give the brain

enough input in order for the brain to develop the necessary patterns.

And modeling neural networks has enabled scientists to prove that unlike what

Chomsky said, where he seemed to suggest that there isn't enough input

for the brain to develop all the necessary rules to deal with language.

In fact, it is enough, and they show how it actually happens.

Leaving aside the fact that not every language deals with

past tense in the same way.

I mean in in Chinese, the verb actually doesn't even change form.

And when the child is born, of course the child can adapt to the patterns of

any language, whether it be the sound or the structure or how words function

in a sentence. Spitzer says that initially the potential is limitless.

The capability is very much limited of the child.

As the child gets better at producing the language, then the capability

improves and the potential, the range of sounds, the range of

structures that the child can deal with, becomes more limited, and

as we get older, then we are less capable of learning new languages to

essentially native-like proficiency.

But the principle is that it's not rules that govern how we speak.

So learning the rules, being instructed in the rules is not going to help us speak.

What we have to do is give our brains enough input, enough exposure,

enough experience, enough events.

So that the brain can gradually evolve its own rules.

And these rules are more efficient than any rules that we attempt to teach.

Spitzer goes into a lot of detail about the hippocampus and the

cortex, which is kind of beyond my knowledge base, beyond my pay grade.

But the gist of it is, as I understand it, and some of you out

there may be more knowledgeable, that short-term memory, even medium-

term memory events are stored in the hippocampus, whereas in the cortex, which

surrounds the brain, there's a much slower process of gradually acquiring

skills, longer-term knowledge, skills, and language, and so forth and so on.

And that the hippocampus is kind of the instructor of the cortex by

giving constant events: a new word, a new sentence, new sounds.

So these things are constantly and repetitively pushed towards the cortex

and the cortex gradually converts this into the sort of explicit knowledge, the

ability to do something, as opposed to our ability to remember specific events.

In fact, we don't retain them in our hippocampus that long.

The hippocampus has to pass them along to the cortex.

And so the important thing is to keep bombarding the hippocampus with

enough events so that it continues to push these towards the cortex.

I don't want to get into too much detail about how the synapses work

because I don't fully understand it. But I think what I can take away from

that is that our effort in language learning should not be focused on

learning rules, studying rules, the way traditional language instruction does.

Instead, we should be allowing words to come in.

Sounds they form, connections to meaning, to connected

meaning, to synonyms, antonyms.

Different contexts where we have seen them.

And over time, all of this stuff develops presumably in our cortex, and we have

a better and better ability to use the language. But it's not based on rules.

So all the time we spend studying rules, it may be interesting. Like if

we've had enough—and I've said this before—if we've had enough experience

with the language and then the grammar explanation comes along to summarize

something we're already familiar with.

That can be satisfying.

I'm not sure that it moves us along very much, but it might.

But conversely, if you haven't had enough experience, enough exposure,

then trying to generalize from a rule is not going to be that successful,

And that is how we designed LingQ.

And in another video, I'm going to talk about how LingQ was designed to

deal with this pattern of learning.

Not that I had read Manfred Spitzer's book at the time, but I just had

observed in my own language learning

the extent to which our brains can pick up on patterns.

Examples? Yes!

Grammar rules?

No! And that had always been my experience in learning languages.

Given enough patterns and examples,

gradually my brain got used to the language.

It turns out that that's right.