Language Isn't Math

Introduction

I spent a long time assuming a language was a system you could work out. Vocabulary was the variables, grammar was the operations, and fluency was what you got once you could run them fast enough.

That is a reasonable thing to assume. It is how I learned to program, and it is roughly how school teaches math. Here are the pieces, here are the rules for putting them together, now go put them together.

It is also wrong, and I think it explains why capable people can spend three years on a language and still freeze up in an actual conversation. A grammar book describes a language after the fact. It is a record of regularities that speakers were already producing without consulting anything, and learning the record is a different activity from acquiring the habit that produced it.

I first heard this put clearly in a Matt vs Japan video called Language Isn't Math, though the claim underneath it is not fringe. Usage-based linguistics and construction grammar both treat a language as a large inventory of learned patterns rather than a rule engine running over a dictionary.

Experts think less than beginners

I wrote about chess a few years ago and one finding from that essay applies here almost word for word. Sonke Ahrens, in How to Take Smart Notes:

Chess players seem to think less than beginners. Rather, they see patterns and let themselves be guided by their experience from the past rather than attempt to calculate turns far into the future.

A beginner looks at the board and tries to compute. A strong player looks at the same board and recognizes it. The complexity did not go anywhere. It got absorbed into recognition, which is fast and close to free.

Native speakers are doing this constantly. They are not assembling a sentence out of parts and checking it against rules. They are reaching for something they have heard several thousand times and bending it a little to fit.

Which is why fluency looks effortless from the outside and feels impossible from the inside. You are calculating what the other person is remembering.

Language comes in chunks

This is the part I wish someone had told me in year one.

A language is not a box of Lego. Most of what people actually say is prefabricated: set phrases and whole clauses that get pulled out as single units. Pawley and Syder made this case back in 1983 in a paper on what they called native-like selection. Of all the sentences that are grammatically legal in English, only a thin slice are ones an English speaker would ever produce, and speakers can tell the difference instantly without being able to say why.

English is full of them. "I'm just messing with you." "It is what it is." "Long story short." You cannot derive any of those from their parts. A learner who tries will produce something legal and obviously foreign.

Japanese made this unavoidable for me. Try saying "I'm walking to the store."

❌ わたしはコンビニに歩いています
✅ コンビニまで歩いて行きます

The first one is not ungrammatical. It is just not a sentence anyone says. I had not failed to apply a rule. I had failed to hear the pattern enough times.

A lot of speech is ritual

Plenty of speech is not carrying literal content at all.

"Goodbye" started as "God be with ye," and compressed over a few centuries through "God b'w'ye" and "Godbwye" into a word almost nobody hears as a blessing. The religious content is gone. The function survived intact: close this interaction without being abrupt.

Japanese learners hit this immediately. よろしくお願いします comes out in translation as some humble request for a favour, which is not what it does; it marks the start of a working relationship. Translate it accurately and you get a sentence that is useless.

So I have started asking a different question. Not what does this mean, but what is this phrase doing here, between these two people.

Maybe the critical period is about tolerance

The standard story is that children have some neural advantage that closes with age. Chomsky's language acquisition device gets invoked here a lot, though notice that even on his account the thing is responding to patterns in the input. It is not responding to grammar lessons. No four-year-old has ever thought about a past participle. They heard "I've been thinking about" a few thousand times.

The research is messier than the popular version suggests and recent work keeps complicating it. But there is one difference between adults and children that has nothing to do with brains at all.

Children get an enormous amount of input, most of it incomprehensible to them at the time, and it does not bother them in the slightest.

Adults get a fraction of that and treat the incomprehensible part as wasted time.

Based on my own experience I think that is most of the gap. A baby listens to noise it cannot parse for two straight years and never once concludes the exercise is inefficient. An adult puts on a podcast, follows none of it, decides nothing is happening, and switches to a grammar app that feels productive and delivers a hundredth of the input.

So what do you actually do

Analysis is not useless and I am not going to pretend it is. Reading about a grammar pattern once, so that you notice it when it goes past, is a good hour spent. Picking what to listen to and building a review system around it is work a child cannot do for itself and you can do well.

The mistake is thinking the analysis is the learning. It is setup. The learning happens later and mostly without you supervising it, as the patterns stack up past the point where you have to think about them.

What I keep getting stuck on is this. If exposure is the mechanism, and we have more access to input than any generation of learners in history, why is nearly everything in this field still built around explanation? My guess is that explanation is easier to sell, easier to schedule, and easier to feel yourself making progress in. But I would like to be wrong about that.

Further Reading: