Grammar as metadata

A kabbalistic account of language

As we all know, spectacular supernatural miracles are happening all the time. Look at the Bible, for instance. Most of the supernatural events aren’t even surprising:

• The Deluge? There’s no lack of documented ancient floods to choose from. It's almost like, in biblical times, people specifically chose to build their civilizations in floodable areas.

• Parting the Red Sea? You can get this to happen in a computer model by fiddling with the wind parameters in the right way.

• A burning bush? I blame Dictamus gymnostylis, a bush that surrounds itself in a cloud of flammable 2-methyl-1,3-butadiene. (How this plant could possibly show up in the Sinai peninsula remains to be explained by biblical botanists).

• A character obtaining secret forbidden knowledge by eating an apple in the presence of a serpent? More likely than you think.

• Sodom and Gomorrah? Possibly inspired by Tall-el-Hamman, a city near the Dead Sea that got wiped out by a cosmic airburst 3600 years ago. [Edit: I’m told the paper was retracted, making Divine Intervention the most likely explanation again]

But there’s one biblical event that I find particularly weird – the Tower of Babel. I get the part where humans build a super giant tower to defy God and poke into Heaven. That sounds like the kind of thing humans do.

The intriguing part is when God punishes humans for their defiance, and now they all speak different languages and can’t understand each other anymore. Surely, this is a metaphor for something. But a metaphor for what?

I have a theory. I think the destruction of Babel happened sometime around 2012.

Grammar as metadata

Grammar, as we all noticed, is absolute bonkers. For example, my birth language, French, does the thing where every random object has a gender assigned to it: forks are female, knives are male, chairs are female, stools are male. Many other languages do this, and they don’t even assign the same genders to the same objects (house is feminine in Spanish but masculine in Hebrew; bridge is feminine in German but masculine in Italian, and so on). This is obviously nuts.

Unless you’re talking about boats, English doesn’t attribute random genders to inanimate objects. But it still has a gendered grammar. If you’re talking about a visit to the doctor, you’ll probably have to mention the gender of the doctor somewhere (“she told me…”), even though it’s completely irrelevant to your point. This is substantially less bonkers than French, but a bit bonkers nonetheless.

I’ve always assumed these were just old, weird, suboptimal traditions, and we cannot change them because language is a difficult coordination game. Designing perfect grammar rules is relatively easy, but getting people to actually use them is pretty much impossible. So we just nod along, and reluctantly infer the gender of the doctors from their voice and clothes.

But I think I was wrong. My mistake was that I misunderstood the nature of grammar: I saw it as a set of logical rules you can use to describe the world. And to be fair, it’s often depicted that way. And it does these things. But that’s not what it is.

Think about language as a social technology, evolved through millennia of cultural selection. Encoding accurate descriptions of the world is clearly a nice feature, but it’s not the only trait being selected for. What else does language do? It conveys information from one brain to another. If a little variation of a language makes it more efficient at conveying information, that variation may be adopted by the whole population, and impose itself as the new grammar.

To illustrate that, here is a fun exercise: let’s build our own language and optimize it for communication.

Imagine the following situation: after a plane crash, you find yourself on a desert island with 4 other persons. Worse, you are all from different countries and you have no language in common.

Fortunately, the plane was taking you to the international Language Creation Conference and the other survivors are all nerds who design fictional languages as a hobby. This is the moment you’ve all been waiting for: it’s time to design a new grammar from scratch.

At first, your rudimentary grammar only has one pronoun: xir. Someone says:

“Xir told xir that xir met xir.”

What could that possibly mean? There are 5 people on the island, and each xir can refer to any set of 1 or more people. That gives us (2⁵-1) × (2⁵-2) × (2⁵-1) × (2⁵-2) = 864,900 possible interpretations.

If instead of having just xir, we had a couple of variations, we could use that to add some metadata to the message. These wouldn’t even make the message any longer. For starters, we could have a singular/plural distinction. Like, if xar is singular and xir is plural, we can have something like:

“Xar told xir that xar met xir.”

Now we have only 5 × (2⁵ - 2) × 5 × (2⁵ - 2) = 22,500 possible interpretations. That’s already a pretty big cut through the combinatorial explosion.

Let’s narrow down the possibilities further: let’s distinguish who is talking (I/we), who is listening (you), and the rest (still xir/xar for now):

We are left with “I told you that xar met xir.”, so 1 × 2³ × 4 × 3 = 96 possibilities, quite a drop from the original 864,900.

In linguistics, this kind of metadata is called a pro-form. English is really basic here – most languages would at least distinguish “singular you” from “plural you”. The most high-tech version would be Sursurunga, a language spoken in the Bismarck Archipelago of Papua New Guinea. They can tell the difference between a few and quite-a-few-but-not-that-many just by picking the right pronoun. They also have clusivity, when you distinguish between we as in “me and the gang” and we as in “me, you, and the gang”:

Table from Wikipedia

At this point, we’ve exhausted most of the criteria that are directly relevant to the message. But we can go further – we can still cram in more metadata by adding completely irrelevant criteria.

You probably see where I am going. The question isn’t whether the criterion is relevant to the message; it’s not even whether the criterion corresponds to some essential reality about the person. The question is: does having pronouns based on one criterion add enough useful metadata for cultural evolution to select it?

And, in that regard, sex-based pronouns are incredibly effective. There are pretty deep evolutionary reason why mammals have two mating types, and why these two mating types are maintained at exactly 50% of the population. In terms of metadata, this is optimal: specifying the gender of someone takes out one whole bit of entropy. It’s also pretty much unambiguous – for better or for worse, people are remarkably good at picking up sexual dimorphism cues. By a strange conjunction of coincidences, Nature gave us a near-perfect criterion to build some effective grammatical pronouns.

Seen through that lens, gendered languages make more sense. If I say "the doctor is in her office", the information conveyed by "her" is not that the doctor is a woman (we don't care about this); it's that, of all doctors, we are referring to the one who is a woman. It's not about specifying the gender; it's about specifying the doctor.¹

If we accept that, we can even justify why French and other languages assign a gender to everyday-life inanimate objects: now that we have a gendered language, we might as well use it as a general ambiguity-reduction mechanism. If you’re holding a fork and a knife, and I tell you “give it to me” in English, you don’t know which one it is. But if I tell you “donne-le moi”, you can immediately tell I’m referring to that masculine, virile, testiculated knife.

I’m not saying that people were thinking about information theory or the Fisherian argument for sex-ratio when they originally came up with gender pronouns. Grammar was just shaped this way by millennia of cultural transmission, mutation and erosion, until it turned into the delicate social technology we know. It’s pretty good at describing the world, articulating thoughts, or communicating ideas. But what it really does, like all things evolved, is more nebulous.

This is where humans start building the Tower.

Puncturing Heaven

Starting around the 16th century and accelerating ever since, a weird metamorphosis happened to our civilization. To use the terms from The Uruk Machine, we went from a society governed by Metis, to a society governed by Episteme.

Metis is the muse of cultural evolution – the God of ritual, imitation and tradition. It’s about replicating things that work, in the sense that they don’t result in immediate gruesome death. The world is assumed to be too complex and magical to be understood.

This went on for centuries and centuries. But, in secret, humans accumulated tiny bits of information about the world until a strange phase transition occurred: one day, we reverse-engineered the world well enough that deviating from tradition no longer resulted in immediate gruesome death. This allowed for more experimentation, hence more knowledge, hence even more precise experimentation. This repeated until most of the visible world had a theory attached to it. We entered the age of Episteme, when most of our decisions are based on our mental simulation of the world.

Metis was a muse, a supernatural being shrouded in mystery. But Episteme is something we built ourselves. It’s a golem. Metis gave us things that work; Episteme promises to give us what we want.

In the Epistemic world, people started to inspect old culturally-transmitted traditions like language, and soon they noticed the shortcomings of grammatical gender.

For starters, what if the irrelevant gender metadata we put into pronouns distorts our understanding of the world? For instance, if you use he as the default indefinite pronoun, people might subconsciously learn a model of the world where everything important is done by men. And, in the Epistemic era, having a distorted model of the world is simply unacceptable.
(From what I gather, the evidence on this is still pretty inconclusive. But I can tell you at least this: in French, frog is feminine and toad is masculine, which leads many French speakers to believe that they are the male and female forms of the same species.)

Also, what happens to the minority of people who don’t fit neatly into the two gender categories? Clearly, the gendered language doesn’t work very well for them. Under the reign of Metis, these individuals were basically ignored by grammar rules. But, in the Epistemic world, we can now ask, can we make a language that works for everyone?

And so, the progressive movement, as the ultimate incarnation of the Episteme mindset, set out to redesign language into something universal and fair. And, together, we built a beautiful, tolerant society where everybody belongs.
Right?

The thunder roars

Of course, this is not how any of this works.

The thing is, we haven’t exactly replaced cultural evolution with modernity. We have just burned down the memetic rainforest, leaving an open field for new forms of psychofauna to proliferate.

Seen from the inside, it feels like we are in control of how we speak, and we get to steer language evolution in a direction that fits our needs. But, seen from the outside, we are humble mesa-optimizers. We can use our wits to come up with new expressions, but these are just candidates in the big jungle of memes. We still don’t get to decide what has high memetic fitness or not. Metis is still in the driver’s seat.

As it turned out, changing our languages’ grammar to make it more inclusive was one of the most deadly toxoplasmas of rage ever created. It’s pretty bad. People started to hate pronouns more than genocide. And English isn’t even that gendered, so the requests to fix English were pretty light. For French, you’d have to basically redesign the whole thing – so people came up with a system with full stops inserted in the middle of words. Needless to say, everybody was so enraged that the government banned it from schools.

And please, don’t get mad. I know you have strong opinions about the pronoun question, and solid arguments about why your preferred grammar is the triumph of Truth, and why the outgroup’s is the end of civilization. The outgroup feels exactly the same. But take a step back. Don’t look at it from the point of view of an opinionated mortal. Look at it from the point of view of a cruel god watching its creatures inside a snow globe as their cities collapse into discord.

Why is grammar such a wonderfully effective culture war topic? I think the main reason is that it's always on display.

If you want to show off your wealth, you can just go around driving an expensive car or wearing a big shiny diamond necklace. But this has gone out fashion, and true high-status people are now using their political opinions as class markers. Having sophisticated opinions shows you're the kind of person who went to a prestigious university, not one of these farmers who still dream of expensive cars and shiny necklaces. This was all predicted by Bourdieu.

The problem is, this only works when the conversation is about certain specific political topics. It’s hard to bring up “marginalized identities” when you’re talking about your coworker’s cheesecake. Of course, you could try to make every conversation about marginalized identities (some people tried), but it’s not easy.

Here is the trick: when your class marker is embedded directly in your grammar, your tribal affiliation is always on display. Everyone can tell which team you are on, even if you are talking about cheesecake.

Once again, grammar gives us a way to sprinkle metadata all over the message.

From the description of a club at a prestigious French university. The club is about playing Lego, but the full stops inside the words give us some unmistakable metadata about the authors’ social context.

To be clear, I don’t think anybody does this deliberately. Based on my experience hanging out in French universities, here’s what I think is happening: I’m sure the Lego club people genuinely want women to feel welcome. I’m also sure writing this way is a net negative for the feminist movement – remember: normies hate it so much that the majority political party is trying to ban it. However, even if the students realize that (not sure if they do), they still can’t stop writing like this, because now that would imply they are the kind of people who support the majority political party – a terrifying perspective for an elite university student.

Once people start using the memes they adopt as parts of their identity, it changes the rules of memetic evolution quite a bit.

It’s tempting to view the different language rules as competitors, each trying to prevail as the Rightful Language. But, from the point of view of cultural evolution, these memes are in a symbiotic relationship. They need each other to survive. In fact, if one meme came to dominate the landscape and become universal, then people would come up with new memes to keep the social distinction going.²

The result is, of course, a stable coexistence of multiple grammars, even among people who grew up learning the same language.

And that's my kabbalistic interpretation of Babel: the tower is a metaphor for memes. We, arrogant humans, tried to engineer memes to our own specifications, surpassing in perfection the naturally-occuring memes cultural evolution had provided for us. But cultural evolution is stronger than us, and here is our punishment: language is now fractured along the fault lines of political tribes.

“And the Lord said, Behold, the people is one, and they have all one language; and this they begin to do: and now nothing will be restrained from them, which they have imagined to do. Go to, let us go down, and there confound their language, that they may not understand one another's speech."

It’s not even that we speak completely different languages – the vast majority of our words is completely unchanged. But the tiny grammatical details that signal allegiance to a tribe are all it takes to make communication impossible. Not because we don’t understand the meaning, but because these words signal that our worldviews are so different that there is no hope we can understand each other.

1. The “metadata” theory of gender pronouns implies something counter-intuitive: the more gender-equal a society, the more useful gender pronouns become. In the land where women raise children while men maintain the sewage system, the sentence "xir takes care of the children while xir repairs the sewers" already contains all the information you might want – we know very well who is doing what. Adding gender pronouns wouldn't bring much more useful information. On the contrary, in the Wittigian post-gender utopia where every task is shared 50-50 between men and women, gender pronouns are maximally useful, so this society would ironically benefit the most from a heavily-gendered language.

2. This happened before in the case of French inclusive writing. The traditional way to write in a gender-neutral way is to put parentheses inside of words instead of full stops: e.g., “acharné(e)”. Curiously, this way to write is totally mundane and well-accepted by normies. As a result, it signals absolutely nothing about your social class, which is why people had to come up with something more radical-looking.