The Cowardice of Linguists will Kill Us All
Let's all judge ourselves, and each other, a little more harshly—and how LLMs can help!
Am I Saying This Right?
A friend of mine, a post-grad in linguistics, describes himself as a rare prescriptivist linguist. Everyone else in his department is a descriptivist. He’s the black sheep, regularly getting into debates with his colleagues.
It is absurd. Why is there even a contradiction between the two?
A scientist doesn’t actively interfere in the thing he or she is supposed to be studying; that part makes sense. If it’s your job to study how people learn and use language, you can’t tell them they’re doing it wrong at the same time.
But if that were the end of it, then he wouldn’t be driven to identify as a contrarian, pushing back against everyone else. In his experience in the field, he’s not supposed to believe that variation in quality of language, and criteria for measuring it, can even exist!
A cancer researcher isn’t forbidden from lamenting the suffering cancer creates in order to study it. It’s merely that their job is not in the direct treatment of cancer patients; that’s the oncologist’s job.
So why are linguists dissuaded from making value judgments on language?
To answer it quickly: evaluations of language facilitate elite gate-keeping and discrimination, and it is preferable for scientists to impartially classify dialects than elect any one as superior to others. Many prescriptive rules seem, and often are, arbitrary. Many famous and once widely-taught rules were also easily demonstrated to be in flagrant contradiction with the observable facts.
Now, these are all, again, great reasons for linguists to be cautious and conservative about the claims they make, given the authoritative weight of their opinions as scientific experts.
But couldn’t it be that their authority is rather out-sized for the limitations of what linguists can actually say? Nobody is asking them to grade English papers, or criticize poetry. But, in general, when people want truth, they immediately seek out “the science” as a heuristic. And literary critics and poets aren’t scientists, so no budding rationalist cares what they think.
Science must be impartial; but everyone can’t be a scientist all the time. The world isn’t a laboratory, and people who spend their lives studying everything around them aren’t living life; they’re avoiding it.
There are some extremely bad consequences from mass fetishization of the scientific method outside the controlled conditions and professional neutrality of the researcher working in their field.
Given that so many widespread societal problems, especially as regarding mental health, can be chalked up to a loss of nuance, failures to communicate, and the instability and loss of meaning itself, maybe we ought to stop prescribing pure non-judgmental, passive descriptivism at large?
We Need to Judge Each Other More
You’ll agree that nobody is perfect. Therefore, you must agree that everyone has room for improvement in many, if not most areas in their life. At least, in areas which can be improved through dedication to training. And maybe the very tool with-which we think, the very stuff of our cognition itself, deserves particular care? Even life-long attention?
Because it seems to me that something terrible has happened. And I mean terrible in a society-destroying sort of way. I mean, so terrible that, thousands of years ago, stories about it were encoded in holy scripture.
Linguists have done a fine job observing how children or new speakers of a language begin acquiring it. Generally speaking, they immerse themselves in an environment full of native speakers, and it just sort of rubs off. It’s mimesis. With some study and practice and immersion, bits and pieces get picked up until regular exposure starts compounding.
For the linguist, this is a fascinating and fruitful observation. They can observe the implicit rules governing syntax and other structures, the meaning of learned tones, etc. in the language everyone is speaking, compare dialects, etc. And they can observe that people naturally pick up and follow these rules without ever having them prescribed by some language policing authority.
Most speakers of any language are not conscious of the very rules they follow. And since it’s the professional occupation of linguists to codify those rules, then if speakers did know the rules of the languages they spoke, they’d be linguists.
But again, we have this overstep of authority; the best political speech writers, or authors, or orators are not linguists! A linguistics degree is not the key to charisma of spoken communication: it does not lead to the masses hanging off your every word. The sorts of rules or prescriptions or criteria for evaluation of good or proper speech are not the science one picks up with a linguistics degree.
So why the hell should anyone care what a linguist says about language acquisition after early childhood?
If you want to improve your thinking, and your speaking, you’ll have to do better than spend your life casually picking up words and phrases from media and the people you talk to. You need to study and always read and listen to people above your skill level. You need to be dictionary-maxxing. You need to read, and re-read, and read again texts which are too complex to parse, to skim. You need to challenge yourself, like you’d challenge your muscles in resistance training.
You need to feel embarrassed at the poor state of your ability to form a subtle thought, or find the perfect metaphor. You need to start flinching at mixed metaphors and lazy words and cheap and exaggerations in the rhetoric you hear around you. You need to let bad rhetoric stop working on you; you need to pull out of the group hive-mind and get alienated for a while. You don’t need to climb a mountain or lock yourself inside to get a taste of the hermit life; just stop being vulnerable to the bad spells of intimacy and emotion everyone is trying to cast on each other every day—
without losing your humanity or capacity for empathy in the process.
The Freedom to Enslave Yourself
What is observable today by everyone, the linguist foremost, is that language constantly changes. It incessantly drifts, too. Like ill-suited partners on their third remarriage, our map and our terrain are as-often divorced and at blows as they are wed and in harmony.
This state of affairs is precisely what makes social engineering and mind control, at scale, possible.
I often feel pressure to celebrate such a state of affairs as an expression of freedom. Of the celebration of a wide variety of forms. Of progress, especially by framed in the metaphor of evolution.
If you know a little biology, then you’re aware of the Cambrian explosion: unconstrained blooms of wide variety of species are usually shortly followed by the mass extinction of most varieties once the selective pressure for survival kick back in. When those enforcing the rules let up on enforcement for a little while, the prudent thing to do is to continue to live by those rules.
Imagine that! Following the rules even without the incentive of immediate punishment. Not having ice-cream for dinner, even though your parents aren’t there to stop you. Its called long term thinking, and requires delayed gratification though. It’s the way out of the behavioural conditioning governing your life. And long term planning and delayed gratification, in your consumer class, is terrible for business!
Those who want to control you want you to express yourself in an unguarded sense, and to chase immediate gratification. That’s how the cybernetic feedback loop controlling you is closed. And the simpler you speak and reason, the easier it is to gain that control.
To the contrary, I’d like to tell you that just like rules for speaking, and for thinking, carry the same sort of implicit contract. Speak and think and listen carefully even if there is not immediate and obvious bad consequences for not doing so. Again, that’s what breaks the loop of the Skinner box they’ve got you in.
Advertisers are Evil
Thoughtful criticism of a service or commodity impossible when the very words used to talk about that service or product are given to you by its marketers
This is my primary criticism of the entire language of the contemporary tech industry. That’s why I always put the term “social media” in scare quotes. It’s the worst sort of garbage word, the sort of garbage word which gives the mistaken impression of referring to a real thing. It exists in relation to a whole bunch of fake, nonsensical words like “platform”, which also does not refer to anything real.
All these words form a fake conceptual structure the tech company marketers have socially constructed. This means, you learned these words uncritically, by picking them up the way a child picks up language.
You hear enough people talking about “social media” and “platforms”, via a total flooding of your environment according to Ellul’s principles of Propaganda, that you just internalized it. You constructed a conceptual structure in your own mind that you now project onto computers in place of having a perception of them. A structure which nearly everyone now mentally projects into virtual existence, merely by the habitual use of all these fake words in co-validating interrelation, mixed into their otherwise coherent sentences.
At least Magritte’s pipe referred to a real world object! Ha! Sheesh!
When I complain about new words and terms, I’m not complaining that the kids today shortened the word charisma to rizz. I think that’s neat. I’m not some prematurely-old man griping about slang.
I’m fighting the power of well-paid social engineers to assemble new structures like these in your head, while distracting you with false consciousness about the meaning of words in the daily culture war bullshit.
When you remain in the state of operant conditioning, closing the feedback loop of emotional disgust and gratifying catharsis, you’ll never catch the engineers planting these control structures into your very ears and mouth.
Evolutionary Metaphors must be critically evaluated
Primarily, you’ve been denied any means of evaluating the quality of language. And, so, of any practical vision of how better language might be cultivated and engineered into higher precision and, dare I say, better forms.
Returning to the evolutionary metaphor, it is often said of our ever-changing language that it is evolving.
No. Dynamic and complex systems don’t just magically always improve. They can tend toward entropy when their regulatory systems breakdown. They can fizzle out after a bursting, bubbling bloom of chaos, or after seizing-up in rigidity and plummeting into a death spiral.
I do not see language as “evolving” right now. There is more to evolution than change; than mutation. Individual words and phrases can be selected for survival over short duration, but the criteria of selection matters, and the present criteria is arbitrary and tends to serve power. The survival of terms like “social media” demonstrates no merit to its usefulness to anybody except the marketers who coined it. It’s a term you resort to because nobody would politely accept the self-recrimination of terms which were more correct.
Language doesn’t evolve. Not as a whole. It was finished long ago. Its usage, though, mutates. It drifts. It adapts to a changing world. It is pulled and tugged and steered. It is engineered. Some changes clarify thought and perception. But I judge that most changes bamboozle and ensnare speakers into conceptual labyrinths, while providing only some momentary catharsis which can be mistaken for clarity or perception.
You know what evolves? The systems and institutions for maintaining and teaching and prescribing higher-and-higher orders of language. The rules of grammar, the figures of rhetoric, the techniques of poetry. The the tools for cultivating good language grew and developed over centuries, from primitive to advanced and, yes, often to stultifying and degenerative excess.
When you read McLuhan on poetry, or on the classical trivium, that’s what you’re reading. When you read his literary criticism, you’re getting the keys for evaluating the various things various eras have been able to do, to achieve, using only language. We’ve been running, as a civilization, experiments in classroom education of language for two millennia. Everything to be learned about instruction and cultivation of language has been learned. Exhaustively.
We easily recognize periods of high technology from periods of low technology; we can appreciate what the fall of Empires means for the “loss” of a technology which requires mass coordination and upkeep. Rome had aqueducts, and then they lost them. A country today which outsources some once-local industry can literally lose that technology when the geopolitical winds of trade shift against them, and everybody who remembers how it was done is long-retired or dead.
Language is such a technology. Poets in the early 20th century were doing things, in English, which hadn’t been done since the 17th century. There are rules. There are techniques. We had them—and we gave them up.
The biggest scam ever perpetrated was upon the most massively literate and intellectual population to ever exist on the plant. It is this: since linguists have studied and explained how children pick up a language merely by being immersing into it, that means that immersion in society and popular media is the way stay in touch with “evolving” and current language.
“People learn by copying others! And they innovate by playing around, and figuring out pragmatically what works!”
No. Hard no. It might be a way that people learn. But it’s been sold to us as the way people learn.
That’s just the start. That’s what babies do. That’s what a student does; where a student exists by virtue of relation to a special arena for learning created and ruled by teachers. And teachers are carefully selected to exemplify a superior mastery to which the student aspires, but has not yet reached. There is an explicit vertical dimension to the very artificial nature of the classroom which has evolved from, lets say, the Academy of the Greek philosophers.
When McLuhan said that classrooms were obsolete, that the world itself had become the classroom, this is precisely what he was indicating. Nobody learns how to make it in this world from the language lessons they learn in school.
It is dangerous to simply learn by passively picking things up out in the real world. All the traps are there. To learn from society is self-imprisonment. A person who learns language only that way, throughout their life, is a hopelessly dependant consumer; the perfect sucker.
But if it’s not school, then where is the environment for learning langauge?
The Anti-Environments for Self-Motivated Language Learning
You know who ascends beyond merely learning by immersion in the lowest common denominators of speech? Politicians. Lawyers. Marketers. Preachers. Gurus. Grifters selling courses in persuasion and seduction and manifestation. I’m not claiming they teach effective language, even if they claim that they do. But they have learned it.
They don’t settle for speaking in the ways which people commonly acquire; they actively develop their tongue for rhetorical speech, and their ears for the most resonant of new metaphors and poetical phrases to steal. They grind and test and study.
Like musicians or producers seeking out the new sound, some people out their are driven to find the best of words. They seek, actively.
“What about writers!” you cry. “What about poets!” “What about lyricists!”
The quality of the sort of study of language I’m describing can be known by its fruits. The writer, or the lyricists you’re thinking of; are they successful?
In our wild jungle of uncultivated language habits, those who speak better than most ascend. They gain in power. They, without loss of generalization, make money. People want to hear them. They buy tickets to attend their public appearances, they copy them and try to become them. They phone into their shows, and quiver with nervousness when talking speaking with them.
At the same time, we must recognize degrees of success in language which are not so lofty. Many, many people become good speakers and writers in a very lop-sided, uneven sort of way.
There are specialisms of language mastery; narrow domains which different groups have colonized and occupied in the electric mass of new post-literacy. Here, I readily admit, evolutionary selection plays a role. For some groups, survival has meant becoming excellent at insults and catty or incisive snap-backs. Others must resort to mastering the language of empathy, and mirroring. Narcissists spend their lives honing their opening set; over the entire duration of your acquaintance with them, the first hour after meeting them is the most interesting and enjoyable time you’ll ever spend together.
But again, this wild and pragmatic form of self-education is only one step better than passive absorption and social construction.
Which brings me back to the ought. How ought we to think about, to teach and use language? In a generalizable, well-rounded way?
The First Blue Marble
Imagine the first person who ever fashioned a globe.
One day in history, someone set out to seek and gather together all the maps they could find. After that arduous task, they assembled them all together into a coherent—if largely speculative—whole. And, once finished, they held, for the first time ever, the world in their hands.
The environment which surrounded them—which had forever surrounded every life form on this planet since the beginning—was now itself surrounded by their palms and fingers. I originally wrote here “the tables had turned,” but this phrase falls woefully short. It must have been more like the sun and the moon and the heavens and the earth had reversed their positions in space; permanently. It was the inauguration of the rupture, or topological recursion, in the fabric of reality which has given us our postmodern cyberspace today.
Today, advanced computing has given us the very same experience in relation to language. Large Language Models represent, for the first time ever, a concentration of all the inherent meaning within language itself as an artifact which we can observe from the outside. We can surround language; loom over it, poke it with a stick. Study it all together, all in one place.
In an essay which I recommend reading in full, Paul Valery anticipated what we will observe from this new reversal of who-surrounds-what a century ago
Indeed, if one considers things from a sufficient height, can we not see language itself as the supreme masterpiece among all literary works—since every creation in literature is nothing more than a recombination of the powers contained in a given vocabulary, according to forms once and for all instituted?
Those who learn the most from LLMs won’t gain that knowledge through asking it questions, treating it as some kind of simulation of a conversation partner. They’ll figure out, in the way that the training process reliably recreates, over and over, the same common “attention heads” or forms of pattern recognition, facets of language which transcend any generation’s or cultures usage of it.
I foresee new competition, or even the usurpation of the scientific authority of linguists over language to follow the discoveries facilitated by what LLMs will bring. Discovies expanding into the domains of literary composition, mythic interpretation, and poetry. Or, better, what study of them will confirm about what we once knew. What we were enticed, by manipulation and social engineering, to forget.
The Purpose of Language
The oft-quoted maxim by cybernetics Stafford Beer that “the purpose of a system is what it does” is more than just a snappy retort to those who defend the good intentions of institutions or methods which fail to deliver favourable results. It is actually one of the core tenants of cybernetics—the rehabilitation of the ancient idea of telos to ascribe the possibility for things to actually have a defining purpose at all:
Teleology has been interpreted in the past to imply purpose and the vague concept of a final cause has been often added. This concept of final causes has led to the opposition of teleology to determinism… Teleology has been discredited chiefly because it was defined to imply a cause subsequent in time to a given effect. When this aspect of teleology was dismissed, however, the associated recognition of the importance of purpose was also unfortunately discarded. Since we consider purposefulness a concept necessary for the understanding of certain modes of behaviour we suggest that a teleological study is useful if it avoids problems of causality and concerns itself merely with an investigation of purpose.
We have restricted the connotation of teleological behaviour by applying this designation only to purposeful reactions which are controlled by the error of the reaction — i.e. by the difference between the state of the behaving object at any time and the final state interpreted as the purpose. Teleological behaviour thus becomes synonymous with behaviour controlled by negative feedback, and gains therefore in precision by a sufficiently restricted connotation.
According to this limited definition, teleology is not opposed to determinism, but to non-teleology.
—Behavior, Purpose and Teleology, (Rosenblueth, Wiener, Bigelow, 1943)
If it can be claimed that language is undergoing a process of evolution, rather than aimless drift into nonsense, or entropic degeneration into aesthetic sensation which would render it a variety of musical expression, then we have to figure out if its what the functional feed back mechanism is matched to its ostensible purpose.
In training an LLM, the guiding feedback is prediction. The mutations to the attention heads which improve the ability for a model to accurately predict what comes next in a sentence survive.
The final cause of the LLM, as engineered by the humans who train them, is then the purpose of predicting how a sensibly begun text would sensibly continue and conclude in a coherent, meaningful way. Again, this is because most of the training data is comprised of sensible, human-meaningful stuff.
Can we not adapt this criteria, then, as a means of evaluation of our own language? I don’t mean that we ought to learn to speak as machines; LLMs lack perception. Computers are dead matter, and like Native American dream catchers, they are now absorbing the spirit of dead words.
What sort of people complete can each others sentences? Can know what the other thinks? Can predict language? Sometimes it’s the smug, manipulative elite predicting the clichéd responses of the well-conditioned fool. And sometimes it’s lovers , in mutually close and intimate attunement, sharing a common perception of their world which is based in truth.
Your thoughts are not happening in your head.
And reliable and affirming and grounding communication is the distribution of good and perceptive relation to ones environment distributed across multiple people. It’s the stable, self-regulating triangle of self, others, and the world at large.
LLMs can’t do that. They are just remixes of dead letters on pages and screens. They are mere cogitos. LLMs are brains in jars so that we don’t have to be.
Retrieve the Past
If you want to get ahead of the curve on A.I., one way forward is a renewed appreciation for, and retrieval of the past. Beginning, first, with the gradual and dedicated commitment to improving one’s own use of words.
You have time; most tech people have too much to unlearn from their own marketing lingo nonsense to ever catch up to Valery, or McLuhan, and far too much hubris to even acknowledge the need.
They don’t know what they’re holding in their own hands; they think that what is to be learned will come from some an anthropomorphic simulation of talking to the LLM. And the normative, Reddit-user quality of prose comprising the narcissistic navel-gazing of the primitive simulation of conversation, the “chat” interface, will plateau their language capacities to ineffectual midwittery for a long, long time.
The important thing is this: The fact that LLMs even work—that they have reliable predictive powers, and that those predictive powers can be directed into the simulation of conversation---testifies to the inherent meaning in reality and for the capabilities of language to capture that meaning. Reality radiates meaning, and language captures it more frequently than not, such that statistical analysis of language can result in an endless, automated fountain of meaning.
And, we’ll discover, that what is and is not arbitrary, what is and is not socially constructed can be statistically inferred, aided immensely by the happy fact that most texts written are attempts to document true or plausible things.
Even absent the gymnastics of newer reasoning models, the fact that text prediction demonstrates reasoning is owed to reason being inherent within most recorded language usage. The general aggregate of the stuff of thought has been captured, like a fly in amber, for our inspection once we develop the instruments to do so.
We needed the microscope to discover germs, and we are only now beginning to develop the equivalent tool for understanding complex models today; but as surely as we now study individual neurons in the brain we will study the inner-workings of models too.
It’s sort of a funny inversion of the story of computers that I’m used to telling. Until now, the hard part was in hiding how the computer works; information wants to be free, hackers usually share everything they know, and it took decades for big tech companies to finally close and lock everything up in the proprietary, inscrutable devices consumers are now laden with outside the liberating world of Free Software. Suddenly, with modern neural networks, we have the device before understanding the specifics inner-mechanisms of the runtime state. Like gardeners before modern biology, we can grow the things; we just can’t explain why or how why they do what they do.
Over the coming years, we will keep lifting the lid on the black box of their inner-workings. And, in doing so, we will rediscover everything about language which we have forgotten in the last mad century of free-range language acquisition under electric conditions.
Prescriptions—or should we say, actual and proper education—in language is the new science.
Don’t mistake my predictions here for optimism; this is a call to arms. This is a fight with winners and losers. I think wordcels have the advantage over the shape rotaters. Language works by resonant metaphor, not the machinic manipulation of conceptual schema like logic.
I feel very lucky to be a long-time rationalist tech guy who swam deeply into an arcane, dead branch of literary criticism in his mid-30s—I think I’m better prepared than most. This blog is called Less Mad, and I must once again rededicate myself to the stated mission. So take up your swords, and your pens, and lets formulate the purpose, our shared telos to cultivating the clarity and sanity which language, well learned and well spoken, can provide.
To raising the standards of common language, by learning from serpents while remaining gentle like lambs.
Speak or be spoken for.