As the publication date of A Language Insufficiency Hypothesis (LIH) draws nearer, I feel it’s a good time to promote it (obviously) and to introduce some of the problems it uncovers – including common misperceptions I’ve already heard. Through this feedback, I now understand some of the underlying structural limitations that I hadn’t considered, but this only strengthens my position. As I state at the start of the book, the LIH isn’t a cast-in-stone artefact. Other discoveries will inevitably be made. For now, consider it a way to think about the deficiencies of language, around which remediation strategies can be developed.
Audio: NotebookLM summary podcast of this content.
Let’s clear the undergrowth first. The Language Insufficiency Hypothesis is not concerned with everyday ambiguity, garden-variety polysemy, or the sort of misunderstandings that vanish the moment someone bothers to supply five seconds of context. That terrain is already well-mapped, thoroughly fenced, and frankly dull.
Take the classic sort of example wheeled out whenever someone wants to sound clever without doing much work:
‘I made a 30-foot basket’.
Video: a woman making a large basket
If you’re a basketweaver, you picture an absurdly large basket and quietly question the maker’s life choices. If you’re watching basketball, you hear ‘score’. If you’re anywhere near the context in which the sentence was uttered, the meaning is obvious. If it isn’t, the repair cost is trivial. Add context, move on, live your life.
Language did not fail here. It merely waited for its coat. This is not the sort of thing the LIH loses sleep over.
The Groucho Marx Defence, or: Syntax Is Not the Problem
Logicians and armchair philosophers love to reach for jokes like Groucho Marx’s immortal line:
‘I shot an elephant in my pyjamas. Why it was wearing my pyjamas, I’ll never know’.
Video: A man and elephant in pyjamas (no sound)
Yes, very funny. Yes, the sentence allows for a syntactic misreading. No, nobody actually believes the elephant was lounging about in striped silk. The humour works precisely because the “wrong” parse is momentarily entertained and instantly rejected.
Again, language is not insufficient here. It’s mischievous. There’s a difference.
If the LIH were worried about this sort of thing, its ambitions would be indistinguishable from an undergraduate logic textbook with better branding.
Banks, Rivers, and the Myth of Constant Confusion
Likewise, when someone in a city says, ‘I went to the bank’, no sane listener imagines them strolling along a riverbank, unless they are already knee-deep in pastoral fantasy or French tourism brochures. Context does the heavy lifting. It almost always does.
Video: Rare footage of me trying to withdraw funds at my bank (no sound)
This is not a crisis of meaning. This is language functioning exactly as advertised.
Where the Trouble Actually Starts: Contestables
The LIH begins where these tidy examples stop being helpful. It concerns itself with Contestables: terms like truth, freedom, justice, fairness, harm, equality. Words that look stable, behave politely in sentences, and then detonate the moment you ask two people what they actually mean by them. These are not ambiguous in the casual sense. They are structurally contested.
In political, moral, and cultural contexts, different groups use the same word to gesture at fundamentally incompatible conceptual frameworks, all while assuming a shared understanding that does not exist. The conversation proceeds as if there were common ground, when in fact there is only overlap in spelling.
That’s why attempts to ‘define’ these terms so often collapse into accusation:
That’s not what freedom means. That’s not real justice. You’re redefining truth.
No, the definitions were never shared in the first place. The disagreement was smuggled in with the noun.
‘Just Ignore the Word’ Is Not a Rescue
A common response at this point is to suggest that we simply bypass the troublesome term and discuss the concrete features each party associates with it. Fine. Sensible. Often productive. But notice what this manoeuvre concedes. It does not save the term. It abandons it.
If meaningful discussion can only proceed once the word is set aside and replaced with a list of clarifications, constraints, examples, and exclusions, then the word has already failed at its primary job: conveying shared meaning. This is precisely the point the LIH is making.
The insufficiency is not that language is vague, or flexible, or context-sensitive. It’s that beyond a certain level of conceptual complexity, language becomes a confidence trick. It gives us the feeling of agreement without the substance, the appearance of communication without the transaction.
At that point, words don’t merely underperform. They mislead.
Naturally, it will make more sense alongside the book. But it may still provide a bit of entertainment – and mild discomfort – in the meantime.
tl;dr: Language is generally presumed to be stable. Words mean what you think they mean, right? A table is a table. A bird is a bird. Polysemy aside, these are solid, dependable units.
Then we arrive at freedom, justice, truth, and an entire panoply of unstable candidates. And let’s not even pretend qualia are behaving themselves.
So when someone says ‘truth’, ‘free speech’, or ‘IQ’, you may suddenly realise you’ve been arguing with a cardboard cut-out wearing your own assumptions. That isn’t just interpersonal mischief. It’s language doing exactly what it was designed to do: letting you glide over the hard problems while sounding perfectly reasonable.
Audio: Short NotebookLM summary of this page content*
Video: Legacy video explaining some features of the LIH.
If that sounds banal, you’ve already fallen for the trap.
Give it a try – or wait until you’ve digested the book. Not literally, unless you’re short on fibre.
Cheers.
Written by Bry Willis
microglyphics
* As I’ve cited previously, the quality of NotebookLM varies – usually in predictable directions. This one does well enough, but it doesn’t have enough context to get the story right (because it was only drawing from this page rather than from a fuller accounting of the LIH). Its trailing comment reveals that it doesn’t grasp that “new words” don’t solve the problem.
Earlier, it suggests that language is intentionally vague. This is not an assertion I make. You can read some of the earlier incarnations, or you can wait for it to be published.
A NotebookLM Cautionary Tale for the Philosophically Curious
Apologies in advance for the didactic nature of this post.
Every so often, the universe gives you a gift. Not the good kind, like an unexpected bottle of Shiraz, but the other kind – the ‘teachable moment’ wrapped in a small tragedy. In this case, a perfectly innocent run of MEOW GPT (my Mediated Encounter Ontology engine) was fed into NotebookLM to generate a pseudo-podcast. And NotebookLM, bless its little algorithmic heart, proceeded to demonstrate every classic mistake people make when confronting a relational ontology.
Audio: The misinterpretation of MEOW GPT: On Progress by NotebookLM that spawned this post.
It’s perfect. I couldn’t have scripted a better example of How Not To Read MEOW GPT if I’d hired a team of Enlightenment rationalists on retainer.
So consider this your public service announcement – and a guide for anyone experimenting with MEOW GPT at home, preferably while sitting down and not holding onto any cherished metaphysical delusions.
Video: Surreal Light through a Prism Clip for no particular reason (No sound)
Mistake 1: Treating a Thick Concept as a Single Glorious Thing
NotebookLM began, earnestly, by trying to uncover the ‘inner architecture of honour’, as if it were a cathedral with blueprints lying around.
This is the central error:
Honour is not a thing. There is no inner architecture.
There are only patterns – drifting, contested, historically mangled patterns – that happen to share a word. If you start with ‘What is honour?’, you’ve already fallen down the stairs.
Mistake 2: Rebuilding Essence From the T0–T3 Layers
MEOW GPT gives you biological (T0), cognitive (T1), linguistic (T2), and institutional/technical (T3) mediation because that’s how constraints emerge. NotebookLM, meanwhile, reconstructed these as ‘layers’ of the same virtue – like honour was a three-storey moral townhouse with a loft conversion.
No. The tiers are co-emergent constraints, not components of a moral particle. If your conclusion looks like a metaphysical onion, you’ve misread the recipe.
Mistake 3: Sneaking Virtue Ethics in Through the Fire Exit
NotebookLM kept returning to:
an ‘internal compass’
a ‘core record of the self’
a ‘lifelong ledger’
a ‘deep personal architecture’
At this point we might as well carve Aristotle’s name into the hull.
MEOW’s stance is simple: the self is not a marble statue – it’s an ongoing social, cognitive, and technical scandal. Treating honour as a personality trait is just the old moral ontology with a new hairstyle.
Mistake 4: Treating Polysemy as Noise, Not Evidence
NotebookLM acknowledged the differing uses of ‘honour’, but always with the implication that beneath the variations lies one pure moral essence. This is backwards. The ambiguity is the point. The polysemy isn’t messy data; it’s the signature of conceptual drift.
If you treat ambiguity as a problem to be ironed out, you’ve missed half the LIH and all of the MEOW.
Mistake 5: Turning MEOW Into a TED Talk
The podcast tried to wrap things up by contrasting honour’s “deep internal permanence” with the ephemerality of digital rating systems.
It’s cute, but it’s still modernist comfort-food. MEOW does not mourn for the ‘permanent self’. It doesn’t recognise such a creature. And digital honour doesn’t ‘replace’ the old patterns; it aggressively rewrites the honour-economy into algorithmic form. If your conclusion sounds like ‘ancient virtue meets modern technology’, that’s TED, not MEOW.
So How Should You Interpret MEOW GPT?
A short cheat-sheet for anyone experimenting at home:
There is no essence. Concepts like honour, truth, integrity, and justice are drift-patterns, not objects.
The tiers describe mediation, not ingredients. They’re co-emergent pressures, not building blocks.
Thick terms lie to you. Their apparent unity is linguistic camouflage.
Ambiguity is structural. If the term looks fuzzy, that’s because the world is fuzzy there.
If a concept feels granite-solid, you’re standing on conceptual quicksand. (Sorry.)
A Friendly Warning Label
Warning: If you believe thick moral concepts have single, universal meanings, MEOW GPT may cause temporary metaphysical discomfort. Consult your ontological physician if symptoms persist.
Philosophics 