Semantic Infrastructure, Insufficiency, and the False Romance of Interoperability
A recent Substack essay, Jessica Talisman’s Language Is the Bridge, makes a claim that is increasingly common in discussions of artificial intelligence, knowledge graphs, ontologies, metadata, and semantic infrastructure: that language is the bridge between human understanding and machine action. The claim is attractive, and not merely because ‘bridge’ is one of those metaphors that allows technical discourse to cosplay as wisdom literature. It captures something real. AI systems, semantic architectures, ontologies, taxonomies, controlled vocabularies, and knowledge graphs do not run on raw reality. They run on structured representations. Those representations require labels, definitions, mappings, alignments, constraints, and interpretive discipline. In that sense, language work is not decorative. It is infrastructural. But the metaphor is also dangerous.
I read from the Wrong Curve: Free Speech, Pseudo-Invariance, and the Grammar of Liberal Rights. This essay is freely available on Zenodo at https://doi.org/10.5281/zenodo.19636760. This segment is the Abstract and the Introduction.
In this essay, I argue that free speech discourse is structured by a category error whose source lies upstream of speech itself: in the treatment of ‘freedom’ as a stable philosophical primitive when it functions, in practice, as an essentially contested concept operating under a systematically inflated presumption of effectiveness.
tl;dr: I don’t believe in free speech.
We’ve all likely heard that the freedom to swing one’s fist ends at the tip of another’s nose. I can accept this without argument for the purpose of this assertion. Your freedom TO violates my freedom FROM.
Audio: NotebookLM summary podcast of this topic.
The problem is that one’s words don’t stop. In some cases, they continue in the manner of pollution that I don’t want my ear holes to be exposed to this noise. In the social media age, this effect is trebled and molests my eyes. This is especially egregious for misinformation and disinformation, which is to say, much of the internet and beyond.
This impact hasn’t been suitably addressed, so I wrote about it. Here, I read.
I had planned to write a blog post on The Remains of the Day, but I posted it on Substack instead because I changed the scope. I also created this podcast on NotebookLM.
Audio: NotebookLM summary podcast of this topic.NotebookLM Infographic on this topic.
I have just started reading Kazuo Ishiguro’s The Remains of the Day, and I must confess that I can’t read like a normal person.
I made it roughly fifteen pages before the analytical lenses clamped on and refused to retract. Stevens – Ishiguro’s impeccable, heartbreaking narrator – opens with a monologue about what makes a great butler, and I found myself not following the argument so much as watching the scaffolding. Because Stevens isn’t really theorising butlers. He’s performing a world. And the performance is so seamless, so grammatically composed, that it almost disguises itself as description.
Audio: NotebookLM summary podcast of this topic.
Here’s the trick. Stevens treats terms like greatness, dignity, and quality as though they named discoverable features of reality. He doesn’t present them as historically contingent, institution-soaked, liable to buckle under interrogation. He presents them as if the universe itself had ratified them, and he’s merely reporting. It’s the calm, definitional tone of a man who believes he’s describing essences when he is actually curating preferences.
Take butler itself. In ordinary usage, it looks occupationally concrete – a person who does a particular job. But in Stevens’s mouth, it stops being a role descriptor and becomes a bearer of metaphysical vocation. The word accrues civilisational mythology, evaluative freight, quasi-sacred weight, until it no longer refers to a man who answers a door but to a figure who embodies an entire moral cosmology. That is a lot for a job title to carry, and the weight does not come from the world. It comes from the grammar.
Then there is dignity. An interlocutor reasonably suggests that dignity might be something like beauty, a weasel word, like aesthetic, perceptual, taste-bound. Stevens refuses the comparison. Dignity must not sound contingent. It must present itself as something sturdier than preference, almost an objective feature of character, something a person has in the way a table has weight.
And then, naturally, he proceeds to explain it in entirely aesthetic terms: comportment, bearing, restraint, and poise – the look of self-command.
He denies the aesthetic register but then relies on it. His explicit philosophy and his operative language diverge. This isn’t a slip. It’s structural. The term needs to disavow its own supports in order to retain authority. Dignity must deny its kinship with aesthetic discrimination precisely because, without that kinship, it has no flesh.
We do this constantly – take the shakiest words and polish them until they look load-bearing. The more pressure the term is under, the more ceremonially we utter it.
NotebookLM Infographic on this topic.
Weasel words
There’s a name for what Stevens is doing, and I’ve spent a perhaps inadvisable amount of time trying to articulate it.
In my own work, A Language Insufficiency Hypothesis, I distinguish between terms that behave differently under pressure. Some are genuinely invariant – they hold stable reference across contexts, and disagreement about them can be resolved by checking. Some are contestable – stable enough for coordination and argument, but insufficiently fixed to secure the same uptake everywhere. And some are fluid – still rhetorically potent, still action-guiding, but drifting across contexts without anyone quite admitting it.
Stevens’s vocabulary sits at the dangerous end of that spectrum. Greatness and dignity are classic contestables, possibly shading into fluids: they retain social force long after referential stability has gone soft. They still feel like they mean something precise. They don’t. And Stevens’ entire identity is organised around treating them as though they do.
This matters because the language isn’t sitting atop an independent self. It’s helping to constitute the self Stevens can intelligibly be. His selfhood is assembled from grammars of rank, service, and propriety. The role vocabulary doesn’t describe a prior person. It builds one.
This brings us to his father
Stevens admits that his father might belong among the great butlers, despite lacking certain attributes Stevens associates with the newer generation. He notes the gap, then writes it off. The missing qualities belong to a different era; they are secondary, accidental, beside the essential point. His father remains the reference.
But notice what has happened. Stevens is not applying a neutral criterion and then discovering that his father qualifies. He’s adjusting the criterion around the father he already reveres. The supposed standard of greatness turns out to be less a fixed measure than a retrospective act of preservation. Firstly, canonise the figure, then tidy the theory.
The father is doing at least three jobs at once: He’s an exemplar; he’s a stabilising emotional anchor; and he’s a concealed limit-case that proves the category is being bent.
That’s how value-laden language typically works. The abstraction pretends to govern the attachment, but really the attachment is governing the abstraction. Stevens wants the dignity of a rule whilst retaining the comfort of a cherished exception. This isn’t hypocrisy in the simple sense. It’s something more interesting: it’s the ordinary mechanics of human categorisation when the stakes are personal.
And of course, the father isn’t merely a person Stevens admires. He’s the living conduit through which Stevens inherits the whole ontology of service. When Stevens prefers his father as the reference point, he’s anchoring the category in the formative figure through whom the category became intelligible in the first place. The father is part of the installation mechanism.
Then there is the tiger
With the reverent caveat that it may be apocryphal, Stevens recounts a story in which an unidentifiable butler known to his father, confronted by a tiger under a dining room table (the details are gloriously improbable), maintains perfect composure. He doesn’t flinch. He doesn’t break form. He remains, in the deepest sense, a butler.
Whether the story is true barely matters. Its function isn’t evidential but exemplary. It condenses a whole metaphysics of dignity into narrative form. It is a saint’s life for the religion of service.
And that’s the key. For the father – and then for Stevens – dignity is not a detachable opinion. It’s a lived orientation. In the face of the tiger, one doesn’t first deliberate about values and then choose restraint. One already inhabits a world in which restraint is what seriousness looks like. The salience structure is preinstalled. The event is interpreted through it automatically.
The story converts dignity from abstraction into legend. It makes composure appear not merely admirable but real. It turns comportment into ontology. And it authorises imitation through lineage: the son inherits not just a standard but a dramatic image of what the standard looks like when tested.
The possible apocryphal status doesn’t weaken this. If anything, it strengthens it. The father has become less a man than a vessel for the category. The tale smooths over ambiguity and turns a contingent life into an emblem. We’re forever embalming values in stories, then pretending the stories prove the values rather than merely rehearse them.
So Ishiguro’s opening chapter is doing at least four things at once.
It presents contestable terms as if they were settled realities.
It shows institutional language masquerading as neutral description.
It reveals a person whose identity has been stabilised – constituted, really – by those terms.
And it lets the reader feel the gap between the serenity of the grammar and the fragility of what it is trying to hold together.
That last part is the killer. Stevens’s calm definitional tone is itself evidence of instability. The more pressure the terms are under, the more polished the delivery. He’s performing maintenance whilst pretending to describe an essence.
I’m only one chapter in. I suspect Ishiguro is going to make this hurt.
I published the first version of this essay in February, arguing that the Frege–Geach problem, that three-score-year-old albatross around expressivism’s neck rests on a category error. Analytic philosophers were polite about it in the way that analytic philosophers are polite about things they intend to ignore. I don’t often revise my manuscripts, opting instead to publish a new and improved version, but the meat of this one remained strong and not worth revisiting as much as fortifying.
The trouble was that I’d dissolved the problem without resolving it. Good enough for me. Others were less convinced. Telling people they’ve been asking the wrong question is satisfying but insufficient without a better one. Version 1.1 tidied the prose. Version 1.2 does the actual work.
The new section (§4, if you’ve already read previous versions) introduces recruitable expressions – a broader class of expressions (moral predicates, thick evaluative terms, epistemic and institutional vocabulary) whose full functional load is attenuated under embedding whilst a thinner inferential profile remains available for reasoning. The standard of practical inferential adequacy replaces the demand for semantic identity: what ordinary reasoning requires is not invariance but inferential sufficiency. And the pattern isn’t peculiar to moral language – a noted goal –, which means Frege–Geach stops looking like a special embarrassment for expressivism and starts looking like one symptom of a general feature of how natural language handles multi-functional expressions under logical stress.
The essay is dissolved as a demand for unrestricted semantic invariance. It is resolved insofar as the behaviour it identifies is explained, predicted, and shown to be general.
The revised paper is available here, near the rest of my manuscripts:
The first step is to stop pretending that ‘truth’ names a single thing.
Philosopher Bernard Williams helpfully distinguished between thin and thick senses of truth in Truth and Truthfulness. The distinction is simple but instructive.
In its thin sense, truth is almost trivial. Saying ‘it is true that p’ typically adds nothing beyond asserting p. The word ‘true’ functions as a logical convenience: it allows endorsement, disquotation, and generalisation. Philosophically speaking, this version of truth carries very little metaphysical weight. Most arguments about truth, however, are not about this thin sense.
In practice, truth usually appears in a thicker social sense. Here, truth is embedded in practices of inquiry and communication. Communities develop norms around sincerity, accuracy, testimony, and credibility. These norms help stabilise claims so that people can coordinate action and share information.
At this level, truth becomes something like a social achievement. A statement counts as ‘true’ when it can be defended, circulated, reinforced, and relied upon within a shared framework of interpretation. Evidence matters, but so do rhetoric, persuasion, institutional authority, and the distribution of power. This is the sense in which truth is rhetorical, but rhetoric is not sovereign.
NotebookLM Infographic on this topic. I prompted NotebookLM to illustrate a 4-layered model that shows how removed language is from encounter, attention, conception, and representation of what we normally consider to be reality. This view is supported by both MEOW and LIH.
Human beings can imagine almost anything about the world, yet the world has a stubborn habit of refusing certain descriptions. Gravity does not yield to persuasion. A bridge designed according to fashionable rhetoric rather than sound engineering will collapse regardless of how compelling its advocates may have been.
This constraint does not disappear in socially constructed domains. Institutions, identities, norms, and laws are historically contingent and rhetorically stabilised, but they remain embedded within material, biological, and ecological conditions. A social fiction can persist for decades or centuries, but eventually it encounters pressures that force revision.
Subjectivity, therefore, doesn’t imply that ‘anything goes’. It simply means that all human knowledge is mediated.
We encounter the world through perception, language, culture, and conceptual frameworks. Every description is produced from a particular standpoint, using particular tools, within particular historical circumstances. Language compresses experience and inevitably loses information along the way. No statement captures reality without distortion. This is the basic insight behind the Language Insufficiency Hypothesis.
At the same time, our descriptions remain answerable to the constraints of the world we inhabit. Some descriptions survive repeated encounters better than others.
In domains where empirical constraint is strong – engineering, physics, medicine – bad descriptions fail quickly. In domains where constraint is indirect – ethics, politics, identity, aesthetics – multiple interpretations may remain viable for long periods. In such cases, rhetoric, institutional authority, and power often function as tie-breakers, stabilising one interpretation over others so that societies can coordinate their activities. These settlements are rarely permanent.
What appears to be truth in one era may dissolve in another. Concepts drift. Institutions evolve. Technologies reshape the landscape of possibility. Claims that once seemed self-evident may later appear parochial or incoherent.
In this sense, many truths in human affairs are best understood as temporally successful settlements under constraint.
Even the most stable arrangements remain vulnerable to change because the conditions that sustain them are constantly shifting. Agents change. Environments change. Expectations change. The very success of a social order often generates the tensions that undermine it. Change, in other words, is the only persistence.
The mistake of traditional realism is to imagine truth as a mirror of reality – an unmediated correspondence between statement and world. The mistake of crude relativism is to imagine that language and power can shape reality without limit. Both positions misunderstand the situation.
We do not possess a final language that captures reality exactly as it is. But neither are we free to describe the world however we please. Truth is not revelation, and it is not mere invention.
It is the provisional stabilisation of claims within mediated encounter, negotiated through language, rhetoric, and institutions, and continually tested against a world that never fully yields to our descriptions. We don’t discover Truth with a capital T. We negotiate survivable descriptions under pressure.
I’ve long had a problem with Truth – or at least the notion of it. It gets way too much credit for doing not much at all. For a long time now, philosophers have agreed on something uncomfortable: Truth isn’t what we once thought it was.
Truth isn’t what we once thought it was
The grand metaphysical picture, where propositions are true because they correspond to mind-independent facts, has steadily eroded. Deflationary accounts have done their work well. Truth no longer looks like a deep property hovering behind language. It looks more like a linguistic device: a way of endorsing claims, generalising across assertions, and managing disagreement. So far, so familiar.
Audio: NotebookLM summary podcast of this topic.
What’s less often asked is what happens after we take deflation seriously. Not halfway. Not politely. All the way.
That question motivates my new paper, Truth After Deflation: Why Truth Resists Stabilisation. The short version is this: once deflationary commitments are fully honoured, the concept of Truth becomes structurally unstable. Not because philosophers are confused, but because the job we keep asking Truth to do can no longer be done with the resources we allow it.
The core diagnosis: exhaustion
The paper introduces a deliberately unromantic idea: truth exhaustion. Exhaustion doesn’t mean that truth-talk disappears. We still say things are true. We still argue, correct one another, and care about getting things right. Exhaustion means something more specific:
After deflation, there is no metaphysical, explanatory, or adjudicative remainder left for Truth to perform.
Truth remains grammatically indispensable, but philosophically overworked.
Image: NotebookLM infographics of this topic. (Please ignore the typos.)
The dilemma
Once deflationary constraints are accepted, attempts to “save” Truth fall into a simple two-horn dilemma.
Horn A: Stabilise truth by making it invariant. You can do this by disquotation, stipulation, procedural norms, or shared observation. The result is stable, but thin. Truth becomes administrative: a device for endorsement, coordination, and semantic ascent. It no longer adjudicates between rival frameworks.
Horn B: Preserve truth as substantive. You can ask Truth to ground inquiry, settle disputes, explain success, or stand above practices. But now you need criteria. And once criteria enter, so do circularity, regress, or smuggled metaphysics. Truth becomes contestable precisely where it was meant to adjudicate.
Stability costs substance. Substance costs stability. There is no third option waiting in the wings.
Why this isn’t just abstract philosophy
To test whether this is merely a theoretical artefact, the paper works through three domains where truth is routinely asked to do serious work:
Moral truth, where Truth is meant to override local norms and condemn entrenched practices.
Scientific truth, where Truth is meant to explain success, convergence, and theory choice.
Historical truth, where Truth is meant to stabilise narratives against revisionism and denial.
In each case, the same pattern appears. When truth is stabilised, it collapses into procedure, evidence, or institutional norms. When it is thickened to adjudicate across frameworks, it becomes structurally contestable. This isn’t relativism. It’s a mismatch between function and resources.
Why this isn’t quietism either
A predictable reaction is: isn’t this just quietism in better prose?
Not quite. Quietism tells us to stop asking. Exhaustion explains why the questions keep being asked and why they keep failing. It’s diagnostic, not therapeutic. The persistence of truth-theoretic debate isn’t evidence of hidden depth. It’s evidence of a concept being pushed beyond what it can bear after deflation.
The upshot
Truth still matters. But not in the way philosophy keeps demanding. Truth works because practices work. It doesn’t ground them. It doesn’t hover above them. It doesn’t adjudicate between them without borrowing authority from elsewhere. Once that’s accepted, a great deal of philosophical anxiety dissolves, and a great deal of philosophical labour can be redirected.
The question is no longer “What is Truth?” It’s “Why did we expect Truth to do that?”
The paper is now archived on Zenodo and will propagate to PhilPapers shortly. It’s long, unapologetically structural, and aimed squarely at readers who already think deflationary truth is right but haven’t followed it to its endpoint.
Read it if you enjoy watching concepts run out of road.
Philosophics 