There is a moment, familiar to physicists and poets alike, when a word stops working. You reach for it the way you reach for a light switch in a power cut — the gesture is right, the word is there, but nothing illuminates. The universe has done something it was not supposed to do, and language, for a breath, cannot follow.

This essay is about that gap: the space between what we discover and what we can say. It is also, quietly, about why that gap matters not just for science communication but for thinking itself — because the words we have shape the thoughts we can reach for.

I. Language as the Shape of Thought

The idea that language does more than name a pre-existing reality — that it cuts reality at its joints, and cuts it differently depending on which language you speak — is both old and contentious. Its modern form begins with Edward Sapir and Benjamin Lee Whorf. Sapir argued in 1929 that the real world is to a large extent unconsciously built up on the language habits of a group — that we are, in his phrase, "very much at the mercy of the particular language which has become the medium of expression for our society" (Sapir, 1929, as cited in Whorf, 1956). Whorf, working from Sapir's framework and from his own detailed study of Hopi and other Native American languages, pushed the claim further: the grammatical system of each language is not merely a reproducing instrument for voicing ideas but rather the shaper of ideas, the program and guide for the individual's mental activity (Whorf, 1956). We dissect nature, he wrote, along lines laid down by our native languages. The strong version of this claim — that language determines thought — has since been substantially qualified. But the weak version has accumulated enough empirical weight that it can no longer be dismissed as romantic speculation.

In a landmark study, cognitive scientist Lera Boroditsky demonstrated that Mandarin and English speakers represent time differently in their mental architecture. English speakers overwhelmingly reach for horizontal spatial metaphors when reasoning about time. Mandarin speakers, while also using horizontal metaphors, show a significantly stronger tendency toward vertical ones — earlier events below, later events above. When participants were primed with vertical spatial images before making temporal judgments, Mandarin speakers were faster to confirm vertical temporal relations than English speakers, suggesting that the vertical framing their language makes habitual had become a genuine cognitive resource (Boroditsky, 2001). Language, it turned out, was not a coat hung over a pre-formed conceptual hook. It was part of the hook.

A more precisely controlled demonstration of the same principle comes from colour perception. Russian makes an obligatory grammatical distinction between lighter blues — goluboy — and darker blues — siniy. English uses a single word for both. Winawer and colleagues tested Russian and English speakers on a speeded colour discrimination task using blue stimuli that straddled the siniy/goluboy boundary. Russian speakers were significantly faster at discriminating blues that fell into different linguistic categories than blues that fell within the same one. Crucially, introducing a verbal interference task — silently rehearsing a string of digits — eliminated the Russian speakers' advantage entirely, while a spatial interference task did not. The effect was specifically linguistic, operating online during perception, not merely a memory artefact (Winawer, Witthoft, Frank, Wu, Wade, & Boroditsky, 2007). The language had carved the colour spectrum at a joint that English left unseparated — and that cut was cognitively real.

The implications deepened when researchers turned to communities whose spatial vocabulary was structured more radically still. Stephen Levinson's (1997) study of the Guugu Yimithirr people of Queensland found that they orient space entirely through absolute cardinal directions — north, south, east, west — rather than egocentric terms like "left" and "right." Speakers of such languages maintain an extraordinary, apparently unconscious awareness of their cardinal bearing at all times, a faculty that speakers of egocentric languages do not develop in the same way.

Boroditsky and Gaby (2010) extended this finding into time among the Kuuk Thaayorre, a neighbouring Australian community with a similar absolute spatial orientation. When asked to arrange temporal sequences — photographs of a person ageing, or a banana being eaten — English speakers arranged them left to right, Mandarin speakers sometimes top to bottom. The Kuuk Thaayorre, regardless of which direction they happened to be facing, consistently arranged them east to west, following the arc of the sun. The language had not just given them a word. It had given them a compass built into cognition itself.

The language had not just given them a word. It had given them a compass — woven into the structure of thought itself.

— On spatial language and mental architecture

This is not to say that speakers of different languages inhabit entirely different realities. The weak version of linguistic relativity — now far better supported than the strong Whorfian version — holds that language is a persistent, probabilistic nudge: it makes certain thoughts more available, certain framings more automatic, certain distinctions more salient (Iskandar et al., 2025). The grammar of your mother tongue is a pair of glasses you never take off. You can still see without them. But the tint is always there.

II. The Universe Refuses to Behave

Physics offers the most extreme stress-test of this idea, because physics keeps discovering things no language was built to describe. The quantum world arrived already grammatically impossible. An electron does not have a definite position until it is measured — not because we lack the instruments, but because no definite position exists to measure. English grammar, with its confident assignment of subjects and predicates and settled locations in space, resists this at every turn.

Cosmology presents an even more vertiginous confrontation. Take the Big Bang — two of the most loaded words in science. The name conjures an explosion, a before and after, a moment when everything began. But this framing is almost certainly misleading. A 2025 survey of physicists gathered for the "Black Holes Inside and Out" conference in Copenhagen found that the only statement about the Big Bang to gain majority approval — from 68% of respondents — was simply that the universe evolved from a hot, dense early state. The stronger claim that it represented an absolute beginning of time did not reach consensus (Chen, Halper, & Afshordi, 2025). The phrase "Big Bang" has a before built into it. The physics may not.

A paper published in Physical Review Letters in 2026 shows how far the mathematics can run beyond ordinary language. Liu, Quintin, and Afshordi demonstrated that within a quadratic gravity framework — one whose coupling constants become well-behaved at arbitrarily high energies — the physics can be described continuously through the very region that classical general relativity treats as a singularity (Liu, Quintin, & Afshordi, 2026). There is no point at which the equations break down and demand a "beginning." The formalism does not require one. The word does.

What does it mean to call something a beginning if nothing begins there? The word carries the full weight of a narrative grammar — stories have openings, causes precede effects, time has a first entry. When physics describes a region that ordinary language insists on calling the start, the difficulty is not only technical. It is linguistic. We do not yet have a word for "the mathematical region our narrative grammar calls a beginning but which the physics treats as continuous." We gesture toward it with equations. The equations work. The words keep lagging behind.

III. When Experts Cannot Agree on a Name

The Copenhagen survey result is worth dwelling on. Here were some of the world's most accomplished physicists, gathered around the very questions that define the frontier of the discipline, unable to reach majority agreement on frameworks that dominate textbooks and grant applications. The cosmological constant, cosmic inflation, string theory — while each commanded support from a significant minority, none won a majority (Chen et al., 2025). The survey authors noted that this should give us pause before describing any of these frameworks as consensus science.

This is not a failure of physics. It is physics being honest about the edge of its own language. At the frontier, there is no settled vocabulary because the concepts are not yet settled. The words we have — "singularity," "inflation," "the beginning" — are placeholders: they name a gap shaped like a thing we have not yet understood. The danger comes when we mistake the name for the understanding, when the placeholder calcifies into a concept and we stop seeing the gap it was always pointing at.

This is the precise opposite of the Kuuk Thaayorre's remarkable precision. Their language encoded a real directional structure of the world with such fidelity that it built a navigational faculty into daily cognition — the vocabulary matched the territory. The challenge at the cosmological frontier is that no vocabulary yet matches the territory, and the ones we have may actively mislead us, quietly pulling our models back toward human-scale intuitions about before and after, inside and outside, beginning and end.

IV. Building Better Words

There is a productive response to this, one that physicists and linguists, in their different ways, have both found. It is to hold the word lightly — to use it as a scaffold you know you will eventually remove, rather than as a wall.

Albert Einstein spent nearly a decade searching before completing his general theory of relativity in November 1915, a theory that overturned Newtonian notions of absolute space and time and replaced them with a geometry of curved spacetime. The search began not with equations but with an image. At sixteen, Einstein imagined chasing a beam of light through space — and noticed that classical physics allowed no coherent answer to what he would see. Maxwell's equations required light to travel at a constant speed regardless of the observer; Newtonian mechanics said he should be able to catch up and see a frozen wave. The contradiction could not be expressed within the available vocabulary. No existing concept covered it. Einstein held the image for nearly a decade until the mathematics caught up (Norton, 2004). What made that possible was a willingness to work ahead of any settled language — to reason through images and thought experiments when formal concepts had not yet arrived. His conviction that imagination matters more than knowledge was a description of method. You cannot wait for the right words before you start thinking.

Mathematical language has real advantages here. Equations make no commitment to a narrative direction. The formalism of quadratic gravity contains no word for "beginning" and is therefore not seduced by one. This is precisely why Liu, Quintin, and Afshordi could write a theory that runs cleanly through the region ordinary language insists on calling a singularity — the mathematics was not infected by the metaphor.

Natural language can approach this discipline too, but it requires conscious effort: holding the ordinary description and the corrective one simultaneously, without collapsing them. This is what physicists mean by physical intuition — not a vague feeling, but the trained ability to navigate the gap between what the equations say and what any sentence will make them sound like.

The research on linguistic relativity suggests that this capacity is not purely individual. It is shaped by the language community you work within. A community that builds more precise scientific metaphors — that stops treating the Big Bang as an explosion in pre-existing space, or quantum states as hidden classical realities waiting to be revealed — thinks more clearly about these problems, not merely communicates more carefully. The words matter because the words are part of the thinking.

V. The Limits of the Framework

Conceptual and linguistic frameworks are not merely passive vehicles for communicating scientific knowledge. They are active structures that shape which hypotheses seem natural, which questions seem worth asking, and which answers count as explanations. This influence is sharpest at the frontier, where inquiry extends beyond the scale of ordinary human experience and what Sellars called the "manifest image" — the framework of persons, causes, and everyday objects through which ordinary language organises reality — provides no reliable guide (Sellars, 1962; Vacura, 2015).

Ludwig Wittgenstein was among the first to press this point philosophically. In the Tractatus Logico-Philosophicus he argued that the limits of one's language define the limits of one's world — that the structure of what can be said constrains what can be clearly thought (Wittgenstein, 1922). His later Philosophical Investigations refined rather than abandoned this: meaning, he argued, arises through use within particular social practices — "language-games" — which means language is also malleable, and that new practices can open new conceptual territory (Pinto, 2024). The implication cuts both ways. Language constrains thought, but new language can expand it.

Wilfrid Sellars gave this a specifically scientific dimension. His distinction between the manifest image and the "scientific image" produced by theory identifies a permanent source of friction in science (Sellars, 1962). The two frameworks are not simply different descriptions of the same facts; they are organised differently and can come into genuine conflict. When physics ventures beyond the manifest image — into quantum superposition, spacetime curvature, or a cosmology without a temporal beginning — it is not adding new facts to a neutral vocabulary. It is entering territory the manifest image was never built to map (Vacura, 2015).

Thomas Kuhn made the historical dimension of this explicit. In The Structure of Scientific Revolutions he argued that major scientific advances are not merely accumulations of new observations but transformations of the conceptual vocabulary through which observations are made and interpreted (Kuhn, 1962). When a scientific lexicon is replaced in a paradigm shift, the old vocabulary can no longer express the new ideas cleanly: the Copernican claim that planets orbit the sun could not be stated within Ptolemaic terms without distortion (Kuhn, 1962). The current difficulty in agreeing on what the Big Bang was — an absolute origin, a boundary condition, or a region the mathematics simply passes through — may signal that cosmology is approaching exactly this kind of threshold.

A common objection at this point is that mathematics transcends these ambiguities — that it is precise in ways natural language is not. But symbols do not interpret themselves. Their meaning depends on the conceptual frameworks, metaphors, and shared practices through which physicists understand what they are doing. The equations of quadratic gravity pass through what we call a singularity; the word "singularity" is still needed to explain what problem the equations address. Mathematics reduces linguistic ambiguity. It does not escape conceptual interpretation.

And if mathematics is itself a kind of language — which in the relevant sense it is — then it carries probabilistic nudges of its own. The clearest historical case is notation: Leibniz's calculus and Newton's described identical mathematics, yet when British mathematicians refused to adopt Leibniz's more expressive notation out of loyalty to Newton, their mathematics fell measurably behind the European continent for the better part of a century (Li, 2024). Notation is not neutral. It makes certain moves feel natural and others cumbersome — precisely what Boroditsky's priming experiments showed natural language doing to temporal reasoning. The mechanism is the same; the medium is different.

The choice of mathematical structures runs deeper still. Most of the tools physicists reach for — differential equations, smooth manifolds, calculus — assume that spacetime is continuous and infinitely divisible. This is not an established fact about the universe; it is a mathematical habit inherited from a specific intellectual tradition. When equations break down at a singularity, part of what has happened is that continuous mathematics has reached a boundary the universe may not actually possess. The quadratic gravity framework addresses this by working in a regime where higher-order curvature terms dominate and the theory remains UV-complete — a different mathematical language, not merely a different solution within the same one.

There is also the question of aesthetic norms. Mathematical communities develop strong, socially enforced intuitions about what a beautiful or elegant result looks like, and these intuitions guide research as reliably as any grammatical habit (Jevtić, Kostić, & Maksimović, 2024). String theory's hold on theoretical physics over several decades was sustained in significant part by mathematical elegance in the near-absence of experimental confirmation — a pattern that Ritson and Camilleri (2015) documented as a sociological as much as a scientific phenomenon. That is a probabilistic nudge as powerful as any grammar, and considerably harder to identify precisely because it presents itself as rigour rather than preference.

The honest position — and the one that follows from taking Wittgenstein, Sellars, and Kuhn seriously — is therefore that mathematics is a less linguistically biased language, not a bias-free one. Equations are not bound to a subject-predicate structure, do not force a before and after, and can describe relationships symmetric in time or holding across spaces of arbitrary dimension in ways that sentences struggle even to gesture at. The formalism of general relativity does not say that time flows forward — physicists supply that when they interpret it. So mathematics genuinely escapes some of the traps of ordinary language. But it brings its own image along.

It is worth being precise about what distinguishes natural language, mathematics, and specialised symbolic systems, since they constrain thought in different ways and to different degrees. Natural language is the most pervasive and the least visible in its influence: it operates before scientific thinking begins, shaping which questions feel natural, which analogies feel apt, and which answers count as satisfying explanations. Whorf's insight — that we dissect nature along lines laid down by our native languages — applies here with full force (Whorf, 1956). Mathematical language escapes the narrative grammar of natural language but substitutes structural assumptions of its own: continuity, smoothness, the algebra of real numbers, the geometry of Euclidean space. These are not neutral containers; they are frameworks that make certain physical theories feel natural and others feel contrived. Specialised symbolic systems — Feynman diagrams, tensor network notation, the language of category theory — occupy a third register. They are purpose-built representational tools, designed precisely to make certain classes of problem tractable and to suppress others. A Feynman diagram does not merely represent a calculation; it shapes which calculations a physicist reaches for first. Each system, then, carries its own probabilistic nudge, its own bias toward certain configurations of the thinkable. The task of the scientist is not to escape all such systems — that is impossible — but to hold them consciously enough to notice when they are steering rather than serving.

Scientific progress therefore depends on more than the accumulation of data. It requires the ongoing development of conceptual vocabularies capable of organising what is found. How such vocabularies emerge, evolve, and reshape inquiry deserves sustained attention — for advances in knowledge often begin, as Kuhn recognised, with advances in the language through which knowledge becomes thinkable.

VI. The Sky Remains

There is something humbling and thrilling about the situation we find ourselves in. The universe is larger than any of the languages we have developed to describe it. This is not a complaint. It is, in a sense, the most important empirical finding of the last century: reality keeps exceeding our vocabulary.

The Kuuk Thaayorre built a compass into their language and carried it in their heads. The physicists at Copenhagen could not agree on what words to use for what they had found. Both facts point at the same thing: the relationship between the structures we can speak and the structures that are there. Sometimes language is a precision instrument, carving cognition to match the world with extraordinary fidelity. Sometimes it is a blunt inheritance — a tool built for an older world that keeps snagging on the new one.

There is a darker reading of the evidence assembled here, and it deserves to be stated plainly. If language shapes thought as persistently as the research suggests — if the grammar of time nudges cognition, if mathematical notation steers a field for a century, if aesthetic norms sustain entire research programmes in the absence of evidence — then language begins to look less like a clumsy tool we constantly outgrow and more like a rigid ceiling we keep pressing against without knowing it is there. On this reading, the history of science is not only a story of imagination overcoming ignorance. It is also a story of frameworks quietly setting the terms of what imagination was allowed to reach for, and of how much it never reached at all.

The history of quantum mechanics offers the most instructive example of this ceiling at work. Niels Bohr insisted that quantum phenomena could only be described using the concepts of classical physics, because those were the only concepts physicists had inherited for communicating physical experience. This was not a temporary measure; for Bohr it was a permanent epistemological constraint — the unambiguous description of any measurement, he argued, must be framed in terms of classical physical theories, and the language of Newton and Maxwell would remain the language of physics indefinitely. Several commentators have since argued that this insistence actively foreclosed the development of genuinely non-classical conceptual frameworks for decades, trapping the interpretation of quantum mechanics within a vocabulary the theory itself had already outgrown (Zinkernagel, 2016). The ceiling was not imposed from outside. It was built into the language in which the questions were being asked.

The consequences were concrete. Hugh Everett III, then a graduate student at Princeton, developed his relative-state formulation of quantum mechanics — what became known as Many Worlds — in 1957, proposing that the universal wave function never collapses and that quantum superpositions simply branch into all possible outcomes simultaneously. The formulation required no classical observer and no wave function collapse: it operated entirely within quantum mechanical language rather than importing classical concepts to handle measurement. Bohr's response, conveyed through Niels Bohr Institute colleagues, was dismissive; the framework did not fit the conceptual vocabulary Bohr regarded as mandatory, and it was largely ignored for nearly two decades. What Everett had done was not wrong the physics — the mathematics was correct. He had built a new language. And the existing community could not initially hear what it said (Kuhn, 1962). The ceiling, in this case, was not resolved by better data. It was resolved by a conceptual vocabulary that did not need the old constraints to function.

Kuhn's paradigm analysis makes the structural point general. Before Copernicus, the question "why does the Earth move?" could not be asked in a way that made astronomical sense, because the lexicon of Ptolemaic astronomy treated terrestrial rest as a starting assumption, not a hypothesis. Before Einstein, the question "what would happen if there were no fixed reference frame?" had no place in Newtonian mechanics — not because physicists were incurious, but because the conceptual vocabulary provided no foothold for the question. In each case, the ceiling was the language itself: not a prohibition, but an absence, a region of thought the framework simply did not make available (Kuhn, 1962).

The evidence does not quite support the conclusion that language is always a ceiling, but it makes the optimistic alternative harder to hold cheaply. What rescues it is precisely the historical record: ceilings do get broken. Einstein held his light-beam image for nearly a decade until the mathematics arrived to express what it implied. Bohr's classical constraint was eventually contested, and alternative interpretations have since developed precisely the non-classical conceptual vocabularies Bohr declared impossible: relational quantum mechanics, in which quantum states are defined only relative to an observer rather than absolutely (Rovelli, 1996), and QBism, which treats the quantum state as an agent's personal probability assignment rather than a description of an observer-independent reality (Fuchs, Mermin, & Schack, 2014). Leibniz's notation displaced Newton's. Liu, Quintin, and Afshordi wrote a theory that passes cleanly through a region ordinary language insists on calling an end. In each case, the framework that felt like a ceiling turned out to be a floor — a foundation for something larger, once someone noticed the walls were not load-bearing.

The word "beginning" cannot hold the sky. Neither, perhaps, can "singularity" or "inflation" or the many other terms pressed into service at the frontier. What can hold the sky — imperfectly, provisionally, with full knowledge of its own limits — is the practice of noticing when a word stops working. Of naming the gap. Of sitting with the discomfort of description failing, long enough to build something better.

That is not a weakness of language. It is what language, at its best, does.

References

  1. Boroditsky, L. (2001). Does language shape thought? Mandarin and English speakers' conceptions of time. Cognitive Psychology, 43(1), 1–22.
    https://doi.org/10.1006/cogp.2001.0748 · ScienceDirect
  2. Boroditsky, L., & Gaby, A. (2010). Remembrances of times east: Absolute spatial representations of time in an Australian Aboriginal community. Psychological Science, 21(11), 1635–1639.
    https://doi.org/10.1177/0956797610386621 · SAGE Journals
  3. Levinson, S. C. (1997). Language and cognition: The cognitive consequences of spatial description in Guugu Yimithirr. Journal of Linguistic Anthropology, 7(1), 98–131.
    https://doi.org/10.1525/jlin.1997.7.1.98 · Wiley
  4. Sapir, E. (1929). The status of linguistics as a science. Language, 5(4), 207–214.
    https://doi.org/10.2307/409588 · JSTOR
  5. Whorf, B. L. (1956). Language, thought, and reality: Selected writings of Benjamin Lee Whorf (J. B. Carroll, Ed.). MIT Press.
  6. Winawer, J., Witthoft, N., Frank, M. C., Wu, L., Wade, A. R., & Boroditsky, L. (2007). Russian blues reveal effects of language on color discrimination. Proceedings of the National Academy of Sciences, 104(19), 7780–7785.
    https://doi.org/10.1073/pnas.0701644104 · PNAS
  7. Chen, A. Y., Halper, P., & Afshordi, N. (2025). Copenhagen survey on black holes and fundamental physics. arXiv preprint arXiv:2503.15776.
    https://arxiv.org/abs/2503.15776
  8. Liu, R., Quintin, J., & Afshordi, N. (2026). Ultraviolet completion of the Big Bang in quadratic gravity. Physical Review Letters, 136(11), 111501.
    https://doi.org/10.1103/6gtx-j455 · APS Journals
  9. Iskandar, G. R., Salim, F., Jihan Azmi Kusmawan, & Rahmadsyah Rangkuti. (2025). Linguistic relativity in Gap × Katseye 2025 Instagram advertisement: A Whorfian weak hypothesis approach. Jurnal Penelitian Inovatif, 5(4), 3181–3196.
    https://doi.org/10.54082/jupin.1947
  10. Sellars, W. S. (1962). Philosophy and the scientific image of man (R. Colodny, Ed.). PhilPapers; Humanities Press/Ridgeview.
    https://philpapers.org/rec/SELPAT
  11. Vacura, M. (2015). Sellars's concept of manifest and scientific image. E-LOGOS: Electronic Journal for Philosophy, 22(2), 94–102.
    https://doi.org/10.18267/j.e-logos.426 · E-LOGOS
  12. Pinto, A. O. (2024). Language and thought in Wittgenstein. Advances in Social Sciences Research Journal, 11(12), 14–23.
    https://doi.org/10.14738/assrj.1112.17871 · Scholar Publishing
  13. Wittgenstein, L. (1922). Tractatus logico-philosophicus (C. K. Ogden, Trans.). Kegan Paul, Trench, Trubner.
    https://www.gutenberg.org/files/5740/5740-pdf.pdf
  14. Kuhn, T. S. (1962). The structure of scientific revolutions. University of Chicago Press.
    https://archive.org/details/thomas-s.-kuhn-the-structure-of-scientific-revolutions
  15. French, S. (2020). Imagination in scientific practice. European Journal for Philosophy of Science, 10(3), 1–21.
    https://doi.org/10.1007/s13194-020-00291-z · Springer
  16. Norton, J. D. (2004). Einstein's investigations of Galilean covariant electrodynamics prior to 1905. Archive for History of Exact Sciences, 59(1), 45–105.
    https://doi.org/10.1007/s00407-004-0085-6 · Springer
  17. Rovelli, C. (1996). Relational quantum mechanics. International Journal of Theoretical Physics, 35(8), 1637–1678.
    https://doi.org/10.1007/BF02302261 · Springer
  18. Fuchs, C. A., Mermin, N. D., & Schack, R. (2014). An introduction to QBism with an application to the locality of quantum mechanics. American Journal of Physics, 82(8), 749–754.
    https://doi.org/10.1119/1.4874855 · AIP Publishing
  19. Zinkernagel, H. (2016). Niels Bohr on the wave function and the classical/quantum divide. Studies in History and Philosophy of Modern Physics, 53, 9–19.
    https://doi.org/10.1016/j.shpsb.2015.11.001 · ScienceDirect
  20. Li, J. (2024). A historical analysis of the independent development of calculus by Newton and Leibniz. Theoretical and Natural Science, 30, 1–5.
    https://doi.org/10.54254/2753-8818/30/20240739
  21. Jevtić, F. D., Kostić, J., & Maksimović, K. (2024). Reflecting on beauty: The aesthetics of mathematical discovery. arXiv preprint arXiv:2405.05379.
    https://arxiv.org/abs/2405.05379
  22. Ritson, S., & Camilleri, K. (2015). Contested boundaries: The string theory debates and ideologies of science. Perspectives on Science, 23(2), 192–227.
    https://doi.org/10.1162/POSC_a_00168 · MIT Press
Labels: Science and Philosophy Language Cosmology Linguistics