When people speak about language, they often imagine that meaning sits inside words like a substance carried in a vessel. If only we could replace “false” words with “true” ones, communication would repair itself. The history of both linguistics and information theory shows something else: semantics does not precede use. It follows frequency.
From signals to words
Joseph Fourier, in the early 19th century, demonstrated that any complex vibration can be broken down into a set of simple, repeating cycles. Each cycle can be described by its frequency (how often it repeats), amplitude (its strength), and phase (its position in the cycle). Norbert Wiener extended this thinking to noise in the 20th century, showing that even seemingly random processes could be measured in terms of how their energy was distributed across different frequencies. Claude Shannon reframed communication around probability, demonstrating that redundancy and predictability govern how information flows through a channel. Together, these insights set the stage for understanding language not as a set of isolated symbols, but as a stochastic, or probabilistic, spectral field: words act less like containers of meaning and more like oscillatory components whose probabilities of appearing in certain contexts generate coherence (Fourier, 1822; Wiener, 1930; Shannon, 1948).
Distributional semantics
Mid-20th century linguists like Zellig Harris and John Firth shifted the focus from words as isolated units to words as elements defined by their statistical environment. Harris (1951) proposed “distributional analysis,” where meaning is derived from the contexts in which words occur. Firth (1957) famously put it: “You shall know a word by the company it keeps.” The principle here is close to spectral analysis: frequency of co-occurrence between words produces structured patterns of use, and those patterns carry meaning. Later, computational linguists operationalised this. Latent Semantic Analysis (Deerwester et al., 1990) demonstrated that if you create a large matrix of how often words appear in documents, and then decompose it mathematically using a method called Singular Value Decomposition (SVD), hidden dimensions of meaning emerge. These are not predefined “truths” but patterns of probability. Probabilistic models extended this logic: Probabilistic Latent Semantic Analysis (Hofmann, 1999) and Latent Dirichlet Allocation (Blei et al., 2003) treated documents as mixtures of latent “topics,” each topic being a distribution over words. These topics are the equivalent of spectral modes: stable patterns within the field of probability.
Neural embeddings and beyond
In the 2010s, word2vec (Mikolov et al., 2013) and GloVe (Pennington et al., 2014) scaled distributional semantics to massive web corpora. These methods produced vector spaces where geometry itself encodes meaning. Yet the vectors are not intrinsic essences. They are compressed representations of the co-occurrence spectra — probability fields shaped by frequency. Modern transformers, introduced in Attention Is All You Need (Vaswani et al., 2017), generalise this principle. Their attention layers adaptively filter across positions in a sequence, highlighting some rhythms while damping others. Functionally, they act like dynamic spectral analyzers, emphasising couplings that stabilise meaning.
Implications for communication
This perspective matters directly for contemporary crises of communication — from social media turbulence to misinformation. The instinct is often to believe that “truth” can overwrite falsehood if only it is repeated enough. But semantics does not overwrite semantics. What stabilises is what repeats. A phrase, even if false, accumulates amplitude through repetition, and when synchronised across multiple channels it can lock into coherence. The probabilistic structure of language is primary; semantics is downstream. The “truth of language” is that it organises meaning by rhythm and probability.
Toward field logic
To treat language spectrally is to shift focus from static content to dynamic distributions. Detection becomes the identification of emergent couplings across contexts. Description becomes a form of filtering that stabilises some resonances and ignores others. Intervention, or interdiction shaping, becomes the modulation of amplitude and synchrony: slowing the viral, amplifying diversity, inserting friction where lock-in would otherwise occur. In this framing, language and technology are not external carriers of truth but fields in which frequency structures guide the semantics that emerge.
The conclusion is stark: meaning is not a fixed property secured once and for all. It is an emergent resonance, sustained by probability. Semantics follows frequency — and to shape communication systems responsibly, whether in science, media, or politics, we must engage at the level where rhythm, repetition, and coupling decide what survives.
References
Blei, D. M., Ng, A. Y. and Jordan, M. I. (2003). Latent Dirichlet Allocation. Journal of Machine Learning Research, 3, pp. 993–1022.
Deerwester, S., Dumais, S. T., Furnas, G. W., Landauer, T. K. and Harshman, R. (1990). Indexing by Latent Semantic Analysis. Journal of the American Society for Information Science, 41(6), pp. 391–407.
Firth, J. R. (1957). A Synopsis of Linguistic Theory 1930–1955. Studies in Linguistic Analysis. Oxford: Philological Society.
Fourier, J. (1822). Théorie analytique de la chaleur. Paris: Firmin Didot.
Harris, Z. S. (1951). Methods in Structural Linguistics. Chicago: University of Chicago Press.
Hofmann, T. (1999). Probabilistic Latent Semantic Indexing. Proceedings of the 22nd Annual International ACM SIGIR Conference, pp. 50–57.
Mikolov, T., Chen, K., Corrado, G. and Dean, J. (2013). Efficient Estimation of Word Representations in Vector Space. Proceedings of Workshop at ICLR.
Pennington, J., Socher, R. and Manning, C. (2014). GloVe: Global Vectors for Word Representation. Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing, pp. 1532–1543.
Shannon, C. E. (1948). A Mathematical Theory of Communication. Bell System Technical Journal, 27(3), pp. 379–423.
Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, L. and Polosukhin, I. (2017). Attention Is All You Need. Advances in Neural Information Processing Systems, 30, pp. 5998–6008.
Wiener, N. (1930). Generalized Harmonic Analysis. Acta Mathematica, 55, pp. 117–258.
Daedelus Kite 