emsenn

Computational Semiotics

4 min read

Computational semiotics studies sign processes in and through computational systems. It asks how software, artificial intelligence, and digital media produce, transform, and circulate signs — and whether computational sign processes constitute genuine semiosis or merely simulate it.

Methods and approach

The field operates at two levels. At one level, it applies semiotic theory to analyze computational artifacts: software interfaces, programming languages, data visualizations, and AI outputs are treated as sign systems whose structure and effects can be studied using the tools of semiotics. At another level, it asks whether computational processes — particularly machine learning — are themselves semiotic processes, and if so, what kind.

Key contributions

  1. Algebraic and functorial formalizations — Joseph Goguen’s algebraic semiotics used algebraic specification and category theory to formalize sign systems, with applications to user interface design and information visualization. Guerino Mazzola’s functorial semiotics uses presheaves and topos theory to formalize sign relations and model creativity. These represent the most mathematically developed branch of computational semiotics. See the formal semiotics survey for detailed analysis.

  2. Machine learning as semiosis — Fernando Tohme, Rocco Gangle, and Gianluca Caterina (2024) showed that the space of learners between objects forms a topos, enabling logical propositions about what a learner knows to be stated in the topos’s internal language. This frames learning as a semiotic process: training data are signs, model updates are interpretants, and the training process is semiosis.

  3. Large language models as semiotic machines — a growing body of work (2024–2025) analyzes LLMs through semiotic theory. Davide Picca (“Not Minds, but Signs,” 2025) argues LLMs are semiotic machines that recombine linguistic forms based on probabilistic associations, drawing on Lotman’s semiosphere. Elad Vromen (2024) uses Saussure’s relational sign system and Derrida’s concept of writing to argue LLMs model language itself, not cognition. David Manheim (“Language Models’ Hall of Mirrors Problem,” Philosophy & Technology, 2025) argues LLMs exist within a hall of mirrors — reflecting linguistic surfaces without indexical grounding — and that developments like persistent memory and mediated interaction with reality move toward genuine Peircean interpretants.

  4. Computational visual semiotics — the FRESCO project (Morra et al., 2024) applies computer vision aligned with visual semiotics to analyze facial image archives. It deconstructs images at three levels corresponding to semiotic categories: plastic (lines, colors), figurative (recognizable entities), and enunciative (point of view, spectator construction).

  5. AI semiotics as a field — Semiotica devoted a special issue (Volume 262, 2025) to “Aspects of AI semiotics: enunciation, agency, and creativity,” guest-edited by Maria Giulia Dondero, Juan Alonso Aldama, and Massimo Leone. The issue addresses whether AI systems enunciate (produce utterances with subjective positioning), whether they have semiotic agency, and what kind of creativity their outputs exhibit. Massimo Leone has also explored the political semiotics of AI (2026), treating AI as an epistemic Other whose probabilistic mode of signification challenges existing semiotic categories.

Open questions

The field’s central question remains unresolved: do computational systems perform semiosis or simulate it? Peirce’s triadic sign requires an interpretant — a meaning produced for an interpreter. If the “interpreter” is a software system that manipulates tokens according to rules, is the resulting interpretant genuine? Or does genuine semiosis require the phenomenological dimension — the experience of meaning — that cognitive semiotics insists on?

The topos-theoretic approach (Tohme, Gangle, Caterina) sidesteps this question by working at the structural level: if the mathematical structure of learning satisfies the formal properties of semiosis, then it is semiosis in the formal sense, regardless of phenomenological questions. The LLM-as-semiotic-machine literature engages it directly, with different authors reaching different conclusions.

Key texts

  • Tohme, Fernando, Rocco Gangle, and Gianluca Caterina. “A Category Theory Approach to the Semiotics of Machine Learning.” Annals of Mathematics and Artificial Intelligence (2024).
  • Picca, Davide. “Not Minds, but Signs: Reframing LLMs through Semiotics.” arXiv:2505.17080 (2025).
  • Manheim, David. “Language Models’ Hall of Mirrors Problem: Why AI Alignment Requires Peircean Semiosis.” Philosophy & Technology (2025).
  • Dondero, Maria Giulia, Juan Alonso Aldama, and Massimo Leone, eds. “Aspects of AI Semiotics.” Semiotica 262 (2025).
  • Morra, Lia, et al. “For a Semiotic AI: Bridging Computer Vision and Visual Semiotics.” Computer Vision and Image Understanding 249 (2024).
  • Peircean Semiotics — the triadic sign model and the question of interpretants structure the field’s central debates
  • Cognitive Semiotics — raises the phenomenological challenge: can there be semiosis without experience?
  • Biosemiotics — the question of whether computational sign processes are semiosis parallels the question of whether biological sign processes are semiosis

See also

  • Joseph Goguen — developed algebraic semiotics for computational applications
  • semiosis — the process whose computational realization is the field’s central question
  • Formalizing Sign Theory — survey of mathematical approaches to semiotics

Graph View

Backlinks