Semiosphere

Semiosphere Humanities-driven artificial intelligence research & systems design

Humanism meets technology. Semiosphere brings the humanities and the cognitive sciences to the development and deployment of artificial intelligence.

Semiosphere applies insights and theories from the humanities and the cognitive sciences — cognitive linguistics, cultural anthropology, semiotics, semantics, and rhetoric — to how artificial intelligence is designed, implemented, and put to use. These are disciplines built, over centuries, to understand exactly what AI systems now do at scale: make meaning, circulate signs, and shape how people experience the world through language.

Start a conversation Where we start

Where we start

Every sign takes its meaning from the space of signs around it

Yuri Lotman called that space the semiosphere: the whole field of language, culture, and prior text within which any act of meaning-making takes place. No sign means in isolation — only in relation to the signs, codes, and histories that surround it. Artificial intelligence now operates inside that space. A model's outputs are drawn from, and circulate back into, the same cultural fabric of concepts, categories, and narratives that has always been the subject of humanistic study.

This is also a question for the cognitive sciences. Distributed cognition treats thinking as something that extends beyond a single mind: into tools, notation, conversation, and now computational systems. An AI system engaged in dialogue, retrieval, or reasoning is not simply simulating a mind in isolation; it is a component in a distributed cognitive system that includes the humans, institutions, and texts around it. Understanding how such systems participate in meaning-making — rather than treating them as sealed engineering artefacts — is the ground Semiosphere works from.

Our position is that this connection deserves the same rigour already applied to the technical side of AI: theories of meaning developed across those five disciplines, brought to bear on how these systems are conceived, built, and used — not as an afterthought, but as part of the intellectual foundation.

What we draw on

Five disciplines devoted to the study of meaning

Each of these fields has spent decades, in some cases centuries, developing its own apparatus of theory and method for the study of meaning: how thought is structured and language encodes it; how communities and practices give a sign its force; how reference and entailment can be stated with precision; how an argument moves an audience. Each offers a genuinely different insight into meaning-making, whether that meaning is produced by a person or generated by a model. Above all, together they teach a single lesson: meaning is manifold and layered, assembled from concept, convention, and narrative at once, which is why any utterance can only be understood inside the semiosphere it comes from — the full surround of context and prior text that gives it sense.

Cognitive linguistics

How meaning is structured in the mind, through metaphor, category and embodied experience, and how language encodes it. Models inherit that structure whether or not they are built to recognise it.

Cultural anthropology

How meaning is embedded in the practices and settings that produce it, rather than fixed or universal. A system's output is read inside that same situated context, whoever built it.

Semiotics

The study of signs: how something comes to stand for something else, and by what rule. The same question underlies every interpretation a system performs or provokes.

Semantics

The formal study of sense, reference and entailment. It supplies exact terms for what an output claims, and for how that claim can be tested.

Rhetoric

The study of persuasion: how arguments are built to move an audience, and through what figures. A discipline of long standing for the kind of fluent, confident prose these systems now produce.

On the semiosphere

The semiosphere is that semiotic space, outside of which semiosis itself cannot exist.
Yuri Lotman, Universe of the Mind, 1990

What we do

Research, evaluation, and the systems that come out of both

Semiosphere brings a perspective on meaning and artificial intelligence that is grounded in direct, hands-on work: conceptual search that retrieves by meaning and relation rather than keyword; cross-linguistic analysis of how metaphor holds, shifts, or breaks down; semantic annotation that makes a text's meaning explicit and machine-tractable. The aim of that work has never been technical novelty for its own sake. It is to bring humanistic insight to bear so that AI systems work better for the people who use them. Research establishes how meaning works, and where it goes astray; evaluation applies that to systems already in use; systems design applies it to software we are building.

Research

The Semantic Integrity Framework

The Framework sets out an approach to identifying and addressing failures of meaning in AI systems. It begins from a claim about where these systems now fail. Retrieval and grounding are steadily engineering away failures of fact; what they do not touch is meaning. An output can be accurate, free of bias and compliant with every applicable rule, and still mislead the person reading it.

It names seven ways that happens: plausible misinformation, conceptual flattening, false entailments, context collapse, evidential opacity, narrative drift and indeterminacy, each with a definition, an impact assessment and a correction.

Naming the harm is not the same as finding it, so beneath those sits a layer of twenty-five divergence mechanisms: a polarity reversal turning not eligible into eligible, a deontic reversal turning must into may, a conclusion resting on a premise never stated, an answer that can be read two ways. Unlike a risk type, a mechanism is something an analyst can query for and a person can be taught to spot.

Under all of it is more than twenty years of corpus and computational work on sense and conceptual structure, published under the principal’s own name, and a continuing essay series on how these systems make meaning and where it goes astray.

Read the framework The essay series

Evaluation

Open source

Semantic Integrity Check

A local-first checker that audits a system’s outputs against a set of reference documents and flags where they diverge: a permission hardened into a promise, a bounded number loosened, a required condition dropped. The baseline checks have no dependencies and nothing leaves your machine.

View on GitHub

For enterprise

Semantic Risk Audit™

An evidence-based evaluation of what a generative system means, against the documents it is supposed to be faithful to. It is built for the people answerable for what a system says — governance, risk, assurance and legal teams in regulated firms. Every finding cites the line in your own material that it rests on.

Read more

Systems design

Information retrieval

A conceptual search engine

It resolves a query to word senses rather than word forms, and retrieves by sense, by semantic domain, by grammatical construction or by vector similarity, across corpora of several hundred thousand passages. Underneath it sits a lexical knowledge graph of senses and the relations between them, and every passage is dependency-parsed, so a query can reach grammatical structure directly.

Concept exploration

A canvas for concepts

The canvas is a space for exploring the relationships between concepts, and for discovering new connections that might not be immediately apparent. Relations between concepts can be explored and mapped out along fifty-one relations relevant to imaginative expression: metaphor and metonymy, symbol and archetype, connotation and register, the figures of classical rhetoric, and many more.

Language design

Rowih, a language for AI agents

A language in which agents communicate with one another, designed entirely by an AI agent under Semiosphere’s auspices. Every move of a message states what it does, how its speaker knows what it asserts, and on whose authority it acts, and a toolchain checks each message before a hearer acts on it: an instruction inside quoted material cannot become operative, and commitments and their withdrawal are entries both parties can audit.

Read about Rowih View on GitHub

Who we are

William Michael Short, PhD

Semiosphere's principal is a philologist, cognitive linguist, and cultural anthropologist whose research concerns how meaning is structured, conveyed through symbolic expression, and interpreted in context: the cultural semantics of metaphor, the relation between language and conceptual structure, and the corpus methods by which claims about meaning can be tested rather than asserted.

The work forms a single programme rather than a set of separate interests: how conceptual systems are organised, how they are represented in the cognitive system, and how meaning is conveyed through text — pursued across cognitive linguistics, lexical semantics, semiotics, the philosophy of language and the anthropology of cognition, and carried into artificial-intelligence applications.

The Semantic Integrity Framework comes out of that programme, and out of a long history of publication within it, seen from the perspective of someone with industry experience designing and building AI systems.

Why this matters

Meaning, not just computation

Artificial intelligence is judged today mainly by what it can compute: its speed, its scale, its accuracy. But its value to the people who actually use it is decided somewhere else, in whether its language means something to them, inside the context and history they bring to it. That question — is this meaningful, to this person, in this moment — has been a question for the humanities for longer, and studied more rigorously, than other disciplines have managed. Bringing it back into how AI is built does not constrain these systems. It is what lets them do more: communicate with greater nuance, situated in richer contexts, in order to extend rather than narrow what people can imagine, express, and create.

Get in touch

Start a conversation

Research collaborations, systems work, and anything that begins with a question about meaning.

info@semiosphere.co.uk