Language as substrate and bridge
The opening claim is that language is the surface shared by humans and systems, and without it there is no usable bridge from weights to the world.
Jessica Talisman argues that language is the bridge between large language models, ontologies, and human action, and that labels, controlled vocabularies, and shared meaning are the practical layer that makes formal systems usable.
The essay argues that language is the substrate and bridge for both LLMs and ontologies. It opens with the octopus thought experiment to show that form alone is not meaning, then moves to labels, controlled vocabularies, taxonomies, and ontology-lexicon work as the engineering discipline that binds formal systems to human practice. The closing logic is that without a shared linguistic layer, even well-formed systems remain inert, because decisions require language that humans and machines can both act on.
There is a reason we call them language models. Language is the substrate and the bridge, the surface with which we interact with systems and interfaces.
The essay is represented as a set of semantic layers: bridge, grounding, labels, lexical anchors, controlled vocabularies, and the translation of knowing into doing.
The opening claim is that language is the surface shared by humans and systems, and without it there is no usable bridge from weights to the world.
The octopus story is used to show that fluent structure without world grounding can still produce nonsense when the situation requires action.
The essay argues that labels are the human-facing layer of ontologies and that missing or weak labels make a formally valid model practically inert.
The article moves from labels to controlled vocabularies, taxonomies, and standards as the successive layers that make agreement and action possible.
The endpoint of the argument is that shared language is the translation device that turns understanding into coordinated action.
The comments page exposed six visible comments, including a fresh Kingsley Uyi Idehen comment, a follow-up reply by Jessica, and a short reply chain around grounding, Frame Theory, and the meaning of language.
Founder & CEO at OpenLink Software | Driving GenAI-Based AI Agents | Harmonizing Disparate Data Spaces (Databases, Knowledge Bases/Graphs, and File System Documents) · 1m
“Language is mankind's greatest innovation :)”
Kingsley Uyi Idehen agrees with the bridge framing and compresses it into a concise statement: language is humanity's greatest innovation, which reinforces the essay's point that language is the shared surface for human and machine coordination.
Reader · 1h
“they don't want to learn, they're so stubborn”
This comment is a joking aside that uses French language politics as a side-channel metaphor for resistance to learning.
Reader · 1h
“I thoroughly enjoyed this piece, Jess, thank you!”
Ramona C. Truta praises the piece and notes a recent conversation about Wittgenstein's math philosophy, connecting the essay to lived scholarly discussion.
Author · 1h
Reply to Ramona C. Truta
“Thanks, Ramona. I’ll look forward to your friend’s book!”
Jessica replies warmly and keeps the thread open to the book reference and related work.
Reader · 9h
“Language is indeed the bridge. But a bridge to what?”
Sadok pushes the argument toward grounding by asking what language bridges to if meaning is not rooted in events before it stabilizes into form.
Reader · 10h
“Frame Theory ... FrameNet ... thank you”
The Strategic Linguist says the post helped them articulate ideas they already recognized, connecting the essay to Frame Theory and FrameNet as a practical semantic resource.
The article implies a practical path: bind symbols to real-world meaning, give them shared labels, and keep the vocabulary controlled enough for coordinated action.
Tie the model's symbols to things users can recognize in the world, not only to internal formal consistency.
Make each class and property readable through labels, comments, and human-friendly descriptions.
Use controlled terms, aliases, and hierarchy so teams and systems can agree on what each label means.
Ensure the model helps people decide and do something, not just validate a file or reason over an empty structure.
Lean on SKOS, FOAF, PROV-O, and related standards to keep language usable across systems and communities.
The FAQ layer captures the bridge argument, the octopus grounding critique, and the practical role of labels and vocabularies in real systems.
The glossary isolates the article's reusable semantic units, including the standards and vocabulary layers it cites as the bridge between model output and human action.
Systems that learn and generate patterns in human language and serve as the starting point for the essay's bridge metaphor.
The common linguistic surface through which humans and machines can coordinate meaning and action.
The point where humans and systems meet, exchange intent, and make decisions possible.
Bender and Koller’s thought experiment showing that fluency without grounding can produce meaningless output.
The distinction between producing well-formed language and actually understanding what the language refers to.
The link between symbols and real-world events, objects, or practices that gives language operational meaning.
Human-readable names that anchor ontology terms, classes, and properties to understandable concepts.
A human-readable label that makes a concept understandable to users and downstream systems.
A managed set of terms and aliases that ensures people and systems mean the same thing.
A hierarchical organization of categories that helps people and systems structure meaning.
A richer vocabulary structure that captures preferred terms, synonyms, and relationships.
The Simple Knowledge Organization System used to model vocabularies and concept schemes.
Friend of a Friend vocabulary for people, organizations, and simple social relations.
The W3C provenance ontology for modeling entities, activities, and agents.
The W3C community work showing why ontology labels need richer lexical information.
The rule that labels and descriptions are the way humans interact with Wikidata data.
The use of language and meaning to make outputs actionable in real operational settings.
The condition in which a system output can be used to make a decision or take a step.
Common terms and labels that let different people and systems act together.
The W3C guidance that labels are lexical anchors, not decorative metadata.
The essay’s footnotes are modeled as graph entities so the source page, the citations, and the discussion layer stay linked together.
Anchored at #footnote-anchor-1
Anchored at #footnote-anchor-2
Anchored at #footnote-anchor-3
Anchored at #footnote-anchor-4
Anchored at #footnote-anchor-5
Anchored at #footnote-anchor-6
Anchored at #footnote-anchor-7
Anchored at #footnote-anchor-8
Anchored at #footnote-anchor-9
Anchored at #footnote-anchor-10
Anchored at #footnote-anchor-11
Anchored at #footnote-anchor-12