The art of thinking · Context card
Frames as a format for concepts
What is a frame, and why do cognitive scientists think concepts are frames?
The short answer
A frame represents a concept as a central node connected by attributes to values, which can carry attributes of their own. Lawrence W. Barsalou proposed in 1992 that frames are the general format of human concepts; the Düsseldorf frame research applies the hypothesis to word meaning and gives it a formal definition.
For: Information architects, terminologists and technical writers interested in concept modeling
Key points
- A frame has a central node for the thing described, attributes that each return one value, and values that may have attributes of their own.
- Constraints link attributes: a value of one attribute can restrict the value of another.
- Frames combine by unification, which is how they model the meaning of a verb together with its arguments.
- The frame hypothesis (Barsalou, 1992) says frames are the general format of concepts — a hypothesis, tested in linguistics, philosophy and psychology.
- Frames are related to ontologies but serve another purpose: they model cognition and meaning, ontologies fix a shared vocabulary for systems.
The context
The idea
Think of the concept pen. A frame for it has a central node — the pen — and attributes that lead to values: its color, its owner, its writing tip, its ink. The ink has a color of its own, so the structure continues: frames are recursive. Each attribute is a function, so it gives exactly one value. Some nodes carry a type condition, for example that the owner must be a person.
Where it comes from
Charles J. Fillmore used frames in the 1970s for the scenes words evoke. Marvin Minsky proposed frames as data structures in artificial intelligence (1974). Lawrence W. Barsalou defined frames as attribute–value structures with constraints and proposed them as the general format of human concepts (1992). The Collaborative Research Centre 991 in Düsseldorf applied this hypothesis to the meaning of words and sentences; the volume Concepts, Frames and Cascades in Semantics, Cognition and Ontology (2021) collects work from this program.
What frames do
Frames make the content of a concept explicit and comparable. Because attributes are functions, two frames can be merged by identifying nodes — the formal counterpart of putting a verb together with its subject and object. Chapters in the volume use frames to explain the readings of verbs, to model how derived words get their meanings and to weight features in category learning.
Frames and ontologies
A cognitive schema is the psychological term for organized knowledge in memory; a frame is a precise format for one concept. An ontology and an information model also connect entities by named relations, but they fix a vocabulary for systems to exchange and check data. Frame attributes give one value each, while ontology properties can have several. The frame idea — a central node, functional attributes, constraints between them — is a useful lens on such models; it does not replace them.
The encyclopedia article Frames gives the full picture with the chapters it draws on.
Questions readers ask next
- Is a frame the same as a schema?
- Related, not the same. “Schema” is the broad psychological term for organized knowledge about familiar situations. A frame in Barsalou’s sense is a specific format: a central node with functional attributes, values and constraints.
- Are frames proven to be how the mind works?
- No. The frame hypothesis is a working hypothesis. Research tests it on word meaning, category learning and perception; the volume this card draws on presents such tests.
- Can I model an ontology as frames?
- Partly. Classes with typed properties resemble frames, and frame thinking can help to make attributes and constraints explicit. Ontology languages, however, follow their own semantics, and properties need not be functional.
Sources
- Sebastian Löbner et al. (eds.), Concepts, Frames and Cascades in Semantics, Cognition and Ontology (open access, CC BY 4.0) — Springer, January 1, 2021
- Sebastian Löbner, Thomas Gamerschlag, Tobias Kalenscher, Markus Schrenk and Henk Zeevat, Introduction (in: Concepts, Frames and Cascades in Semantics, Cognition and Ontology, open access, CC BY 4.0) — Springer, January 1, 2021
- Sebastian Löbner, Cascades. Goldman’s Level-Generation, Multilevel Categorization of Action, and Multilevel Verb Semantics (in: Concepts, Frames and Cascades in Semantics, Cognition and Ontology, open access, CC BY 4.0) — Springer, January 1, 2021
- Robin Cooper, Perception, Types and Frames (in: Concepts, Frames and Cascades in Semantics, Cognition and Ontology, open access, CC BY 4.0) — Springer, January 1, 2021
- Samuel D. Taylor and Peter R. Sutton, A Frame-Theoretic Model of Bayesian Category Learning (in: Concepts, Frames and Cascades in Semantics, Cognition and Ontology, open access, CC BY 4.0) — Springer, January 1, 2021
Review log and changes
Every context card is checked against its sources before it is published, and again whenever it changes; the date under the byline is the last review. Corrections (something was wrong) and additions (something was missing) are logged below with date and time (Berlin time). Typos, formatting and link fixes are not listed.
Reviewed
No corrections or additions since publication.