The hierarchy of knowledge

Built from the Middle:

The Hierarchical Content of Conceptual Knowledge in Lakoff and Johnson

A Scholarly Essay

September 2026

I. The Question

Lakoff and Johnson spent decades arguing that language is not a thin label on a finished world but a record of how concepts are formed and linked. A previous essay in this series treated one linking device: metaphor. The present essay sets metaphor aside and asks about content. What, on their account, does the hierarchy of knowledge contain? How are those contents arranged? How is the arrangement built?[1]

The answer is not a single tree of genera and species. It is a network with several kinds of vertical and lateral organization, all of them visible in ordinary words. There is a taxonomic stack of kinds. There is a nesting of frames, in which one concept is intelligible only against a larger structured scene. There is a radial structure, in which a central case anchors variants that do not all share one set of features. There are whole-language classifications that sort the universe into a handful of classes. The hierarchy is large and interconnected because these organizations overlap. A word sits in a taxonomy, inside a frame, as a node in a radial network, and sometimes as a member of a classifier class, all at once.

II. How the Hierarchy Is Built: From the Middle Out

The construction does not start at the top. It starts at what Rosch and Lakoff call the basic level.[2] In a stack such as animal – dog – retriever, or furniture – chair – rocking chair, the middle term is privileged. It is the highest level at which members share an overall perceived shape, the highest level at which a single mental image can stand for the class, the level of a distinctive motor program (petting, sitting, walking), the level of short common names, and the level at which most of a person’s knowledge of the domain is stored. Children master basic-level cuts first. Superordinate and subordinate cuts come later.[3]

That is a claim about content, not about rhetoric. What a mind first knows about the living world is not organism and not Canis lupus familiaris, golden retriever, three years old. It is dog: a shape one can image, a way of moving, a cluster of what the thing does and what one does with it. Knowledge is then built up by grouping several basic-level kinds under a thinner superordinate (animal keeps little shared shape) and down by adding distinguishing measurements inside the basic kind (retriever keeps the dog-shape and adds a function and a look). The hierarchy is grown from a cognitively dense middle, not deduced from a highest genus.[4]

The same middle-out growth holds for actions and properties. Walk, eat, and talk are basic-level actions; move is superordinate; saunter is subordinate. Hot, tall, and hard are basic-level properties. A conceptual system is already hierarchical at this grain before any more elaborate model is added.

III. Taxonomic Content

Once basic-level kinds exist, they can be assembled into a taxonomy: a propositional idealized cognitive model in which a higher category is a whole and the next lower categories are its parts, with little overlap at a given level.[5] The content of the stack is inclusion of kinds.

Superordinate	Basic level	Subordinate
animal	Dog	retriever
furniture	Chair	rocking chair
tool	Hammer	claw hammer
vehicle	Car	sedan

What is stored at each rung differs in kind, not only in amount. At the basic level the content is rich: shape, parts, behavior, typical setting, what one does with the thing. At the superordinate level the content is sparse and often functional or biological rather than imagistic: an animal is a living thing that moves itself; furniture is what one puts in a room to use. At the subordinate level the content is the basic-level cluster plus a few extra specifications. That is why most everyday reasoning and most ordinary nouns live in the middle. The top of the taxonomy is useful for science and for wide inference (“if it is an animal, it is alive”). The bottom is useful for precision (“hand me the claw hammer”). The middle is useful for living.

IV. Frames: Knowledge That Requires a Scene

A second hierarchy is not a stack of kinds but a nesting of scenes. Fillmore’s frames, which Lakoff takes up as idealized cognitive models, are structured packages of content: roles, relations, typical sequences, background assumptions.[6] A word does not carry its meaning alone. It profiles an element against a frame that must already be in place.

Take bachelor. The dictionary line “unmarried adult male” is not the content. The content is that line plus a cultural frame of a standard male life: marriage is available, expected, and freely chosen; the man is independent, not a priest, not the Pope, not Tarzan raised outside the institution. Against that frame the word fits. Off that frame it misfires. The hierarchy here is embedding: the concept sits inside a life-cycle model, which sits inside a model of marriage, which sits inside a model of adult social life.[7]

Take sophomore. The word names a person in the second year of a course of study. That specification is almost empty until it is set inside the frame of a four-year institution. Year, student, school, and course of study are already nested contents. The word picks a slot. The frame supplies the rest of the knowledge.

A restaurant visit is a still larger package: entering, being seated, ordering, eating, paying, leaving. Menu, tip, and check are intelligible only inside that sequence. The hierarchy of content is the nesting of roles inside a script, and of that script inside still wider models of commerce and hospitality. Taxonomic inclusion (“a menu is a kind of document”) is true and thin. Frame inclusion (“a menu is what one reads before ordering, in a restaurant”) is where most of the knowledge lives.

V. Cluster Models and Radial Categories

A third organization of content is radial. It is still hierarchical—there is a center and a periphery—but the links are not kind-inclusion.

Lakoff’s analysis of mother is the standard exhibit.[8] The central case is a cluster of models that usually coincide: the birth model (she gave birth), the genetic model (she supplied the genes), the nurturance model (she raises the child), the marital model (she is the father’s wife), the genealogical model (she is one generation up). In the stereotype those five contents stack on one person. The word’s dense center is that stack.

Around the center stand variants that keep some of the models and drop others. A birth mother may not nurture. An adoptive mother nurtures and occupies the marital and genealogical slots without birth or genes. A surrogate mother may give birth without being the genealogical or nurturing mother. A stepmother occupies some social slots without birth. These are not failed mothers. They are non-central extensions, understood by relation to the center. The content of the category is the central cluster plus a conventional set of deviations. The deviations are learned one by one; they are not generated by a single rule.[9]

That is a hierarchy of content with a different geometry. The vertical relation is not “kind of” but “variant of.” What is inherited from the center is not a necessary feature bundle. It is a contrast: this case is a mother in such-and-such a respect, and the respect is intelligible only because the clustered center is already known.

VI. A Whole-Language Classification: Dyirbal

The largest example of hierarchical content in Women, Fire, and Dangerous Things is the Dyirbal classifier system, which gives the book its title.[10] Every noun in the language is preceded by one of four classifiers. The four classes are not a scientific taxonomy. They are a culturally built partition of the universe.

The basic schema is simple. Bayi: human males, and animals with them. Balan: human females, water, fire, fighting. Balam: edible plants and the plants that bear them. Bala: everything else. That schema already has a hierarchy of content: two human genders, a food class, a residue. What makes the system a network rather than a list is the set of chaining principles that fill the classes.[11]

Domain of experience puts things that belong to one lived sphere together. Myth and belief put the sun with women, because the sun is a female deity married to the moon (which is bayi, with men). Fire goes with the sun, hence into balan. An important property—danger—then pulls dangerous things toward the same class. Birds go with women because women’s spirits are said to become birds. The hairy mary grub goes with the sun because its sting is like sunburn. The title’s trio—women, fire, and dangerous things—is not a joke about random association. It is the end-state of a content hierarchy built by a basic schema plus motivated links.

A speaker who learns Dyirbal does not memorize an inventory. The speaker learns a small set of central contents and a small set of linking principles, and can then place new items. That is how a large classification is built: not by listing every member, but by growing a network from a few dense centers.

VII. Other Large Networks of Content

Two of Lakoff’s case studies show the same architecture inside a single English word or construction.

The preposition over is a radial network of spatial contents. A central schema is something like “above-and-across”: a plane flies over the mountain. From that center, variants drop contact, drop the path and keep the end-point, add covering, add excess (“over ten dollars”), add repetition (“do it over”).[12] Each sense is a spatial or scalar configuration. The hierarchy is a family of contents radiating from one imageable scene. To know the word is to know the central scene and the conventional extensions, not a single abstract feature.

English there-constructions form a network of grammatical contents: existence (“there is a problem”), deictic pointing, presentational arrangement, and further variants.[13] Again there is a central use and a family of motivated extensions. The “knowledge” here is knowledge of how a construction places an entity in a scene. It is hierarchical because the extensions inherit structure from the central case and add a twist.

Across these examples the building procedure is the same. Start with a dense, imageable, or culturally central case. Group it with neighbors by kind (taxonomy), by shared scene (frame), by variant-of (radial structure), or by lived domain and belief (classifier chaining). Store most of the rich content at the basic level or at the radial center. Let thinner superordinates and more specific subordinates, and let peripheral variants, inherit what they need and add only what they must. Language records the result as words, classifiers, and constructions. The hierarchy is the record.

VIII. What Kind of Whole Is This?

It is tempting to picture one great tree. That picture is the folk model of taxonomy, and Lakoff treats it as one ICM among others—the classical category stacked on itself. The system he and Johnson actually describe is a web in which several hierarchies share nodes.

A chair is a basic-level kind under furniture, a role in a room-frame, a variant in a radial network of seats, and, in some languages, a member of a gender or classifier class. Mother is a kinship term in a genealogical taxonomy, a cluster of models, a radial center, and a role in the frames of birth, household, and law. Knowledge of the word is knowledge of all of those placements at once. That is what “large, interconnected hierarchy” means in this program: not one ladder from Being down to this cup, but many ladders and many wheels sharing rungs.

The interconnection has a further consequence for content. Because most knowledge is stored at basic-level kinds and at radial centers, inference tends to run through those dense nodes. To think about a collie, one thinks with dog and then adds. To think about a stepmother, one thinks with the central mother-cluster and then subtracts and replaces. Superordinates and peripheral variants are reached through the middle, not the other way around. The architecture of storage is the architecture of use.[14]

IX. Conclusion

Lakoff and Johnson’s hierarchy of knowledge, considered as content rather than as metaphor, is built from the middle out. Basic-level kinds—dog, chair, walk, hot—carry the richest structure: shape, action, typical setting. Taxonomies grow up and down from those kinds. Frames nest those kinds inside scenes, so that bachelor, sophomore, and menu mean what they mean only as slots in a larger package. Radial categories grow a family of variants around a clustered center, as with mother and over. Classifier systems such as Dyirbal’s four classes grow a partition of the universe from a few central contents and a few linking principles. Words sit in several of these organizations at once.

The decades of work were the mapping of those organizations in language after language, construction after construction. What the mapping shows is not that human knowledge is a neat Porphyrian tree. It is that a mind stores most of what it knows in a limited number of dense centers—basic-level kinds, frame cores, radial prototypes—and then links those centers upward, downward, and sideways into a network large enough to cover a world. The hierarchy is the network. The content is what the centers hold, and what the links pass on.

Works Cited

Dixon, R. M. W. Where Have All the Adjectives Gone? and Other Essays in Semantics and Syntax. Berlin: Mouton, 1982.

Fillmore, Charles J. “Frame Semantics.” In Linguistics in the Morning Calm, edited by The Linguistic Society of Korea, 111–137. Seoul: Hanshin, 1982.

Lakoff, George. Women, Fire, and Dangerous Things: What Categories Reveal about the Mind. Chicago: University of Chicago Press, 1987.

Lakoff, George, and Mark Johnson. Metaphors We Live By. Chicago: University of Chicago Press, 1980.

Lakoff, George, and Mark Johnson. Philosophy in the Flesh. New York: Basic Books, 1999.

Rosch, Eleanor. “Principles of Categorization.” In Cognition and Categorization, edited by Eleanor Rosch and Barbara B. Lloyd, 27–48. Hillsdale, NJ: Erlbaum, 1978.

[1]George Lakoff, Women, Fire, and Dangerous Things: What Categories Reveal about the Mind (Chicago: University of Chicago Press, 1987); George Lakoff and Mark Johnson, Metaphors We Live By (Chicago: University of Chicago Press, 1980); George Lakoff and Mark Johnson, Philosophy in the Flesh (New York: Basic Books, 1999).

[2]Lakoff, Women, Fire, and Dangerous Things, 31–54, synthesizing Eleanor Rosch’s experimental work. See Eleanor Rosch, “Principles of Categorization,” in Cognition and Categorization, ed. Eleanor Rosch and Barbara B. Lloyd (Hillsdale, NJ: Erlbaum, 1978), 27–48.

[3]Lakoff, Women, Fire, and Dangerous Things, 46–54. Three-year-olds master basic-level categorization before superordinate categorization.

[4]Ibid., 12–13, 32–39.

[5]Lakoff, Women, Fire, and Dangerous Things, 68–76, 283–288, on propositional ICMs, including the taxonomy as a type of ICM.

[6]Charles J. Fillmore, “Frame Semantics,” in Linguistics in the Morning Calm, ed. The Linguistic Society of Korea (Seoul: Hanshin, 1982), 111–137; Lakoff, Women, Fire, and Dangerous Things, 68–76.

[7]Fillmore’s classic discussion of bachelor is taken up in Lakoff, Women, Fire, and Dangerous Things, 70–71, 85.

[8]Lakoff, Women, Fire, and Dangerous Things, 74–76, 79–84, 91.

[9]Ibid., 83–84, 91–114.

[10]Ibid., 92–104, following R. M. W. Dixon, Where Have All the Adjectives Gone? (Berlin: Mouton, 1982).

[11]Lakoff, Women, Fire, and Dangerous Things, 93: domain-of-experience principle, myth-and-belief principle, and important-property principle.

[12]Ibid., 416–461, Case Study 2, on over, following Claudia Brugman.

[13]Ibid., 462–585, Case Study 3, on there-constructions.

[14]Lakoff and Johnson, Philosophy in the Flesh, 9–15, 77–117.