Internal Languages Conjecture in Higher Categories

• The Internal Languages Conjecture is the assertion that fragments of type theory, such as MLTT and HoTT, function as internal languages for structured (∞,1)-categories up to homotopy.
• The finite-limit fragment demonstrates that type-theoretic models with Σ-types and intensional identity types are equivalent to finitely complete (∞,1)-categories via a chain of Dwyer–Kan equivalences.
• The locally cartesian closed refinement extends the framework by incorporating Π-types, aligning type theory with locally cartesian closed quasicategories through rigidification and homotopical techniques.

The Internal Languages Conjecture, in recent higher-categorical work, is the claim that appropriate fragments of Martin–Löf Type Theory or Homotopy Type Theory are, up to homotopy, the internal languages of suitably structured $(\infty,1)$-categories. In this program, objects play the role of contexts, morphisms the role of substitutions, and categorical structure such as finite limits or local cartesian closure corresponds to type formers such as $\Sigma$-, $Id$-, and $\Pi$-types. The intended result is not merely a sound interpretation of syntax in semantics, but a Dwyer–Kan equivalence between the homotopy theory of type-theoretic models and the homotopy theory of the corresponding higher categories (Kapulkin et al., 2017, Cherradi, 3 Sep 2025).

1. Classical background and general formulation

For an ordinary $1$-category $\mathcal C$, an internal language is a formal system in which objects of $\mathcal C$ play the role of contexts, morphisms $\Gamma \to \Delta$ are substitutions, and categorical structure corresponds to type formers. In the classical setting, simple type theory is the internal language of cartesian closed categories, while extensional Martin–Löf Type Theory with $\Sigma$- and $\Pi$-types and some extensionality is the internal language of locally cartesian closed categories. The higher-categorical conjecture generalizes this pattern from extensional dependent type theory and ordinary categories to intensional type theory and $\Sigma$0-categories (Kapulkin et al., 2017).

The conjecture is usually stated in homotopy-invariant form. Syntactically presented type theory with specified type formers is expected to generate an initial $\Sigma$1-category with those structures; models of the type theory are expected to be equivalent, in the homotopy-theoretic sense, to $\Sigma$2-categories with the corresponding structure; and there should be a homotopy-equivalence between the homotopy theory of type theories and the homotopy theory of the corresponding $\Sigma$3-categories (Kapulkin et al., 2017).

A useful summary of the main fragments treated so far is the following.

Setting	Type-theoretic fragment	Semantic target
Finite-limit $\Sigma$4-categories	$\Sigma$5-types and intensional $\Sigma$6-types	finitely complete $\Sigma$7-categories
Locally cartesian closed $\Sigma$8-categories	$\Sigma$9-types, intensional $Id$0-types, extensional $Id$1-types with functional extensionality	locally cartesian closed quasicategories
Univalent $Id$2-categorical finite-limit setting	$Id$3	univalent categories with finite limits
Univalent $Id$4-categorical LCCC setting	$Id$5	univalent locally cartesian closed categories

This stratification suggests a stepwise program: finite limits correspond to a fragment with $Id$6 and intensional identity types, while local cartesian closure requires adding dependent products.

2. The finite-limit fragment

A precise higher-categorical instance was established by Kapulkin and Szumiło in "Internal Languages of Finitely Complete $Id$7-categories" (Kapulkin et al., 2017). The paper proves a finite-limit fragment of the conjecture: Martin–Löf Type Theory with dependent sums, intensional identity types, and structural rules, but without $Id$8-types, universes, $Id$9-types, or higher inductive types, has the same homotopy theory as finitely complete $\Pi$0-categories.

On the type-theoretic side, the models are packaged as comprehension categories with $\Pi$1-types with strong $\Pi$2-rule, weakly stable $\Pi$3-types, and all objects fibrant. This structure is called a categorical model of type theory. On the categorical side, the target is $\Pi$4, the homotopical category of quasicategories with finite limits, taken as a model of finitely complete $\Pi$5-categories (Kapulkin et al., 2017).

The main theorem states that the $\Pi$6-category of categorical models of Martin–Löf Type Theory with dependent sums and intensional identity types is equivalent to the $\Pi$7-category of finitely complete $\Pi$8-categories. Concretely, the proof proceeds through the chain

$\Pi$9

where $1$0 is the homotopical category of Joyal’s tribes, $1$1 is the homotopical category of fibration categories, and the arrows are Dwyer–Kan equivalences (Kapulkin et al., 2017).

This result fixes the semantic content of the finite-limit fragment. A context is an object of the base category, a type over $1$2 is represented by a fibration $1$3, a term is a section of that fibration, $1$4-types are modeled by composition of fibrations, and intensional identity types are modeled by factorization of diagonals into an anodyne or weak equivalence followed by a fibration. In this sense, intensional identity types encode the homotopical structure of the model through path objects

$1$5

The paper is careful about scope. It does not directly relate raw syntax or contextual categories to $1$6, and it does not prove the Initiality Conjecture. The theorem is formulated for semantic categorical models rather than for an initial syntactic theory (Kapulkin et al., 2017).

3. The locally cartesian closed refinement

A stronger version was established in "Internal languages of locally cartesian closed $1$7-categories" (Cherradi, 3 Sep 2025). This work proves the internal languages conjecture for locally cartesian closed quasicategories by exhibiting a DK-equivalence between the relative category of $1$8-tribes and the relative category of locally cartesian closed quasicategories.

A quasicategory $1$9 is locally cartesian closed if it has finite limits and, for every morphism $\mathcal C$0, the pullback functor

$\mathcal C$1

admits a right adjoint

$\mathcal C$2

The relevant type theory is Martin–Löf Type Theory with dependent sums, intensional identity types, and dependent products satisfying functional extensionality and the $\mathcal C$3-$\mathcal C$4-rule. The semantic intermediary is Joyal’s notion of a $\mathcal C$5-tribe, namely a tribe with internal products along fibrations such that the induced functor $\mathcal C$6 preserves anodyne maps (Cherradi, 3 Sep 2025).

The main theorem is the DK-equivalence

$\mathcal C$7

where $\mathcal C$8 is the relative category of comprehension categories modeling this fragment of type theory, and $\mathcal C$9 is the relative category of locally cartesian closed quasicategories and locally cartesian closed functors (Cherradi, 3 Sep 2025).

The conceptual correspondence is direct. $\mathcal C$0-types correspond to dependent sums; $\mathcal C$1-types correspond to the right adjoints $\mathcal C$2 to pullback; identity types correspond to path objects and diagonal factorizations; and functional extensionality corresponds to preservation of anodyne maps by $\mathcal C$3. Lemma $\mathcal C$4 in the paper states that if $\mathcal C$5 is a $\mathcal C$6-tribe, then its associated comprehension category admits $\mathcal C$7-types satisfying $\mathcal C$8-rule and function extensionality, and conversely a comprehension category with such $\mathcal C$9-types yields a $\Gamma \to \Delta$0-tribe (Cherradi, 3 Sep 2025).

This theorem extends the finite-limit fragment by showing that dependent products do not merely add more syntax; they raise the semantic target from finitely complete $\Gamma \to \Delta$1-categories to locally cartesian closed ones. The proof also shows that the homotopy theory is insensitive to strict versus weak preservation of internal products, via a rigidification argument using path tribes and semi-cubical tribes (Cherradi, 3 Sep 2025).

4. Univalent and $\Gamma \to \Delta$2-categorical analogues

The higher-categorical conjecture sits alongside a $\Gamma \to \Delta$3-categorical internal-language tradition. Seely originally claimed an equivalence between locally cartesian closed categories and extensional Martin–Löf type theories with $\Gamma \to \Delta$4- and $\Gamma \to \Delta$5-types, but substitution via pullbacks was validated only up to isomorphism. Hofmann repaired this by passing to split fibrations, and Clairambault and Dybjer reformulated the result as a biequivalence between the bicategory of locally cartesian closed categories and the bicategory of democratic categories with families with extensional identity types, $\Gamma \to \Delta$6-types, and $\Gamma \to \Delta$7-types (Weide, 2024).

Van der Weide’s "The internal languages of univalent categories" develops this theorem in univalent foundations and extends it to several classes of toposes (Weide, 2024). The paper replaces categories with families by full univalent comprehension categories, because in a univalent setting the collection of types in a context need not be a set. A comprehension category consists of a category of contexts, a displayed category of types, a cleaving, and a comprehension functor preserving Cartesian morphisms. Democracy means that every context is, up to isomorphism, a single extension of the terminal context.

In this framework the paper proves biequivalences

$\Gamma \to \Delta$8

where $\Gamma \to \Delta$9 is the bicategory of univalent categories with finite limits, $\Sigma$0 is the bicategory of univalent locally cartesian closed categories, and the comprehension-category side carries unit, binary product, equalizer, $\Sigma$1-, and $\Sigma$2-types as appropriate (Weide, 2024).

The same paper extends the correspondence to pretoposes, arithmetic pretoposes, $\Sigma$3-pretoposes, elementary toposes, and elementary toposes with $\Sigma$4. Its summary table gives the following type-theoretic signatures: finite limits correspond to $\Sigma$5; local cartesian closure adds $\Sigma$6; pretoposes add $\Sigma$7; arithmetic pretoposes add $\Sigma$8; and elementary toposes add $\Sigma$9 (Weide, 2024).

These univalent results are not the higher-categorical conjecture itself, but they are structurally parallel. They show that internal-language theorems can be organized bicategorically, that univalence allows one to avoid splitness assumptions, and that syntax–semantics equivalence can be extended modularly through local properties.

5. Core technical mechanisms

The higher-categorical proofs rely on a cluster of homotopical structures designed to mediate between syntax-like semantic models and quasicategories. A fibration category, in the sense used by Brown, is a category with weak equivalences and fibrations, a terminal object with all objects fibrant, pullbacks along fibrations, factorization of every morphism as a weak equivalence followed by a fibration, and $\Pi$0-out-of-$\Pi$1 for weak equivalences. Homotopies are defined by path objects, and homotopy pullbacks are defined by combining strict pullbacks with factorization (Kapulkin et al., 2017).

A tribe, in Joyal’s sense, has fibrations and anodyne maps, with every map factoring as an anodyne map followed by a fibration, and with anodyne maps stable under pullback along fibrations. Tribes carry a canonical fibration-category structure in which weak equivalences are homotopy equivalences. One of the decisive results of the finite-limit proof is that the forgetful functor

$\Pi$2

is a Dwyer–Kan equivalence. This identifies two different presentations of the same homotopy theory and makes it possible to pass from type-theoretic structures to quasicategories via known comparisons (Kapulkin et al., 2017).

The syntax–semantics translation is implemented through comprehension categories. From a categorical model of $\Pi$3, one obtains a tribe whose fibrations are comprehension maps and whose anodyne maps are extracted from identity-type factorization. Conversely, from a tribe $\Pi$4, one forms a comprehension category whose fiber over $\Pi$5 consists of fibrations $\Pi$6; $\Pi$7-types arise from composition of fibrations, and identity types arise by factoring the diagonal $\Pi$8. The two constructions are homotopy inverses (Kapulkin et al., 2017).

The locally cartesian closed proof adds further rigidity issues. A $\Pi$9-tribe is a tribe equipped with internal products along fibrations, but functors may preserve this structure only up to weak equivalence. To control that defect, the proof introduces path tribes, semi-cubical tribes, semi-cubical frames, and a rigidification lemma showing that preservation of dependent products up to equivalence can be replaced, in the homotopy theory, by preservation up to isomorphism. This is the technical bridge from “loose $\Sigma$00-tribes” to strict $\Sigma$01-tribes and then to locally cartesian closed quasicategories (Cherradi, 3 Sep 2025).

Across both proofs, DK-equivalence is the central comparison notion. It requires equivalence on homotopy categories and weak equivalence on derived mapping spaces, typically accessed through hammock localization. Cisinski’s approximation properties provide a practical criterion for proving DK-equivalences between fibration categories, and these criteria are repeatedly used in the semisimplicial and semi-cubical comparison steps (Kapulkin et al., 2017, Cherradi, 3 Sep 2025).

6. Scope, limitations, and divergent usages

The current state of the conjecture is fragmentary but substantial. The finite-limit fragment is established for Martin–Löf Type Theory with dependent sums and intensional identity types, and the locally cartesian closed fragment is established for Martin–Löf Type Theory with dependent sums, intensional identity types, and dependent products satisfying functional extensionality. What remains outside these theorems includes universes, univalence, $\Sigma$02-types, higher inductive types, and a direct proof of the Initiality Conjecture connecting raw syntax or contextual categories to the target $\Sigma$03-categories (Kapulkin et al., 2017, Cherradi, 3 Sep 2025).

A common misconception is to treat the finite-limit theorem as a full HoTT–$\Sigma$04-topos correspondence. It is not. The finite-limit result only classifies finitely complete $\Sigma$05-categories, and the locally cartesian closed result only reaches locally cartesian closed quasicategories. The papers explicitly place universes, univalence, higher inductive types, and $\Sigma$06-topos semantics in future work (Kapulkin et al., 2017, Cherradi, 3 Sep 2025).

There are also broader internal-language programs with different semantic targets. In categorical semantics of computation, "Central Submonads and Notions of Computation: Soundness, Completeness and Internal Languages" proves a $\Sigma$07-equivalence between CSC-theories and CSC-models, where the models are cartesian closed categories equipped with a strong monad and a central submonad. This is an internal-language theorem for monadic notions of computation rather than the higher-categorical MLTT conjecture (Carette et al., 2022).

Finally, the same phrase appears in a distinct automata-theoretic literature. In "Positional $\Sigma$08-regular languages", the Internal Languages Conjecture refers to structural characterization of objectives for which all associated games admit optimal positional strategies. For $\Sigma$09-regular objectives, the paper proves a complete characterization: positionality is equivalent to recognition by a deterministic fully progress consistent signature automaton, equivalently by a deterministic parity automaton that is $\Sigma$10-completable, equivalently by the existence of well-ordered monotone universal graphs for all cardinals (Casares et al., 2024). This usage is mathematically unrelated to the type-theoretic conjecture, but the shared terminology reflects a common ambition: to characterize a semantic class by an intrinsic formal language.