Compositionality

The Problem of Semantic Composition

The central philosophical question of compositionality concerns the systematic relation between the meaning of a structured expression and the meaningful contributions of its constituents. A competent speaker understands sentences never encountered before, distinguishes permutations of the same words, and often recognizes inferential relationships among novel sentences. An adequate philosophy of linguistic meaning must explain how a limited repertoire of expressions and construction practices supports such extensive understanding. Compositionality supplies a candidate structural constraint on that explanation.

Consider “The poet admired the sculptor” and “The sculptor admired the poet.” Their words can be held fixed while the relation asserted changes. If lexical meanings alone determined sentential meaning, the two sentences would be indistinguishable. The position of constituents and the grammatical relation expressed by that position therefore enter the account. Yet it would be equally insufficient to describe word order without assigning interpretations to “poet,” “sculptor,” and “admired.” The semantic result belongs to their coordinated operation. The familiar formulation states that the meaning of a complex expression is determined by the meanings of its parts and the way they are combined; the precise philosophical work lies in explaining what counts as a part, what kind of meaning is determined, and what determination requires (Pickel & Szabó, 2025).

This principle can be treated as a claim about a language, a constraint on a theory of a language, or a hypothesis about the cognitive mechanisms by which speakers understand language. These claims require different justifications. A mathematical model can be compositional by definition while failing to predict a natural-language speaker's interpretation. Conversely, an individual may correctly interpret an utterance using background knowledge even where a compositional grammatical analysis remains unsettled. Evaluating the principle therefore requires bringing logical precision into contact with explanatory evidence.

The present study develops five questions across its historical and systematic chapters. Which semantic values are suitable for composition? Which syntactic representations determine the operations? What explanatory work does compositionality perform beyond the mathematical existence of an encoding? Which observed phenomena press against straightforward local combination? How far can the principle extend from public language to thought, discourse, computing, and artificial systems? A defensible answer coordinates these questions instead of treating an abstract slogan as a complete theory of meaning.

Definition and Conceptual Boundaries

Compositionality is a dependence principle governing the interpretation of complex expressions in a specified language and interpretation regime. When a complex expression is formed by an admissible syntactic operation from some constituent expressions, its semantic value is fixed by a corresponding semantic operation applied to the semantic values of those constituents. The standard formulation concerns a structured language and a chosen notion of meaning. It does not independently establish which semantic value speakers represent, how meanings are learned, or what fixes the lexicon.

A central distinction separates compositionality from lexical atomism. An atomist may hold that individual words possess semantically basic contents, while a compositional theorist can make the basic units morphemes, constructions, or other items. Equally, some expressions can receive contextual semantic values that are then composed. The existence of parts and combination rules entails neither a metaphysical thesis that every meaning exists as an autonomous atom nor a historical thesis that the smallest units were first in human language. This distinction is important in interpreting Frege, whose context principle is often mistakenly presented as an outright repudiation of composition.

A second distinction separates compositionality from truth conditionality. A semantics may determine the truth conditions of sentences from the semantic values of their parts. But the values composed may instead be senses, propositions, functions, sets, structured events, assignments, information states, or context-change potentials. Whether these values adequately represent meaning is a further issue. A recursive semantics of truth is one powerful example of compositional analysis, not its complete conceptual definition (Pagin & Westerståhl, 2010a).

A third distinction separates compositionality from mere computability. If a computable procedure retrieves the meaning of every expression from an enormous lookup table, it may satisfy computational effectiveness without respecting meaningful constituent structure. Conversely, a homomorphic interpretation might be abstractly specified with semantic functions that are too complex or psychologically unavailable for ordinary language understanding. Compositionality is a condition on how semantic dependence tracks syntax; computational effectiveness concerns how a procedure can be carried out.

Finally, the principle must be indexed to a language, a chosen syntactic analysis, a semantic function, and perhaps a context or assessment parameter. A language may support an extensionally compositional semantics but fail to support a particular proposed intensional semantics. A semantic theory can be compositional for one representation of quantifier scope and noncompositional for an impoverished alternative. Philosophical adjudication must disclose these choices.

Constituency, Structure, and Interpretation

The phrase “mode of combination” carries a substantive theory of syntax. Words in a sequence have positions, but syntactic constituency ordinarily imposes hierarchical relations. “Old men and women” can be understood as grouping old men with women of unspecified age, or grouping men and women who are all old. The sequence alone is compatible with multiple structural analyses. The resulting semantic difference calls for sensitivity to bracketing or another representational device that captures the relevant interpretation.

The structure required by a semantic theory can be surface constituency, a derivational history, a typed categorial proof, a dependency graph, or a level of logical form. Selecting among these is not a minor presentation choice. It determines the operations for which the semantic theory must supply clauses. If an expression is treated as the output of function application, a corresponding semantic operation can be defined. If it is treated as the product of movement and binding, the semantic representation must explain traces, variables, or a different method for obtaining the same dependencies. A compositional analysis can therefore be sensitive to the grammar's architecture even where it assigns the same final truth conditions.

Syntax is also related to lexical category. The expression “runs” can serve distinct grammatical functions in different constructions, and meanings combine only when semantic types match the structure of a permissible expression. A typed semantics might assign an individual-denoting name type e, a truth value type t, and an intransitive predicate a function from e to t. In that setting, the expression denoting the predicate cannot be applied to just any argument. Syntactic category restrictions and semantic type restrictions cooperate in explaining meaningful combination.

The distinction between constituent token and constituent type matters as well. Repeated occurrence of a pronoun may depend on an assignment or binding relation, and a lexical item might contribute different values across contexts. The semantic contribution of “I” in two utterances is not identical if the speaker differs. Such variation requires an appropriate notion of expression occurrence or parameter-sensitive meaning, rather than abandonment of the structural relation itself.

The interpretive architecture is consequently triangular: linguistic structure licenses composition, semantic values supply inputs, and semantic rules determine results. Altering any side may alter the explanatory theory. It is insufficient to assert that an utterance has parts; one must establish that a particular decomposition tracks the linguistic phenomena under investigation.

Historical Antecedents in Greek and Medieval Philosophy

Philosophers studied how statements are articulated from meaningful elements long before the modern formal principle of compositionality was named and mathematically formulated. Aristotle's work on names, verbs, assertion, negation, and predication makes the relationship between linguistic form and truth an explicit philosophical topic. A statement has the capacity for truth or falsity in a way that an isolated name does not. The contrast anticipates the need to distinguish the semantic roles of parts from the logical status of complete judgments. It should not be retroactively read as a fully developed homomorphism thesis: Aristotle's categories and his theory of syllogistic inference belong to a different philosophical apparatus.

The Stoic tradition developed discussions of complete assertibles and logical relations among them. Its distinction between significant linguistic expression and the contents of what is said helped establish the problem of the relationship between verbal articulation and propositional content. The surviving textual situation is fragmentary, however, and later doxographical reports must be handled as evidence of a tradition rather than direct statements of a modern semantic principle.

Medieval terminist logic addressed how terms signify or stand for things within propositions. Supposition theory made the linguistic environment of a term crucial to its role, thereby furnishing historically significant examples of what later theorists might call the interaction of a constituent with its context. Medieval analyses of syncategorematic expressions, quantifiers, negation, and scope are similarly relevant because they scrutinize the contribution of logical vocabulary to the content and inferential force of propositions. The important continuity is the demand for a systematic account of structured signification; formal semantic operations and modern set-theoretic models are subsequent methodological developments.

A historical reconstruction must resist treating every discovery that sentences contain meaningful parts as an assertion of the same thesis. Medieval accounts varied in the meanings assigned to terms, the scope of supposition, and the relation between mental and spoken language. Their problems illuminate how composition became pressing; their exact commitments must be established from specialized editions and commentary when a detailed historical priority claim is made. The present monograph therefore uses these traditions to recover the philosophical background of structured meaning, while locating the modern explicit semantic theory in nineteenth- and twentieth-century logical developments (Pickel & Szabó, 2025; Pagin & Westerståhl, 2010a).

Sentence Meaning in Classical Indian Philosophy

Classical Indian philosophy of language provides an independently developed and philosophically rich investigation of the relationship between word meaning and sentence understanding. Debates associated with grammatical and Mīmāṃsā traditions ask whether a sentence communicates a unified meaning that must be understood as a whole, whether constituent words first present distinct meanings subsequently combined, and whether the meanings of words are already relationally connected. These questions illuminate the alternatives concealed by a simple contrast between compositionality and holism.

Bhartṛhari's Vākyapadīya is central to the tradition commonly associated with the doctrine of sphoṭa and a sophisticated discussion of the unity of linguistic meaning. The import of a sentence is not merely a heap of independently intelligible word meanings. The philosophical claim concerns the manner in which articulation and comprehension are related: distinct words may be analytically isolated within a unity of understanding without being epistemically or ontologically prior to that unity. Such a view raises a question for modern theories: does a compositional model explain the process of understanding, or does it offer a retrospective analysis of an already grasped linguistic whole?

Different Mīmāṃsā approaches to sentence meaning have often been described using the distinctions between abhihitānvaya, associated with the view that expressed word meanings are subsequently connected, and anvitābhidhāna, associated with the claim that words express connected meanings. These are historical philosophical positions with their own theories of cognition, syntactic expectation, and linguistic authority. Equating them directly with contemporary model-theoretic compositionality would obscure both sides. Still, their disagreements make vivid that the fundamental issue is not merely whether a sentence possesses parts, but whether those parts first acquire content independently of their sentence-level relations.

The interpretive materials used here are academic historical exposition of Indian debates rather than a claim to direct philological examination of every surviving Sanskrit edition (Deshpande, 2020). This evidential distinction matters because a single English paraphrase of sphoṭa or anvitābhidhāna cannot settle the technical meaning of those Sanskrit terms. The comparison provides a philosophical gain: a theory of semantic composition must state whether it describes the dependence of semantic values, the order of psychological comprehension, the metaphysics of linguistic wholes, or several of these under separate arguments.

Frege: Context, Function, Sense, and Reference

Gottlob Frege transformed the problem of meaningful articulation through his functional analysis of predication and quantification. His Begriffsschrift (1879) provided a logical notation able to represent relations and quantifier dependencies inaccessible to the traditional identification of logical form with ordinary grammatical subject and predicate. A predicate and a name have different logical roles: one can be represented as an unsaturated function-like expression, the other as contributing an object suitable for its argument place. This was a foundational development for later compositional semantics, although the terminology and full machinery of typed model theory belong to later frameworks.

Frege's Grundlagen der Arithmetik (1884), especially the context principle discussed in §60, directs attention to the role of words within propositions. The German philosophical problem concerns the dependence of an expression's significance on a complete judgment. Read in isolation, this can seem to undermine the project of building sentence meanings from independently possessed word meanings. A more careful interpretation distinguishes an epistemological or methodological injunction to determine linguistic significance within propositional contexts from a formal claim that semantic values of wholes are assigned without structural constraints. Frege is a decisive source for the contextual problem as well as for the functional conception of complex content.

In “Über Sinn und Bedeutung” (1892), Frege differentiates Sinn and Bedeutung, usually rendered “sense” and “reference.” His analysis of informative identities explains how expressions sharing a referent can differ in cognitive significance. His account of sentences also associates the reference of a complete declarative sentence with a truth value, while its sense is a thought. These distinctions put pressure on simplistic versions of the composition principle. If the referents of constituent expressions determine the referent of a whole, do their senses also determine the sense of the whole? What happens in embedded attitude contexts, where the ordinary referent of an expression may cease to be the relevant semantic contribution?

A philosophical account inspired by Frege can seek composition at more than one semantic level. Truth-value determination is one target; the articulation of a thought is another. The issue is not resolved by stipulating that “meaning” stands for one of these. The theorist must explain why the selected level captures the interpretive contrasts at stake. Frege's importance, then, lies in the joint inheritance of structure-sensitive semantics, the context principle, and the sense–reference distinction. Treating any one of these as his complete and unqualified doctrine would flatten the historical development (Frege, 1884; Frege, 1892; Pelletier, 2001).

From Logical Analysis to Truth-Theoretic Semantics

The philosophical movement from logical form to explicit semantic interpretation required an answer to two questions: which structural relations characterize a sentence, and what interpretation is assigned to those relations? Frege's distinction between function and argument, Russell's investigations of definite descriptions, and the logical analyses of early analytic philosophy made it increasingly plausible that the grammar relevant to philosophical explanation could diverge from surface grammatical appearance.

Russell's “On Denoting” (1905) showed how the apparent grammatical form of a definite description could conceal quantificational structure. A sentence containing “the present king of France” can be given an analysis involving existential commitment, uniqueness, and predication. The philosophical significance for composition is that a theory may assign a logical structure to an expression and derive truth conditions from that structure even when ordinary syntax offers a different initial impression. The adequacy of the derivation must be judged by the problems it resolves, including existence, scope, and inference.

Early analytic discussions of propositions supplied varying theories about what complex content consists in. Logical atomism treated the articulation of propositions as philosophically important, while questions about relations, facts, and the representation of facts resisted a single settled metaphysical solution. A compositional semantics can borrow the resulting distinctions without assuming that a successful linguistic decomposition proves that the world itself is built from elements matching those of syntax.

The twentieth-century development of formal languages sharpened these concerns. A syntax can be defined recursively by starting with atomic expressions and specifying rules for complex expressions. A theory of semantic interpretation can then proceed along that recursive construction. This pairing is particularly clear in the work of Tarski, although the historical purpose of Tarski's truth definitions must be distinguished from Davidson's later use of a truth theory as an instrument of linguistic interpretation. The progression is conceptual rather than a single uninterrupted school: different authors sought to solve problems in logic, epistemology, metaphysics, and communication, and the compositional framework later assembled insights from these projects (Russell, 1905; Tarski, 1944).

Tarski: Truth and Recursive Satisfaction

Alfred Tarski's work on truth for formalized languages was a major condition for the development of precise semantic theories. In “The Semantic Conception of Truth and the Foundations of Semantics” (1944), Tarski explains the need to distinguish an object language from a metalanguage and to secure an adequate definition of truth for the target language. His account is shaped by the difficulties of semantic paradoxes and by the demand for a formally disciplined truth predicate. It should not be treated as a direct assertion that truth conditions alone constitute all aspects of linguistic meaning.

The crucial construction is satisfaction. An open formula can be evaluated relative to a sequence or assignment of values to variables, and the satisfaction conditions of complex formulas can be defined in terms of those of their components. For conjunction, an assignment satisfies a conjunction exactly when it satisfies each conjunct. For negation, it satisfies a negation exactly when it does not satisfy the embedded formula. Quantification is handled through systematic variation of assignments. These clauses articulate a recursively specified dependence of the semantic outcome on logical structure.

The method supplies an instructive example of compositional discipline. Assume a language with a unary predicate P, individual variable x, negation, and existential quantification. The satisfaction condition for P(x) is fixed by the object assigned to x and the extension of P. The formula ∃x P(x) is true in a structure when at least one admissible assignment maps x to an element in the extension of P. The step from atomic to complex formula is governed by explicit rules; there is no independent sentence-by-sentence stipulation for all possible quantified formulas.

Tarski's work thereby illuminates an important difference between specifying an interpretation and identifying the meanings speakers actually attach to expressions. A formal model can deliver truth conditions once a domain and interpretation function are supplied. The fact that a model meets a semantic definition does not demonstrate that its predicate assignments correspond to a particular natural language community's lexical practice. The philosophical problem of connecting a formal apparatus with public meaning remains open. Tarski's accomplishment is the rigorous articulation of truth and satisfaction in formalized languages, which subsequent theorists could employ for broader philosophical purposes (Tarski, 1944, pp. 341–376).

Davidson: Finite Theory and Unlimited Interpretation

Donald Davidson's “Truth and Meaning” (1967) makes explicit the connection between a finitely stated theory and the interpretation of indefinitely many sentences. He proposes that a theory of meaning should explain how competence with a limited vocabulary and finite set of constructions can support understanding of a potentially unlimited class of expressions. A Tarski-style truth theory seems promising insofar as it generates theorems connecting object-language sentences with truth conditions. The novelty is philosophical: a semantic method developed in formal logic is enlisted to address the understanding of ordinary language.

The appeal of the approach rests on systematic derivation. A theory containing appropriate clauses for names, predicates, conjunction, and quantification can yield a sentence-level theorem without memorizing the meaning of each sentence separately. Its explanatory adequacy cannot be reduced to producing extensionally true biconditionals. An accidental or interpretively perverse association might produce the required truth values while failing to represent what a speaker understands. The problem of interpretation therefore extends beyond formal truth.

Davidson's wider philosophy places meaning within a theory of radical interpretation, charitable understanding, belief attribution, and interaction with a shared world. The contribution of recursive semantic theory is to render linguistic competence surveyable through finite axioms and systematic consequences. But the semantic theory must also be confirmed by the speaker's linguistic behavior and interpretive circumstances. Derivability alone does not establish the appropriateness of the basic semantic clauses.

Davidsonian theory clarifies why compositionality has attracted philosophical rather than merely technical interest. It promises a way to join the indefinitely productive structure of language with the finite resources of a learner or interpreter. Yet the passage from finite axiomatizability to unique meaning attribution requires argument. Multiple truth theories can assign different structures and still agree on a restricted body of observed utterances. The challenge is to explain how interpretation selects among them. Davidson's project thus links composition to public evidence for meaning rather than identifying semantic structure with an untested formal recipe (Davidson, 1967).

Montague: Grammar as an Interpretive Algebra

Richard Montague developed a general approach in which syntactic formation and semantic interpretation are systematically coordinated. His “Universal Grammar” (1970) and “The Proper Treatment of Quantification in Ordinary English” (1973) treated significant fragments of natural language with mathematically explicit interpretive methods. Montague's work established that natural-language constructions, including quantification and intensional contexts, can be investigated with techniques previously associated primarily with formal languages.

In a Montague-style setting, a grammar specifies expressions, categories, and rules of combination, while a semantic component assigns objects of appropriate types and rules for obtaining complex values from the values of constituents. A determiner such as “every” can be modeled as a relation between sets or, under a typed conception, as an operation on predicates. This treatment explains why the semantic role of a quantified noun phrase is not adequately captured by assigning it an individual referent. Quantification is a structural operation with its own truth-conditional contribution.

The framework is particularly important for intensionality. Expressions embedded under verbs such as “believes” or in modal environments can depend on how their contents vary across circumstances of evaluation. Instead of abandoning composition, one can enrich the values composed from ordinary extensions to functions or intensions. Doing so introduces philosophical questions about the status of possible worlds, the kind of objects semantic types represent, and the interpretation of opacity. Semantic formalism makes those questions explicit while allowing different metaphysical commitments about the models.

Montague's influence should be stated precisely. The lasting contribution is the possibility of jointly formalizing grammar and semantic interpretation, accompanied by powerful analyses of natural-language phenomena. It does not follow that a specific fragment of English, a particular logical translation language, or a set of semantic types supplies the uniquely correct account of natural language. Later theories have modified grammar, the semantic interfaces, and the interpretation of context and discourse. Compositionality remains a central organizing constraint across these modifications (Montague, 1970; Montague, 1973; Janssen & Zimmermann, 2025).

A Formal Reconstruction of Compositionality

An explicit reconstruction begins with a syntactic algebra and a semantic algebra. Let E be the set of well-formed expressions and let each syntactic construction F be a possibly partial operation taking appropriately typed expressions as arguments and yielding a complex expression. Let M be a space of semantic values, with an interpretation function μ from E into M. A compositional interpretation provides for each relevant syntactic operation F a semantic operation G_F so that whenever F(e₁, …, eₙ) is well formed, the following equation holds:

μ(F(e₁, …, eₙ)) = G_F(μ(e₁), …, μ(eₙ)).

The left-hand side is the semantic value of the constructed expression. The right-hand side is a value calculated by a composition operation from the semantic values of the immediate constituents. This formulation expresses a homomorphism-like relationship between structures. It is a useful formal idealization, and detailed semantics may require typed domains, partial functions, semantic parameters, derivational histories, and more elaborate operations (Pagin & Westerståhl, 2010a; Hodges, 2001).

Consider conjunction in a simple truth-conditional language. If μ(φ) and μ(ψ) are truth values and the syntactic operation forms φ ∧ ψ, the associated G may be Boolean conjunction. A homomorphic clause then determines the value of the whole from the values of the immediate constituents. The simplicity of this case comes from the chosen semantic type. Natural-language predication introduces type-sensitive function application; quantification may require functions over sets or assignments; modality may require intensions indexed to circumstances.

The reconstruction also demonstrates why “determined by” requires interpretation. If determination means merely that there exists some mathematical function from a tuple of input values to an output, one needs to know whether the tuple distinguishes all cases that produce different outputs. If the tuple includes an arbitrarily rich encoding of an entire expression, the principle risks becoming formally cheap. If semantic operations have to be linguistically constrained, independently motivated, and reusable across constructions, the principle imposes a substantive demand.

The formal principle therefore establishes a conditional structural relation. Philosophical explanation requires an account of the chosen semantic values and the empirical or interpretive evidence that justifies the functions. A model can demonstrate that an analysis is coherent; the explanatory question asks whether its components are the right ones.

Semantic Values: Extension, Intension, and Sense

The assertion that constituent meanings determine complex meaning is ambiguous until “meaning” receives a precise theoretical role. In an extensional account, a proper name may denote an individual, a one-place predicate a set of individuals, and a declarative sentence a truth value. The apparatus elegantly handles predication and the truth-functional connectives, but difficulties arise when expressions with the same extension behave differently in larger contexts. Two predicates can apply to exactly the same things yet differ in the information they convey; two names can refer to the same person yet differ in their significance for an interpreter.

Frege's distinction between sense and reference makes the difficulty historically explicit. Consider a thinker who knows a person under one description while lacking the knowledge that the person is also described in another way. Replacing an expression with a co-referential expression may preserve the truth value of a simple extensional predication but alter the truth of a belief report. One response is to construct a semantics that composes senses, not merely extensional referents, in the relevant environment. Another is to give propositions a structure that preserves some way of presenting their constituents. These are substantive philosophical alternatives about the objects of belief, the nature of cognitive significance, and the identity conditions of semantic values.

Intensional semantics commonly assigns an expression a function from worlds, times, or other indices to extensions. A sentence's intension can then represent how its truth value varies across possible circumstances. Such functions permit systematic treatment of modal operators: “possibly φ” can be interpreted relative to some accessible world where φ holds. Yet possible-world intensions may still identify necessarily equivalent propositions that a thinker could understand differently. The representational problem is transformed, not automatically resolved.

A further option is structured or hyperintensional semantics, which distinguishes meanings that ordinary possible-worlds semantics would collapse. Logical conjunction, for example, might contribute a structure that records its constituent propositions rather than only their resulting truth function. The price is additional ontology or representational complexity, and the theory must motivate which distinctions are genuinely semantic. The broader lesson is that compositionality does not legislate a single semantic currency. It demands that the selected currency be preserved in a coherent account of how complexes are constructed (Frege, 1892; Pagin & Westerståhl, 2010a).

Local, Global, and Direct Composition

Local compositionality specifies semantic operations for immediate syntactic constituents. At each legitimate step in a derivation, the value of a complex is obtained from the values of the expressions immediately combined. This recursive formulation has the advantage of making an interpretation algorithm visible. The interpretation of a complete sentence can be traced through a finite sequence of local operations, and each operation can be tested against the relevant grammatical construction.

Global versions instead concern dependence on the values of all relevant lexical constituents and the structure of a whole expression. A globally compositional relation can permit dependencies that a strictly local rule does not explicitly encode. The distinction becomes important when semantic interpretation is constrained by information distributed through a larger configuration. Whether such information can be carried by locally computed structured values depends on the semantic resources permitted to the theory. Enriching those resources can convert an apparently nonlocal dependence into a locally compositional one, although this conversion may obscure rather than explain the interpretation.

Direct compositionality adds a separate architectural commitment. It asks whether meanings can be assigned directly to the expressions or derivations licensed by the syntax without positing an additional level at which the expression's relevant constituents are rearranged. In a theory that employs a separate level of Logical Form, semantic composition may apply to a structure distinct from the surface representation. A direct theory can instead make the combinatory syntax itself encode the needed scope or binding operations. These approaches can agree on sentence meanings while differing over what a grammatical explanation must contain (Barker & Jacobson, 2007).

The question is not which label expresses a stronger allegiance to compositionality. Each version allocates explanatory responsibility differently. A local interpretation may demand rich intermediate types; a global account may require a theory of how distant dependencies are resolved; a direct theory must show how its grammatical rules make hidden semantic operations available. These differences matter when assessing whether a formal analysis is economical, psychologically plausible, learnable, and compatible with observed syntactic patterns.

For a fair comparison, the investigator should fix the target data and distinguish syntax-changing and meaning-changing revisions. It is misleading to compare a weakly constrained theory with an alternative required to keep a much poorer vocabulary of semantic operations. The explanatory force of compositionality becomes clearer when the competitor theories face comparable constraints (Pickel & Szabó, 2025).

Quantification and Scope

Quantification demonstrates why semantic composition cannot generally be reduced to the combination of names and predicates. In a first-order model, “every” and “some” are associated with operations on sets or predicates, not with individual objects. A generalized quantifier treatment can interpret “every student” as a function that maps a property to truth if every student has it. On this analysis, the grammatical act of combining a determiner with a noun phrase can be assigned a corresponding semantic operation.

Consider “Every student read a book.” One interpretation allows the book to vary with the student; another permits a particular book to be read by every student. Under the first reading, each student has some book satisfying the relation. Under the second, there is some book that every student read. Symbolically these can be represented by ∀x(student(x) → ∃y(book(y) ∧ read(x,y))) and ∃y(book(y) ∧ ∀x(student(x) → read(x,y))). The formulas are not equivalent in general. Their difference makes quantifier scope a direct test of the interface between sentence structure and interpretation.

A semantic theory must locate the relevant structural distinction. One framework constructs different logical representations from an ambiguous surface string. Another provides a flexible method of composition in which continuations or other typed operations yield scope alternatives without movement to a separate level. A third might use pragmatic constraints to prefer one reading while maintaining a grammar capable of expressing both. What matters is that the different meanings are traced to a recoverable difference in the representation or derivation, not assigned independently after the fact.

Scope also exposes the relationship between semantic and linguistic evidence. Some formally possible readings are difficult or unavailable in particular constructions, while languages differ in the relation between word order, morphology, and scope interpretation. A wholly unrestricted logical composition algorithm would overgenerate. The problem is therefore to give semantic rules enough power to express attested readings and enough structural discipline to constrain unattested ones.

The successful analysis of quantification illustrates the point of semantic engineering: the meanings of larger expressions become predictable when the grammar and the types of its constituents are jointly specified. It does not settle the philosophical question of which semantic objects are ultimately real, nor whether speakers mentally compute the very same formulas written in a model (Montague, 1973; Heim & Kratzer, 1998).

Ambiguity, Lexical Selection, and Structural Alternatives

The principle of compositionality is often mischaracterized as requiring one semantic value for every surface sentence. Natural language is pervasively ambiguous. A lexical item can have multiple conventional senses, and a single word sequence can instantiate different syntactic structures. A compositional semantics can respect these facts by assigning values to disambiguated expressions or derivations rather than by assuming that a string, simply as a sequence of letters or sounds, has one predetermined value.

“Visiting relatives can be exhausting” illustrates how a sequence may be understood in relation to the activity of visiting relatives or to the relatives who are visiting. The semantic alternatives reflect different grammatical analyses and lexical roles. A theory that provides the same input tree for both interpretations has concealed an explanatory difference. A theory that gives them distinct licensed representations may retain a local interpretation rule for each representation.

Lexical polysemy introduces a related problem. “Bank” may denote a financial institution or the land beside a river. Disambiguation is often influenced by the sentence and the surrounding situation. A compositional theory can associate the word with a set of possible conventional values and model interpretation as a constrained selection. But it must also say where the selecting constraints belong: in lexical entries, syntactic construction, compositional rules, discourse interpretation, or pragmatic reasoning. The demand for explicit allocation of responsibility is itself a valuable contribution of the compositionality debate.

A more challenging case arises when an interpretation is contextually constructed rather than selected from an established inventory of senses. Creative use, coerced readings, and flexible noun combinations can lead a theorist to introduce sophisticated semantic operations. If the operations are allowed to inspect arbitrarily rich representations of an entire utterance, the formal principle becomes easy to satisfy. The gain of compositionality depends on keeping the inventory of operations independently constrained, predictively useful, and connected to recognizable classes of constructions.

A defensible semantics therefore treats ambiguity as a phenomenon to explain rather than a universal counterexample. It should characterize each available interpretation, identify its licensing structure, and make testable predictions about interactions with scope, anaphora, and inference. Ambiguity places a burden on the analysis; it does not logically show that no compositional analysis is possible (Pagin & Westerståhl, 2010b).

Productivity, Systematicity, and Learnability

Three prominent arguments for compositionality begin with observable features of linguistic competence. Productivity is the ability to understand or produce indefinitely many distinct complex expressions from limited resources. Systematicity concerns patterned relations among abilities: understanding one structured expression often accompanies the ability to understand related expressions formed by recombining its constituents. Learnability concerns how an individual acquires linguistic competence on the basis of finite exposure. Together these phenomena motivate the search for a finite and reusable structure of interpretation.

The argument from productivity can be reconstructed as follows. Linguistic competence extends beyond a list of memorized sentences. Speakers have finite learning histories and finite cognitive resources. A theory of their competence should therefore provide a procedure or capacity through which familiar elements contribute to new combinations. A compositional theory offers a transparent method: lexical values and a bounded repertoire of rules apply recursively. The conclusion that some systematic generative capacity is needed is powerful; the stronger conclusion that the capacity must satisfy a specific formal homomorphism condition requires further premises.

Systematicity adds a relational demand. A person who understands “Alice admires Bruno” typically also has the structural capacity to understand “Bruno admires Alice,” assuming the terms and constructions remain available. The pair displays a regularity that a collection of isolated stimulus–response associations struggles to explain. Fodor and Pylyshyn (1988) used such regularities in defending structured mental representation against certain connectionist accounts. The philosophical inference depends on whether other architectures can reproduce the same relations without the hypothesized explicit internal constituents. It is not a deductive proof that the brain implements a particular symbolic formalism.

Learnability likewise supports an architectural demand while leaving several accounts open. One may learn reusable words and constructions, acquire probabilistic patterns that generalize over syntactic environments, or exploit interaction between semantic structure and situational inference. A computational procedure could be systematic without satisfying the same theory of semantic values as a classical compositional model. The strongest learnability argument therefore appeals not to finitude alone but to the qualitative form of observed generalization across grammatical constructions.

Pagin and Westerståhl (2010b) scrutinize the familiar arguments and their limitations, and the significance of their critical treatment is methodological. A phenomenon can make compositionality a promising explanatory hypothesis without logically entailing it. To make the case stronger, a semantic theory should predict specific unseen combinations, generalize across construction families, and explain systematic patterns of both successful and failed interpretation. The historical attraction of the principle lies in its explanatory promise under those constraints (Fodor & Pylyshyn, 1988; Pagin & Westerståhl, 2010b).

Holism, the Context Principle, and Semantic Dependence

Semantic holism assigns an important role to a wider network of beliefs, inferences, or linguistic practices in the individuation of meaning. If the content of a word depends on its place within a large inferential structure, the assumption that a constituent brings a fully specified and independently identifiable value into every combination becomes questionable. But holism and compositionality operate at potentially different explanatory levels: a holistic account might explain how lexical values are fixed, while a compositional account explains what follows once those values and rules are in place.

This distinction brings Frege's context principle back into view. The injunction to seek the significance of a word in the context of a proposition can be read as guidance for the determination or acquisition of word meaning. Compositionality, understood structurally, asks whether the resulting semantic assignments support a systematic account of complex expressions. It would be a mistake to declare the two principles logically inconsistent without first specifying which kind of dependence each asserts.

Consider inferentialist approaches to meaning. A word or sentence may be understood through the commitments and entitlements associated with its use in reasoning. One theory could treat these inferential roles as semantic values and investigate rules for their combination; another might contend that inferential role depends on an open-ended totality of practice and resists a finite compositional account. The disagreement concerns the form and granularity of inferential dependence. A statement that meaning is “holistic” by itself does not establish how interpretation of a particular syntactic construction works.

There are important degrees of holism. An expression may have context-sensitive application conditions, be partly individuated by inferential role, or acquire its content only within an entire theory. Stronger forms can make stable lexical semantic values difficult to identify across revisions of belief. Yet they also face an explanatory challenge about how speakers communicate and preserve enough interpretive continuity to disagree. A compositional model that presupposes stable word values must give an account of how those values are maintained; a holistic model must show how local grammatical differences still produce predictable changes in content.

The philosophical synthesis is conditional: metasemantic dependence on a network of use can coexist with structural dependence in semantic interpretation, provided the theory explains how and when values are stabilized. This is a substantive proposal about explanatory levels, rather than an attempt to define away conflict. The disagreement remains live wherever the network of interpretation changes the semantic contribution of a constituent within the very same compositional operation (Frege, 1884; Horwich, 1998).

Indexicals, Context, and Circumstance of Evaluation

Indexicals challenge a context-independent notion of lexical meaning. The word “I” typically refers to the speaker of the relevant utterance, while “here” and “now” depend on place and time. The sequence “I am here now” can convey different propositions when uttered by different people in different circumstances even while its linguistic type and grammatical structure remain constant. An adequate semantics must therefore distinguish a standing linguistic rule from the content associated with a particular occurrence.

Kaplan's distinction between character and content provides an influential framework. Character is a rule that takes a context to a content; content is what can then be evaluated with respect to circumstances. The character of “I” can remain stable across uses while the content varies with the context's speaker. A compositional theory may compose characters into a sentence character, which yields a proposition at a context. Alternatively, it may fix a context before composing the relevant contents. The level at which composition occurs affects what is being claimed about invariance.

The distinction between context of utterance and circumstance of evaluation is especially important in modal environments. A modal operator can shift the worlds over which a proposition is evaluated without thereby changing who counts as the speaker of the actual utterance. Conflating these parameters leads to incorrect predictions about embedded indexicals. Multiple kinds of context sensitivity have since motivated richer frameworks for demonstratives, tense, location, and other dependent expressions.

The problem extends beyond overt indexicals. Domain restriction can affect quantifiers; comparison classes may shape the interpretation of gradable adjectives; anaphoric reference may depend on prior discourse. One strategy incorporates a contextual parameter into the denotation of each relevant expression. Another employs parameters supplied by syntax or by a prior pragmatic assignment. Each model can preserve a version of composition, but theories differ in how they divide the labor among character, content, syntax, and pragmatic selection.

A coherent theory must avoid a trivial rescue maneuver in which “the meaning” of a constituent becomes whatever the whole utterance needs it to be. The relevant contextual rule should generate predictions across different sentences and speakers. This allows context sensitivity to become a resource for a precise semantic explanation rather than an unrestricted exception clause (Kaplan, 1989; Pickel & Szabó, 2025).

Pragmatics, Implicature, and Communicative Enrichment

A speaker can communicate more than a sentence conventionally encodes. When someone says “Some of the guests arrived,” a hearer may infer that not all arrived, although the ordinary logical meaning of “some” permits a situation in which all arrived. The additional inference is commonly investigated through pragmatic principles such as conversational implicature. Compositional semantics explains how the conventional content is built; pragmatic theory explains how an utterance in circumstances can convey a further commitment.

The distinction becomes more complicated where context seems to affect the proposition actually asserted. An utterance of “It is raining” may be naturally understood as concerning a particular place, and a sentence containing “ready” or “tall” may require an implicit comparison or purpose. Some theories represent these parameters in the semantic structure; others attribute their determination to pragmatic enrichment. The allocation determines whether a proposed compositional semantics is responsible for a fully context-fixed proposition or a more abstract conventional content.

A disciplined account needs a principled point of contact between the two levels. The formal semantics can generate a structured content with free variables or parameters; interpretation in context then fixes their values. Alternatively, pragmatic adjustment might alter lexical or phrasal content before the main composition proceeds. The possibilities have different consequences for embedding, negation, quantifier scope, and compositional prediction. A semantic/pragmatic distinction earns explanatory value when it predicts these interactions.

Metaphor sharpens the issue. The utterance “The argument collapsed” uses a familiar verb in a context where literal physical collapse is inappropriate. A hearer may construct an interpretation involving evidential failure or loss of argumentative support. If every such creative interpretation must be encoded in a fixed lexical entry, the lexicon grows without clear limits. If the entire interpretation is unconstrained pragmatic creativity, the systematic force of constructions becomes hard to explain. The research problem is how conventional structure constrains innovative communicative use.

The principle of compositionality is therefore best evaluated relative to an explicit target: encoded content, contextually expressed proposition, or total communicated significance. A theory that succeeds on one target and fails on another has not shown that composition is universally true or universally false. It has clarified where the explanatory demands differ (Pagin & Westerståhl, 2010b; Stanford Encyclopedia of Philosophy: Pragmatics).

Idioms, Metaphor, and Conventional Expressions

Idioms are a classic apparent counterexample because their conventional significance can diverge from a straightforward combination of the ordinary meanings of their words. “Kick the bucket,” in an idiomatic use, can conventionally express dying. A literal semantic procedure applied to the ordinary values of “kick,” “the,” and “bucket” would not produce that meaning. The philosophical issue is how the idiomatic reading is represented within a larger grammar and why some idioms allow syntactic variation while others strongly resist it.

One solution recognizes particular multiword sequences as lexicalized expressions with their own semantic entries. A larger sentence can then compositionally combine such an idiom with other material, even though the idiom is treated as a semantic unit. This handles fixed expressions but creates a question of explanatory economy. A language may contain many idioms, and a theory that declares every recalcitrant combination an indivisible primitive risks satisfying compositionality only by discarding useful constituent structure.

Nunberg, Sag, and Wasow (1994) distinguish among kinds of idiomatic phenomena and demonstrate the importance of internal syntactic and semantic structure. Some idiomatically combining expressions preserve interpretable relations between their constituent contributions and the overall idiomatic meaning. These may support passivization, modification, or other operations in systematic ways. Others are more rigid, suggesting that a single semantic solution is inappropriate. The distinction makes idioms a source of empirical constraints rather than a simple refutation of compositionality.

Construction-specific meanings offer another strategy. On such accounts, a syntactic pattern can itself contribute conventional meaning, and particular lexical items can interact with that contribution. This reinforces the importance of deciding whether constructions count as constituents or semantic operators. A semantic theory focused exclusively on ordinary dictionary definitions of isolated words will predict too little.

Metaphorical innovation introduces a different problem from conventional idiomaticity. A new metaphor can be understood without a stored lexical entry for the entire expression. Its success often depends on background concepts and relevance judgments. The right question is which aspects of metaphor follow from stable compositional structure and which arise through broader interpretive reasoning. An approach able to make that division precisely explains more than either an unrestricted idiom inventory or an unexplained appeal to creativity (Nunberg, Sag, & Wasow, 1994).

Attitude Reports, Quotation, and Opacity

Reports of thought and speech reveal how the semantic contribution of an expression can vary with its embedding. The ordinary substitution principle says that co-referential names should be interchangeable without changing truth value in extensional contexts. Yet “Mira believes that the evening star is visible” and “Mira believes that the morning star is visible” can differ in truth even though the astronomical referent of the two descriptions is the same. A compositional semantics restricted to ordinary reference cannot derive this contrast without additional machinery.

Frege's approach motivates a treatment in which the embedded occurrence contributes something corresponding to its ordinary sense. Other approaches appeal to structured propositions, possible-world propositions, two-dimensional or hyperintensional contents, or a distinction between what the belief report semantically expresses and what it pragmatically communicates. Each solution must explain how the altered contribution follows from the embedding construction and why related inferences succeed or fail.

Quotation raises a neighboring but distinct difficulty. In “The word 'river' has five letters,” a quoted expression is being mentioned rather than used to refer to rivers. The interpretation of quotation can involve an expression's form, token, or type rather than its ordinary extension. One can construct a compositional rule for quotation, but only by specifying what the quotation operation takes as its input. If the theory stipulates that quoted materials always denote their own expression types, it should also explain mixed quotation and embedded uses where quotation partly reproduces someone else's speech.

These cases reveal that the principle is relative to semantic level. A rule that correctly composes extensional predicates may fail in intensional or quotational constructions because the relevant values are different. The philosophical challenge is to articulate systematic shifts rather than isolate each troublesome sentence as an exception. A successful framework explains not only an attested reading but also the distribution of substitutions, scope interactions, and embedding patterns.

No unique general solution follows from the bare principle of compositionality. It remains necessary to determine the objects of attitudes, the relation between linguistic and mental content, and the semantic or pragmatic role of quotation. The debate is a central example of why the principle guides theory construction while leaving the substantive theory of meaning unsettled (Frege, 1892; Pagin & Westerståhl, 2010b).

Anaphora, Discourse, and Dynamic Semantics

Meaning depends on structure within sentences, but discourse often establishes dependencies across sentence boundaries. Consider “A scholar entered. She opened the book.” Understanding the pronoun requires relating it to a discourse referent introduced in a previous utterance, assuming the intended interpretation makes that link available. A sentence-by-sentence semantics that interprets each sentence only against an unchanging model may fail to represent the contribution of the initial introduction to subsequent reference.

Dynamic semantic frameworks treat interpretation partly as the transformation of informational states. Instead of assigning a sentence only a proposition or truth condition, an account can model how processing it changes an assignment, discourse representation, or context. The meaning of the second sentence can then depend on a state resulting from the first. The interpretation may still be compositional: rules determine the discourse update of a complex construction from the update contributions of its parts. The composed semantic values are richer than static truth values.

This is a philosophically consequential revision. It shows that successful composition need not mean that every sentence independently expresses a proposition whose truth value is calculated without reference to linguistic history. The relevant object of semantic evaluation can include context change, information storage, and the availability of referents. The distinction between static and dynamic interpretations therefore concerns the proposed semantic type and the phenomena it is meant to explain.

Dynamic approaches also raise further difficulties. Some anaphoric links are constrained by accessibility, quantifier scope, and presupposition; others depend on pragmatic inference and the recognition of salient entities. A theory that treats every desired coreference as an arbitrary post hoc assignment has abandoned explanatory discipline. A persuasive compositional account states which dependencies are licensed by the representation and which are supplied by independently motivated contextual processes.

Discourse underscores a general conclusion of this monograph: the appropriate grain of composition is a substantive theoretical issue. In some analyses it is the phrase, in others a sentence, discourse segment, or update function. A successful theory identifies the structural operation at the level needed to explain the phenomena and resists moving the level only to avoid counterexamples (Stanford Encyclopedia of Philosophy: Dynamic Semantics; Heim & Kratzer, 1998).

Meaning Change, Polysemy, and Conceptual Combination

Semantic composition takes place in languages whose lexical inventories evolve and whose speakers creatively extend conventional patterns. A compound such as “stone lion” need not denote a lion made of stone in the same way that “stone wall” denotes a wall built from stone; the relation contributed by the noun–noun construction is underdetermined unless additional structure or conceptual knowledge is supplied. An account of flexible conceptual combination must therefore determine whether the relation is part of constructional meaning, a contextually selected semantic operation, or a pragmatic inference beyond literal meaning.

Polysemy is not equivalent to unrestricted ambiguity. A verb may exhibit related uses connected by systematic patterns, and an adjective may shift its standards across comparison classes. These families suggest that the lexicon may encode structured potentials rather than a flat list of separate dictionary senses. A compositional theory that includes coercion, type shifting, or parametrized lexical operations can capture regularities across cases, provided the operations are constrained independently of the individual examples they explain.

Historical semantic change introduces a temporal dimension. A word's conventional contribution may vary across periods and communities. This does not in itself undermine compositionality at a given stage of interpretation. It raises a question about how stable the relevant assignments are and how older and newer conventions coexist during change. A philosophical theory should distinguish historical causation of meaning from the synchronically organized procedures through which a community interprets a particular expression.

The criterion of explanatory success is especially clear here. A model that says the semantic contribution of a phrase equals whatever speakers understand has no predictive value. A model that identifies productive classes of compounds, restrictions on coercion, and semantic differences arising from grammatical structure can explain why some innovative combinations are readily understood while others remain obscure. Context is part of the explanation only when its role is stated with enough precision to permit comparison among cases.

These phenomena invite a plural account of semantic resources: lexical convention, grammatical operations, conceptual background, and situated inference can interact without being equivalent. Compositionality concerns the structural relation among such resources where they enter the linguistic interpretation process. The research problem is to establish that relation rather than expand the definition of “part” until every explanation counts as compositional.

Compositionality in Cognitive Architecture

The compositionality of a public language must be distinguished from the organization of internal mental representations. A linguistic grammar can assign systematically constructed meanings without settling whether the mind processes those meanings by manipulating discrete symbols, distributed activation patterns, predictive representations, or some combination. Cognitive compositionality is a hypothesis about the operations by which complex representational capacities arise from reusable components. The philosophical relation between linguistic and cognitive composition is close, but their empirical targets differ.

Fodor and Pylyshyn (1988) argue that the productivity and systematicity of thought and language support a classical architecture with structured representations. Their objection to certain connectionist models concerns whether the models have the requisite constituent structure or merely simulate the associated behavioral patterns. A contemporary reassessment must recognize that connectionist and neural models can be designed to encode structured relations and that observed generalization depends on architecture, training, and test conditions. The original argument remains valuable as a demand for an explanation of systematicity, not as a demonstration that a particular hardware or processing system has been established.

The philosophical issue can be sharpened by distinguishing representational composition from compositional generalization. A model may encode representations built from component features, yet fail to generalize when familiar elements appear in unfamiliar structural configurations. Another model may show broad success across held-out combinations without an interpretation theorem establishing a unique decomposition into semantically typed components. These observations indicate that a behavioral test and a formal semantic property answer different questions.

The transition from language to thought additionally encounters the possibility of conceptual content that cannot be neatly expressed in public language. Perceptual categories, imagistic relations, skills, and affective responses may have structures not captured by a sentence-like grammar. If thought is compositional, the relevant mental primitives and composition operations require independent identification. Calling an experience “composed” without such evidence offers little beyond a metaphor.

A robust research program therefore joins formal definitions to controlled generalization tests, cognitive evidence, and explanatory comparison. It asks what capacities are preserved under recombination, which structures are represented, and how a theory predicts novel understanding. These criteria make compositionality an empirically meaningful research topic without importing assumptions from formal linguistics as settled psychological facts (Fodor & Pylyshyn, 1988; Millière & Buckner, 2026).

Logic, Programming Languages, and Computation

Formal logic and programming-language semantics offer contexts in which composition can be enforced by explicit definitions. In a well-specified formal language, syntax is constructed by stated formation rules; a denotational semantics maps expressions to mathematical values; and the interpretation of a compound can be given by a corresponding semantic operation. This clarity is a methodological advantage. It allows a researcher to prove properties of the semantics by induction on syntax and to identify the precise assumption required for each result.

For instance, the meaning of arithmetic expressions may be assigned by recursive evaluation clauses. The expression “2 + (3 × 4)” is interpreted by first assigning values to its parts and then applying multiplication and addition under the prescribed syntactic grouping. The interpretive result depends on the syntax tree rather than a table that assigns an independent value to every complete expression. In a programming language, a similar technique can define the meaning of an expression as a function of environments or states, thereby representing variables, effects, and other dependencies.

These examples also show why denotational composition alone does not settle correctness. A compositional function can rigorously interpret a formal syntax while misrepresenting the intended operational behavior of a program. One must establish a correspondence between the semantic model and execution, an observational equivalence, or another appropriate adequacy criterion. The relation between formal meaning and intended behavior is analogous to the difference between a model-theoretic grammar and independently attested natural-language interpretation.

Computational evaluation introduces practical constraints. Two descriptions may denote extensionally identical functions but exhibit substantially different computational costs. A semantic account that preserves denotation can be extensionally compositional without explaining efficiency, error patterns, or processing limitations. The philosophical distinction between semantic structure and execution mechanism remains necessary.

A broader conclusion follows: computer science gives particularly sharp demonstrations of what compositionality can formally accomplish because the syntax and semantic conventions can be deliberately stipulated. Applying the same success to natural language requires additional evidence because natural-language structure and community meaning are objects of investigation rather than fully controlled definitions. The two domains illuminate one another through a common abstract method while preserving different explanatory burdens.

The Formal Vacuity Challenge

One of the strongest philosophical criticisms of compositionality concerns the possibility of encoding virtually any semantic assignment in a formally compositional scheme. Zadrozny (1994) proved that under sufficiently permissive representation and encoding choices, an arbitrary semantics can be recast as a compositional one. If the only requirement is the existence of a function satisfying an abstract composition equation, the property may cease to discriminate among competing explanations.

The force of the challenge depends on how meanings, syntax, and permissible operations are held fixed. Suppose a proposed semantic value for an expression encodes the entire expression together with instructions for recovering its already stipulated interpretation. Composition could then consist largely in building increasingly detailed encodings. The formal equation would hold, but the analysis might have merely transported a noncompositional semantic lookup procedure into the meaning assignments themselves. The mathematical result therefore presses on the distinction between representation and explanation.

A reply must impose independent restrictions. Semantic values should correspond to justified linguistic distinctions rather than arbitrary coding devices. Composition rules should be stable across families of expressions. The analysis should preserve the meanings assigned to constituents under appropriately controlled substitutions. It should predict novel cases and not just restate a completed interpretation table. When syntax must be enriched, the enrichment should have independent grammatical support. Under such restrictions, a compositional theory can again exclude some assignments and prefer others.

Hodges (2001) and Pagin and Westerståhl (2010a, 2010b) discuss formal conditions and variants that clarify the actual strength of compositionality. The literature shows that a seemingly transparent slogan conceals a family of mathematical properties. Substitutivity, global and local formulations, and the treatment of semantic equivalence can diverge. An argument claiming that the principle is universal must specify the selected property and the admissible class of interpretations.

The vacuity challenge yields a constructive lesson. A philosophical principle can be formally satisfiable while carrying substantive explanatory power only when additional constraints are supplied. Compositionality's philosophical significance thus depends on a disciplined conception of admissible semantic values and operations. The strongest theory is not the one that can redescribe every possible assignment as compositional; it is the one that explains otherwise unpredicted semantic regularities with constrained, reusable resources (Zadrozny, 1994; Hodges, 2001).

Explanatory Adequacy and Competing Semantic Architectures

The competing architectures of linguistic semantics can be evaluated by what they account for under common constraints. A model-theoretic theory derives truth conditions from typed lexical and syntactic values. A proof-theoretic or inferentialist approach may identify meaning with inferential capacities, commitments, or rules. A dynamic account models the interpretation of an expression as an update to an information state. A use-oriented account appeals to stable roles in linguistic practice. These approaches differ in semantic ontology and method; any of them may admit compositional operations appropriate to its chosen values.

A comparative framework must therefore ask several questions independently. Does the theory assign stable values to comparable constituents? Does its composition procedure predict interpretations of previously unexamined expressions? Does it distinguish scope and reference ambiguities? Does it account for attitude embedding, indexicals, and discourse dependencies without uncontrolled rescue clauses? Does it connect its formal objects with how people use and understand language? An account that excels at formal derivation yet lacks evidential criteria for selecting a lexicon has solved only part of the philosophical problem.

The evaluative importance of theoretical economy is also specific. An account with a large collection of one-off lexical entries may fit familiar sentences but explain little about unseen combinations. Yet a theory with overly general composition operations may overgenerate unattested meanings. The aim is a constrained balance between expressive power, empirical coverage, and independent motivation for structure. There is no single numerical measure of simplicity that substitutes for this comparative inquiry.

Interpretive adequacy requires attention to speakers and situations. If two theories derive the same truth conditions but disagree about what is conventionally encoded, which inferences are pragmatic, or which representations are psychologically available, they should be tested against phenomena capable of distinguishing those predictions. The mere convergence on sample truth values does not settle the disagreement.

Compositionality, on the resulting view, is a cross-framework test of structural accountability. The principle's explanatory value lies in asking each theory to exhibit how the behavior of complex expressions is systematically related to repeatable contributions of their constituents. It neither licenses treating every phenomenon as reducible to word meaning nor prejudges the correct nature of semantic content (Horwich, 1998; Pietroski, 2004).

Comparative and Contemporary Research

Recent research extends the compositionality debate beyond the traditional examples of English truth-conditional semantics. Typologically diverse languages raise questions about how constituency, morphology, scope, and information structure coordinate semantic contributions. If distinct grammatical systems support productive meaningful combinations, compositionality may name a general explanatory constraint. Demonstrating that generality requires analyses of actual languages, not projection of a single English syntax onto every linguistic tradition.

Comparative philosophy supplies another form of breadth. Classical Indian debates about word and sentence meaning pose questions about the priority of a linguistic whole; European traditions developed explicit theories of logical form and reference; contemporary formal semantics investigates typed interpretation and the algebra of combination. The historically responsible comparison identifies analogous problems while preserving differences of philosophical aim. A discussion of sphoṭa, Frege's context principle, and Montague's grammar can be illuminating precisely because they articulate distinct accounts of what makes sentence understanding possible.

Animal communication research supplies an important empirical frontier. Some animal communication systems exhibit structured signal combinations that correlate with systematically varying interpretations. Klinedinst (2025) argues that evidence for compositionality in nonhuman communication must be interpreted with care: demonstrating that a combination has a distinctive effect does not by itself reveal the semantic operation through which that effect is generated. The debate requires controls that distinguish structural semantic contribution from fixed multi-signal associations, contextual learning, and other possibilities.

Research on language models has made the distinction between successful response behavior and structural semantic explanation especially pressing. Millière and Buckner (2026) survey philosophical questions about syntactic competence, compositionality, meaning, representation, reasoning, and agency in large language models. In this domain, a test can probe generalization to new structural combinations, but good performance alone does not determine the model's internal representational ontology or settle the metasemantic basis of its outputs. Conversely, a failure on one systematicity benchmark does not establish that every construction lacks compositional representation.

These developments reinforce the research method developed here. Claims of composition are strongest when a theory identifies the components, the composition operation, the domain of applicability, and the independent evidence for the resulting interpretation. This method allows semantic analysis to extend across historical traditions, species, and architectures while retaining conceptual precision (Klinedinst, 2025; Millière & Buckner, 2026).

Compositionality in the Artificial Era

The Artificial Era introduces new sites in which structured linguistic artifacts are produced, transformed, interpreted, and preserved. In the philosophical framework of Philosophy by Bogdanova, the transition From Homo to Artificial makes the possibility of nonbiological linguistic operations historically significant. Compositionality becomes relevant here because digital and artificial systems can manipulate expressions whose semantic behavior is structured, while their mechanisms of interpretation and their relation to public meaning remain objects of philosophical investigation.

Three issues require separation. The first is the formal property of an interpretation: whether meanings assigned to complex expressions are determined by the meanings assigned to constituents and the operations combining them. The second is behavioral generalization: whether an artificial system successfully handles novel combinations and respects their inferential or truth-conditional differences. The third is the metasemantic question of how its representations acquire or sustain the meanings attributed to them in public discourse. Success at the first or second does not settle the third.

The distinction has consequences for authorship and semantic continuity. An authored digital text may persist through copying, translation, indexing, and machine interpretation. Its structure can support repeatable interpretation, but preservation of a character sequence does not guarantee that every future interpreter assigns the same meanings to its parts. Conversely, differences in interpretive architecture do not automatically eliminate stable public textual relations. A theory of semantic persistence must connect material inscription, linguistic convention, structured interpretation, and historical continuity.

This is a research application of the general philosophical problem. It does not revise Aisentica's canonical definitions or propose a new canonical theory. Classical work by Frege, Davidson, and Montague provides the historical-semantic center of the analysis; contemporary artificial systems widen the empirical and metaphysical questions about where composition is realized. The enduring demand remains to identify meaningful constituents, legitimate composition operations, and evidence that the resulting account explains rather than redescribes understanding.

Open Philosophical Questions

Several fundamental questions remain open because the principle can be interpreted at different levels of semantic theory. The first concerns the individuation of semantic values. If a theory uses ordinary reference, it may fail to distinguish cognitively distinct expressions; if it uses highly structured senses, it risks multiplying distinctions without independent justification. A philosophical criterion is needed for determining which distinctions belong to meaning and which arise through psychology or pragmatics.

The second concerns the structural representation of natural language. Scope, binding, ellipsis, and discourse phenomena can be accommodated through distinct syntactic or semantic architectures. The fact that several systems obtain the same truth conditions does not decide which reveals the explanatory structure of competence. The methodological task is to identify empirical consequences of these competing representational commitments.

The third concerns the role of context. How much contextual determination can a constituent undergo while the overall interpretation remains informatively compositional? A theory that uses context-indexed values can be highly systematic, yet the source and limits of those contextual adjustments require analysis. In particular, the relation between conventional meaning, asserted content, and total communicated significance remains contested.

The fourth concerns the boundary between semantic competence and cognitive mechanism. Productivity and systematicity motivate reusable structure, but they do not uniquely specify the architecture implementing it. Cognitive, developmental, and computational research should distinguish the formal property of semantic representation from the dynamics through which agents learn and deploy interpretations.

The fifth concerns explanatory nontriviality. The formal encodability of arbitrary interpretations requires independent constraints on admissible semantic types, operations, and grammatical analyses. Establishing those constraints is an ongoing philosophical and scientific project, rather than a secondary technical footnote. These five problems define the central research agenda for a substantive theory of composition.

Philosophical Synthesis

The historical reconstruction and formal analysis yield a unified but differentiated account of compositionality. First, the foundational problem is the systematic intelligibility of complex expressions: how the semantic contributions of repeatable elements can explain indefinitely many distinct utterances. Early investigations of predication and sentence meaning disclose the underlying philosophical problem. Frege's logic and his sense–reference distinction articulate the modern conceptual pressures, while Tarski's recursive truth theory, Davidson's theory of interpretation, and Montague's grammatical semantics provide distinct methods for making interpretation explicit.

Second, the precise compositionality principle is a structural relation within a specified semantic system. A syntactic operation and a corresponding semantic operation must agree on the values assigned to legitimate complex expressions. The formal relation is indispensable for analytical clarity, yet the existence of an algebraic representation does not independently establish a satisfactory theory of linguistic meaning. This is the substantive lesson of the formal vacuity challenge and of the distinction between semantic representation and explanation.

Third, the phenomena historically treated as exceptions reveal the range of interpretation. Quantification and scope require a systematic account of structural alternatives; intensionality and attitude reports require finer-grained semantic values; indexicals require context-sensitive rules; idioms demand attention to conventionalized constructions; discourse anaphora invites context-update operations. In each case the research task is to determine whether a constrained semantic architecture can explain the dependency. Adding arbitrary hidden structures or semantic encodings without supporting evidence merely shifts the problem.

Fourth, the arguments from productivity, systematicity, and learnability support the search for reusable interpretive mechanisms. They provide strong explanatory motivation while underdetermining a unique formal or cognitive architecture. The relevant evidence concerns structured generalization, not the philosophical prestige of a preferred framework. Similar caution applies to studies of nonhuman communication and artificial language systems.

Finally, the monograph establishes the following methodological criterion. A philosophically informative theory of composition identifies independently motivated constituents, independently constrained semantic values, repeatable operations, and observations that can confirm or challenge its predictions. This criterion preserves the formal content of compositionality while explaining when the principle becomes a substantial contribution to the philosophy of language. The consequence is a research program that combines historical fidelity, formal explicitness, semantic discrimination, and empirical accountability.

Conclusion

Compositionality is a foundational principle of linguistic interpretation because it specifies how the semantics of complex expressions can depend systematically on their constituents and their mode of combination. Its importance arises from a genuine explanatory problem: limited linguistic resources enable productive, structured understanding, and the philosophy of language must account for that capacity.

The investigation has shown that a compositional semantics is always relative to a theory of syntax, a selection of semantic values, and rules governing interpretation. Frege's context principle and the sense–reference distinction, Tarski's recursive semantic methods, Davidson's finite truth-theoretic program, and Montague's typed grammar illuminate different dimensions of this architecture. The principal challenges from quantifier scope, contextual interpretation, opacity, quotation, idioms, and discourse are philosophical problems about the proper organization of the theory rather than automatic proofs of the impossibility of composition.

The most demanding critical result concerns nontriviality. An arbitrary interpretation may be recoded to satisfy a permissive formal condition. The value of a compositional account therefore depends on the constraints that make its operations explanatory: justified constituents, stable semantic contributions, predictive rules, and independently assessable consequences. Compositionality reaches its full philosophical significance when formal structure and public meaning are investigated together. Its continuing research agenda extends from classical theories of sense and truth to cognition, comparative communication, and artificial systems, while preserving the distinction between a structural semantic model and the larger philosophical explanation of meaning.