The Reasons for Semantic Compositionality Markus Werning [email protected] Department of Philosophy Heinrich Heine University Düsseldorf c Markus Werning – p.1/27 The Reasons for Semantic Compositionality Overview Compositionality — the formal notion Productivity Systematicity Interchangeability of synonyms salva significatione c Markus Werning – p.2/27 The Reasons for Semantic Compositionality The Colloquial Notion The meaning of an expression is determined by the meanings of its parts. c Markus Werning – p.3/27 The Reasons for Semantic Compositionality Frege Frege, in the only posthumously published manuscript Logic in Mathematics, writes what later, to many, serves as the template for a formal definition of the principle of Semantic Compositionality: (...) thoughts have parts out of which they are built up. And these parts, these building blocks, correspond to groups of sounds, out of which the sentence expressing the thought is built up, so that the construction of the sentence out of parts of a sentence corresponds to the construction of a thought out of parts of a thought. And as we take a thought to be the sense of a sentence, so we may call a part of a thought the sense of that part of the sentence which corresponds to it. c Markus Werning – p.4/27 The Reasons for Semantic Compositionality The General Message The quotation is, of course, worded in Frege’s jargon. A closer look, however, reveals a message that can be put in more general terms and might be convincing even to someone who refuses to adopt thoughts and senses into his universe of discourse. To uncover this message, I shall distinguish three aspects of Frege’s statement: (a) The part whole relation in the linguistic realm. (b) The meaning function from the linguistic realm into the realm of meanings. (c) The correspondence between the linguistic part-whole relation and a part-whole relation in the semantic realm. c Markus Werning – p.5/27 The Reasons for Semantic Compositionality The Part-Whole Relation between expressions First, he claims that there is a part-whole relation between sentences and less complex items of language. It has been common since to regard the part-whole relation in question not as a relation of mereological constitution in the sense that a mereological part of an object is tokened whenever and wherever the object is tokened. We will here conceive the part-whole relation in question as a relation of syntactical constitution, instead. Syntactical parts need not be mereological parts. In English the expression ‘not’, e.g., is a syntactical part of the expression ‘can’t’, although it’s not a mereological part thereof — one may even say that ‘go’ is a syntactical part of ‘went’. c Markus Werning – p.6/27 The Reasons for Semantic Compositionality Syntactic Constitution Let us assume a language with grammar G. The grammar specifies a set of atomic terms and a list of syntactic rules such that the set of terms T of the language is determined. The syntactic part-whole relation can now be defined as follows: A term s of a language with the set of terms T and grammar G is called a syntactic part (or constituent) of a term t of T if and only if (a) there is a Turing computable partial function α from the n-th Cartesian product of the set of terms T into the set of terms T such that t is a value of the function α with s as one of its arguments and (b) there is a syntactic rule in the grammar G according to which α(s 0 , ..., sn−1 ) is a term of the language if α is defined for (s0 , ..., sn−1 ) and if s0 , ..., sn−1 are terms of the language. In this case the function α is called a syntactic operation justified by the grammar G. In short, a term and any of its syntactic parts stand in the relation of value and argument of a syntactic operation. c Markus Werning – p.7/27 The Reasons for Semantic Compositionality Partial Order Since the set of rules of a language is closed under conjunction, every iteration β(..., α(...), ...) of two syntactic operations α and β is itself a syntactic operation. The transitivity of the relation of syntactic constitution is thus warranted. The reflexivity of the relation follows from the fact that the identity mapping is a syntactic operation of every language. Often it is also presumed that syntactic constitution be anti-symmetrical, i.e.: a term s is identical with a term t if s is a syntactic part of t and t is a syntactic part of s. In the case of transitivity, reflexivity and anti-symmetry, syntactical constitution would be a partial order in the set of terms. c Markus Werning – p.8/27 The Reasons for Semantic Compositionality Some Technicalities We allow terms to have variables ξ, ξ0 , ξ1 , etc. as syntactic parts. The set of grammatical terms GT (G) is a set of terms such that the terms of the set do not contain any variables. GT (G) is closed under syntactic constitution. The only purpose for the introduction of variables is to specify the positions in which certain grammatical terms occur as syntactic parts within other grammatical terms. In our notation t(p0, ..., pn−1 |ξ0 , ..., ξn−1 ) is the term one obtains if one simultaneously replaces each occurrence of the variable ξi in the term t by the term pi , for each index i (0 ≤ i ≤ n − 1). Given that a non-grammatical term t contains the variables ξ0 , ..., ξn−1 only, and given that p0, ..., pn−1 are variable-free, the term t(p0, ..., pn−1 |ξ0 , ..., ξn−1 ) is grammatical and the grammatical terms p0, ..., pn−1 figure in it at the positions ξ0 , ..., ξn−1 as syntactic parts. c Markus Werning – p.9/27 The Reasons for Semantic Compositionality Meaning Functions We assume that the introduction of terms and syntactic operations allows us to disambiguated language with respect to lexical ambiguities caused by homonymy (‘John went from bank to bank’) and syntactic ambiguities (‘John saw the girl with a telescope’). We can now construe meaning as a surjective function from the set of grammatical terms into a set of meanings: µ : GT (G) → M. A grammatical term of the language is called µ-meaningful if the term is in the domain of the meaning function µ. Any µ-meaningful terms p and q of the language are said to be µ-synonymous (in symbols: p ≡µ q) if and only if they have the same µ-meanings. c Markus Werning – p.10/27 The Reasons for Semantic Compositionality Semantic Compositionality — the Formal Notion Let µ be a meaning function for a language with grammar G. Then µ is called compositional if and only if, for every syntactic operation α of the language, there is a function µα (called a semantic operation) such that, for every µ-meaningful term α(t0 , ..., tn−1 ) whose syntactic parts t0 , ..., tn−1 are also µ-meaningful, the following equation holds: µ(α(t0 , ..., tn−1 )) = µα (µ(t0 ), ..., µ(tn−1 )). A language is called compositional just in case it has a compositional meaning function. c Markus Werning – p.11/27 The Reasons for Semantic Compositionality The Equivalence of the Colloquial and the Formal Notion of Compositionality The formal notion of compositionality turns out to be equivalent to our introductory formulation according to which the meaning of an expression is determined by the meanings of its parts. We only have to translate ‘... is determined by :::’ as ‘... is the value of some semantic operation with ::: as arguments’, to take ‘expression’ to mean term, and to specify ‘part’ as meaning syntactic part. c Markus Werning – p.12/27 The Reasons for Semantic Compositionality Is Productivity a Reason for Compositionality? A language is called productive just in case the grammar of the language comprises no more than finitely many atomic terms, the grammar of the language contains syntactic rule such that the generation of potentially infinitely many terms is allowed, the meaning function of the language is computable. Many authors claim that only compositional languages are productive. c Markus Werning – p.13/27 The Reasons for Semantic Compositionality Counterexample: Holophrastic Quotation Assume the grammar of a language justify the the following syntactic operation of quotation: q:T →T s 7→ ‘s’ such that µ(q(s)) = s. The inclusion of this operation in a language warrants that the language is productive. This account of quotation is called holophrastic because it takes well-formed phrases as unanalyzed wholes and sets them in quotation marks. Holophrastic quotation violates compositionality, provided the language to be considered contains synonyms. c Markus Werning – p.14/27 The Reasons for Semantic Compositionality Proof Assumption: µ(pPaul and Peter are brothersq) = µ(pPeter and Paul are brothersq). But: pPaul and Peter are brothersq (1) 6= pPeter and Paul are brothersq. (2) = µ(q(pPaul and Peter are brothersq)) (3) = pPaul and Peter are brothersq. (4) = µ(q(pPeter and Paul are brothersq)) (5) = pPeter and Paul are brothersq. (6) In accordance with holophrastic quotation we get: µ(p‘Paul and Peter are brothers’q) µ(p‘Peter and Paul are brothers’q) From (2), (4), (6) we get: µ(q(pPaul and Peter are brothersq)) 6= µ(q(pPeter and Paul are brothersq)). (7) c Markus Werning – p.15/27 The Reasons for Semantic Compositionality continued Because of compositionality, there is a function µ q , such that µ(q(pPaul and Peter are brothersq)) = µq (µ(pPeter and Paul are brothersq)). (8) Substitution of identicals according to (1): µ(q(pPaul and Peter are brothersq)) = µq (µ(pPaul and Peter are brothersq)). (9) Another application of compositionality: µ(q(pPaul and Peter are brothersq)) = µ(q(pPeter and Paul are brothersq)). (10) This contradicts (7). c Markus Werning – p.16/27 The Reasons for Semantic Compositionality Systematicity Many authors cite the systematic correlation of linguistic capacities and mental capacities as a reason for semantic compositionality. Minds must have the capacity to compose contents/meanings, so it is argued. Otherwise, they would not show a systematic correlation between representational/linguisitc capacities: If a mind is capable of certain intentional states in a certain intentional mode, it most probably is also capable of other intentional states with related contents in the same mode. Mutatis mutandis: If a mind is capable of understanding certain linguistic expressions, it most probable is also capable of understanding other expressions with related meanings. c Markus Werning – p.17/27 The Reasons for Semantic Compositionality Examples of Correlated Capacities the capacity to think that the dog is chasing the cat and the capacity to think that he cat is chasing the dog. the capacity to think that if the cat runs, then the dog will and the capacity to think that if the dog runs then the cat will; the capacity to see a visual stimulus as a square above a triangle and the capacity to see a visual stimulus as a triangle above a square; the capacity to prefer a green coat to a red shirt and the capacity to prefer a red coat to a green shirt. the capacity to understand the sentence ‘John loves Mary’ and the capacity to understand the sentence ‘Mary loves John’. c Markus Werning – p.18/27 The Reasons for Semantic Compositionality Does Systematicity Presuppose Compositionality? The systematic correlation of both contents and meanings seem, indeed, to imply more than mere syntactic recombination on the level of natural language or a language of thought. The capacity to think that a child with a red coat is distracted by an old herring is not correlated with the capacity to think that a child with an old coat is distracted by a red herring. The thoughts ought to be correlated, though, if the fact that one is a syntactic re-combination of the other was sufficient for systematic correlation. Notice that both thoughts are syntactically combined from exactly the same primitives by exactly the same operations. One may, however, well have the capacity to think of red coats and old herrings even though one lacks the capacity to think of red herrings. The two thoughts fail to be correlated because r ed herring is idiomatic and — as a consequence — semantic compositionality is violated. c Markus Werning – p.19/27 The Reasons for Semantic Compositionality Instability of Systematicity Systematicity seem to be too instable a phenomenon as to justify a general need for compositionality. Consider the following examples: (a) The man watches the show. (b) (?) The show watches the man. (c) The woman feels hatred. (d) (?) Hatred feels the woman. c Markus Werning – p.20/27 The Reasons for Semantic Compositionality Interchangeability of Synonyms salva significatione The principle of interchangeability of synonyms salva significatione says that the substitution of synonyms for expressions in any linguistic context leaves unchanged the meaning of the context. The principle can be regarded as the meaning (or intensional) counterpart of the principle of extensionality, also called the principle of interchangeability of co-extensionals salva veritate. It claims that the substitution of co-extensional expressions for each other leaves unchanged the truth value of the embedding linguistic context. While the principle of extensionality is violated in intensional contexts — contexts like ‘It is necessary that ...’, ‘S believes that ...’ — the principle of interchangeability salva significatione even pertains to those cases. c Markus Werning – p.21/27 The Reasons for Semantic Compositionality The Primary Reason for Compositionality The following theorem proves the equivalence between the principle of semantic compositionality and the principle of interchangeability of synonyms salva significatione. Here, meaning functions are called substitutional salva significatione if they abide by the principle that the substitution of synonyms for expressions in any linguistic context leaves unchanged the meaning of the context. c Markus Werning – p.22/27 The Reasons for Semantic Compositionality Theorem Let µ be a meaning function for a language with grammar G, and suppose that every syntactic part of a µ-meaningful term is µ-meaningful. Then the following are equivalent: (a) µ is compositional. (b) µ is substitutional salva significatione, i. e., if s is a term and p0 , ..., pn−1 , q0 , ..., qn−1 are grammatical terms such that s(p0 , ..., pn−1 |ξ0 , ..., ξn−1 ) and s(q0 , ..., qn−1 |ξ0 , ..., ξn−1 ) are both µ-meaningful and, for all m < n, p m ≡ µ qm , then s(p0 , ..., pn−1 |ξ0 , ..., ξn−1 ) ≡µ s(q0 , ..., qn−1 |ξ0 , ..., ξn−1 ). c Markus Werning – p.23/27 The Reasons for Semantic Compositionality Proof (a) ⇒ (b). Assuming (a), we prove (b) by induction on the complexity of s. In case n = 0, s is a µ-meaningful term and the conclusion s ≡µ s is trivial. We now consider the case where s is the term α(t0 , ..., tm−1 ). In this case we get s(p0 , ..., pn−1 |ξ0 , ..., ξn−1 ) by substituting the terms pi for ξi , with 0 ≤ i < n, in all syntactic parts t0 , ..., tm−1 of the term s. We analogously proceed with s(q0 , ..., qn−1 |ξ0 , ..., ξn−1 ) and thus have: s(p0 , ..., pn−1 |ξ0 , ..., ξn−1 ) = α(t0 (p0 , ..., pn−1 |ξ0 , ..., ξn−1 ), ..., tm−1 (p0 , ..., pn−1 |ξ0 , ..., ξn−1 ), and s(q0 , ..., qn−1 |ξ0 , ..., ξn−1 ) = α(t0 (q0 , ..., qn−1 |ξ0 , ..., ξn−1 ), ..., tm−1 (q0 , ..., qn−1 |ξ0 , ..., ξn−1 ). Since s(p0 , ...|ξ0 , ...) and s(q0 , ...|ξ0 , ...) are assumed to be µ-meaningful, their syntactic parts ti (p0 , ...|ξ0 , ...) and ti (q0 , ...|ξ0 , ...), respectively, are also µ-meaningful. c Markus Werning – p.24/27 The Reasons for Semantic Compositionality continued By induction hypotheses we may, therefore, presume that ti (p0 , ...|ξ0 , ...) ≡µ ti (q0 , ...|ξ0 , ...). According to (a) the µ-meanings of s(p0 , ...|ξ0 , ...) and s(q0 , ...|ξ0 , ...), respectively, are a function of the meanings of their syntactic parts. Thus, the identity of the µ-meanings of the parts of both terms implies the identity of the µ-meanings of both terms. c Markus Werning – p.25/27 The Reasons for Semantic Compositionality continued (b) ⇒ (a). (a) follows at once from the special case of (b) where s has the form α(ξ0 , ..., ξn−1 ). For, in that case (b) just claims the functionality of the relation µα = {h(µ(ξ0 ), ..., µ(ξn−1 )), µ(α(ξ0 , ..., ξn−1 ))i|(ξ0 , ..., ξn−1 ) ∈ dom(α)}. c Markus Werning – p.26/27 The Reasons for Semantic Compositionality Conclusion Whereas productivity and systematicity are unlikely to provide sufficient justification for semantic compositionality, the principle of interchangeability of synonyms salva significatione may well do. c Markus Werning – p.27/27 The Reasons for Semantic Compositionality
© Copyright 2026 Paperzz