Logical consequence (also entailment) is one of the most fundamental concepts in logic. It is the relationship between statements that holds true when one logically "follows from" one or more others. Valid logical arguments are ones in which the conclusions follow from its premises, and its conclusions are consequences of its premises. The philosophical analysis of logical consequence involves asking, 'in what sense does a conclusion follow from its premises?' and 'what does it mean for a conclusion to be a consequence of premises?'[1] All of philosophical logic can be thought of as providing accounts of the nature of logical consequence, as well as logical truth.[2]
Logical consequence is taken to be both necessary and formal with examples explicated using models and proofs.[3] A sentence is said to be a logical consequence of a set of sentences, for a given language, if and only if, using logic alone (i.e. without regard to any interpretations of the sentences) the sentence must be true if every sentence in the set were to be true.[4]
Logicians make precise accounts of logical consequence with respect to a given language by constructing a deductive system for , or by formalizing the intended semantics for . Alfred Tarski highlighted three salient features for which any adequate characterization of logical consequence needs to account: 1) that the logical consequence relation relies on the logical form of the sentences involved, 2) that the relation is a priori, i.e. it can be determined whether or not it holds without regard to sense experience, and 3) that the relation has a modal component.[5]
Main article: Logical form |
Given Γ is a set of one or more declarative sentences.
Since
and
it follows that
It is therefore of the first importance to clarify the term logical form and explain how the logically relevant form(s)[6] of a sentence can be established.
The logical form of sentences can be revealed by means of a formal language enabling the following definition of entailment. Roughly, if S1 and S2 are interpretations of two sentences θ and ψ in a formal language of classical logic, then S1 entails S2 if and only if not (θ and not ψ) is true under all interpretations.
More precisely, if Γ is a set of one or more sentences and S1 is the conjunction of the elements of Γ and S2 is a sentence, Γ entails S2 if and only if not (S1 and not-S2) is a logical truth. S2 is called the 'logical consequent' of Γ. S1 is said to 'logically imply' S2.
Not (S1 and not-S2) is a logical truth if θ and Ψ are closed well-formed formulae (often denoted 'wff'), wffs (sentences) in a formal language L in classical logic, and I is an interpretation of L, and θ is true under I if and only if S1 and Ψ is true under I if and only if S2, and not (θ and not Ψ) is logically valid.
A closed wff Φ in L is 'logically valid' if and only if Φ is true under all interpretations of L. Hence
Thus if Γ = {"Roses are red", "Violets are blue"}, S1="Roses are red and Violets are blue" and S2 = "Violets are Blue" then Γ entails S2 because not(S1 and not-S2), "It is not the case that roses are red and violets are blue and violets are not blue" is a logical truth.
Not(S1 and not-S2) is a logical truth because there are two closed wffs, P&Q and Q in a formal language L in classical logic and there is an interpretation I of L, and P&Q is true under I if and only if roses are red and violets are blue, and Q is true under I if and only if violets are blue, and ¬((P&Q)&¬Q) is logically valid. ¬((P&Q)&¬Q) is logically valid because it is true under all interpretations of L (note that ¬ means not).
It will be noted that, on these definitions, if (i) S1 is inconsistent (self-contradictory) or (ii) not-S2 is inconsistent (self-contradictory) then (S1 and not-S2) is inconsistent (not consistent) and hence S1 entails S2.
It is of considerable interest to be able to prove that Γ entails S2 and hence that Γ/S2 is a valid argument. Ideally, entailment and deduction would be extensionally equivalent. However, this is not always the case. In such a case, it is useful to break the equivalence down into its two parts:
A deductive system S is complete for a language L if and only if implies : that is, if all valid arguments are deducible (or provable), where denotes the deducibility relation for the system S. NB means that X is a semantic consequence of A in the language L, and means that X is provable from A in the system S.
A deductive system S is sound for a language L if and only if implies : that is, if no invalid arguments are provable.
Many introductory textbooks (e.g. Mendelson's "Introduction to Mathematical Logic") that introduce first-order logic, include a complete and sound inference system for the first-order logic. In contrast, second-order logic — which allows quantification over predicates — does not have a complete and sound inference system with respect to a full Henkin (or standard) semantics.
Since
a proof that not (θ and not Ψ) is logically valid would be a proof that Γ entails S2.
It can be easily demonstrated, for example by means of a truth-table, that ¬((P & Q) & ¬Q) is a tautology and hence true under all interpretations and hence logically valid. Moreover, if T is a consistent theory in L and ¬(θ ∧ ¬Ψ) is a theorem in T (written ⊢T¬(θ ∧ ¬Ψ)) then ¬(θ ∧ ¬Ψ) is logically valid and, consequently, all interpretations of ¬(θ ∧ ¬Ψ) are logical truths, including not(S1 and not-S2). Hence Γ entails S2 if ⊢T¬(θ ∧ ¬Ψ) and T is consistent.
Entailment is one of a number of inter-related terms of logical appraisal. Its relationship to other such terms includes the following see e.g. Strawson (1952)[7] Section 13, 'Entailment and Inconsistency', pp 19 et seq) where S1 and S2 are sentences, or S1 is the conjunction of all the sentences in some set of sentences Γ, S1 entails S2 if and only if:
A formula A is a syntactic consequence[8][9][10][11] within some formal system FS of a set Г of formulas if there is a formal proof in FS of A from the set Г.
Syntactic consequence does not depend on any interpretation of the formal system.[12]
A formula A is a semantic consequence of a set of statements Г
if and only if no interpretation makes all members of Г true and A false.[13] Or, in other words, the set of the interpretations that make all members of Г true is a subset of the set of the interpretations that make A true.
The difference between material implication and entailment is that they apply in different contexts. The first is a statement of logic, the second of metalogic. If p and q are two sentences then the difference between "p implies q" and "p is a proof of q" is that the first is a statement within formal logic, the second is a statement about it. Entailment is a concept of proof theory, whereas material implication is the mechanics of a proof.[14]
Entailment is one form but not the only form of inference. Inductive reasoning is another. Scientific method involves inferences that are not solely entailment. Entailment does not encompass non-monotonic reasoning or defeasible reasoning. See also
Modal accounts of logical consequence are variations on the following basic idea:
Alternatively (and, most would say, equivalently):
Such accounts are called "modal" because they appeal to the modal notions of necessity and (im)possibility. 'It is necessary that' is often cashed out as a universal quantifier over possible worlds, so that the accounts above translate as:
Consider the modal account in terms of the argument given as an example above:
The conclusion is a logical consequence of the premises because we can't imagine a possible world where (a) all frogs are green; (b) Kermit is a frog; and (c) Kermit is not green.
Modal-formal accounts of logical consequence combine the modal and formal accounts above, yielding variations on the following basic idea:
Most[who?] logicians would probably agree that logical consequence, as we intuitively understand it, has both a modal and a formal aspect, and that some version of the modal/formal account is therefore closest to being correct.
The accounts considered above are all "truth-preservational," in that they all assume that the characteristic feature of a good inference is that it never allows one to move from true premises to an untrue conclusion. As an alternative, some[who?] have proposed "warrant-preservational" accounts, according to which the characteristic feature of a good inference is that it never allows one to move from justifiably assertible premises to a conclusion that is not justifiably assertible. This is (roughly) the account favored by intuitionists such as Michael Dummett.
Main article: Non-monotonic logic |
The accounts discussed above all yield monotonic consequence relations, i.e. ones such that if A is a consequence of Γ, then A is a consequence of any superset of Γ. It is also possible to specify non-monotonic consequence relations to capture the idea that, e.g., 'Tweety can fly' is a logical consequence of
but not of
For more on this, see belief revision#Non-monotonic inference relation.
It is impossible to state rigorously the definition of 'logical implication' as it is understood pretheoretically, but many have taken the Tarskian model-theoretic account as a replacement for it. Some, e.g. Etchemendy 1990, have argued that they do not coincide, not even if they happen to be co-extensional (which Etchemendy believes they are not). See "The Blackwell Guide to Philosophical Logic",[16] for a good introduction.