Whither Compositionality?
One possible reason that the theory of semantics which the model recommends is largely not noticed is that it presupposes the falsity of the widely held principle of compositionality. My purpose here is very briefly to explain and defend this rejection.
Natural languages familiarly have finite vocabularies whose words speakers string together to compose an effectively infinite wealth of sentences. The semantic values of words are meanings. The semantic values of sentences are the thoughts or propositions which sentences express, and their truth-values.
The principle of compositionality (PoC) is that the semantic values of composite linguistic structures (phrases and sentences) are a function of -are wholly determined by- the semantic values of their parts and how they are put together.
The short statement of the thought is this: philosophers think there needs to be a systematic explanation either of how people are able, as a matter of scientific fact, to understand arbitrary new sentences, or of how they are able to be justified in understanding arbitrary new sentences as they do, as a matter of epistemological fact. The thought about scientific explanation is incorrect: there does need to be an explanation in the terms of natural science, but these do not include semantic terms. Natural science does not need there to be a compositional semantics. The thought about epistemological explanation is also mistaken, being grounded in a pervasive but incorrect conception of justification.
Why People Believe the Principle of Compositionality
Superficially, PoC is mere common sense. You know what 'black' means and you know what 'dog' means; your understanding of the phrase 'black dog' is made possible by and determined by those two bits of knowledge together with your knowledge of English grammar. You know what 'chartreuse dog' means, too, though you may never have heard the phrase. What could be more obvious?
It's altogether natural to think that this bit of common sense can be parlayed into a non-trivial, informative theory of how we understand language; a scientific or quasi-scientific theory, if you like. Indeed, it has been thought that something like this has to be possible, in order to explain the effectively infinite ability of speakers to understand things said, given our manifest intellectual finitude.
It's this latter understanding of PoC, in its scientific or quasi-scientific guise, that I reject. My goal here is to remind us why it need not be true. The case for its falsity is made by the model
.
Why Question it? Take 1: Science.
The initial thought, then, is that in striving to come up with a compositional semantic theory we are trying to give a scientific explanation of people's judgements about what words say, and whether what they say is true.
The first point to get clear on is that a compositional semantic theory, understood this way, takes the form of a set of rules which govern speakers. It might tell you, for example, that if 'the dog' refers to a particular thing (dog) and 'is black' predicates the property of being black of a thing and the particular dog in question has the property of being black, then a speaker may infer that "The dog is black." is true.
In this, PoC is different than, say, a theory of the movement of the planets, whose form is a set of natural laws to which planets conform. A rule specifies what may or may not be done, but it does not imply that a thing bound by it will never fail to conform to it. We all occasionally fail to follow rules. When we do, we make a mistake. Our failure does not imply we are not bound by the rule, just that we have lapsed.
A natural law, on the other hand, does imply that a thing bound by it will always behave as it specifies. If a thing fails to conform to a law, then either the law does not apply to the thing or the law is false
.
So, rules and natural laws are quite different things. Natural science is formulated in terms of the latter, PoC, the former. If the thought is that somehow a compositional theory can be a scientific theory in its own right, this seems to be a confusion. That something is bound by a rule is not empirically testable, as natural science requires its claims to be.
There is, though, a more defensible way of understanding the project. We may suppose -a theory may stipulate- that a person is a follower of a rule only if they follow it correctly some quite large majority of the time - that they not be vulnerable to a proportionately large number of mistakes. This being the case, PoC's claim can be taken to be simply that there must be in the brain some set of mechanisms relevant to the processing of sentences and scrutable by natural science which are closely isomorphic to the regularities implied by a compositional semantics, when its rules are correctly followed. Formulating the rules of a compositional semantics would then potentially be scientifically useful because it might specify regularities natural science can test.
Does this redeem PoC's scientific credentials? Let's suppose this is true; that there is in fact some quite complex set of brain structures governing in a precise way behaviours corresponding to judgements about the truth of things said, with the kind of consistency and interrelations that impress the PoC advocate.
What we should care about in this case is not in the first instance the rules but rather the physical structures they mirror - these latter are what, properly speaking, explain people's behaviour, in the sense of 'explain' relevant here. Indeed, if there are any systematic deviations from what, so to speak, the logic of semantics proposes and what we in fact say, our interest should always be in the discovery of the structures rather than the rules themselves, if this is how we are conceiving the project. From a scientific standpoint, the rules are just a heuristic expedient. Obsessing about the fine details of the rules would be justified only if there are good reasons to think those fine details are reflected in our cerebral apparatus. If PoC is to be defensible on the current grounds, it should be acknowledged to lose all scientific value as soon as and wherever its rules outstrip or diverge from what's physically between our ears. This is just a matter of scientific clarity.
There is an exact analogy here to our thinking about the explanation of organisms' evolved substructures. Because evolution favours organisms with effective mechanisms for homeostasis, it's often convenient and productive to think of those mechanisms as having a function or purpose in the maintenance of the life of the organism, and to seek to understand the presence of their respective sub-components in terms of their contributions to the fulfillment of the purpose. And this is of course fine. But to fail to see the necessary disposability of these explanations is to fail to grasp the whole point of Darwin's great achievement, which was to extirpate teleological explanation from our understanding of the biological world. Claims made in the vocabulary of functions and purposes are always expedients which are strictly-speaking false.
It's worth footnoting in this context that the stipulation we made, that there exists some set of semantic-judgement-governing brain-structures which the divination of a compositional semantics will illuminate, looks pretty unlikely. Even if we allow that the recent crop of large language models do not exactly replicate how our brains work, they very plausibly give us some insight. If they do, then the remoteness of the manner of their working from anything envisioned by compositional semantics effectively kills-off the line of thought just essayed.
The important upshot of the foregoing is that the felt need to contrive a compositional semantics does not issue from the requirements of natural science. Science is in order as it is, without any such theory being contrived.
Why Doubt it? Take 2, Social Anthropology.
There is another, quite different way of understanding the aspirations and claims of PoC. Consider for a moment the rules of logic and the different branches of mathematics. In these cases we have clear examples of sets of rules -clearly not physical laws- the careful, systematic and rigorous study of which for their own sakes needs no defending. Specifically, we surely don't study them in the hope that doing so will be paid-off with insight into applicable brain structures, though of course they may have other practical values. The knowledge got from such study is systematic (as systematic as it comes), but it is not natural science. The hopeful thought here is that the rules of compositional semantics are comparable to those of logic and mathematics. We are looking to uncover a system of rules, but in the first instance only for its intrinsic interest.
According to this understanding, there is plainly no imperative that a fully adequate compositional semantics be discoverable. Trying to find one may be interesting and worthy as an exercise comparable and adjacent to philology, of purely cultural interest. Here again, there need be no presumption of the undertaking's eventual success. No broader undertaking will be undermined if there turns out not to be any tidy, unifying system of rules to be uncovered.
Why Doubt it? Take 3, Reason.
A final way of understanding compositional semantics is that it follows from the contention that it is only if the rules of semantics can be resolved to a mathematically precise structure, that we can make sense of speakers as being fully justified in the semantic determinations they make. The goal of semantics, viewed this way, is at once to show to be generally rational -justified or reasonable- the semantic judgements speakers characteristically make, and to regiment how we ought to make such judgements. Considering some complex, difficult to understand sentence, we might find that speakers differ in their judgements about it, and may want to figure out what would make one person right and another wrong. The activity of deducing specific rules which would show to be correct one or the other person would be the paradigm of semantic theorizing, according to this view.
The philosophical thought which drives this understanding and which intersects with my line of questioning, is that a compositional semantics is essential to our understanding of ourselves as rational beings. Without such a theory, peoples' judgements about the truth of sentences will have to be seen ultimately to be arbitrary and indefensible. Anyone committed to the idea that we are at least some of the time justified in our semantic judgements -as I, for one, am- may think themselves committed to PoC for this reason. This being said, if this view is correct, then the necessity of the success of the project of compositional semantics trades on the necessity of the underlying conception of justification. Commitment to PoC is a consequence of commitment to traditional epistemology.
The response to this is that the model affords us a quite different conception of justification than the traditional one. It is a conception which renounces the picture pervading so much of philosophy, of a universal, systematic set of rules. What it provides in its stead is a conception where justification is bound to context and 'comes to an end' according to the requirements of conversation and situation. It's important to emphasize in this context that the model's alternative understanding of justification and norms in general isn't merely tacked-on with properties stipulated to fulfill the brief. It is, rather, a direct consequence of the simple premises of the model. This all being the case, the felt need for there to be a fully rigorous PoC fails on this understanding of its point, too.