Concepts are widely taken to be among the most basic elements of theories of mind (Margolis and Laurence 2014). They play a ubiquitous explanatory role in psychological theories of cognition, especially in the areas of reasoning. Concepts are frequently appealed to in accounts of judgment, categorisation, and inference, as well as higher order behaviour such as learning and decision-making. This widespread appeal to concepts both within and across disciplines implies a degree of credence in conceptual monism: the thesis that all concepts manifest the same structure. For example, the method of conceptual analysis that is pervasive in philosophy treats concepts as elements of the same semantic kind: the same devices of reasoning that can be applied to morality can be applied to science, language, and other domains.
That said, the precise nature of concepts is a subject of dispute. Various ontological accounts of concepts have cast them as abstract objects, abilities, or mental representations (idem). On the latter view of concept, words in natural languages derive their meaning from subpropositional mental representations (Margolis and Laurence 2000). Not all concepts correspond to simple lexical items in natural languages, but those that do– lexical concepts– occupy the attention of most theories of concepts among philosophers who assume the mental representation view of concepts (idem).
Concepts may be simple or complex depending on whether they possess structure. There are two general accounts of what exactly «structure» consists in (idem):
Furthermore, concepts may also be categorised by mereological hierarchy (idem):
Having laid out the basic distinctions in the concepts literature, we are now in a position to review the four popular theories of concepts (idem).
Definitionism: Concepts are structured mental representations that encode a set of necessary and sufficient conditions for their application (where possible) in sensory or perceptual terms.
Prototype theory: Concepts are structured mental representations that encode the attributes that objects in their extension tend to possess.
Theory theory: Concepts are representations whose structure consists in their relations to other concepts as specified by a mental theory.
Conceptual atomism: Concepts are primitive and lack semantic structure.
Each theory has its own explanatory merits and shortcomings which turn on background commitments and desiderata. I aim to bring some of these commitments to light. My argument will proceed as follows:
Consider P4 my thesis. P1 follows a basic formula of intractability posited in the metaphilosophy literature (Hájek 2016). P3 builds on this formula by introducing a stalemate condition. I will first establish P2 by a brief overview of the state of the concepts discussion. I will then argue for P4 by examining three different considerations which govern whether a containment model or an inferential model is appropriate, and, by extension, which of the four theories best describes any particular concept. Separately and together, each of these bears on the structure of concepts in ways that are irreconcilable with a monistic ontology of concepts.
A theory of concepts should explain our ability to reason not just within domains which may be internally diverse, but also across disparate subject areas. Conceptual monism seems intuitively well suited for this job: universal structure is what allows us to transcend content dissimilarity. But this allure alone doesn’t amount to sufficient reason for accepting conceptual monism; without independent justification, such a move would be ad hoc. Do we have that?
Theories of concepts across the board are each met with considerable challenges. Yet, each of these theories demonstrates merits in qualitatively different aspects. Of course, in principle any of these theories could be amended in order to address their most salient criticisms. The state of things as such can be taken to suggest either that we just haven’t found the right model (or argument for a model), or that no one model can meet all of these desiderata. But if no independent means of adjudicating between different theories of concepts is forthcoming, then there is no reason to suppose a single theory of concepts will be able to explain the range of ways in which concepts are used in reasoning, natural language, and so on (Margolis and Laurence 2014). And indeed, this seems to be the case.
It’s not just that conceptual monism doesn’t seem to be on the horizon– it’s never coming. In what follows, I will examine three determinants of conceptual structure: existing epistemologies and practices, the referents of concepts, and theories of mind. From these arise the general desiderata for a theory of concepts. However, these various requirements, though individually plausible, preclude a unified ontology. In summary, I will argue that:
- A theory of concepts should integrate nicely into our existing epistemologies and practices, but this cannot be done if we’re stuck with a monist ontology of concepts.
- Ontic dissimilarities demand different epistemological treatment which are manifest in conceptual structure.
- Heterodox approaches in theory of mind, particularly concerning informational processing, predict less parsimonious conceptual ontology.
Concepts & us
Existing epistemologies & practices
Let’s start with the obvious: a theory of concepts should nicely integrate into our ideas of science, history, language, and folk psychology. In other words, a theory of concepts should not require us to dramatically revise our existing epistemologies or practices just to avoid unintuitive consequences. This criterion underlies the most damaging objections to the four previously discussed theories:
These are just four areas of human concern where concepts are put to work. A given context of application implies a range of epistemic, pragmatic, and sometimes moral values. In this way, the specific task at hand– whether it be descriptive work or normative work, explanatory work or predictive work– calls for different conceptual structures.
Theoretical vs. operational species concepts
In the natural sciences, a distinction is often drawn between, on one hand, theoretical or ontological concepts, and, on the other hand, operational or epistemological concepts (Mayden 1997; Zachos 2016). While concepts of the former type aim to elucidate the salient constitutive and causal properties of an explanandum, concepts of the latter category serve solely as diagnostic criteria. The attempts by various authors to elucidate a sound species concept in the philosophy of biology illustrates this distinction. Consider the following three examples (Ereshefsky 2010):
- A species is distinguished by the sharing of core genes.
- x and y are the same species IFF x and y exhibit ≥97% sequence similarity in genes for 16S rRNA.
- x and y are the same species IFF x and y exhibit ≥70% reassociation.
(1) stipulates a criterion which is central to being a species. (2) and (3) merely designate properties that can be used to detect members of this category. These properties need not be constitutive or causally central, but may merely be [tightly] correlative. While determining whether any particular concept is theoretical or operational is often controversial, and the two may well be co-constitutive, the distinction does call attention to a meaningful difference in real-world research agendas.
Philosophy of biology has long grappled with its own form of Plato’s problem as regards a theoretical species concept, leading many authors to embrace nominalism and pluralism about species (idem, Doolittle and Olga 2009). This rising zeitgeist is associated with a shift in the conceptual structure of species. Previously, philosophers of biology were [implicitly] interested in delineating a containment model concept of species on which tokening «species» entails tokening, say, the sharing of core genes. However, the move towards nominalism and pluralism involves, in the first place, a renouncement of the possibility for this kind of grounding (at least as far as species is concerned), and in the second place, a sober humility. Pluralism in this sense entails Rortian ironism about the particular concepts one employs– the admission that no one of these operational concepts can be privileged insofar as they actually track species best (Ereshefsky 2010). However, it is not the case that anything goes; there is room for local comparisons. For example, some microbiologists favour 16S rRNA because it is more stable across generations, and more ubiquitous (idem). Others prefer DNA:DNA rehybridisation for delimiting species because it correlates better with phenotypic similarity (idem). The incommensurable advantages and disadvantages of each method foreclose universally privileging of any of these criteria, but nonetheless permit rational selection in different research contexts (idem). In this way, accepting pluralism of operational concepts implies accepting an inferential model of conceptual structure: a biologist may token species, but not necessarily ≥70% DNA-DNA reassociation.
The containment model (and by extension, definitionism) just didn’t work for biology in this case. The lesson that should be applied to the philosophy of concepts more generally is that specific contexts of applications demand different conceptual structures. Therefore, conceptual structures should be judged to be appropriate based on less general use cases rather than by the extent and breadth of domains which seem to exhibit these structures. If we are to retain the notion of concepts– and there are plenty of motivations for doing so– the question should not be «what are most concepts– definitions, prototypes, theories, etc.?» but rather «what is this concept?»
Differences in ontology should entail downstream differences in epistemology. Viz., the nature of the object of inquiry also determines conceptual structure. This is because fundamental ontic differences might be better captured by certain conceptual structures rather than others. It will be seen how this plays out.
One such difference consists in whether an explanandum involves a univariate or a multivariate aetiology. Consider two health conditions: pseudomembranous colitis and major depressive disorder. We can give a fairly straightforward account of the pathophysiology of the former which conforms to the containment model: tokening pseudomembranous colitis requires tokening the pathogen Clostridium difficile. On the other hand, because the pathogenesis of major depressive disorder implicates not only biological but also psychological and social causes (none of which are privileged), we must appeal to an inferential model to explain major depressive disorder. Thus, a psychiatrist might token major depressive disorder without tokening decreased monoamines. The contrast between the two cases stems from the intrinsic difference in complexity between the actual objects of inquiry moreso than the state of knowledge in both areas.
The cost of pretensions
Many concepts in the natural sciences are supposed to track natural kinds, whether they be subatomic particles, elements, or diseases. A natural kind is a category of things that mirrors the structure of the natural world rather than parochial human interests and actions (Bird 2018). In general, the fact that a concept has a referent which is external to itself introduces constraints that favour certain conceptual structures rather than others. That is to say, the metaphysical commitments of these concepts render them vulnerable to certain kinds of objections.
According to the desideratum of conceptual continuity, concept talk should be intelligible across paradigms. This can be established by appeal to causal reference-fixing or through epistemological bridging (theory-neutral observation language, persistent epistemic goal). Concepts that purport to capture NKs incur the burden of ensuring conceptual continuity in order to avoid ontological explosion (e.g. positing a distinct NK for every paradigm).
The problem of ignorance and error finds a tension between individual justification and successful reference. Many individuals seem to be able to use a concept despite not being able to offer a complete and correct account of its meaning (Margolis and Laurence 2000). So the descriptivism implicit within definitionism can’t be the right story of how individuals use concepts. But that doesn’t mean that descriptivist accounts of reference have nothing to offer. Perhaps descriptivism isn’t the right story of how individuals use concepts to refer, but maybe it’s part of the story. The insight of descriptivism as applied to NK concepts (as opposed to concepts whose referents are not NKs, or those that lack external referents altogether) is that in some sense it matters whether the stories they tell are actually true. Because NK concepts purport to describe the natural world, they better do it well. What is «true» is defined by their aforementioned metaphysical commitments, and can be analysed in terms of (1) whether the referent in question actually exists and (2) whether or not the concept accurately captures the causal properties of this entity. A concept that makes more or less accurate predictions about a phenomenon, and which readily inspires feasible operational concepts is likely to persist in future explanatory frameworks. The problem of ignorance and error should not motivate us to foreclose on necessary and sufficient criteria, but to question their supremacy and understand their limits. Just because a tool or heuristic does not apply in all cases does not mean it’s bankrupt. While necessary and sufficient criteria may not be, well, necessary or sufficient, they seem nonetheless important.
So: on one hand, it seems that since most people do successfully use NK concepts in order to refer despite being either under-informed or mistaken, reference-fixing must be minimally causal. On the other hand, it seems important that epistemic agents are privy to the reality of the objects they refer to with their concepts, which is what we have when we are normative about descriptions and when we have conceptual continuity. I don’t know about you, but I would like it both ways. The way that we might be able to have it both ways might be by rejecting the assumption that one type of conceptual structure can buy us both of these things.
Theory of mind
So whether a conceptual structure is appropriate depends not only upon the intended contexts of application but also the nature of the referent itself. One final factor which I argue has plausible bearing on conceptual structure is theory of mind. Cognitive architecture, too, determines conceptual structure. My intention here is not to argue for any one theory of cognition, but to draw attention to different approaches and their implications for conceptual structure. Conceptual monism seems to almost be an a priori commitment of «top-down» research programs in cognitive science– for instance, those that start from semantics. On the other hand, «bottom-up» approaches which place theoretical emphasis on cognitive phylogenesis are less predisposed towards conceptual monism.
To be sure, an account of productivity and systematicity is an appealing desideratum for a theory of concepts, but its necessity as such is undermined by objections to representational theory of mind in psychological and cognitive science (Margolis and Laurence 2014). RTM attempts to understand cognition in terms of the processing of mental representations (Rescorla 2017), but critics worry that its proximity to folk psychological explanations misleads future research (Margolis and Laurence 2014).
One such way in which RTM may be construed as overly simplistic involves its picture of cognition as detached from the immediate embedded environment: the body (Wilson and Foglia 2017). At the very least, corporeal mediation between the brain and the external environment must have some effect on neural informational input (Clark and Webb). Our brains are not just floating in the universe: they are in bodies. A coevolutionary story of the brain and the body requires looking beyond the cranium in trying to understand cognition. Both the external environment and the body plausibly exert causal and constitutive influence on mental activity. Perhaps advocates of RTM might be able to respond to these objections, but it’s hard to see how conceptual structure could be uniform if informational input is modulated in exceedingly complex and nonlinear, continuous and diachronic ways distributed over multiple levels of organisation (idem).
Products of dedicated processing
Moreover, specialised cognitive architecture may further determine conceptual structure. If the mind is to a significant extent comprised of discrete neural circuits tailored to certain types of content (Cosmides and Tooby 2013, Robbins 2017), then it seems plausible that concepts might be structured differently depending on the content. Or, to put it another way, concepts may be structured differently depending on the cognitive apparatuses that create them. These neural circuits could be evolutionarily designed modules (Cosmides and Tooby 2013), or they could arise ontogenetically from a process of proliferation and pruning (Buller and Hardcastle 2006). Either way, there remains reason to believe that conceptual structure is pluralistically overdetermined.
The assumption of conceptual monism has stagnated the philosophy of concepts. A review of the rich extra-theoretical environment that concepts inhabit reveals diverse ways in which conceptual structure is overdetermined: from the diverse tasks that we put them to, to the types of phenomena they are supposed to capture (if any), to our beliefs about the relationship between neural processing and semantics. These influences cannot be accommodated with the notion of a common structure underlying even just most concepts. Future study of concepts should be sensitive to these undercurrents.
As it stands, it seems like no «theory of everything» for reasoning is forthcoming. If conceptual monism is false, how can we hope to explain the generality of human reasoning? To be honest, I don’t know. I would like to know. But conceptual monism doesn’t seem to be the way to get there. One could say we’re back to square one, but I would say we’ve never left.