Building on a treasure trove of previous work, Stephen Laurence and Eric Margolis have produced a richly detailed and well-motivated resurrection of concept nativism and the rationalism–empiricism debate. Their enormous book is segmented into four almost self-standing parts. I’ll discuss some of the merits of each in the order that they appear, before closing with some general remarks about the reach and promise of the book as a whole.

Given that the book argues for a particular version of concept nativism, it may come as a surprise that concepts and their purported nativism are not actually addressed in much detail until about a third of the way into the book. But the long setup in part 1 is useful and important. For nearly 200 pages, Laurence and Margolis argue that too many theorists have construed the debate between rationalism and empiricism as something either silly or settled (in favour of empiricism). In particular, they argue that the debate fails to reduce to any overly simplistic genetics-versus-environment dichotomy, or any other outdated variation on the nature–nurture debate. At the same time, they hold that the debate is not resolved by any straightforwardly empiricist form of interactionism, which they regard instead as an uncontroversial truism. Carefully rejecting and debunking many versions of this general attitude, Laurence and Margolis instead argue that we should reframe the debate in terms of our psychological acquisition base, from which all further psychological traits are acquired.

According to Laurence and Margolis, empiricists posit a sparse base with very few starting elements, perhaps only some low-level attentional biases, some basic sensory-motor representations, and some domain-general learning mechanisms. Experience must provide the rest. Rationalists, by contrast, posit a rich base, stocked with abstract representational structures, along with several domain-specific learning mechanisms to acquire more. Experience is still involved, but the mind comes much more prepared than empiricists allow. So, on this construal, rationalists and empiricists disagree on the elements of our psychological ‘starter pack’, but they agree on its role: the base elements enter into psychological processes that draw on experiential inputs to produce further psychological traits, but these elements are not themselves products of such processes. They are acquired in some other, non-psychological manner. In this way, then, Laurence and Margolis provide a construal of the debate that renders it substantive, unsettled, and ripe for empirical testing.

One of the most important contributions in this first part is the claim that concept nativism is compatible with learning concepts. According to Laurence and Margolis, concept nativism need not be committed to the existence of any innate concepts. According to their construal of rationalism, concept nativism is compatible with concept learning, as long as the concepts are acquired via a characteristically rationalist learning mechanism. This follows from the fact that the acquisition base might include non-psychologically acquired mechanisms as well as non-psychologically acquired representational structures. In particular, the base might include domain-specific learning mechanisms, which enable the acquisition of representations in only one particular content domain (or several closely related domains). Thus, if our base is stocked with a learning mechanism specifically geared towards the acquisition of spatial representations, for example, then the resulting spatial concepts would be learned rather than innate, and yet the account of their origin would still be rationalist.

Critics and sceptics might worry that Laurence and Margolis’s re-construal of the debate makes it much too easy to be a rationalist: one needn’t even be committed to any innate representations! Indeed, in the course of setting up their re-construal, Laurence and Margolis even develop a rather interesting account of abstraction that is compatible with their brand of rationalism. And yet, the re-construal is well motivated and extraordinarily detailed. It offers both rationalists and empiricists of all sorts a well-developed framework in which to build and dispute specific proposals about the psychologically basic elements from which all further psychological traits flow. And despite the focus on the acquisition of concepts among humans, the framework easily lends itself to theorizing the acquisition of psychological traits of all sorts among a great variety of cognizing creatures. In this respect alone, the book offers an extraordinarily rich launching point for theorizing mentality at large.

According to Laurence and Margolis, a large variety of human concepts across a large variety of content domains are most plausibly acquired by domain-specific learning mechanisms that trace back to our psychological acquisition base, and some are even among the representational structures included in the base. The focus in part 2 is thus on detailing the many fascinating empirical findings that appear best explained by their version of concept nativism.

Laurence and Margolis distinguish and illustrate seven versions of their argument to the best explanation, arguing that concept nativism is best positioned to explain many remarkable findings concerning the early abilities of infants and toddlers, analogous abilities in a wide variety of animals, cross-cultural findings (including cross-cultural variations), neurological findings (including cases of moderate plasticity), apparent ease of learning and corresponding difficulty in unlearning, the need for initial representational access in order to learn anything further about a domain, and various unexpected ‘quirks’ that cannot be explained within empiricist frameworks.

In each case, the data are fascinating and highly suggestive. Taken as a whole, the data are also impressively broad. Over the course of almost 200 pages, Laurence and Margolis present in rich and highly accessible detail many different studies and findings that concern various abstract content domains (including logic, quantity, geometry, causality, modality, purpose, distance, and time), as well as some natural domains (including food, animals, biological motion, danger, disgust, and death), and several more human and social contents (including faces, beliefs, emotions, language, and social categorization).

One of the major strengths of part 2 is the extent to which the seven arguments overlap and reinforce one another. Consider, for example, just one of the above examples: the ability to attribute beliefs to others. Over the course of their discussion, Laurence and Margolis show that infants as young as six-months old can pass non-traditional false-belief tasks, which minimize the processing burdens of more traditional tasks; that the ability to mentalize is present in many different cultures, including those non-WEIRD (western, educated, industrialized, rich, and democratic) cultures that specifically withhold attributing mental states to others, such as the Yasawan in Melanesia and the South Pacific; that the ability is not at all impaired among people with congenital blindness, who lack visual access to the facial and bodily expressions that serve as significant perceptual cues for the mental states of others; and yet the ability appears absent, or significantly impaired, in many people diagnosed with some versions of autism spectrum disorder. Most suggestive is the fact that the ability to mentalize appears to be all but impossible to acquire without some initial representational ‘foothold’. Without some pre-existing sense of mentality, it’s hard to see how an empiricist learner could possibly bridge the gap between merely perceiving the behaviour of others and attributing imperceptible causally potent inner states to them. This one example thus illustrates aspects of the arguments from early development, cultural universality, neural wiring, and initial representational access.

Mental states are just one of many content domains that rationalists and empiricists can disagree on. The real contribution in this part is thus the articulation of the seven argument forms. In each case, the argument is well illustrated by very suggestive data, but each might also be enriched and strengthened (or undermined) by future empirical work. Laurence and Margolis have thus provided empirical researchers an invaluable roadmap for future studies.

In order to bolster their rationalist account of concepts and their acquisition, Laurence and Margolis turn in part 3 to discussing and then rejecting several prominent empiricist accounts of concept learning. Laurence and Margolis mostly focus on the same content domains for which they have provided a rationalist explanation in part 2, but offer a few new cases to further enrich the empirical detail of their arguments. The discussion of socio-moral cognition, in the context of neo-associationism, and of physical formidability, in the context of embodied cognition, are especially intriguing. Once again, the data are presented in rich accessible detail, and the conclusions drawn against empiricism are well motivated.

In some cases, Laurence and Margolis simply argue that the target empiricist framework has misconstrued the rationalist position, or misinterpreted the relevant data. In other cases, however, they argue that the supposedly empiricist framework is quite compatible with rationalism. In particular, they argue that both neural network models and embodied cognition frameworks, long associated with empiricist accounts, can be incorporated into rationalist theories such as their own. Indeed, they maintain that the accounts become more plausible, and more aligned with the empirical work that support them, once stripped of their empiricist presuppositions. Thus, neural network models are shown to be more amenable to the data on human cognition once we allow the initial state of a network to include richer initial representational inputs or connection weights that pre-dispose the model to certain outputs over others (or both). And theories that centre embodied cognition are shown to be quite plausible once we allow that the tight interplay between body and environment is sometimes better understood as a biological adaptation, rather than something that is experientially acquired through domain-general learning.

Despite their negative conclusions, Laurence and Margolis provide an excellent reconstruction of prominent empiricist accounts and what they have to contribute to our understanding of concept learning. Once again, both empiricists and rationalists of all sorts have much to gain by engaging with the careful discussion of the empirical details and the theoretical lessons that Laurence and Margolis draw.

Having spent almost 500 pages showing that many concepts can indeed be learned, albeit via rationalist learning mechanisms, Laurence and Margolis close the book in part 4 by detailing their objections to Fodor’s lifelong rejection of concept learning and the consequent ‘radical’ version of concept nativism, using this part to also articulate the true extent of their comparatively moderate account.

They begin by reconstructing the subtleties of how Fodor’s position changed over the course of his writings on the subject, indicating several weaknesses in his reasoning along the way. Here, they focus especially on his claims concerning hypothesis testing, showing that it is neither impossibly circular nor the only way a concept can be learned.

Most important for concept theorists is their discussion of the ‘acquisition by composition’ model of concept learning, according to which complex concepts can be learned only by a composition of simpler concepts. On pain of regress, the model implies the existence of some set of basic concepts, simple and innate, from which all other concepts are composed. Against this model, Laurence and Margolis detail several alternative modes of concept learning, often relying on the idea of a concept ‘template’: a schematic representational structure that can be saturated by experiential learning. Thus, an animal template might be saturated by perceptual encounters with animals in the local environment, allowing Australians to learn concepts for koalas and kangaroos, and Southern Africans to learn concepts for klipspringers and kudus, without composing them from simpler innate concepts.

Most notable, of course, is the fact that such perceptual learnings, particular to experiences in local environments, are rationalist, since the learning mechanisms involve innate representational structures specifically geared towards the acquisition of concepts in a particular domain. Once again, then, sceptics might worry that Laurence and Margolis have stretched the notion of ‘rationalism’ beyond recognition.

This impression is bolstered by Laurence and Margolis’s further claim in this part that virtually every concept will be at least minimally rationalist on their account. As an illustration, they note the fact that if our concept of ‘object’ is innate—or acquired by some rationalist mechanism, as seems quite likely—and if it is part of the acquisition of all further object concepts, then all those latter concepts would trace back, in part, to the concept of ‘object’. And this means that the acquisition account for each of these concepts must be rationalist too, at least to this extent.

Of course, in very many cases, Laurence and Margolis hold that the acquisition story is more robustly rationalist than this, but it does create the impression that in making room for a more moderate rationalism, distinct from Fodor’s radical version, Laurence and Margolis have ruled out correspondingly moderate forms of empiricism. Then again, perhaps such ‘minimal rationalism’ is compatible with a moderate concept empiricism. In this case, their re-construal offers concept theorists of every allegiance a way to explore more hybrid, in-between accounts of concept learning, incorporating elements from both characteristically empiricist and characteristically rationalist approaches, in whatever combination best captures the current data.

Taken as a whole, the book is an extraordinary achievement. It is accessible enough to introduce novices to many important aspects of the overall debate, with wonderfully suggestive examples, useful historical overviews, and generous reconstructions of many different theoretical approaches. At the same time, Laurence and Margolis have provided a robust starting point for further theoretical and empirical work, paving the way for experts across philosophy and the cognitive sciences to further deepen our understanding of the origins of our conceptual capacities. Despite being rooted in a centuries-old debate, there is much more work to be done.

Alnica Visser
Texas Christian University
alnica.visser@gmail.com