Stop me if you've heard this one before: The Chomskyan hammer and the Skinnerian nail

Abstract

The target article signal boosts important ongoing work across the cognitive sciences. However, its theoretical claims, generative value, and purported contributions are – where not simply restatements of arguments extensively explored elsewhere – imprecise, noncommittal, and underdeveloped to a degree that makes them difficult to evaluate. The article's apparent force results from engaging with straw rather than steel opponents.

Batman: Then why do you want to kill me?

Joker: Kill you? I don't want to kill you. What would I do without you? Go back to ripping off mob dealers? No, no, no. No, you… you complete me. (Nolan, 2008)

For many a hammer, everything is a nail. For many a philosopher of mind, everything is a chance to rehearse Kant's criticism of Hume, Chomsky's criticism of Skinner, and Fodor's criticism of every empiricist, holist, or, in the pages of BBS (1985), relativist who rubbed him the wrong way. Thus the target article repeats, again and again across different domains, a well-worn argumentative maneuver in psychology and philosophy: “this mental phenomenon you're trying to explain in terms of a simpler process – be it associations, model-free learning, neural nets, or icons – must instead be explained by a more complex process, which performs language-like computations.”

The theoretical payoff of this selective tour through case studies is remarkably modest. For Quilty-Dunn et al. do not deny that the simpler, nonlinguistic processes exist and have real effects. They deny that the nonlinguistic processes explain everything. Repurposing the old joke, we've established what kind of theorist you are – a pluralist – and now we're just haggling over the details. The haggling, in this case, recalls trench warfare. On some fronts (like artificial intelligence), neural nets make stunning advances, even as the language-of-thought hypothesis (LoTH) plants its flag on other patches of cognitive terrain hitherto claimed by nonlinguistic theories. The broader import of this unsystematic assemblage of localized skirmishes is unclear. Is LoTH “the best game in town,” or one game among others, which the mind perhaps plays somewhat more often than some think?

The authors deny that so-called “system 1” (target article, sect. 6, note 11) is purely associative. It's true that associative interpretations of implicit bias continue to hold sway in pop-psych discourse, but hardly anyone paying attention to what the authors rightly call a “near-deluge” (target article, sect. 6.2, para. 5) of research on propositional effects in implicit social cognition continues to defend the extreme associationist views targeted by the authors. Many of us never did (Brownstein & Madva, 2012; Del Pinal, Madva, & Reuter, 2017; Gawronski & Bodenhausen, 2006, pp. 706–707; Madva, 2016, p. 2681, 2019; see also Brownstein, 2018, Chs. 2–3). This is not to say all our predictions panned out, but to question the ease with which other pluralist approaches are pigeonholed into the dreaded empiricist/associationist/behaviorist position in these recurring debates (e.g., Kurdi, Morris, & Cushman, 2022b, p. 3). Indeed, according to Mandelbaum (2022, sect. 8), we represent “a revival of associationist theories in philosophy,” citing a paper that is explicitly orthogonal to the association–proposition debate (Madva & Brownstein, 2018, sect. 6.1; see also Kurdi, Mann, Charlesworth, & Banaji, 2019; Phills, Hahn, & Gawronski, 2020). With apologies to Voltaire, one senses that if modern-day associationists did not exist, modern-day Fodorians would have to reinvent them. With apologies to Taylor Swift, I would very much like to be excluded from this narrative.

In any case, the downfall of pure-associative models has not occasioned the uncontested reign of propositional alternatives. Leading propositional theorists continue to uncover effects more naturally explained by nonpropositional processes, or at least uneasily assimilated into prevailing propositional theories (e.g., Van Dessel, De Houwer, Gast, Roets, & Smith, 2020; see also Byrd, 2021). As a recent meta-analysis by Kurdi, Morehouse, and Dunham (2023, p. 1) explains, no current theory is well-poised to predict and explain the disorienting array of findings, and the time for “existence proof demonstrations” of propositional effects has passed. Yet in lieu of synthesizing the disarray, the target article consists in just such a grab bag of existence proofs, trumpeting all and only recent successes for propositional approaches – while ignoring evidence of their shortcomings and boundary conditions, and deferring long-standing concerns about how LoTs are implemented in the brain and integrated with other processes.

The authors nevertheless advertise LoTH's “unificatory power” (target article, sect. 1, para. 7), specifically its provision of a lingua franca mediating between psychological domains (perception, higher-order thinking, so-called “system 1,” etc.). But if each of these domains involves proprietary LoTs and who knows how many other representational formats, the question still remains how these diverse representational formats interact with each other (within each psychological domain, rather than between domains). If non-LoTs interface with LoTs after all, what explanatory traction is gained by noting how LoTs pop up in lots of distinct psychological locales? And if a thousand other representational formats are abloom across the mind (target article, sect. 2, para. 2), why couldn't some of them mediate between domains, too? What we have here is not unification but proliferation, not explanation but more to explain.

No doubt the authors would cite their six core LoT properties as significant theoretical contributions. But the conceptual and causal interrelations of these properties (which are invoked in seemingly random combinations from one case study to the next) are muddled at best. Do they represent a homeostatic property cluster, as the authors claim, or are they tied more tightly together? The authors stress that representations involving discrete constituents need not be structured like sentences, but they “usually interpret” sentence-like representations “as requiring” discrete constituents (target article, sect. 2, para. 8). They then grant that successive properties on their list necessitate others, for example, predicate–argument structures and logical operators “requiring role-filler independence” (target article, sect. 2, para. 9). To the extent that property B requires property A, it is completely trivial to predict that A will show up wherever B does, and only slightly less trivial to predict that B will appear alongside A above chance. The mere prediction that properties “should tend to cooccur” (target article, sect. 2, para. 12) is weak, vague, and unconstrained, allowing theorists to underscore cooccurrences and ignore (or explain away) noncooccurrences. We “usually require” fewer degrees of freedom from our theoretical frameworks. We are also compelled to ask whether the six properties offer anything substantively novel or illuminating, or simply stick new labels on the analytic entailments contained in the original LoT view.

The target article at times positions itself as a lone voice of logic in an associationist wilderness, fighting the good fight for a nearly forgotten rationalist cause while flanked on all sides by zombie empiricisms that refuse to stay dead. Yet the article's principal value consists in signal-boosting others' important ongoing work. The question, then, is what it would mean to take up the authors' proposals over and above what the exemplary researchers being cited are already doing.