No biological trait can be simply assumed as a “constraint.”

Viewing the Now-or-Never bottleneck as an evolutionary constraint to which language adapts – C&C's central idea – is unwarranted. In evolutionary theory, biological constraints – as opposed to constraints imposed by physics and chemistry, which are not subject to biological evolution – cannot simply be assumed; they must be understood in terms of trade-offs among selective pressures. Clearly, birds' wings evolved under aerodynamic constraints rather than vice versa. However, biological traits such as memory are not exempt from evolving. In proposing a bottleneck to which everything else in the system must adapt while the bottleneck itself remains fixed and independent (Fig. 1 in the target article), C&C implicitly assume that it cannot evolve.

To justify this assumption, C&C should have offered evidence of stabilizing selection pressures that act against genetic variants coding for a broader or narrower bottleneck, and thereby affecting cognition and, ultimately, fitness. Alternatively, they might have assumed that the biological mechanisms underlying the memory bottleneck cannot be genetically variable – an odd assumption, which runs counter to substantial evidence in humans of (a) a range of verbal memory decay rates (Mueller & Krawitz Reference Mueller and Krawitz2009), including in particular the longer verbal working memory span in individuals with Asperger's (Cui et al. Reference Cui, Gao, Chen, Zou and Wang2010); (b) heritable variation in language and in word memory (Stromswold 2001; van Soelen et al. Reference van Soelen, Brouwer, van Leeuwen, Kahn, Hulshoff Pol and Boomsma2011) and in working memory (Blokland et al. Reference Blokland, McMahon, Thompson, Martin, de Zubicaray and Wright2011; Vogler et al. Reference Vogler, Gschwind, Coyne, Freytag, Milnik, Egli, Heck, de Quervain and Papassotiropoulos2014); and (c) variation in perceptual memory across species (Lind et al. Reference Lind, Enquist and Ghirlanda2015; Mery et al. Reference Mery, Belay, So, Sokolowski and Kawecki2007). Given that heritable variation in a trait means that it can respond to selection (e.g., Falconer Reference Falconer1981), it is likely that the bottleneck can evolve, and that it is what it is because individuals with longer or shorter verbal working memory had lower biological fitness.Footnote ¹

If language is supported by domain-general mechanisms, verbal memory is even less immune to evolution

If the emergence of language constitutes a recent and radical departure from other cognitive phenomena, it is in principle possible that working memory evolved and stabilized prior to and separately from the “increasingly abstract levels of linguistic representation” (sect. 3.2, para. 2) posited by C&C. However, there are good arguments in support of a domain-general view of language (e.g., Chater & Christiansen Reference Chater and Christiansen2010). In particular, linguistic representations and processes are hardly as modular as C&C assume (Onnis & Spivey Reference Onnis and Spivey2012). Furthermore, theories of neural reuse (Anderson Reference Anderson2010) point to the massive redeployment of existing mechanisms for new functions, resulting in brain regions coming to be involved in diverse cognitive functions. If circuits that support language continue contributing to nonlinguistic functions (including working memory), a memory bottleneck is not a prior and independent constraint on language, but rather a trait that continues to evolve under multiple selective pressures, which include language.

The bottleneck may be the solution, not the problem

As we have suggested (Goldstein et al. Reference Goldstein, Waterfall, Lotem, Halpern, Schwade, Onnis and Edelman2010; Lotem & Halpern Reference Lotem and Halpern2008; Reference Lotem and Halpern2012; Onnis et al. Reference Onnis, Waterfall and Edelman2008), a limited working memory may be an adaptation for coping with the computational challenges involved in segmentation and network construction. (Importantly, regardless of whether this specific hypothesis is correct, entertaining such hypotheses is the only way of distinguishing a function from a constraint; cf. Stephens & Krebs Reference Stephens and Krebs1986, Ch. 10.) A recently implemented model that includes this hypothesis has been tested on tasks involving language, birdsong, and foraging (Kolodny et al. Reference Kolodny, Edelman and Lotem2014; Reference Kolodny, Edelman and Lotem2015a; Reference Kolodny, Lotem and Edelman2015b; Menyhart et al. Reference Menyhart, Kolodny, Goldstein, Devoogd and Edelman2015) The model includes a time window during which natural and meaningful patterns are likely to recur and thus to pass a test for statistical significance, while spurious patterns decay and are forgotten. We stress that rather than acting as a constraint, the duration of the window must co-evolve with the mechanisms influencing the distribution of data so as to increase the effectiveness of memory representations (Lotem & Halpern Reference Lotem and Halpern2012).

We do agree with C&C regarding some of the consequences of the memory bottleneck, such as the need for online incremental construction of hierarchical representation. Indeed, our model effectively implements what C&C call “Chunk-and-Pass” (Kolodny et al. Reference Kolodny, Lotem and Edelman2015b).Footnote ² We believe, however, that the ultimate constraint on learning structure (such as that of language) in time and space is not the memory bottleneck in itself, but rather the computational challenges of chunking the data and of building hierarchies.

Biological communication is about affecting behavior, not pumping bits

Our final point focuses on the communicative function of language. Viewing a memory window as a communication “bottleneck” suggests that massive amounts of information must flow through the channel in question. However, the real function of a message is to influence the rich network of connotations and interconnections already present in the listener's brain (cf. Edelman Reference Edelman2015, sect. 2.3). Communication is about generating adaptive behavioral changes (Burghardt Reference Burghardt, Johnston, Moulton and Turk1970; Green & Marler Reference Green, Marler, Marler and Vandenbergh1979) – the listener gleans from it cues relevant to decision-making. For this, a signal must be informative and reliable in the given context (Leger Reference Leger1993); the amount of information is not the main issue (except as a signal of quality, as in complex courtship songs; Lachmann et al. Reference Lachmann, Számadó and Bergstrom2001). This implies that evolutionary selection in language is for how messages fit into the information already represented by their recipient; a bottleneck may not impose significant constraints here.