Talking to Ourselves: Inner Speech and Natural Language as a Language of Thought
By Wade Munroe
We talk to ourselves. Sometimes we do so out loud. However, frequently, we do so without making a sound. We use inner speech. In my contribution to Neurocognitive Foundations of Mind, I argue that inner speech modulates ongoing mnemonic, attentional, and sensorimotor processing, and, in certain cases, serves as the necessary vehicle for conceptual processing. Language isn’t merely for communication; it’s for cognition (writ large). Additionally, I argue that recent work on inner speech can supplement embodied/grounded approaches to conceptual processing (GACP), especially in their explanations of abstract concepts (Borghi, 2023; Dove, 2022).
Although GACP are a diverse lot, they constitute a programmatic study of conceptual processing in relation to sensorimotor systems, the body, and the (physical/social) environment.[i] According to GACP, conceptual processing is (in part) functionally reliant on sensorimotor, affective, and interoceptive systems. GACP stand in stark contrast to classical, amodal views of conceptual processing and, in particular, the language of thought hypothesis (LOTH) (Fodor, 1975; Quilty-Dunn, Porot, & Mandelbaum, 2023). Traditionally, advocates of LOTH posit (what I will call) a Classical LOT: a lingua mentis that possesses a grammar distinct from natural language. Instead of functionally relying on sensorimotor systems, Classical LOT operates in a central processing unit, separable from memory units and ‘sandwiched’ (Hurley, 2001) between systems that subserve perception and action.
While Classical LOT is wanting for empirical evidence, there is now considerable evidence for GACP (Barsalou, 2008). However, GACP have a harder time accounting for the explanatory bread and butter of Classical LOT: (i) abstract conceptual processing and the much discussed (ii) systematicity and (iii) productivity of thought (Fodor & Pylyshyn, 1988). As I argue in the first half of the chapter, insofar as GACP are wanting in their explanation of (i)-(iii), recent work on inner speech can pick up the explanatory slack. Assuming there is an explanatory need for a LOT, natural language suffices.
For the sake of space, I focus on (i) as opposed to the well-trodden (ii) and (iii). The token objects/properties/relations that fall under abstract concepts—concepts like FREE WILL, JUSTICE, and PRIME NUMBER—are not straightforwardly perceivable or manipulatable and can be spatiotemporally unbounded, as in the case of mathematical objects (at least on certain realist metaphysics) (Borghi, 2023; Dove, 2022). For instance, we can perceive and interact with token dogs, but we can’t perceive or interact with the prime number, 523.[ii] However, insofar as our concepts are grounded in perception and action, we should expect them to be, in part, grounded in our experiences—both overt and covert—of parsing and producing language (Borghi, 2023; Dove, 2022). In turn, just as the conceptual processing of a category involves, in part, the offline use of modality-specific systems that subserve perceiving and interacting with category tokens, we should expect conceptual processing to involve, in part, the offline use of speech planning and comprehension systems that subserve parsing and producing utterances about category tokens and types. In other words, we should expect conceptual processing to involve inner speech, were inner speech, as I define it, is the offline operations of the language faculty, that is, operations that (roughly) don’t causally result in or result from overt speech.
Looking to relevant empirical work, several meta-analyses find that language areas are the most likely to exhibit increased activation with abstract concepts (e.g., Binder, Desai, Graves, & Conant, 2009; Wang et al., 2018). As I argue, in the limiting case, certain abstract concepts will rely on the language faculty out of psychological necessity because we lack the means to represent the category independent of language. For example, as I discuss, our ability to think about and operate over integer values and quantities necessarily relies on the language faculty (Munroe, 2023). Unless we humans acquire a numeric code to represent positive integers—e.g., a language with a generative number vocabulary that includes recursive rules for combining lexical primitives to generate expressions of successive integers—we will lack the cognitive means to think about and operate over integer values or quantities (Frank, Everett, Fedorenko, & Gibson, 2008).
Importantly, a numeric code, like English number words, doesn’t merely function as a means of communicating mathematical knowledge. Instead, our thoughts about and operations over integers are vehicled by the offline use of the code. The idea that natural language can serve as a LOT figured prominently in the work of philosophers like Gilbert Harman (2015) and, arguably, throughout the history of generative linguistics (Hinzen & Sheehan, 2013). For instance, take Chomsky’s claim that the ‘Basic Property’ of language is the “generation of an unbounded array of hierarchically structured expressions mapping to the conceptual-intentional interface, providing a kind of ‘language of thought’—and quite possibly the only such LOT” (Chomsky, 2015, p. 13). Arguably, the popularity of Classical LOT sidelined the idea of a natural language LOT in cognitive science for some time. However, recent work on the ability of inner speech to modulate mnemonic, attentional, and sensorimotor processing provides fresh empirical support for a natural language LOT that complements GACP. But we should be clear about the scope of the claim that natural language can serve as a LOT: In certain limiting cases of abstract concepts, e.g., integers, inner speech is the sole medium of conceptual processing, but the claim is not that all conceptual processing reduces to inner speech (cf. Gauker, 2011). Instead, the claim is that—just as conceptual processing is, in part, functionally reliant on sensorimotor, affective, and interoceptive systems—conceptual processing is, in part, functionally reliant on the language faculty.
In the second half of the chapter, I critically examine recent neurocognitive models of inner speech. While Classical LOT lacks a plausible implementation story, with LOTH advocates traditionally arguing for the autonomy of implementation-level and computational/algorithmic-level explanations of cognition, inner speech can be situated within neurocognitive models of overt speech. However, as I demonstrate, these models still largely assume that language plays the solely communicative function of mapping non-linguistic, conceptual representations onto (motor commands for) acoustic blasts and vice versa. If we are to cast off Classical LOT in favor of a natural language LOT, a series of issues need to be resolved in current neurocognitive models.
Bibliography
Barsalou, L. W. (2008). Grounded cognition. Annu. Rev. Psychol., 59(1), 617-645.
Benacerraf, P. (1973). Mathematical truth. The Journal of philosophy, 70(19), 661-679.
Binder, J. R., Desai, R. H., Graves, W. W., & Conant, L. L. (2009). Where is the semantic system? A critical review and meta-analysis of 120 functional neuroimaging studies. Cerebral Cortex, 19(12), 2767-2796.
Borghi, A. M. (2023). The freedom of words: abstractness and the power of language: Cambridge University Press.
Chomsky, N. (2015). What Kind of Creatures Are We? Columbia University Press: Columbia University Press.
Dove, G. (2022). Abstract concepts and the embodied mind: Rethinking grounded cognition: Oxford University Press.
Fodor, J. A. (1975). The Language of Thought (Vol. 87): Harvard University Press.
Fodor, J. A., & Pylyshyn, Z. W. (1988). Connectionism and cognitive architecture: A critical analysis. Cognition, 28(1-2), 3-71.
Frank, M. C., Everett, D. L., Fedorenko, E., & Gibson, E. (2008). Number as a cognitive technology: Evidence from Pirahã language and cognition. Cognition, 108(3), 819-824.
Gauker, C. (2011). Words and Images: An Essay on the Origin of Ideas. Oxford: Oxford University Press.
Harman, G. H. (2015). Thought: Princeton University Press.
Hinzen, W., & Sheehan, M. (2013). The philosophy of universal grammar: OUP Oxford.
Hurley, S. (2001). Perception and action: Alternative views. Synthese, 129, 3-40.
Meteyard, L., Cuadrado, S. R., Bahrami, B., & Vigliocco, G. (2012). Coming of age: A review of embodiment and the neuroscience of semantics. Cortex, 48(7), 788-804.
Munroe, W. (2023). Semiotics in the head: Thinking about and thinking through symbols. Philosophy and Phenomenological Research, 107(2), 413-438.
Quilty-Dunn, J., Porot, N., & Mandelbaum, E. (2023). The best game in town: The reemergence of the language-of-thought hypothesis across the cognitive sciences. Behavioral and Brain Sciences, 46, e261.
Wang, X., Wu, W., Ling, Z., Xu, Y., Fang, Y., Wang, X., . . . Bi, Y. (2018). Organizational principles of abstract words in the human brain. Cerebral Cortex, 28(12), 4305-4318.
[i] A host of theories fall under the broad, GACP banner. See (Meteyard, Cuadrado, Bahrami, & Vigliocco, 2012) for further discussion.
[ii] The problem is analogous to Benacerraf’s problem for causal accounts of knowledge (1973).