Pompe-Alama suggests that to lessen the grip of the illusion, we must pay attention to the low level realization of our thoughts. That is of course a goal of many cognitive neuroscientists, but as Pompe-Alama well recognizes, it is a difficult one to achieve. Unlike perception and action, both which can be correlated with measurable external phenomena (perception with the stimuli occurring in the external world; action with elicited motor activity), thoughts are seemingly spontaneous, and largely uncoupled from immediate environmental stimulation and control. The unpredictability of the content and occurrence of our thoughts, together with the fact that we have no idea how they are realized in neural activity (and thus which aspects of the remarkably complex signals we can record from the brain are relevant), has the consequence that thoughts promise to be extremely difficult to measure scientifically. What exactly are we supposed to look for in signals from neural tissue that is supposed to correspond to propositional thoughts as opposed to other (non-propositional) forms of mental representation? Unless we discover some means of answering this question, it will be difficult to determine empirically whether other animals have the capacity for propositional thought or not.
Taking a reductive approach, Pompe-Alama says “the question of how far thinking relies on our capacity to speak or use language can be replaced by the question of which brain areas and input-output relations we find involved in the faculties mentioned above” (this collection, p. 6). She suggests that the progress we have made in understanding the neural basis of language processing could help us resolve the debate about whether human and nonhuman cognitive processes are fundamentally different. Work in cognitive science has shown that a network of brain areas seem consistently linked with processing of natural language. Pompe-Alama suggests that we could approach the question of whether human thought is primarily linguistic by determining with functional imaging whether these areas are consistently active during human propositional thought. This will not be determinative, for reasons I sketch here. Most importantly, even if we do see activity in these areas, it will not serve to answer the question of whether human thought is fundamentally linguistically-based. Suppose phenomenal inner speech typically accompanies our thought, and it is dependent on activity in these areas. This may be because our thoughts are fundamentally linguistic, but it could also be merely a causal consequence of the deeper thought processes, without constituting them or being a necessary component of them at all. Thus, if we consistently saw activity in language-relevant areas, it might not be reflective of the fundamental nature of our thought. Suppose, on the other hand, that we failed to see such activation (and suppose we knew that inner speech was dependent on activation of language areas). This could be due to the low signal-to-noise ratio of the methods, or to the fact that language pervades brain representation and is not restricted to the areas that we typically see “light-up” in a language task, or it could indicate the non-linguistic nature of thought. In this domain, negative results are not decisive. Thus, the question of whether language centers are always active during human propositional thought will not resolve the issue.
That said, significant progress is being made in understanding at least some aspects of the representational coding of thought contents. The object perception literature demonstrates that cognitive neuroscience has achieved much in the last few years, due to work with both noninvasive fMRI in humans and invasive recording in humans and nonhuman primates. In particular, we have gained much greater insight into the representational coding of faces, with access to regional information about coding of representational aspects of face identity, similarity, expression, and so on (see e.g., Haxby et al. 2014, and Freiwald & Tsao 2011). Other work suggests that the visual cortex represents sematic features in the form of a cortical map (Huth et al. 2012). Although this kind of work is in its infancy, novel analytical and modeling techniques promise to continue to yield a deeper understanding of how our brains represent semantic properties. An important result stemming from this kind of research is evidence of the extensive homologies between neural processes of visual representation in humans and nonhuman primates (Sha et al. in press; Kiani et al. 2007). These homologies seem to extend in large part to complex cognitive processes such as decision-making (Gold & Shadlen 2007). At the neural level, we have no evidence of qualitative differences in neurological processing between humans and nonhuman primates, nor evidence that we and they possess radically different representational frameworks. Nonetheless, none of the work mentioned explicitly targets propositional contents, and very little extant work has looked at the combinatorial or structural properties of these mental representations. In my own view, answers to these difficult questions will not come from bottom-up approaches alone or even in large part. Only a high-level theory of brain function is likely to make real headway on this issue. It will be interesting to see whether new work in predictive coding (see Clark this collection; Hohwy this collection; Seth this collection) allows for new ways of approaching these fundamental questions.