Do we know what we think before saying it to ourselves?
Marta Jorba
Pompeu Fabra University
LOGOS Research Group
Self-talk, or talking to oneself, seems to present a puzzle. If, for instance, I tell myself, “I think I will read the Odyssey this summer”, it seems that, in order to have the intention to communicate this thought to myself, I must already know what I think. Yet if I already know what I think, it is unclear what I gain by telling it to myself. Conversely, if I do not already know what I think, it is unclear how I can have the intention to communicate that thought to myself in the first place. This is the puzzle of self-talk—with a venerable history going back to a similar formulation in Plato’s Theaetetus—and it can be stated as follows (Deamer, 2021):
- self-talk exists;
- we know our communicative intentions; and
- what we say in self-talk conveys information to an interlocutor (in the case of self-talk, ourselves) about our communicative intentions.
The puzzle arises for broadly Gricean accounts of communication, according to which a speaker conveys information about her own mind to the receiver (Grice, 1957). Applied to self-talk, either out loud (private speech) or silently (inner speech), the question is whether we can give information to ourselves about our own mental states. The obvious source of the puzzle is that the speaker and the hearer are the same person, so what is the purpose of self-talk if the person who wants to convey a piece of information is the same as the one who is meant to receive it? It seems, then, that the three premises can’t be held simultaneously.
Focusing on the inner speech form of self-talk from now on, one possible solution is to deny (ii) and endorse an inferentialist or interpretivist view on the role of inner speech in self-knowledge (see Jorba 2026 for an overview and discussion of different accounts). For the inferentialist, self-knowledge proceeds by the interpretation of utterances in inner speech, similarly to what we do with others’ utterances (Carruthers 2009; Deamer 2021). For Carruthers (2014), we have access to thought contents when they get into the global workspace, and they can only be globally broadcast by being bound into perceptual vehicles. Thus, inferentialist views use the perceptual vehicles as evidence from which we infer what we are thinking. These views abound in the literature, mostly inspired by Ryle’s (1949) famous observation that we know what we are thinking by “overhearing” our own silent monologues (see different versions of inferentialism in Byrne 2011; Jackendoff 1996; Prinz 2011, a.o.). Here, the apparent paradox can be avoided because the thought is not something that we need to know in advance in order to produce the inner utterance. Rather, the utterance provides evidence from which we determine what our thought is. Premise (ii) can thus be blocked.
The inferentialist solution to the puzzle faces at least a couple of important problems. In particular, not all conscious thought appears to involve inner speech. We can have visual or spatial thoughts, for example, without putting them into words. There are also cases in which people report little or no inner speech (see recent research on anendophasia by Nedergaard & Lupyan 2024), as well as reported cases of Unsymbolized Thinking (UT; Hurlburt & Akhter 2008; see Vicente & Martínez-Manrique 2016 and Vicente & Jorba 2019 for accounts of UT in terms of the speech production mechanism). It seems difficult to maintain that these subjects cannot know the relevant episodes of thinking. Such views would have to either reductively explain non-linguistic forms of thinking in terms of linguistic forms of thinking or restrict the epistemological view to just one kind (or specific cases) of thinking. Inferentialist views also face the problem of how sensory and semantic representations are bound together (Langland-Hassan 2014; and see Bermúdez 2018 or Munroe 2023 for responses to the problem).
A second prominent way to proceed is to put pressure on premise (iii). Geurts (2018), for instance, denies that the primary role of inner speech is to convey information and instead proposes that what we do in self-talk is to entrench commitments: promises, statements, questions, etc. Such a solution abandons the idea that we talk to communicate information. However, the view has been criticized because of its limited scope, as it has problems accounting for questions or assertions, and the commitment entrenched here seems to be based on some more basic mental event (Deamer 2021). Denying (iii) still leaves it open whether (ii) is tenable. Defenders of (ii) are introspectionists and claim that immediate knowledge of thought contents is available to the subject, amounting to a form of cognitive phenomenology (see Bayne and Montague 2011 and Jorba & Moran 2016 for overviews and Jorba & Vicente 2014 for a defense).
Notice that the solutions considered so far imply different views on the relation between thought and inner speech: denying (ii) amounts to saying that we don’t have access and knowledge of a pre-constituted thought before expressing it in language (and perhaps such a thought does not even exist); denying (iii) amounts to saying that inner speech cannot have the epistemic role some theories attribute to it. These solutions suggest that the answer to “Do we know what we think before saying it to ourselves?” must be either yes or no, either denying the possibility of accessing thought contents (by denying ii) or rendering inner speech epistemically superfluous (by accepting ii).
I want to suggest that this either/or kind of solution to the puzzle excludes the following relevant case: there is a fully determined non-linguistic conscious thought that is then linguistically further specified and expressed in inner speech. Such a case undermines the need to choose either (ii) or (iii) and would exemplify a way in which both premises are acceptable, thus precluding the puzzle from arising. The case of a fully determined non-linguistic thought can be an imagistic thought or an Unsymbolized Thought—let us focus on UT, which presents an experience of thinking without any words or images. This thought, although determined in itself in terms of syntax and semantics (see Vicente & Jorba 2019, for an account of this determination) could receive further linguistic specification. One dominant model that explains the process of linguistic specification is the speech production mechanism (Levelt 1989; Jeannerod 2006; Perrone-Bertolotti et al., 2014, a.o.) according to which we formulate a message (thought), select the words, encode their sounds, and articulate them—in the case of inner speech, outer articulation is inhibited, resulting in “the voice in our head”. The idea of our case is that the process of linguistic specification or “putting the thought into words” amounts to knowing the thought in a different way from the way in which one knew the UT. (Consider an analogy with experiencing an emotion and giving it a name: linguistic labeling allows us to know the emotion in a different way, or simply allows us to know the emotion). Also, giving the thought “the sensory clothing” allows it to play a different functional role than what the UT could initially play, because it is more specified in terms of the selected words and auditory-phonological properties. The process of linguistic specification can thus be seen as an epistemic gain or contribution without having to deny that we already knew the content of the UT (as is argued by Hurlburt and colleagues, and defended by introspectionists). There is thus an epistemic gain in at least these two ways. Or, in terms of communication: there is transmission of information, but not between two agents, but within a subject through different steps of the process of linguistic specification. Crucially, then, we can both have access to the thought prior to its linguistic specification and also gain something epistemically once it has gone through the further stages of the speech production process, namely, word selection and acquisition of auditory-phonological properties.
A full defense of the case presented exceeds the space of this post, but its consideration allows us to point to a way to avoid the need to dispense with one of the two premises of the puzzle and thus the either/or strategy presented above. Moreover, the plausibility of this case is also compatible with allowing for cases in which thoughts are fully constituted by their expression in inner speech (and so with no experience of UT)—resulting in a picture that encompasses at least three possible relations between conscious thought and inner speech: 1) there is just UT; 2) the UT receives further linguistic specification (the case presented), and 3) the inner speech expression constitutes the very thought (there is no previous UT). The answer to our question “Do we know what we think before saying it to ourselves?’” can then be both yes (option 1), no (option 3), and yes, but we can also epistemically gain something by linguistic specification (option 2). This picture adds complexity to the ways of addressing the puzzle by exploring the epistemic implications of the possible relations between conscious thought and inner speech.
References
Bayne, T. & Montague, M. (2011). Cognitive Phenomenology: An Introduction, in T. Bayne and M. Montague (eds.). Cognitive Phenomenology. New York and Oxford: Oxford University Press: 1–34.
Bermúdez, J. L. (2018). “Inner Speech, Determinacy and Thinking Consciously About Thoughts”. In Peter Langland-Hassan and Agustín Vicente (eds.). Inner Speech: New Voices, Oxford University Press.
Byrne, A. (2011). “Knowing That I Am Thinking”. In A. Hatzimoysis (ed.). Self-Knowledge. Oxford: Oxford University Press.
Carruthers, P. (2009). “How We Know Our Own Minds: The Relationship Between Mindreading and Metacognition”, Behavioral and Brain Sciences 32: 121–138.
Carruthers, P. (2014). “On Central Cognition”, Philosophical Studies 170: 143–162.
Deamer F. (2021). “Why Do We Talk To Ourselves?”, Review of Philosophy and Psychology 12(2):425-433
Geurts, B (2018). “Making Sense of Self Talk”, Review of Philosophy and Psychology (2):271-285.
Grice, P. (1957). “Meaning”, The Philosophical Review, 66: 377–88.
Hurlburt, R. T. & Akhter, S. A. (2008). “Unsymbolized Thinking”, Consciousness and Cognition 17: 1364–1374.
Jackendoff, R. (1996). “How Language Helps us Think”, Pragmatics and Cognition 4, 1–35.
Jeannerod, M. (2006). Motor Cognition: What Actions Tell the Self. Oxford: Oxford University Press.
Jorba, M. & Vicente, A. (2014). “Cognitive Phenomenology, Access to Contents, and Inner Speech”, Journal of Consciousness Studies 21 (9–10): 74–99.
Jorba, M. & Moran, D. (2016). “Conscious Thinking and Cognitive Phenomenology: Topics, Views and Future Developments”, Philosophical Explorations 19 (2): 95-113.
Jorba, M. (2026). “Inner Speech and Introspection”, in Giustina, A. (Ed.). Routledge Handbook of Introspection. London. Routledge, pp. 452-468
Langland-Hassan, P. (2014). “Inner Speech and Metacognition: In Search of a Connection”, Mind & Language 29(5): 511–533.
Levelt, W. (1989). Speaking. MIT Press.
Munroe, W. (2023). “Thinking Through Talking to Yourself: Inner Speech as a Vehicle of Conscious Reasoning.” Philosophical Psychology 36 (2): 292–318.
Nedergaard, J. S. K. & Lupyan, G. (2024). “Not Everyone Has an Inner Voice: Behavioral Consequences of Anendophasia.” Proceedings of the Annual Meeting of the Cognitive Science Society 45.
Perrone-Bertolotti, M., Rapin, L., Lachaux, J.-P., Baciu, M., & Loevenbruck, H. (2014). “What is that little voice inside my head? Inner speech phenomenology, its role in cognitive performance, and its relation to self-monitoring”, Behavioural Brain Research, 261, 220–239.
Prinz, J. (2011). “The Sensory Basis of Cognitive Phenomenology”, in Bayne, T. & Montague, M. (eds.). (2011). Cognitive Phenomenology. Oxford: Oxford University Press, 174–196.
Ryle, Gilbert (1949). The Concept of Mind. Hutchinson.
Vicente, A & Martínez-Manrique, F. (2016). “The nature of unsymbolized thinking”, Philosophical Explorations,19 (2), 173–187.
Vicente, A. & Jorba, M. (2018). “How Far Does the User-Illusion Go? Dennett on Knowing What We Think.” Teorema XXXVII (2): 205–221.
Vicente, A. & Jorba, M. (2019). “The Linguistic Determination of Conscious Thought Content”, Noûs 53 (3): 737–759.
I think the question becomes much easier if we stop describing the brain in terms of sender, message, and receiver. Neural processing takes place within a recursively interconnected system. Highly connected hub regions form what network neuroscience calls a rich club organization, where different streams of neural activity converge and, through extensive feedback loops, influence distributed networks again. Such a zone of high functional and causal density can be understood as the neurophysiological realization of what I call the causal core. In my model, this core is primarily realized within the thalamocortical system, where sensory and other neural processes converge, are recursively processed, and are kept temporally coherent.
From this perspective, inner speech does not require a second internal addressee. A initially nonlinguistic neural state enters this recursive organization, recruits linguistic networks, and returns to the overall process in a modified form. The internally formulated thought is therefore neither a mere repetition nor a message sent to a second self. It is a further processing of the same overall state. This explains how a thought can already be present and yet become more precise, more differentiated, or differently accessible through linguistic formulation.
The apparent paradox therefore arises largely from modeling inner speech as communication. Once it is understood as a recursive process within a rich club organized causal core, much of the paradox disappears.
The same process can be described at the psychological level without reducing it to physiology. Psychologically, a thought need not first exist as a fully verbalized proposition that is then communicated to oneself. It may exist as an already meaningful but not yet linguistically articulated state. Inner speech then gives this state a more explicit conceptual form. In saying something to myself, I therefore do not inform a second internal listener. I articulate, differentiate, and stabilize what I am already thinking. The psychological description and the neurophysiological description refer to the same process in different descriptive grammars: one as the recursive reorganization of neural activity, the other as the progressive articulation of thought.