A Note on Heterophenomenology

In comments to some previous posts, Marcin Milkowski raised a spirited defense of Dennett’s heterophenomenology (HF) as a correct methodology of data from first-person reports (about mental states).  Among other statements, he made the following:


“HF is not about inferring beliefs from verbal behavior in any setting.  It’s about verbal reports as produced during psychological research when the subjects aren’t lying, ironising etc. but just, say, pressing a red button or a green one.  I cannot find any place where Dennett would say that verbal reports are the only evidence we have about beliefs.  He would say (and I agree) that this is the best evidence we could get (as there are no mindreading machines (yet?)).”


I don’t think I said that according to Dennett, verbal reports are the only evidence we have about beliefs.  So let’s agree that he doesn’t say that.  Even so, it seems to me that Marcin’s interpretation of Dennett is not entirely correct.  So I propose a close reading of one of Dennett’s latest statements of his view [in the context of a response to Levine 1994, who maintains that “conscious experiences themselves are the primary data to which a theory must answer; for a  critique of that view, cf. this post]:


“… a nesting of proximal sources is presupposed as we work our way from raw data to heterophenomenological worlds:

(a) “conscious experiences themselves”;

(b) beliefs about these experiences;

(c) “verbal judgments” expressing those beliefs;

(d) utterances of one sort or another.

What are the “primary data”?  For heterophenomenologists, the primary data are the sounds recorded when the subjects’ mouths move, or (d) the utterances, the raw uninterpreted data.  But before we get to theory, we can interpret these data, carrying us via (c) speech acts to (b) beliefs about experiences.  These are the primary interpreted data, the pretheoretical data, the quod erat explicatum … for a science of consciousness. …  Sticking to the heterophenomenological standard, and treating (b) as the maximal set of primary data, is a good way of avoiding a commitment to spurious data”  (Dennett 2005, Sweet Dreams, pp. 44-5).


I think it’s fair to make the following points:


1. Nowhere in this passage or any other passages that I am familiar with does Dennett require that subjects not be lying, “ironising”, or otherwise misleading their listeners about their beliefs.  Dennett says that we take the (apparent) utterances of subjects about their mental states and interpret them as (apparent) expressions of belief.  Furthermore, Dennett says we give the subject dictatorial authority on what it is like to be her.  If anyone knows of passages where Dennett tells us how to distinguish utterances that are good guides to a subject’s beliefs about mental states from those that aren’t, I would be very interested to know.  Until then, I maintain that this might be a problem for HF, because HF, as Dennett describes it, has no apparent way to insure that its data are not “spurious”.


2. I disagree with Dennett that the step from utterances to beliefs is “pretheoretical”.  In the context of a science of mind, “belief” is a theoretical term, so interpreting utterances as expressions of belief requires a theoretical assumption, which may be questioned.  Even if we set aside the possibility that “belief” will be eliminated from our theoretical vocabulary, there are still worries.  In a thin sense of “belief”, beliefs are only attributed from the intentional stance without necessarily existing in people’s heads.  In this sense, you believe everything you sincerely say you believe, because, say, sincerely asserting that p is constitutive of believing that p (sometimes, Dennett says things roughly to this effect).  Under such a thin notion of belief, HF makes perfect sense (barring worries about how to tell when people are sincere).  But I don’t think this thin notion of belief is enough to underwrite a sound methodology of data from first-person reports.  In a thick sense of “belief”, beliefs are states actually occurring in your head.  In this sense, you may sincerely assert that p without having a belief that p (perhaps because you have a belief that not p but are deluding yourself that you have the opposite belief, or even just because you lack a belief pertaining to p).  In my opinion, the science of mind is after beliefs in the thick sense, or whatever more refined theoretical counterpart of belief scientists will find in our brains.  So if we are developing a scientific methodology, we cannot always reliably infer beliefs from assertions (contrary to HF).


3. I think there is a methodology that is close in spirit to HF—because it is still a third-person methodology—but is more open minded and more sophisticated than HF.  We still start with (d), but then we are not forced to infer (descriptions of) beliefs and only beliefs as data.  We can (defeasibly) infer whatever (descriptions of) mental states our best evidence tells us we can infer.  It may be that we infer (descriptions of) beliefs, but we may also infer (descriptions of) desires, chunks of information stored in working memory, conscious experiences, or whathaveyou.  If we infer (descriptions of) beliefs, our descriptions may or may not match the subject’s descriptions, and the same is true of any mental states.  For instance, when my daughter tells me she is not hungry, I rarely if ever infer that she has a belief that she is not hungry (as mandated by HF).  Sometimes I infer she is not hungry, whereas other times I infer she doesn’t like what she is eating.  It depends on what other evidence I have.  Furthermore, I have evidence that my inferences are pretty reliable, even though they go beyond what HF allows (though in other ways, they are more prudent).  In short, what data about mental states we extract from first-person reports should depend on the total evidence we have (linguistic, behavioral, and neurological).  (I defended a view like this in my 2003 JCS article, and I offered a refined version of the same view in my recent talk at the PSA meeting.)



  1. Marcin Miłkowski


    the full answer will take me some time (I’m currently trying to discuss with Mark Bickhard on interdisciplines, https://www.interdisciplines.org/adaptation/papers/2),
    and I was on a major cogsci conference in Poland this weekend, and have to prepare a completely new presentation for a workshop next weekend… This said, I will try to answer gradually, but I’m still here to defend my views spiritedly ;).

    First of all, Dennett does use ‘belief’ in a thin sense. HF is to be an introduction to the science about consciousness, and deciding on the subpersonal level using HF verbal reports is simply unjustified. And the thick notion of belief requires theorizing about subpersonal mental representations (like Mentalese or connectionist vector spaces or whatever). This theorizing is of utmost importance for the cognitive science, and science of consciousness especially, but HF will never ever let us decide which representation is there (if any!). Of course, the question is how to encode person’s expressed beliefs but this doesn’t have to be verbal or propositional at all (for example, a robot displaying some dots on a CRT, in Ron Chrisley’s version of synthetic phenomenology). Let’s not conflate our notation with the actual mental representation in the brain.

    Second, the passage you’ve chosen is not the best thing Dennett’s ever written. I mean I translated ‘Sweet Dreams’ into Polish and I asked him about this very passage which isn’t the clearest 🙂
    BTW the newest paper by Dan Dennett on HF is to be published in a special HF issue of Phenomenology and Cognition edited by Alva Noe. I’ll see if there are ceteris paribus clauses there.

  2. In heterophenomenology language plays a central role, we understand others´ heterophenomenological world by means of describing/interpreting speaker´s utterances, but language traditionally considered is a kind of smoking screen distorting the “experiences themselves” subjects have. We are capable of cancel out what we really want to mean with something, we can lie, ironize, use metaphors…all of this instances making difficulty to access the experiences themselves. Despite its enormous complexity and being ignited human society, language cannot guarantee its reliability, it is not an honest system of communication.

    What if heterophemonology, without changing this main property (language as a sign of consciousness) is complemented, enrich by recent advances in “motor cognition” (Jeannerod 2006, Pulvermüller et al. 2005), where even language is tought to be reenacted, simulated by the motor system as the vocal gesture we issued when we speak the same words and sentences as if we were the speaker, just because language processing is associated with the activation of sensory-motor programs. In that case, the sentences we hear are more related with us, belong to us, with what we are, because we can match the vocal actions (gesture of the vocal tract when someone speak) with our capacity/ability to produce that same vocal actions when we speak (motor theory of speech perception).

    If this is correct, beliefs are interpeted as declarative memories, and therefore pre-theorical because whenever we hear a lexical sound or we see a mouth in motion they elicit in us a past experience of that stimuli (the unovoidable belief formation processes) to be implemented in a causal resoning task (if the stimuli match our expectation_declarative memory_then our expectation are correct, say, our beliefs held)

    Moreover, what it is simulated is not only the vocal gesture, the semantics as well, actions words or visual words activate the same underlying neural mechanisms as the correponding reality of the words uttered, that is, motor or visual areas; put in another way, action words (pick, kick…) recruit motor centers to intepret that words and visual words (animals) its corresponding visual areas.

    In this sense, heterophenomenology save the initial conceptualization made by Dennett plus a sophistication added by the rapidly advancing pace of cognitive science in general and the discovery of mirror system in particular.

  3. Marcin Miłkowski

    Note that you don’t have to make any change in HF in order to take gestures into account. Dennett explicitly takes button presses as expressing beliefs, so language is understood quite thinly. Anyway, no gesture can replace the full verbal communication. Try to gesture the Homer’s Iliad for me 😉

  4. …or no verbatim can replace the subtlety of gestures, we can say. Try to speak… with flatten pitch or tone, without hands celebrating the waves of language, without any facial expression remarking how you feel when everybody else says some or yourself, substracting colourfull intonation in every question…

    Language is action and action is language, and there is more in language than merely letters written on paper, symply because letters on paper induce us to act (thinking how i´ll feel being Achilleus, perhaps reproducing his values in my life)

  5. Marcin Miłkowski

    The claim “Language is action and action is language” sounds great. Yet it is flatly false. Hammering a nail can bear no communication value, so not all action is language. Of course, maybe by action you understand “thinking” but then again, driving a car without thinking about it wouldn’t be a linguistic phenomenon, would it? Anyway, language is an abstract system, and action is a concrete process, so there’s some basic category mistake in your claim. If you meant linguistic communication, and not language, then the first sentence is trivially true, but the second is still false.

    And there’s an interesting phenomenological argument. Blind people use speech synthesizers. Most early speech synthesizers sounded horribly to untrained normal people (no prosody, flat tone of voice). So engineers began to model prosody, and current speech synthesizers sound quite nice. However, blind people prefer early models because they can speak much more rapidly than normal human beings (like 600 words per minute, as I was told) and this can match the speed of speed reading (without imagining any tone of voice). The message gets earlier and is understood better by users of such synthesizers. So, sometimes less enactment means more message. And this is exactly against your position, isn’t it?

  6. First of all, i still thinking the slogan “language is action and action is language” is mostly correct but of course if we try to find fractures in the idea by means of comparing every instances in which an action is not language per se… of course yo´ll find it, because up to the moment is still an slogan with metaphorical overtones but with many, many, valid evidences.

    Hammering something gives to the motor cortex of an observer too much information about the kinematics of the action itself, pragmatic information to interpret the action… and possibly the same mechanisms or related mechanisms are in application to interpret language, particularly words involving names about tools (Martin, Ungerleider and Haxby 2000). In fact, the syntax of language or symblic thought is very similar to the engram or motor scheme of whatever action you perform or watch (Gallese 2003).
    You put the example of driving a car in wich there is action but not language phenomena, some might say that even there is no consciousness at all because it is an automatic behaviour, so, how the driver is able to drive his car! Magic, a mystery. No. The over part of language is not present but probably the driver is internally verbalizing his worries and we can detect that by means of electromyography. But even if that not persuade you, think that we can form and idea or intention to act but neverthless not acting, that´s mean that i haven´t the idea to act. No.

    Secondly, there is a long tradition dating back to Locke passing by H. Jackson that conceive thinking as action, in fact in its more overt part, is action par excellence, say, speech: the most subtle of all discrete motor behaviours and the fastest (diadochokinetic rate of 6-9 syllables or 18 phonemes) Morover, speech involve more motor fibers than any other human motor behaviour (Fink 1986).
    As such, it has an abstract and a concrete nature. Before language is heard by a listener, in the speaker´s mind language underway a vast cascade of operations translating meaning (thought, abstract) into sensory-motor events (speech, concrete).

    You´ve also mentioned the case of blind people prefering synthesizers without any realistic modelling of natural human voice. Although, i think is too dare to assert that all blind people prefer that kind of machine to more technological adavance text-to-speech machines, blind people or more precise the lack of sight, make people more socially handicapped not only in their functionality, but also in their communication; just for the fact that they unfortunately cannot acces well what are the most elusive stimuli that contribute to language and perhaps is at its evolutionary basis _gesture, expression, body posture_ and therefore they have difficult to manage the social world becoming in some sense like autistics individuals (with all the precautions that this assertion convey).

  7. I recently came up with a technique for describing consciousness that had sufficient power to take me through the perspective change that Buddhists refer to as “enlightenment”. Everything I have ever read suggests that what I have done is not possible. Until now the necessary conceptual tools (most particularly fractal self-similarity) were not known to me.

    The question to which I am currently seeking an answer is as follows.

    Preamble: I have defined two entities, as follows:

    Self-similar definition (SSD): one or many levels of X, described as X

    Fuzzy SSD (FSSD): boundaries assumed only for convenience of description

    Question: Does this concept already exist in mathematics (set theory?), linguistics, neurology, or some other field of endeavour?


    I am a self-trained amateur, not a scholar.

    I find that at least one other person has used the term “Fuzzy SSD”. I am not laying claim to that term.

    Thank you.

  8. gualtiero piccinini


    Thanks one more time for your comment.

    I agree that Dennett uses ‘belief’ in a thin sense. I just think this is a problem, for the reason I gave. I also don’t think the personal/subpersonal distinction is useful in psychological theorizing.

    The Dennettian passage I’ve chosen may not be the best thing Dennett has written but he actually repeated it verbatim in a 2003 article in JCS, so he must feel pretty good about it.

    At the time it was submitted, I refereed Dennett’s JCS article. In my report, I raised something along the lines of the objection I raised in this post. Dennett’s only change in response to the referee report was footnote 1, in which he says that the interpretation of the subject’s reports as expressing beliefs is “as neutral as possible between different theories of what is actually happening in the subject”.

    Needless to say, I don’t think it’s neutral enough and I think we can do better for the reasons I gave.

  9. We have to be cautious with not conflating utopias, wishfull thinking, science with pseudoscience. Within the “status quo” in neuroscientific research and more broadly in stablished scientific “modus operandi”, the neuroscientist Giulio Tononi is working with dynamic mathematical models to describe transient cosncious states and their neural underpinnings.

    For more exotic, although serious approach to consciousness, is the recent edited volume by Jack Tuszynski “The Emerging Physics of Consciousness”

  10. Marcin Miłkowski

    Hi Gualtiero,

    I’ll begin with a quote from Dan’s newest piece:

    In fact two other essays in that issue [of JCS – MM] – essays I had not seen when I wrote mine–are even more vivid examples of the unwitting re-invention of heterophenomenology. Piccini (2003), after labeling me an “introspection agnostic” who is seen as “rejecting introspective reports as sources of evidence” (p142), insists that “we do have means to evaluate the accuracy of introspective reports” (p147) and then goes on, in a section entitled “The Epistemic Role of Introspective Reports in Science” to describe, quite accurately, the assumptions that heterophenomenologists use to turn the raw data of verbal reports into data about what it is like for the subject. He quotes Jack and Roepstorff (2002) as saying we should adopt a “second-person” perspective, in which subjects are treated “as responsible conscious agents capable of understanding and acting out the role intended” (p149) as if this contrasted with heterophenomenology.

    Then the fact is that your proposal is taken by Dennett simply as another instance of HF. Why so?

    1. HF does not imply treating all reports as infallible, see this remark:

    „Heterophenomenology, I argue, is a cautious, controlled way of taking subjects seriously, as seriously as they could possibly be taken without granting them something akin to papal infallibility (…)”

    Anyway, the only (and the only viable) assumption that allows to treat spurious reports as spurious is using intentional stance that requires clear interpretation methods and standards. Dennett is quite clear about that. I can try to find other passages if you want but this is a little bit redundant if you remember what it means to use intentional stance. Psychologists are quite efficient at doing that, by the way. No empirical psychology can be done without it, so I wouldn’t worry at how to make this possible.

    2.HF treats reports as expressing beliefs, not as positing beliefs:

    „When a subject says “There is a tree in front of a house” he is not reporting a belief, he is expressing a belief (see the discussion of reporting and expressing in CE, pp303-9). His sentence is about the tree, not his inner state–but of course we can learn something about his inner state from his utterance, just as we can learn something about a person’s inner state when he asks “Where’s the men’s room?” or requests “Pass the mustard, please.” If the subject says “I see a tree in front of a house” he is, officially, reporting a mental state (his current vision state) by expressing a belief about it.”

    So I would say that HF worlds are composed only from thin beliefs.

    I’ll look more for some more passages but I think these hints are quite interesting for you.

  11. Marcin Miłkowski

    Annibal writes:
    “Hammering something gives to the motor cortex of an observer too much information about the kinematics of the action itself, pragmatic information to interpret the action… and possibly the same mechanisms or related mechanisms are in application to interpret language, particularly words involving names about tools (Martin, Ungerleider and Haxby 2000).”

    Well, so let me rephrase my doubts. Do you suppose that a person that never learnt any language (a deaf person, for example, that uses only basic gestures) could be ascribed any linguistic processes while hammering? It could be well true that hammering in normal persons triggers an internal monologue or whatever, and it is of course true that using linguistic tools is acting. But let’s not be overly reductive, as this posited identity of action and language is even more fishy than type-type identity theory of mind and brain. Or you don’t mean any specific natural language by “language”, and by “action” you mean “an act of using language”, but that would be simply misleading way of putting this.

  12. It´s a good thing to interpret critically contemporary Dennett´s work, scholarly; and you are a good scholar, but one of the goals or missions, or even obligation of further generations, is to recieve great philosphers´ legacy and upgraded it to advance conceptually the discipline.
    Dennett heterophenomenological program is based on many assumptions (failure of classical phenomenological approach to understand first person perspective, behavioural tradition, Ryle´s influence, holism, functionalism, autonomism…)

    I think we can save many instances of Dennets´legacy with a litle bit of cutting-edge motor neuroscience and those allied sciences (ranging from developmental psychology, psychopathology, social neuroscience, ethology…) dealing with issues concerning social behaviour accepting their findings, and if that move us to accept some fishy reductionism, well, i have not prejudice, i´m only trying to understand what is over the table.
    In response to your last sentences in the post is exactly that what i meant with “language” and “action” and i believe is not misleading, on the contrary, is just pursuing the basic foundations of embeddded generality and flexibility of natural language with the striking similarities of embeded generealities and flexibilities of motor acts which overlapping neural apparattus controls both of them paradoxically.

  13. gualtiero piccinini


    Thanks for the quote from Dennett. I presume the paper you quoted is currently unpublished.  It’s nice to see that Dennett seems to agree with what I say in my paper about the methodology of first-person data, although he distorts what I say by redescribing it in terms that I find ambiguous and overly restrictive (“what it is like for the subject”).

    Re 2: I don’t think the distinction between expressing a belief and reporting a belief makes a difference here.  If beliefs are real things, both an expression and a (true) report of a belief should result in the positing of a belief.  I agree that Dennett uses ‘belief’ in a thinner sense than that, but I see this as just one more problematic aspect of HF.

    Re 1: I agree that HF does not treat first-person reports as infallible, but it doesn’t follow that HF is the correct way to think about the methodology of first-person data or that HF is identical to the view I defend.  Dennett says scientists should not take reports at face value, but rather as expressions of beliefs about the subject’s mind.  I say, scientists need not take reports as expressions of beliefs about the subject’s mind.  It’s better to interpret them and evaluate them directly in light of our best theories and best evidence about the mind and the current circumstances.

    Both Dennett and I are opposed to so-called “first-person science”.  If HF is defined as third-person use of first-person reports, then, sure, I’m in the HF camp.  But then we need to say more precisely how to turn first-person reports into valid data.  Dennett has a story about that, and I take ‘HF’ to be the name for that story.  I have told my own story, which in many ways goes beyond what Dennett says.  As Anibal points out, we can’t stop at interpreting the work of our predecessors.  We should strive to improve on it.

    If Dennett finds it convenient to say that I’ve just reinvented HF, so be it.  The fact that he says it doesn’t make it so.

    Final comment:  for the record, I’ve read a lot of Dennett’s work and I think he is one of the best philosophers of mind out there.

  14. Marcin Miłkowski


    first, I don’t want to argue that everything that Dan Dennett ever said is true (as this is, to the best of my knowledge, false, as is the case with human beings in general). My point is that HF is really a thin theory. You could think of HF as type not as a token, and various introspective methodologies as token or implementations. That’s why HF is pretty general.

    Anyway, the only point where you and Dan disagree is the subtle problem of how to treat subject’s reports. Of course, it would be absurd to treat any report at face value, for example reports about causes of behavior cannot be treated as infallible. But