In Part One, I explained the critiques of Dignāga’s epistemology offered by Candrakīrti and Jayarāśi. In Part Two, I’ll consider whether these arguments create serious problems for Dignāga’s epistemology.
I suspect that the two arguments are different in that Candrakīrti focuses on the object of knowledge (prameya) while Jayarāśi’s focus is on the means of knowledge (pramāṇa). Whether Candrakīrti and Jayarāśi are giving the same argument, they are both exploiting what they see as a self-referential problem in Dignāga’s epistemology: if the content of Dignāga’s epistemological theory is correct, then that theory itself could never be established. These arguments both fit what Graham Priest calls an Inclosure Schema (see Priest 2002, which includes Priest’s and Jay Garfield’s co-authored article applying this schema to some of Nāgārjuna’s arguments). But does Dignāga’s theory really have this problem?
Dignāga might claim that both means of knowledge could be the object of an inferential cognition. Jayarāśi considers such an objection from a Buddhist opponent:
But then someone might object that the ascertainment of two [pramāṇas] is due to conceptualization. This is not correct. Even that conceptualization does not grasp two [pramāṇas], because it concludes in the cognition of itself. Or if it did grasp [two pramāṇas], then the [Buddhist] position would be abandoned. (Tattvopaplavasiṃha, 3.3a)
If an inferential cognition could have a perceptual cognition as its object, Buddhists would abandon their position that perception is free from conceptualization. Jayarāśi may also be alluding to the Buddhist position of momentariness: since the conceptual cognition terminates in a single moment, it cannot last into a second moment in order to apprehend perception as well. Buddhists may reply that inferential cognitions can be about perceptual cognitions without revoking the non-conceptual status of the perceptual cognition, or maybe an appeal to exclusion (apoha) might be used to show that inferential cognitions can have some causal relation to perceptual cognitions of ultimately real particulars.
However, even if inferential cognitions could somehow be about perception, I think that Candrakīrti and Jayarāśi raise a serious problem for Dignāga. Dan Arnold discusses a similar issue with help from Wilfrid Sellars’s Myth of the Given. The idea is that non-conceptual states cannot justify anything insofar as justification requires something conceptual that is capable of being entered into what Sellars calls the “logical space of reasons.” If a cognition is non-conceptual, it can’t be used to justify a belief such as the belief that there are two means of knowledge. At best, it might cause such a belief. Dignāga or Dharmakīrti might say that a causal relation is good enough. However, as Arnold says,
… if perception’s privileged status is a function of its having been caused by its object, and if discursive cognitions are defined by their adding something (insofar as their content involves, by definition, some object that is not immediately present), then how can one ever be sure that what one is thinking about, when entertaining some proposition, is in any sense the same thing that was perceived? (Arnold 2005, 38)
Arnold suggests that this leads to a self-referential problem:
… one might ask what reasons could be given, in their own account, to support the correlated beliefs that only causally efficacious objects are ‘real’ and … that only directly caused cognitions are finally veridical. … The truth of their own statement of this claim is something that could be known only inferentially; but their whole epistemological set-up leads any inferential knowledge to be regarded as suspect. (Arnold 2005, 42)
In other words, Dignāga cannot claim that perception is both epistemically fundamental and non-conceptual. We can’t be sure that our inferences allegedly about perception are in fact caused by perception – while inference might yield knowledge about inference, we can never be sure that it yields knowledge about perception. Therefore, the problem that Arnold identifies deepens objections like Jayarāśi’s and Candrakīrti’s.
Furthermore, if Arnold is right that one of Dharmakīrti’s most important commentators, Dharmottara, attempted to solve this problem by claiming that even perceptual cognitions might have some sort of propositional, conceptual content (Arnold 2005, 42-48), this would show that the kind of problem Candrakīrti and Jayarāśi raise was an issue taken seriously by philosophers in the tradition after Dignāga and Dharmakīrti. This suggests that Candrakīrti’s and Jayarāśi’s considerations are problems that Buddhist epistemologists ought to take seriously. In fact, some did take these problems seriously.
Do you think the sorts of self-referential issues raised by Candrakīrti and Jayarāśi are serious problems for Dignāga’s epistemology?
Works Cited
Arnold, Dan. (2005). Buddhists, Brahmins and Belief: Epistemology in South Asian Philosophy of Religion. Delhi: Motilal Banarsidass.
Jayarāśi Bhaṭṭa. (1994). Tattvopaplavasiṃha. Ed. and Trans. Eli Franco. In: Perception,Knowledge and Disbelief: A Study of Jayarāśi’s Scepticism, Second Edition. Delhi: Motilal Banarsidass.
Priest, Graham. (2002). Beyond the Limits of Thought. Oxford: Oxford University Press.
I would agree with Arnold (as described by you) if nirvikalpa pratyakṣa were established only via negativa, insofar as otherwise we would have no access to reality. However, Buddhist epistemologists can also count on the immediate access to their perceptual awareness of, say, a patch of blue, which is immediately available as svayaṃprakāśa.
Thanks, Elisa. I think these arguments work even if we grant that pratyakṣa establishes itself by being svaprakāśa There’s still a problem of how we could talk about pratyakṣa as being one of the pramāṇas along with anumāna. Even if pratyakṣa can be self-established in a purely non-conceptual way, we could never form a theory about it, since that theory would always be inferential and by that very fact not established by perception. And if the Buddhists want to found their theory on pratyakṣa, then they have no business talking about it or trying to engage in conceptual inferential theorizing about it. The self-referential problem is this: if the theory is right, then the theory can’t be right, or at least we can’t really speak or think about the theory.
Uddyotakara makes an argument similar to yours, Ethan, to the effect that if perception is non-conceptual, how can Buddhists theorize about it? Potter has claimed that the argument is sophistical, IIRC, or something like that.
Thanks, Matthew. I will look for that argument from Uddyotakara. Plenty of people have accused Candrakīrti and Jayarāśi of being sophistical, but I think they’re doing something interesting even if their primary aim isn’t to establish positive theses in epistemology. It’s hard to see why someone would call Uddyotakara sophistical, though!
It’s in his commentary on Nyayasutra 1.1.4. Thanks for your great posts on this topic.
Yes and no, Ethan, insofar as the theory is no longer nirvikalpa (it could not!), but its vikalpa-status is based on a nirvikalpa one and this continuity grants its partial validity (unlike what happens in the case of sheer vikalpas, such as dreams, illusions and the like). Please notice that I wrote on purpose that the theory, like all vikalpas, can only be said to be partially true (in this sense, I think also a Buddhist epistemologist would have to partially agree with you).
That’s a good point that the Buddhists would partially agree. They would have to admit that all theory is conventional and conceptual, but it’s hard to see how you could know that the conceptual theory is even partially based on non-conceptual reality. You can make an argument that there’s a causal relation or maybe the mere existence of such a causal relation is enough, but what Arnold is getting at is that it seems very hard to know that a particular concept has any causal relation to a particular perception.
The problem raised by Mills becomes particularly relevant at this point because it concerns the third of the three problems identified earlier in this essay: whether there is a basis for designation, whether there is a basis for metaphorical application, and whether there is an epistemic basis for distinguishing valid cognition from error. The first two questions concern what makes conventional phenomena intelligible at all; Mills’s discussion bears directly on the third, asking what makes a cognition valid rather than merely an occurrence within the stream of experience. The arguments in Parts I and II of his discussion of Candrakīrti and Jayarāśi are therefore relevant here because they expose a difficulty in the Dignāga–Dharmakīrti account of pramāṇa: how can a conceptual epistemological theory establish that a non-conceptual cognition is valid without simply presupposing what it is supposed to explain?
Dignāga distinguishes direct perception (pratyakṣa) from inference (anumāna), defining direct perception as non-conceptual and granting it a privileged epistemic status. Inference, by contrast, is conceptual and establishes its object through a reason. The difficulty is that Dignāga must give a conceptual account of what direct perception is and why it qualifies as a valid cognition. This creates a potentially self-referential problem. If conceptual cognition cannot directly apprehend what non-conceptual perception apprehends, how can a conceptual epistemological theory establish that perception has the epistemic characteristics attributed to it?
The problem becomes sharper when the validity of a particular perception is considered. A non-conceptual cognition does not itself contain the judgment “this is a valid perception.” If pratyakṣa is genuinely non-conceptual, such a judgment must arise subsequently, through conceptual cognition. But then the question becomes how that subsequent cognition is known to have correctly identified the preceding perception. If its own validity requires further justification, the epistemologist appears to face either a regress or a gap between the immediate epistemic event and the conceptual theory that describes it. Mills’s discussion of Candrakīrti and Jayarāśi brings this difficulty into focus: the epistemologist cannot simply assume the validity of the perceptions on which the epistemological account itself depends, but neither can perception conceptually establish its own status as valid cognition.
The force of this criticism changes, however, once it is understood that the epistemological model in question need not posit externally existing objects. If there are no external objects standing outside consciousness and serving as the ultimate objects of perception, perception need not be explained as a relation between an external object and an internal representation of that object. What is directly present is the appearance itself: a patch of blue, a sound, a sensation, a visual form. Such an appearance can be understood as a dependently arisen event within the cognitive stream. Its being directly experienced does not depend upon the existence of an external object corresponding to it.
This removes one important source of the difficulty. The question is no longer how a non-conceptual cognition reaches an independently existing object behind experience. The question becomes how a subsequent conceptual cognition correctly cognizes a preceding non-conceptual appearance. That remains a genuine epistemological problem, but it is a different problem. The non-conceptual cognition does not need to contain the conceptual judgment “I have just had a valid perception.” The direct cognition simply occurs. A subsequent cognition can recognize, remember, describe, investigate, or infer about what occurred. The fact that this subsequent cognition is conceptually constructed does not by itself make it invalid.
The distinction is therefore between the epistemic event and the epistemological theory about that event. The experience of blue is one thing; the claim that this experience was a non-conceptual valid perception is another. The former belongs to phenomenology, while the latter belongs to epistemology. The fact that the second is conceptually formulated does not entail that it has no legitimate relation to the first. A Yogācāra account without external objects can explain that relation in terms of causal and phenomenological continuity within the cognitive stream. The subsequent conceptual cognition need not correspond to an external object hidden behind the appearance. It can arise dependently upon the preceding cognitive event. The concept is not the appearance itself, but neither is it therefore arbitrary.
This also clarifies the distinction between pratyakṣa and anumāna. When a sound is directly experienced, that direct cognition does not itself establish the proposition “sound is impermanent because it is produced.” The latter is an inferential cognition. It formulates a proposition and establishes it through a reason. The direct experience of sound and the inference concerning its impermanence are therefore different epistemic events. There is nevertheless no contradiction in holding that the second can be validly related to the first. The sound is directly experienced; its impermanence is subsequently established through inference.
The same applies to a conceptual theory of perception. The claim that direct perception is non-conceptual and can function as a pramāṇa is not itself a perception. It is a conceptual formulation grounded in reflection upon cognitive events and inferences concerning their characteristics. Its conceptual character does not automatically make it epistemically defective. What would have to be established is the much stronger claim that no conceptual cognition can ever validly cognize or theorize about a non-conceptual cognition. The distinction between non-conceptual perception and conceptual inference does not by itself establish that conclusion.
This brings the issue back to the distinction between correction and validity. Nāgārjuna’s mirage example establishes an epistemic fact: a mistaken cognition can arise and a subsequent cognition can correct it, without either cognition possessing svabhāva. What the example does not itself provide is a complete epistemological theory explaining the criterion by which the subsequent cognition is known to be genuinely corrective rather than merely another error. A later cognition could itself be mistaken. Correction is therefore not yet the same thing as validity.
Nor does this gap imply that there is no valid cognition. It shows instead that an account of validity requires more than an account of the possibility of correction. The Vigrahavyāvartanī gives us a negative result—the rejection of an independently established epistemic foundation—together with a minimal positive result—the preservation of dependent functional efficacy. What remains less explicit is a general criterion for distinguishing reliable conventional cognition from error.
This distinction is particularly important in light of Kamalaśīla’s criticism of lokavyavahāra in his Sarvadharmaniḥsvabhāvasiddhi. Widespread worldly acknowledgment can explain how a convention is used, but it cannot by itself distinguish a shared error from a valid cognition. Nor can the mere fact that a cognition functions successfully establish its validity in every case. A hallucination can produce behavior; a mistaken inference can lead to an action; a collectively accepted convention can nevertheless be false. Conventional operation and epistemic validity are therefore related but distinct.
The Dignāga–Dharmakīrti tradition takes up precisely this further question. Its significance for the present discussion is not that it supplies an ultimate foundation of knowledge of the sort Nāgārjuna rejects. Rather, it provides an explicit account of the conditions under which cognition succeeds or fails. Direct perception and inference are themselves dependently arisen cognitive events; what makes them pramāṇa is not an intrinsically existing epistemic status but their capacity to establish their objects non-deceptively under the appropriate conditions.
This is especially clear in the distinction between perception and inference. Direct perception does not establish the proposition “sound is impermanent because it is produced.” The perception of sound is one cognition; the inferential establishment of its impermanence is another. The latter is a conceptual cognition whose validity depends upon the logical relation between the reason and what is to be established. A cognition can therefore be conceptually constructed without being merely arbitrary, just as a non-conceptual cognition can occur without providing a conceptual account of its own validity.
This gives a more precise formulation of the issue left open by Nāgārjuna. The question is not whether cognition occurs, whether it can function conventionally, or whether valid cognition requires an intrinsically existing object. Nāgārjuna has already shown that dependent cognition, designation, error, correction, and functional efficacy do not require svabhāva. The question is what makes a cognition non-deceptive with respect to its object. That is the question to which the Dignāga–Dharmakīrti tradition gives a systematic answer.
The significance for the present argument is therefore considerable. Nāgārjuna’s rejection of an independently established pramāṇa does not entail the rejection of epistemology. It removes the demand that epistemology terminate in an intrinsically established foundation. A conventional epistemology can instead be relational: appearances arise dependently; cognitions arise in dependence upon those appearances; subsequent cognitions can recognize, correct, conceptualize, and infer; and valid cognition can be distinguished from error according to the conditions under which cognition successfully establishes its object.
The problem raised by Mills consequently does not undermine this possibility. It identifies the difficulty of giving a conceptual account of non-conceptual cognition, but that difficulty should not be confused with the absence of a valid relation between them. Once external objects are not posited, the problem is not how cognition crosses the gap between consciousness and an independently existing world. It is how a later cognition can be validly related to an earlier appearance within a dependently arisen cognitive process. That requires an epistemological account of the relation among cognitions, but it does not require an ultimate object or an ultimately established epistemic foundation.
This returns the discussion to the larger distinction between intelligibility and validity. Nāgārjuna’s account of dependent designation explains how conventional phenomena can be designated, related, and cognized without requiring an independently established svabhāva beneath them. His account of dependent cognition shows how error and correction are possible without such a foundation. But neither establishes by itself a general criterion for why one cognition is valid and another erroneous. That additional question is precisely where a developed pramāṇa theory becomes relevant.
The resulting picture is therefore neither foundationalist nor nihilistic. Phenomenologically, something appears and is experienced. Epistemologically, subsequent cognition can distinguish, conceptualize, infer, and correct. Ontologically, neither the appearance nor the cognition needs to possess svabhāva. The absence of svabhāva removes an ultimate foundation; it does not remove the dependent relations through which cognition occurs or the epistemic conditions under which cognition succeeds or fails. In this sense, the Dignāga–Dharmakīrti project can be understood not as restoring the foundation Nāgārjuna rejected, but as developing an account of validity within a dependently arisen cognitive world.