Dharmakīrti's Dual Anumāna and the Social Fabric of Knowledge — Epoche C2
Two names for one inference, and why the distinction is not the one it looks like Dharmakīrti, the seventh-century Buddhist epistemologist, divides inference into two: svārthānumāna , the inference a person draws for his own purposes, and parārthānumāna , the inference stated for the purposes of another. The division is structural to his corpus — the Nyāyabindu , his short manual, gives each a chapter of its own after the chapter on perception, and the four chapters of his major work, the Pramāṇavārttika , open with the first and close with the second. Because the second is by definition a public performance, the division has looked to several recent readers like a foothold for a social epistemology: an admission, inside a system usually presented as an austere defence of individual certainty, that knowledge is completed only when it can be laid before others and made to bind them. The published version of this essay took that view. It is worth taking seriously, and it does not survive contact with what Dharmakīrti says about parārthānumāna in the sentence immediately after he defines it. What follows sets out the machinery in enough detail to see why, and then argues that the social element in Dharmakīrti is real but located elsewhere, in two places the original argument did not visit. One correction of vocabulary first, since it obscures the sources. The terms are svārthānumāna and parārthānumāna — inference for one's own purpose and for the purpose of another , from artha , purpose or object. The forms svānumāna and parānumāna , used in the earlier version of this essay, are not the terms of art and will not find the passages. What a pramāṇa is for Dharmakīrti, and what makes it awkward to call him a foundationalist A pramāṇa is an instrument or episode of knowing that warrants what it delivers. Dharmakīrti admits exactly two, following Dignāga (c. 480–540), his predecessor by roughly a century: perception ( pratyakṣa ) and inference ( anumāna ). The restriction to two is not parsimony for its own sake; it falls out of a claim about objects. There are, on this view, two kinds of object and no third. There is the svalakṣaṇa , the unique particular — momentary, causally efficacious, and the only thing that is real in the strict sense — and there is the sāmānyalakṣaṇa , the general character, which is a construct of the conceptualising mind and has no existence apart from it. Perception takes the first; inference takes the second. Dignāga's definition of perception, in the first chapter of the Pramāṇasamuccaya , is kalpanāpoḍha : free of conceptual construction. Perception is therefore mute — it delivers the particular but cannot say anything, because saying requires generality. This alone should make one hesitate before calling Dharmakīrti a foundationalist in the sense the word carries in post-Cartesian epistemology, where the foundation is a class of beliefs that are self-warranting and from which the rest is derived. His foundational cognition cannot serve as a premiss, because it has no propositional content to lend. And his criterion of validity points the other way from self-warrant altogether. In the second chapter of the Pramāṇavārttika he defines a pramāṇa as avisaṃvādi jñānam — cognition that does not belie, that does not let one down — and glosses non-belying by arthakriyā , the successful accomplishment of the purpose for which one acted on it. The mark of a warranted cognition is thus downstream of the cognition, in what happens when one acts, and it is defeasible: the water one sees may prove to be a mirage, and the cognition is then retrospectively demoted. There is a further strain in the system, which Dharmakīrti embraces rather than hides. Inference is bhrānta , erroneous with respect to its object, because its object is a construct that nothing in the world answers to; it is nevertheless a pramāṇa , because it is reliably non-belying. The most articulate and communicable of our cognitions is precisely the one that does not touch the real. John Dunne's reconstruction of the system's layered method makes the point that Dharmakīrti argues at several levels of analysis at once, and that the ontological commitments of the lower levels are provisional throughout. What makes a reason good: three characters, and a wheel with nine spokes Inference of either kind proceeds through a hetu or liṅga , an inferential reason or mark, which establishes a sādhya , the property to be proved, in a pakṣa , the subject under discussion. The traveller sees smoke on a mountain and concludes there is fire: smoke is the hetu , fire the sādhya , the mountain the pakṣa . Two further classes are needed. The sapakṣa is the set of loci independently known to possess the sādhya — a kitchen hearth, for fire. The vipakṣa is the set of loci known to lack it — a lake. Dignāga's criterion, inherited by Dharmakīrti, is the trairūpya , the triple character of a good reason. Writing $H$ for the reason, $S$ for the property to be proved and $p$ for the subject, its three clauses are: $$\text{(i)}\quad H(p) \qquad \text{(ii)}\quad \exists x\,\bigl(S(x) \wedge H(x)\bigr) \qquad \text{(iii)}\quad \forall x\,\bigl(\neg S(x) \rightarrow \neg H(x)\bigr)$$ That is: the reason belongs to the subject ( pakṣadharmatva ); it is present in at least one similar instance ( sapakṣe sattvam ); and it is absent from every dissimilar instance ( vipakṣe asattvam ). The asymmetry between (ii) and (iii) is the substance of the criterion and is easy to miss. Presence in the sapakṣa may be partial — smoke need not occur at every fire — because a mark that occurred only sometimes where the target property is found would still, if it never occurred where the property is absent, license the conclusion. Absence from the vipakṣa must be total, since a single fire-free smoke destroys the inference. Dignāga made this asymmetry visible by enumeration, in the device known as the hetucakra , the wheel of reasons, set out in the short treati