Chapter 2: Inference for Oneself
Inference divides into inference for oneself and inference for others, and this chapter takes the first: seeing the object by means of a reason complete in the three modes. What inference reaches is the universal-character and never the particular, which is why its result and its object both differ from perception’s. Five refutations follow, against the Vādavidhi, the Nyāya, the Vaiśeṣika, the Sāṃkhya and the Mīmāṃsā.
Condensed from Fazun’s Chinese translation-and-exposition of the Pramāṇasamuccaya and its auto-commentary, and rendered into English. Root verses are given as Fazun gives them; the prose beneath summarises his exposition. On the provenance and the limits of this rendering, see the series introduction.
Chapters: 1. Perception · 2. Inference for Oneself · 3. Inference for Others · 4. Example and Pseudo-example · 5. Apoha · 6. Sham Rebuttals
Volume Two
Chapter on Inference-for-Oneself
Two main divisions: establishing one’s own position, and refuting rival views. The first divides into inference in general and inference-for-oneself in particular.
1. Inference in general
Inference is of two kinds.
Inference comprises inference-for-oneself (svārthānumāna) and inference-for-another (parārthānumāna), expounded in this chapter and the next respectively.
2. Inference-for-oneself in particular
Two parts: the meaning of inference, and the distinction of its members.
(i) The meaning of inference
Inference for oneself is seeing the object from a reason that possesses the three modes.
The result [of inference] is as before. The objects and natures of those two [valid cognitions] are not the same.
The specifically characterised phenomenon (svalakṣaṇa) is not what [inference] reveals, since what is apprehended [by each] is different. [Inference] is other than that [direct perception].
[The svalakṣaṇa] is indicated through [expressions like] “seen” and so forth; [inference] does not express its own nature [directly].
Because [the object] is known from names, it reveals a different aspect — this [inferential cognition] is a twofold mental cognition.
[Objection:] Because even what is not a universal is seen [by inference]—
[Reply:] That is not so, because what is indicated is a universal.
A trustworthy statement, as a means of proof, excludes what is other than that.
If [you say] it is established as non-different — that is not so, since difference is also possible.
A trustworthy statement, being equally non-deceiving, is itself inference.
The definition: by a reason possessing the three characteristics one observes the object one wishes to infer. The pramāṇa-phala is as set out in chapter 1 — cognition arises by way of the grasped (grāhya) and grasper (grāhaka) aspects, and the result is thereby made manifest.
Perception and inference differ in two respects. Their objects differ: perception takes the own-mark, inference the universal-mark. Their natures differ: perception is non-conceptual cognition, inference conceptual.
[Question] Why is inference divided in two while perception is not? Reply: again because of the object. The own-mark cannot be displayed, so it admits no division into a form kept to oneself and a form communicated. Were verbal designations attachable to perception’s object, those very sounds would turn the cognition into inference.
[Objection] But inference does arise about objects of perception — one infers touch from colour. Reply: it does not approach such an object in perception’s manner. On the basis of a previously observed co-presence, one drops the perceptual mode and moves from the universal of the colour to the universal of the tangible. The particular tangible as actually touched is never made manifest. The two pramāṇas therefore do not share an operative sphere.
[Objection] We say “I saw such-and-such”; how then is the object of perception undisplayable? Reply: no contradiction. The object is indicated through modes such as seeing, hearing, noticing and knowing, not by stating its own-nature. What visual consciousness sees, mental consciousness afterwards cognises as “blue” and names blue. (Xin-hui takes this same passage as the objection rather than the reply — the point being that inference too would then take the own-mark. The answer runs: both cognitions bear on one colour, but visual consciousness as perception takes its own-mark, whereas the mental consciousness that is inference reaches it only through a name, and so takes only the universal, “the other mode.“)
“Twofold mental cognition”: the mental cognition that understands for oneself is inference-for-oneself; the one that declares it to another is inference-for-another. The sequence is: visual consciousness takes in an image that excludes non-blue and thereby posits blue; mental consciousness then knows it as blue and fixes the term “blue.” Since the own-mark is not displayable, what is understood as own-mark remains perception’s object, while the universal-mark is inference’s object and is what language expresses.
[Objection — Vaiśeṣika] Not every inference takes a universal for its object: from tangibility one infers the own-nature of wind, saying “what is tangible here is such-and-such, not visible to sight.” Reply: not so; a universal is still expressed. One is not inferring wind as an individual nature but indicating a locus of tangible-type qualities, that is, marking the substance-locus over and above the bare universal of substance.
[Objection] Then let it be established by excluding everything else: exclude all the rest and the own-nature of wind is inferred. Reply: the conclusion fails, and exclusion by negation cuts both ways. If “wind” were independently established, it would obtain even without the negation. (Variant reading: “once excluded, it would no longer exist.“) Abandoning the inference from a quality-differentia to the quality-universal in favour of an inference from quality-differentia to substance-differentia cannot be made valid. Touch is a differentia within quality; wind a differentia within substance. Inference has a fixed principle: the narrower may prove the wider, never the wider the narrower. From a specific quality one may infer a generic quality, and from “there is a quality” one may infer “there is a substance,” since quality depends on substance. One cannot infer any particular substance unless it is the unique locus of that quality. Touch inheres in many kinds of substance, so it cannot single out wind; negating “this touch is not earth’s” equally negates “it is not wind’s.” Nor does one see the specific quality that would belong to wind and not to earth. Nor will bare non-perception negate. Even inferring, from “visible and resistant,” that something is no quality of mind (mental qualities being intangible), one still could not establish “necessarily no tangible object.” The visible must infer the visible and the invisible the invisible. Since the quality attaches to many properties, deciding which substance it belongs to becomes a cause of doubt.
A trustworthy statement, as a means of proof, excludes what is other than that.
Having inferred the substance-universal and given up inferring the particular, it is the valid cognition of trusted utterance that excludes the other substances and settles that it is wind. Things like “wind” are established only by instruction.
[Objection] Then scripture (āgama) and inference come to no difference, and inference too should take the own-mark. Reply: they do differ, but the current accounts of the difference all fail.
- Some say scripture needs no example because it is commonly accepted by the world. Then inferring fire from smoke, which is also commonly accepted, would count as testimony.
- Others say verbal cognition treats the verbal meaning as not different from the object and does not directly infer the object. Let them explain how, in that “not different” mode, the object is reached at all. If the meaning of “tree” is just the meaning already cognised by “substance,” then by a distinct mode one knows the object as “turned back from non-substance” — but that is known through the meaning, not through the sound. On that account every word would be a pramāṇa, since even an action-word conveys its object through the implied meaning. (Without uttering “substance” one uses “tree” to know the tree as substance; equally, without saying “tree” one could know it is a tree from “poplar,” “willow” and the like.)
- Others say only sound is speech, and it is by its own nature as sound that the meaning is reached — anything beyond the sound as cause being inference. But verbal cognition has no inferential connection. For a visible object it shows the name; for an invisible one it is merely conceptual and cannot reach the object’s own distinction. Even śabda-pramāṇa, that is, cannot grasp the own-mark.
[Objection] If words like “heavenly realm” cannot denote their meaning, how do they function as inference? Reply: “a trustworthy statement, being equally non-deceiving, is itself inference.” For matters extremely hidden and beyond the senses, only reliable scriptural pronouncement gives access; they are the sphere of neither perception nor ordinary inference. Because one accepts such a pronouncement, practises accordingly, and obtains the corresponding result without deception, the pronouncement too is called inference — not because it directly denotes the own-mark of its object.
The chapter cites the principle “all verbally denoted meanings have perception as their precedent,” which also blocks any inference of the Sāṃkhyas’ pradhāna. Inference’s object, therefore, is never the own-mark.
(ii) The members of an inference
Two parts: the proving-member (能比支) and the member-to-be-proved (所比支). The first divides into the valid three-charactered reason and the fallacious reasons that lack one or more characteristics.
(a) The valid reason with the three characteristics
[The reason] is [present] in the subject of inference and in what is similar to it, and absent in what is of a dissimilar nature.
Here, “[being] known” is also [implicitly] obtained, since [the definition] is stated in the context of the means of knowing.
The thing-to-be-inferred is the dharma-bearer as qualified by the dharma-predicate. Once it has been seen by perception or inference, one establishes by way of the universal that the reason is present in the similar class, either throughout it or in part. The particle is a specifier: read “[the reason exists] only in the similar class,” not “the similar class exists exclusively wherever the reason is.” The next clause tightens this: in what lacks the property-to-be-proved the reason is unconditionally absent, and nowhere else. The dissimilar class is only what lacks the property-to-be-proved; other dharmas and contrary dharmas do not constitute it.
[Objection] The definition should say explicitly that the reason is “the cogniser” (能知). Reply: unnecessary, since it is stated under the topic of the cogniser. Moreover a reason-as-cogniser, being the cause of a cognition, must wait until its own content is ascertained, unlike a producer, which produces its effect without waiting to be cognised. Speaking of the reason that makes known therefore already implies the cognition it makes. When all three characteristics are complete, the reason is valid.
(b) Fallacious reasons lacking [one or more] characteristics
[A reason with] only one or only two modes is shown by implication not to be a [valid] reason.
“Sound is permanent, because it is produced”; “…because it is corporeal”; “…because it is not an object of valid cognition”; “…because it is non-corporeal”; “…because it is audible”; [and] “Sound is impermanent, because it is apprehensible by the eye.”
Six fallacious reasons, three with a single characteristic and three with two:
| Example | Defect | |
|---|---|---|
| One characteristic | ”Sound is permanent, because it is produced” | pakṣadharmatā only; absent in the similar class, present in the dissimilar |
| ”Sound is permanent, because it is corporeal” | present in the similar class only; absent in the subject, present in the dissimilar | |
| ”Sound is permanent, because it is not an object of valid cognition” | absent in the dissimilar class only; absent in subject and similar class alike | |
| Two characteristics | ”Sound is permanent, because it is non-corporeal” | in subject and similar class, but not excluded from the dissimilar |
| ”Sound is permanent, because it is audible” | in the subject, present in the dissimilar too, absent from the similar class | |
| ”Sound is impermanent, because it is apprehensible by the eye” | in the similar class and the dissimilar, absent from the subject |
(2) The member-to-be-proved
Two parts: the proper exposition, and the resolving of doubts.
Some assert that through a non-mistaken reason, another property is to be inferred.
Since the property and the property-possessor are [already] established, some assert that the connection is what is to be inferred.
If the reason is [already] established in the property, then what else does it infer beyond that?
In that case, why not accept the property-possessor itself as what is to be inferred?
Moreover, the connection does not exist as two [separate relata].
Likewise, [it] would become the sixth case [i.e. genitive].
[The connection] is not stated [explicitly], because the meaning is [already] expressed.
The reason is not connected with this [bare connection].
As regards the property [predicate], a non-mistaken reason is displayed [first] in other [loci]; once established there [in the example], it makes known the property-possessor [subject] that possesses that [predicate].
Two rival accounts of what an inference proves. The first: smoke is seen together with fire, not with the fire-possessing locus, so what smoke proves is fire alone. The second: fire and locus are both already universally accepted, so neither is the probandum; smoke infers their connection.
Against the first: if smoke as reason is already obtainable apart from the predicate “fire,” what further dharma is it to prove? And if the answer is that it reminds one of the locus to which fire and smoke are joined, then accept the dharma-bearer itself as the probandum — for it is not fire’s presence that one was ignorant of.
Against the second: the connection is not something existing as two relata. “Where there is smoke, there is fire” is a way of speaking, not an assertion that there is a connection of fire or of smoke. Were the connection the probandum, the statement would have to take a genitive — “the fire-connection of this place” — rather than “in this place there is fire”; and “connected to this fire” is expressed in the nominative. Pleading that it is said by implication fails, since the reasoning above has already shown the connection is not what smoke proves. The invariable relation “without fire, no smoke” must first be displayed elsewhere, in the example; only then, in a fresh locus, does bare smoke display and establish fire. If the non-errant relation had not been set up elsewhere first, seeing smoke here would tell one nothing. What the reason displays is the common basis.
The conclusion: neither the bare dharma nor the connection is the probandum. What is proved is the dharma-bearer as qualified by the dharma — “sound is impermanent,” “there is fire in that place” — not bare “impermanent” or bare “fire.”
Two parts: doubts about the proving and proved members, and doubts about inherence (和合系屬).
Further analysis also applies to the property that is to be indicated following the reason.
A reason does not cognise a property with many aspects in every respect. Having determined the connection with [the predicate], [it] obtains [the predicate] as separated from other properties.
Quality, smell, fragrant smell, and so forth — by the sequence of those distinctions, having excluded non-substance and so forth, [one] gradually cognises the utpala (blue lotus) and so forth individually.
If the reason were to operate through the door of positive proof, as in perception, then either nothing would be cognised, or everything would be cognised.
If [you say inference operates] from a universal — that [universal] would become other than that [particular]. Because the substrates are not [all] seen, a single [universal] not different [from them] would also not be seen.
Through this partial mode, [the reason] is not mistaken regarding the reason-possessor. Even though its properties are many, [inference] obtains the other [i.e. merely the probandum] alone.
[Question] If the reason infers the thesis-property, does it infer every property and difference of the subject? Reply: no. It establishes only what is non-errantly related to it. Inferring fire from smoke, one does not infer how that fire burns or how hot it is; with respect to such particulars the reason is errant.
What it does obtain is an exclusion. Fire is obtained only as “turned back from non-fire”; substance-hood and quality-hood likewise only as “turned back from non-substance” and so on. Smoke is seen at no non-fire locus and only at a fire locus.
The blue-lotus ladder illustrates the graded exclusion: being a quality excludes non-substance; being a smell excludes non-earth substances; being a fragrance excludes the universal of bad smell; and the particular fragrance excludes all but the blue lotus. Each stage is cognised by ruling out. The whole passage shows that an inferential mind reaches its object by excluding others.
Were it otherwise — were the reason to operate by positive display as perception does — then either nothing or everything would be known, with no discrimination. One does not see every fire, so on that model one should not acquire this fire either; and if one grasped as one sees, the particular brightness and fierceness would have to be cognised too.
[Objection] Then let the reason operate by way of a universal, excluding its non-fire. Reply: apart from the individual fires there is no separate fire-nature; and even if there were, it would not be seen. Not having seen every substratum, one cannot have grasped a common universal formed of many dharmas; whoever claims to see the universal is in the same case as one who has not grasped its substrata. And if grasping one substratum were grasping all, the universal would be as many as the substrata.
Only the narrower part of smoke stands in a non-errant relation to fire: smoke-nature and its grey colour are what acquire the fire. Substance-hood and the like are errant and too wide. A narrower reason proves a wider thesis; a wider reason cannot prove a narrower one, and incurs the fault of an inconclusive conclusion. (Fazun notes that the commentarial sense requires the second line to read “one does come to cognise the reason-possessor”; “cannot establish the reason-possessor” is awkward.)
The reason-possessor that is connected with whichever reason — that reason-possessor is what is cognised here. The specific distinctions are not what is cognised, because regarding those there is mistake.
What is cognised is the object connected with the reason — the reason-possessor — not its specific distinctions, for [cognition] regarding those [specifics] is thoroughly mistaken. These are interstitial verses.
The reason-possessor is the thesis-subject, and its connection — its universal dharma, such as “substance” — is wider than the subject. That is what the reason makes understood; its specific differences, such as fire’s heat, are not, since smoke is errant with respect to them.
Fazun observes that on his rendering the second verse merely repeats the first. Xin-hui’s exposition avoids the redundancy: the first verse concerns the universal and particular of the thesis (the reason makes the subject’s universal understood, not its differences), the second the universal and particular of the reason (what is connected with the reason is its universal feature, and that universal by itself does not make the subject understood — “being produced” in relation to the category “substance” — whereas what the reason differentiates does). Fazun judges Xin-hui’s reading more accurate.
C. Resolving doubts about inherence
A. Analysing the doubt
Although the connection resides in two [terms], the reason and reason-possessor are co-present — like support and supported, not like [the Vaiśeṣika] inherence (samavāya).
The reason-possessor exists solely in the reason; the other exists solely in the reason-possessor. The certainty of the connection of reason and reason-possessor is established by reversal [i.e. contraposition].
[Something may be] accepted as pervaded by the reason; in that respect it is a reason-possessor. [But] from being a pervader, [the reason] does not make [the reason-possessor] known — as [one cannot infer] a cow from [something merely] having horns.
Because non-mistakenness regarding what is to be negated depends on the connection, therefore even though the reason pervades the reason-possessor, [pervasion alone] is not the agent [of inference].
[Objection — Vaiśeṣika] Connection, like inherence, holds between both relata at once; so the subject should also be a reason and the reason a subject. Reply: although the connection pertains to both places, the two reside together as support and supported, each role fixed. What is supported is not the support. Connection by itself does not make two items mutually support-and-supported, and only that relation yields reason and reason-possessor. Vaiśeṣika inherence is precisely undifferentiated: however it connects to one thing, it connects to the second in the same way. Two further contrasts: the Buddhist connection is asymmetric (“without fire, no smoke,” but not “without smoke, no fire”); and it is posited only nominally, on the inseparability of the two, whereas the opponents make inherence a real entity distinct from its relata.
“The reason is found only where the reason-possessor is”: every reason has the subject as its thesis and must follow upon the thesis-property; there is no reason lacking it. Smoke, occurring only where fire does, can therefore indicate fire’s substance-hood as well, but not its brightness or intensity, which are particulars. Just as smoke-nature is a revealing cause, so is its grey colour; substance-hood is not, being too wide.
[Objection] Could the reason not also be reason-possessor, as producedness and impermanence are mutually related? Reply: granted the reason pervades its thesis, and in that respect the thesis is a reason-possessor — but pervasion is not itself the condition for being a reason. Horned-ness pervades cow, yet cow is not cognised by horned-ness; it is cow that cognises the pervader. Each exclusion is relative to the connection under consideration: producedness pervades impermanence, yet it cognises impermanence only by excluding permanence, and impermanence cognises producedness only by excluding non-producedness. Pervasion alone is never the agent of inference.
B. Summary verses
“Produced” pervades “destructible,” but not by means of “non-produced.” It is not a pervader of “non-permanent”; rather, it shows [impermanent] through the meaning of [the reversal of] “non-produced.”
“Having horns” pervades “cow,” so it excludes “non-horned”; [yet] just as “having horns” pervades the bull, it does not reverse [i.e. establish] the meaning “cow.” These are interstitial verses.
Reason and thesis being equal in extension, “producedness” excludes the non-produced within the impermanent. Fazun suspects his line “it is not pervading” is corrupt and prefers Xin-hui’s rendering: because producedness pervades impermanence, the sense of producedness does not state “non-impermanent,” that is, it states just “impermanent.” “Having horns” is wider in sense, so it neither excludes the sense of “cow” nor expels the sense of “non-cow.” Fazun marks the point as needing further study.
(2) Refutation of rival views
Five parts.
1. Refutation of what the Vādavidhi (《論軌》) says
From the Vādavidhi: “Inference is said to be the cognition that sees the object of [the relation of] ‘non-occurrence without [the reason].‘”
If [the Vādavidhi] accepts [inference as] merely one’s own apprehension of seeing “non-occurrence without [the cause]” — [but] the cognition of that is not stated. If stated as something other, then in what way and regarding what?
If [they say] “the object” [refers to] the probandum — then what need is there for “non-occurrence without”?
From seeing smoke and so forth in fire and so forth, what cognition is there of any other inferendum?
Since many properties exist in the subject, how can one state “the cognition of that” [without specifying which]?
Four objections to the definition. First, if “seeing the without-which-not” refers only to seeing one’s own mode, the definition has not yet said that one cognises the inferendum. Second, if it does extend to the inferendum, the definition fails to say how and where it is seen — the relation is seen in another place, and that should have been stated. Third, if one supplements it with “regarding the probandum,” so that inference is whatever object of the probandum is seen at the time of proving, then “without-which-not” is redundant, since first stating the relation and then stating the cognition of the inferendum come to the same thing. Fourth, having seen smoke and inferred fire, what further inferendum is cognised by way of “without-which-not”? The subject has many properties; the definition says neither which of them smoke cognises nor by which part smoke reveals the fire.
2. Refutation of the Nyāya (正理派)
Connection is not an object of apprehension by the sense faculties.
Regarding this [term] “possessing the prior” — since all [inferences could be so described], [the designation] is not appropriate.
Because of mistake, it does not possess [the prior].
Regarding “possessing the remainder” too — is [the designation] appropriate?
It would not be cognised from any other connection either. Because it would [merely] become [inference] “possessing the remainder,” it would not be cognised from any other connection.
[Inference based on] commonly observed [correlation] is likewise [subject to the same faults].
The result itself is not obtained in something other [than the cognition].
If [they say] the cognition of the reason is valid cognition — then the [threefold] classification would also not be established.
The Naiyāyikas hold that inference is preceded by perception and is of three kinds: possessing-the-prior (pūrvavat), possessing-the-remainder (śeṣavat), and seeing-the-universal (sāmānyato-dṛṣṭa).
Against “preceded by perception”: the connection between reason and reason-possessor is not a sphere of sense-consciousness, and neither are the two relata grasped as relata by perception. The premise fails at the outset.
Against “possessing-the-prior”: the term is ambiguous. If “possessing” means “like the former,” the label fails, because all the relata are present in any inference — the later reason can take the inferendum as object just as inference can take the connection as object — and an inference based on something unobserved cannot be called “like.” If “possessing” means the later cognition has the earlier dharma, so that the effect possesses its prior cause, then cognising the effect from the cause is errant and no valid cognition: the cause may be present while obstacles block it or auxiliary causes are missing, and then no effect arises.
Against “possessing-the-remainder”: read as “like the other,” the later cognition sharing an object with a later perception would equally qualify, and is errant. Similar colour does not make taste similar. Read as the cause possessing the effect, the cognition too would be “possessing-the-other” — but then, if the effect is not grasped, one could not grasp the without-which-not and infer the cause, whereas such inferences plainly occur, and “produced” could not yield “impermanent.” What actually happens is that a cognition bearing the aspect of the effect arises through the connection; cognitions of reason and reason-possessor arise only through the gate of connection.
Against “seeing-the-universal”: the same faults, since it too is meant to infer an extremely hidden object through the cause-effect connection.
Against the pramāṇa-phala structure: if every inferential cognition is itself the pramāṇa, there is no further result on that object; a cause with no effect cannot be a valid cognition, since agent and action always have a result. If instead the reason-cognition is the pramāṇa and the inferential cognition its phala, the two have different objects, and one cognition cannot be pramāṇa for one object and phala for another — as chapter 1 established. On that account their own distinctions of reason-possessor are unestablishable, since the reason itself becomes the object.
Finally, the restriction “only possessing-the-prior has three kinds, since it takes objects of all three times” is unreasonable: every inference operates on objects of the three times. And “like” states no connection at all, so the threefold division has no basis.
3. Refutation of the Vaiśeṣika (勝論)
The Vaiśeṣikas assert: “[This is] the effect of this, its cause, its connection, the co-occurrence in one object, and its contradiction” — these arise from reasons.
Since the [Nyāya’s] cause-and-effect [account] has already been refuted, [their] “conjunction” is likewise established as vacuous. “Conjunction” holding of both [terms] indiscriminately is no [relation of] difference [between reason and inferendum]. “Fire in conjunction with smoke” [would entail that] not by a mere part [of smoke does one infer fire], but the whole of smoke must be the cognitive indicator, and particular [features of] fire must be what is cognised. With “combination-in-a-single-locus” and the rest it is the same. “Contrariety” too is [found everywhere], [so] it is [no distinctive relation either]. The reason’s other universals are not reasons; nor are [its] particulars the thing-possessing-the-reason; [were it] as the [Vaiśeṣikas’] other [relata], then everything would turn into cogniser and cognised. — This is a summary verse.
Even if one observes place and time, one cannot [thereby] know the reason [just as reason]; since the connection has no [such] differentiating feature, and since other [things] exist, there are not two [separate relations needed]. “Contrariety” is [then] not a reason; a non-connected [relatum] cannot generate the cognition of a reason. [Their] saying “[the effect] arises from the cause” does not mean all [causes] are [cognitive] reasons. [Their view] establishes [merely] the presence of a remembering [of the cause]; “effect” etc. are not declared [as reasons] in the treatise. The last [item, “combination in one locus”] is no [true] indicator either — to no purpose, since others would also [do the job].
The cause-effect account falls with the Nyāya’s possessing-the-prior and possessing-the-remainder, refuted by the same reasoning.
The opponents posit two connections: conjunction (相應, saṃyoga), as of fire and smoke, and inherence (和合, samavāya), as of cow and horn.
Against conjunction: one infers by merely recalling the without-which-not relation, without separately grasping any conjunction; and if the connection were not apprehended at all, the reason could not display the probandum. Worse, if conjunction were the cause of inferential cognition, fire would have to be the cause of a cognition of smoke, since conjunction holds of both equally, and both would be cogniser. The opponents’ own escape concedes the point: they say a red-hot iron ball or a live coal shows fire without smoke, so fire is a doubtful and therefore invalid reason, whereas smoke must be conjoined with fire. But conjunction admits no such differentiation — whatever holds of one relatum holds of the second. Only “without-which-not” makes one of the pair a reason and not the other, and only it supplies the test for errancy.
A second consequence: if fire is connected to smoke by conjunction, then no narrower part of smoke does the inferring — every part, down to smoke’s substance-hood, becomes the reason, since nothing falls outside the conjunction; and every particular of fire, its fierceness and brightness, becomes the probandum.
The same faults attach to the rest. Inherence: cow and horned should not be reason and inferendum on their own terms, or else the particular “yellow cow” becomes probandum and the universal “horned” the prover. Combination-in-one-locus: head and feet combined in one body are not thereby reason and inferendum. Contrariety: even the unrealised four great elements, and realised particular and universal alike, would become reason and reason-possessor, since no part of anything lacks some contrariety.
The summary verse gathers this: smoke’s other universals, its substance-hood and quality-hood, are not cognitive reasons; fire’s differences, its fierceness and heat, are not the reason-possessor to be proved. On the Vaiśeṣika view all cause-effect dharmas alike become cogniser and cognised, since nothing differentiates them.
[Objection] Here one is not proving fire’s existence by smoke, which is already established, but showing that where there is smoke the fire-conjunction is present; otherwise the word “here” would be idle. Nor may one say fire infers smoke, or fire’s fierceness, since then a portion of the thesis would become the reason. (Compare the Nyāyamukha: “When fire is established by smoke, or touch by fire — what does this mean?” Answer: one is not proving fire or touch as thesis but proving the conjoined thing; otherwise a portion of the thesis would function as reason; and one does not mean to prove the existence of fire or touch, which are already commonly known.) Reply: what determines reason and reason-possessor is not conjunction but only without-which-not.
Observation of place and time works the same way. One first sees at a distant locus that where there is no fire there is no smoke; and at the moment wood is kindled, that if fire is absent then smoke does not arise. Inherence, meanwhile, is not the reason for inferring that a cow has horns, since it is nowhere directly seen. Grant any of these relations and one still cannot cognise the reason by it: what acquires the inferendum is solely the seeing of without-which-not.
Combination-in-one-locus is said to be of two kinds, effect-with-effect and cause-with-cause, exemplified by colour-and-touch and hand-and-foot. The reply is the same: it is nowhere actually seen, and inferring touch from colour would equally license inferring colour from touch, since the connection admits no differentiation. The fixed enumeration also fails, since further cases exist.
Contrariety is said to be of four kinds, visible-and-invisible and so on. None is a reason. The conjunction of wind and cloud is contrary to raining, but the raining is not the reason for the wind-cloud conjunction, nor is non-raining contrary to it. Where there is no connection, there is no reason-cognition, and the account over-extends.
On the formula “this is the cause of this”: merely saying “cause” will not do. If arising from a cause made something a reason, the inferential cognition would not be the effect; and not every producing cause is a cognising reason, or the Self and inherence, which they count as productive causes, would be reasons. Since memory intervenes, the reason-cognition is not the immediate cause of inference at all. Their maxim “the reason must be preceded by cognising” therefore establishes only that a recollection is present, not that a connection has been cognised.
If connection alone were the cause of inference, “effect” and the rest should not have been listed in the treatise, yet the treatises do list causedness, effectedness and transformation, and speak of causes of non-being, of impermanence and of permanence — none of which the sūtras declared. If only the displayer is the cause, the last member in the causal sequence, conjunction and the rest, cannot be an indicator, since it differs from the primary cause. And since other connections also reach the object, saying “this is the indicator of this” has no function.
4. Refutation of the Sāṃkhya (數論)
Since [perception] observes only its own object, [and] since without knowing [the connection] one does not reach the inferendum — [the Sāṃkhya account fails]. Since there is no distinct [universal] and it does not reside in two [relata alike], there is [on their view] no genuine reason-and-possessor. One cannot speak of a “conformative accompaniment” (anvaya) as separate from the “specific” (viśeṣa); since that accompaniment is not “other,” not being other it is not the [Sāṃkhyas’] fifth [kind of relation]. Even supposing a portion of prakṛti [were the reason] — it is not of the nature of generator, and “the specifics” are also not the generated; yet from it [the Sāṃkhyas] claim to acquire them all generally. The “oneness” and so forth of prakṛti and the rest — by what is it grasped through the “accompaniments” of the specifics? Therefore it is not a cogniser. So too “property-and-property-possessor” (財-有財) — in every case these are not cognitive indicators; the sense is exhibited only through other connections. So it is rightly said [that their system ends up with a pramāṇa] without connection. If only [the pairs] “property and property-possessor” etc. were taken to be graspings of the thing-possessing-the-reason, then, apart from the connection, they would be cognised by something else, [which implies] every cogniser must be [a cogniser of] that-which-indicates-and-that-which-is-indicated. If it is one [single relation], then it is not seven; and [their] non-inferential [kind] is just [reducible to] memory — so [their] fixed listing is untenable.
Again it is said: — You speak of seven kinds of connection [but] only two kinds of inference. [Your] three “preceded-by-the-former” inferences — where are they subsumed among these [two]? [Your] “no result” [on inference] is not the [valid] reason-account; any other connection is inapplicable. It must be stated “with observation”; what exists also exists with observation. [The Sāṃkhya’s inferences] are also established through “other” [connections]; [by them] one can reach the single “essence-meaning” (頂髻義, the principal meaning). Names of “specifics” have no [separately existing referent]; [so] the “meaning of puruṣa” (the self) is not cognised [in their system at all].
The Sāṃkhyas hold that inference arises from one relatum connected to another by perception increased to establishment, the connections being of seven kinds; where the reason-cognition is indeterminate, one advances to the second member, and from the perceived universal the undecided particulars are shown.
Against “from one perception”: perception observes only its own object. The ear and the rest do not operate on the universal and particular of “cow”; perception grasps only the own-mark. If they retort that “perception” here names the phala, the mental transformation, and that its object is therefore also perception’s, the reply is the same as in chapter 1: nothing can be pramāṇa for one object and phala for another, and the term “perception” cannot be applied to objects of another kind.
Against “one, connected”: by what mode is the connection meant? If by property-and-property-possessor, that is a relation holding between relata, and the recollection arising from seeing the property first would be idle. If by some other mode, it has not been stated. If by the similar cause, no — since one does not grasp the connection itself, one must first grasp the reason (smoke) present on the subject and only then recall “without fire, no smoke.”
Only knowing without-which-not yields decisive cognition of the thesis-object. If the connection were the cause of inference, the probandum should be cognised by its mode alone, and fire should infer smoke, the connection being symmetric. Without-which-not carries no such fault, because it does not reside in both: only smoke fails to arise without fire, not fire without smoke.
Grant the seven connections and the reasons still fall short. Harm-and-harmed is precisely a case of no connection: mongoose and snake are not reason and reason-possessor, and whichever defeats the other, no contradiction follows.
Against inferring pradhāna: when the “existing-hood” of pradhāna is cognised from the accompaniments of the specifics, by which connection is it cognised? If by cause-effect, ask whether it is generator-and-generated or indicator-and-indicated. If generator-and-generated, the accompaniment is not really present, since the reason-possessor is the plurality of specifics and the reason is pradhāna; and one cannot say that some extra accompaniment exists apart from the specifics to serve as reason. If they answer that accompaniment is not other than the specifics, then it is not any of the five conjunctions they posit, those being other. On their own showing pradhāna’s unity is grasped through the accompaniments of the specifics, so generator-generated is not what reveals it — and the ten items of the topknotted pupil are established the same way. Property-and-property-possessor is likewise no indicator: any portion becomes reason and reason-possessor only through some other connection, or else all relata become mutually cogniser and cognised.
Against the enumeration: if the relation is indicator-and-indicated and only one kind is indicator, the sevenfold division fails, since the other six would not be indicators, and any further acquirable connection would become one. The same answer meets “co-going” (pradhāna and the accompaniments): if one, then not seven; and it could not reach objects beyond the senses in any case.
Against the two kinds of inference: they deny that seeing-the-particular is inference, counting it memory. Then their fixed division of seeing-the-universal into two is arbitrary — by cause-effect the kinds would be innumerable, and property-and-property-possessor, co-going, and harmer-harmed would not be inferences at their first arising, since they arise neither as cause-effect nor as taking-the-particular.
Against possessing-the-prior: it is errant, since obstacles on the cause may leave the effect absent, and an errant reason is a cause of doubt, not a cause of inference. Calling it “inference” nominally does not repair it. Only what depends on the non-errant is the cause of inference, and that is without-which-not, not any connection with an effect.
Against the claim that only seeing-the-universal reaches super-sensuous objects: the same is established through other connections, since pradhāna’s existing-hood is acquired by seeing specifics; so the restriction fails. And on their own account puruṣa — its existence, its plurality, its otherness or non-otherness — cannot be known by any inference at all, since puruṣa is not a cause.
The seven Sāṃkhya connections (each with a worldly and a doctrinal example): (1) property-and-property-possessor — king and servant; pradhāna and puruṣa; (2) own-nature-and-transformation — curd and milk; pradhāna and mahat; (3) effect-and-cause — chariot and its parts; sattva etc. transforming into sound etc.; (4) indicator-and-indicated — potter and pot; puruṣa and pradhāna’s activity; (5) limb-and-limb-possessor — branch and tree; sound etc. and the great elements; (6) concurrent (俱行) — mandarin ducks; sattva etc.; (7) harmed-and-harmer — snake and mongoose; sattva etc. as limbs and the limb-possessor. Fazun notes the five “conjunctions” are not explicitly found in the texts and remain to be investigated.
The ten items of the topknotted pupil: existence, unity, being-of-a-purpose, being-for-another, otherness, non-agency, conjunction, separation, plurality of puruṣas, abiding.
5. Refutation of the Mīmāṃsā (觀行派)
“Because inference etc. are preceded by perception” — so [you hold] the [fundamental] scripture has not stated them; but then memory, desire, anger, striving, and the like should also be pramāṇas [on the same basis]. The expositors [of the Mīmāṃsā] have drawn [their doctrine] from other books and collected it in one place; what they expound — “inference and the rest” — is not [really] preceded by perception. A “possessor-of-object” is not [grasped by] a bare sense-cognition; in the object, no such [inherent connection] exists. “Possession” is not a valid reason for inference, [and] what is possessed, by its own nature, is not capable of making anything known. It is not something else either, nor is it perception. The refutations given above against the [other] inferential [theories] equally refute [the Mīmāṃsā’s] too. A definite object-cognition is of two sorts: inference [properly so called] and the other, which is fallacious. “A dark-one is not in the house” does not [by itself] convey that he is outside; rather, by the implied sense of the word “house,” one comes to know that he is outside.
The Mīmāṃsakas hold that inference and the other pramāṇas have perception as their precedent and, being similar to it in mode, were not separately stated in the sūtras; the commentators expounded them with distinctions.
Reply: on that ground memory, desire, anger and striving would also be pramāṇas, being equally preceded by perception. In fact the commentators extracted these from other books and gathered them in one place; inference is not preceded by perception. A connection depends on many dharmas, so it cannot be the object of a non-conceptual sense-cognition, and grasping the connection is exactly what inference requires and perception does not. Inherence, combination-in-one-locus, contrariety and conjunction are none of them perception, though this school treats possessors of conjunction as perceptions. Possession is not a reason of inference — fire and smoke are not inferences by virtue of it, the two relata being undifferentiated on that view — and what is possessed, by its own nature, assists no inference; nor does perception.
Their accounts of inference, scriptural transmission (声量) and analogy (upamāna) need no separate refutation: what little they state is what the Vaiśeṣikas hold, and the earlier refutations cover it, since they have displayed no other genuine connection. Two items are treated briefly:
Cognition-through-the-object (arthāpatti) is of two sorts, definite and indefinite. The definite is simply inference: from one item being boiled one knows the food in the same pot is boiled, cognising by the “being-boiled” reason. The indefinite is no valid cognition but a cause of doubt — “because it is produced by effort, it is impermanent” runs alongside “because it is not produced by effort, it is permanent,” and nothing settles between them.
Cognition from non-fact (abhāva). Some say “the dark-coloured one is not in the house” indicates that he is outside, and count this a separate pramāṇa. Reply: the negation does not indicate it; the implication carried by the word “house” does. Were both house and outside absent, one could say only “he is not.” So this is inference by implication, not a cognition from mere negation. The related claim, that refuting the opposing category establishes one’s own, is ruled out by the same reasoning.