Inference (རྗེས་དཔག་, anumāna)
One of the two types of valid-cognition in Buddhist epistemology. Inference is the conceptual, mediated cognition of objects through reasoning based on evidence (signs/reasons, རྟགས་, liṅga/hetu).
Definitions
Dignāga (Pramāṇasamuccaya)
མཚན་གཞི་ (Definiendum): རྗེས་དཔག་ (anumāna) — inference
མཚན་ཉིད་ (Definiens): Cognition of a hidden object (ལྐོག་གྱུར་, parokṣa) based on evidence (རྟགས་, liṅga) satisfying the triple mark (ཚུལ་གསུམ་, trairūpya)
མཚོན་བྱེད་ (Illustration): Inferring fire from smoke; inferring impermanence from production
The triple mark (trairūpya) constitutes the defining criterion of valid evidence:
- Property of the position (ཕྱོགས་ཆོས་, pakṣadharmatā) — the evidence is a property of the subject
- Positive concomitance (རྗེས་འགྲོ་, anvaya) — the evidence is present in similar cases
- Negative concomitance (ལྡོག་པ་, vyatireka) — the evidence is absent in dissimilar cases
Source: pramanasamuccaya, II–III; dreyfus-recognizing-reality, p. 214
Dharmakīrti (Pramāṇavārttika)
མཚན་གཞི་ (Definiendum): རྗེས་དཔག་ (anumāna) — inference
མཚན་ཉིད་ (Definiens): Consciousness that apprehends something hidden (ལྐོག་གྱུར་, parokṣa) on the basis of evidence satisfying the three criteria, where the evidence establishes a necessary connection (pervasion, ཁྱབ་པ་, vyāpti) between reason and property to be proved
མཚོན་བྱེད་ (Illustration): Inferring fire from smoke on a mountain pass
Dharmakīrti shifts the emphasis from the triple mark as a formal criterion to pervasion (vyāpti) as the underlying logical relation. An inference is correct only if the reason is necessarily pervaded by the property to be proved (dreyfus-recognizing-reality, p. 214; westerhoff-dignaga-dharmakirti, pp. 226–228).
Source: pramanavartika, I; dreyfus-recognizing-reality, p. 214
Geluk Collected Topics (Tutor’s Collected Topics, via Perdue)
མཚན་གཞི་ (Definiendum): རྗེས་དཔག་ (anumāna) — inference (Perdue: “inferential cognizer”)
མཚན་ཉིད་ (Definiens): An ascertaining cognition (Perdue: “determinative knower”, ངེས་ཤེས་) that, depending on its basis — a correct sign (རྟགས་ཡང་དག་) — is incontrovertible (མི་སླུ་བ་) with regard to its object of comprehension, a hidden phenomenon (ལྐོག་གྱུར་)
མཚོན་བྱེད་ (Illustration): The inference realising the impermanence of sound by the sign “product”; the inference realising the presence of fire on a mountain pass by the sign “strongly billowing smoke”
English: Inference is ascertaining (hence always conceptual), arises in dependence on a correct sign (the three modes — see below), and is incontrovertible — one will not turn away from its content so long as one remembers. Its object of comprehension must be a hidden phenomenon (i.e. not directly perceived by the cogniser at that moment).
Source: perdue-buddhist-reasoning-debate, p. 301, citing the Tutor’s Collected Topics
The Collected Topics treatment lists three sub-types (Perdue p. 302): inference through belief, inference through renown, and inference by the power of the fact (the standard logical inference). This matches Dharmakīrti’s threefold division but is presented as a bsdus grwa internal scheme rather than directly cross-referenced to PV.
Types
- Inference for oneself (རང་དོན་རྗེས་དཔག་, svārthānumāna) — the internal cognitive process of drawing a conclusion. Best understood as a mental model: the reasoner constructs a model of situations satisfying all premises and tests whether the conclusion holds in all of them (westerhoff-dignaga-dharmakirti, p. 229)
- Inference for others (གཞན་དོན་རྗེས་དཔག་, parārthānumāna) — the formal, linguistic expression of an argument to convince others. Not merely a perspectival difference from inference for oneself — structurally distinct (westerhoff-dignaga-dharmakirti, pp. 228–229)
Pre-Dignāga History
Katsura traces the development of Indian inference theory from two distinct traditions (katsura-dignaga-lectures):
The debate tradition developed Points of Defeat (nigrahasthāna) — 15 in the Carakasaṃhitā, 22 in the Nyāyasūtra — which led to the examination of “bad” reasons (hetvābhāsa), and ultimately to the theory of the triple mark. Buddhist logicians contributed significantly: Nāgārjuna’s Upāyahṛdaya/Prayogasāra and Vasubandhu’s three vāda manuals (Vādavidhi, Vādahṛdaya, Vādavidhāna) preceded Dignāga’s own Nyāyamukha.
The epistemology tradition moved from the Carakasaṃhitā’s five pramāṇas and the Nyāyasūtra’s four to Dignāga’s two. The Nyāyasūtra gives only an ostensive definition of inference through three kinds: pūrvavat (from present to past), śeṣavat (from present to future), and sāmānyato dṛṣṭam (from present to present, based on similarity) (katsura-dignaga-lectures).
The Triple Mark (trairūpya)
dignaga defined valid inference as characterised by a triple mark (westerhoff-dignaga-dharmakirti, p. 227):
- pakṣadharmatā — the subject of inference (pakṣa) is characterised by the reason (hetu). E.g., the mountain has smoke
- anvaya — positive concomitance: there is at least one similar case (sapakṣa) with both reason and property to be established. E.g., a kitchen stove has smoke and fire
- vyatireka — negative concomitance: there is no dissimilar case (vipakṣa) with the reason but without the property to be established. E.g., a lake has neither smoke nor fire
The relation encoded in conditions 2–3 is pervasion (vyāpti, ཁྱབ་པ་). dharmakirti requires only two parts for inference for others: assertion of pakṣadharmatā + statement of pervasion — no need to even state the thesis (westerhoff-dignaga-dharmakirti, p. 228).
Not Formal Logic
anumāna is not an attempt to construct a formal logic. It merges a logical and an epistemological dimension: the triple mark indicates how properties in the world are related, and if they are so related, drawing the inference will result in knowledge. The role of examples is epistemic — the positive example assures us the properties are real; the negative example shows due diligence in checking the inference against the world (westerhoff-dignaga-dharmakirti, pp. 227–230).
Both kinds of inference are tokens, not types — concrete events, not abstract structures. But not psychologistic: the underlying worldly facts determine validity (westerhoff-dignaga-dharmakirti, p. 229).
Indian inference is also fallibilist — a clear contrast with deductive inference in the Western sense. And externalist: knowledge comes from standing in a specific relation to the object known, whether or not one knows that this relation obtains (westerhoff-dignaga-dharmakirti, p. 230).
Sapan’s Account of Correct Reasoning
sakya-pandita in Ch.10 of the Treasury defines a correct reason as one endowed with three modes (ཕྱོགས་ཆོས་, རྗེས་ཁྱབ་, ལྡོག་ཁྱབ་) — notably, the earlier Tibetans defined it as one that is three modes, a subtle but important distinction (sonam-introduction-treasury).
Three types of correct reasoning:
- Same nature (svabhāvahetu, རང་བཞིན་གྱི་རྟགས་) — affirming
- Effect (kāryahetu, འབྲས་རྟགས་) — affirming
- Non-cognition (anupalabdhihetu, མ་དམིགས་པའི་རྟགས་) — negating
In Ch.11 on inference for others, Sapan defines the parties in formal debate: defender (སྔ་རྒོལ་), challenger (ཕྱི་རྒོལ་), and witness (དཔང་པོ་). He admits four types of response (contradictory, uncertain, unestablished, acceptance) — the earlier Tibetans having enumerated only three (sonam-introduction-treasury).
A Good Reason
A “good reason” (saddhetu) requires more than formal validity. It must also have true premises (soundness), a relevant doubt (saṃśaya) together with the desire to know (jijñāsā), and strategic considerations about when to make the argument (westerhoff-dignaga-dharmakirti, p. 231).
The Validity Problem
A deep tension exists in dharmakirti’s system: inference deals with the generally-characterised (conceptual constructs), yet Dharmakīrti’s ontology holds that only the specifically-characterised (individuals) are real. How can inference be genuinely valid if it never touches reality directly?
This is one of the central philosophical problems explored in dreyfus-recognizing-reality (Chapter 18). The Ge-luk moderate realist answer provides ontological support for inference by granting universals a degree of reality. The Sa-gya answer maintains that inference is valid practically despite being ontologically unfounded in real universals.
Members of a Logical Proof (Nyāyapraveśa Commentary)
Śaṅkarasvāmin’s Nyāyapraveśa (a summary of Dignāga’s system, translated from Sanskrit by Dr. Berzin) provides the classic Sautrāntika account of logical proof (sgrub-bya, Skt. sādhana). The Buddhist logical proof has three members (ya-gyal, Skt. avayava) — fewer than the Nyāya system’s five (berzin-science-of-mind, Commentary on “A Manual for Engaging in Logic”):
- Thesis (bsgrub-bya, Skt. sādhya): states the subject (phyogs, pakṣa) and the property to be established
- Reason (gtan-tshig, Skt. hetu): states a property of the subject that is a reason — must satisfy the triple mark
- Examples: must include both a homogeneous example (mthun-phyogs, Skt. sapakṣa) and a heterogeneous example (mi-mthun-phyogs, Skt. vipakṣa)
Example (the canonical proof of the impermanence of sound):
- “Sound is impermanent.” [thesis]
- “Because of being something produced (byas-pa-nyid, Skt. kṛtakatva).” [reason]
- “Whatever is produced is seen to be impermanent, like a clay jug.” [concordant example + positive concomitance] / “Whatever is permanent is seen to be something unproduced, like space.” [heterogeneous example + negative concomitance]
Homogeneous item (mthun-phyogs): something with both the property to be established and the property given as the reason — e.g., a clay jug (produced and impermanent).
Heterogeneous item (mi-mthun-phyogs): something that lacks both the property to be established and the reason — e.g., space (not impermanent and not produced).
Being something produced (byas-pa-nyid): specifically, being something that has arisen immediately from effort (brtsal ma-thag-tu byung-ba-nyid, Skt. prayatnānantarīyakatva). This is narrower than simply “affected by causes and conditions” — lightning is caused but not produced from effort; karma ripens from constructive behaviour but not immediately.
Semblances of a Thesis
A semblance of a thesis (phyogs-kyi ltar-snang) is one that appears valid to the unknowledgeable but is flawed. Nine types are listed (berzin-science-of-mind, Commentary on “A Manual for Engaging in Logic”):
- Contradictory to direct perception (Berzin: “bare perception”, མངོན་སུམ་) — e.g., “Sound is not something that can be heard”
- Contradictory to inference (Berzin: “inferential cognition”, རྗེས་དཔག་) — e.g., “A clay jug is permanent”
- Contradictory to textual tradition — e.g., “Sound is permanent” (contradicts Vaiśeṣika’s own scriptural tradition)
- Contradictory to worldly common sense
- Contradictory to one’s own statement
- An unestablished characteristic
- For an unestablished item to be characterised
- Both an unestablished characteristic and item
- A characteristic having a mutually established connection with the item
Semblances of Valid Cognition (Inference)
A knowing that has been preceded by a semblance of a reason is a semblance of inference (Berzin: “semblance of inferential cognition”) — it arises inferentially but from a fallacious reason (berzin-science-of-mind). A conceptual knowing in reference to an external object is a semblance of direct perception (Berzin: “semblance of bare perception”; mngon-sum-gyi ltar-snang) — it appears to be direct but operates through the medium of a conceptual category rather than taking an individually characterised phenomenon as its object of experience.
The Geluk bsdus grwa Presentation (Perdue)
perdue-buddhist-reasoning-debate gives the most accessible English statement of how the Tibetan Collected Topics tradition (drawing on Pur-bu-jok Jam-ba-gya-tso’s Tutor’s Collected Topics) systematises inference.
Definition of the correct sign
“A correct sign is that which is the three modes (ཚུལ་གསུམ་)” (p. 146).
Critically, a correct sign is the three modes — not something that has them. This is the Geluk version of the distinction sakya-pandita reverses in PYN Ch. 10 (Sapan: “endowed with three modes”; see Treasury discussion at sonam-introduction-treasury).
The three modes in the standard proof “the subject sound is an impermanent phenomenon because of being a product” are:
- Property of the subject (ཕྱོགས་ཆོས་, pakṣadharmatā): sound is a product
- Forward pervasion (རྗེས་ཁྱབ་): whatever is a product is necessarily an impermanent phenomenon
- Counterpervasion (ལྡོག་ཁྱབ་): whatever is not an impermanent phenomenon is necessarily not a product
Each mode is given a Collected-Topics-style definition keyed to the specific proof (Perdue pp. 147–158). Validity is epistemological-verification-relative: the sign is the property of the subject only for a person who has ascertained that the subject has that quality. A syllogism of the form “S is R because of being R” is therefore invalid, not by form but because the four ascertainments must be done in sequence and would collapse (p. 150).
The three types of correct signs
Perdue follows Manchester-Rogers’ Tibetan Logic in dividing correct signs by entity into three (pp. 179–183):
| Type | Sanskrit / Tibetan | Classic example | Purpose |
|---|---|---|---|
| Effect sign | kāryahetu, འབྲས་རྟགས་ | “fire because of smoke” on a mountain pass | trains cause-and-effect understanding |
| Nature sign | svabhāvahetu, རང་བཞིན་གྱི་རྟགས་ | “sound is impermanent because of being a product” | trains understanding of the nature of impermanent phenomena |
| Sign of nonobservation | anupalabdhihetu, མ་དམིགས་པའི་རྟགས་ | “sound is empty of being permanent because of being a product”; “the self of persons is empty of inherent existence because of being a dependent-arising”; “no smoke on that island because no fire” | set forth “for the sake of realising that all things are selfless” |
Two patterns of nature sign are recognised: MI pattern (sign and similar class are mutually inclusive — products and impermanent phenomena) and inclusion pattern (sign is properly included in the similar class — cats among mammals).
Consequences (thal ‘gyur) as the principal refutation tool
Distinct from the syllogism. The consequence’s formulator is bound only to fair derivation from the opponent’s stated position, not to the validity of the conclusion. A consequence has a point of clarification (གསལ་བ་) where a syllogism has a thesis (“promise”, དམ་བཅའ་). Standard use: against a stated pervasion “every P is Q,” the challenger flings “then the subject, this-P-which-is-not-Q, would be Q because of being P.” Three patterns of false pervasion are catalogued (Perdue pp. 417–419): broader-reason (indefinite reason), no-overlap (contradictory reason / perverse pervasion), partial-overlap. Of these, “indefinite reason” maps onto the Treasury’s ma nges pa’i rtags (inconclusive sign) and “contradictory reason” onto ‘gal rtags — connecting the Collected Topics fault-list directly to the Sa-gya doubt/wrong-cognition classification in non-valid-cognition.
Sources
- dreyfus-recognizing-reality — Chapter 18 (Can inference be valid?), Chapters 1, 7, and throughout
- westerhoff-dignaga-dharmakirti — triple mark, inference for oneself/others, fallibilism, examples (pp. 225–231)
- katsura-dignaga-lectures — pre-Dignāga history of the two traditions (vāda and pramāṇa), development of Points of Defeat, integration via svārthānumāna/parārthānumāna
- sonam-introduction-treasury — Sapan’s three types of correct reasoning, “endowed with” vs. “is” three modes, four types of response in debate, consequence vs. autonomous syllogism
- berzin-science-of-mind — members of logical proof (Nyāyapraveśa commentary); homogeneous and heterogeneous examples; being-something-produced; semblances of thesis and inference
- gorampa-pramanavartika — detailed scholastic treatment of the three perfect reasons (result, nature, nonapprehension) and the indispensable relationship, with the three criteria glossed as “property of the section” + “part pervades” + implied counter-pervasion (on PV 1.1); nonapprehension extended to eight and (via Pramāṇaviniścaya) ten subtypes
- perdue-buddhist-reasoning-debate — Ch. 6 (three modes, epistemological-verification requirement); Ch. 8 (three types of correct signs by entity, with examples for each); Ch. 22 (consequence/thal ‘gyur as the principal refutation tool and its relation to false-pervasion patterns)