The Nyayapravesa by Sankarasvamin, a disciple of Dignaga, is a primer of Buddhist logic. Translated by Xuanzang in 647, it became the foundational text of Chinese yinming; Kuiji's great sub-commentary launched the Tang study of logic.
It is organized by "eight gates and two benefits": proof, refutation, and their semblances (for others); perception, inference, and their semblances (for oneself). The first four concern debate; the last four concern cognition.
Its heart is the three-membered proof—thesis, reason, example. By cutting the old five members ("application" and "conclusion") down to three, Dignaga radically streamlined Indian logic. A reason must satisfy three characteristics or it collapses into one of the "pseudo-reasons."
Formal logic: the three members resemble major premise + minor premise + conclusion, but the "example" is decisive—it supplies both a general rule (if produced, then impermanent) and an instance (a pot). It is thus deduction fused with induction, not pure syllogism. Aristotle systematized the syllogism at roughly the same era; Dignaga achieved his equivalent independently in the East.
AI reasoning: the taxonomy of "pseudo-reasons" is a fallacy detector. LLM reasoning chains often commit the "inconclusive reason" (a premise with no necessary link to the conclusion)—the three-characteristics test is a classical template for validating a chain.
Traditional: Nalanda debate proceeded by three-membered proof and refutation. At the great assembly of Kanauj, Xuanzang posted his "True Consciousness-Only" proof for eighteen days; none could refute it.
Modern: when writing a proposal or making a technical call, force the three members—what is my thesis? my reason? does the reason satisfy "whatever has this reason has this thesis" (positive example) and "whatever lacks this thesis lacks this reason" (negative example)? Missing any member, the judgment fails. A weapon against gut-feel decisions.
The Nyayamukha is Dignaga's own (c. 480–540) logical masterwork, translated by Xuanzang in 649; with the Nyayapravesa it forms the pair of Chinese yinming treatises. "Gateway to logic" means the entry to correct inference.
Its great contribution is the "nine reasons"—exhaustively enumerating a reason's distribution across similar and dissimilar instances into nine combinations, sorting valid from pseudo-reasons. Only "definitely present in similar cases, wholly absent in dissimilar cases" (the 2nd and 8th cases) is valid. This is a mechanical, enumerable decision procedure.
From it Dignaga distilled the three characteristics of a valid reason: pervading the thesis's subject; definitely present in similar cases; wholly absent in dissimilar cases. All three present, the inference is sound.
Computer science: the nine reasons are a 3×3 enumeration of "similar {present, absent, both} × dissimilar {present, absent, both}"—essentially a truth table / decision table testing validity by exhaustion. This is isomorphic to the truth-table method and to branch coverage in programs. In the sixth century Dignaga was already doing "combinatorial enumeration plus case adjudication."
Popper's falsifiability: "wholly absent in dissimilar cases" demands that no counterexample exists anywhere the thesis fails. This is nearly falsifiability: the strength of an argument lies in its exclusion of counterexamples.
Traditional: with the nine reasons as his yardstick, Dignaga tested the proofs of rival schools one by one, purging "inconclusive reasons" and recasting Indian logic.
Modern: facing a correlation claim ("use this tool and efficiency rises"), apply "wholly absent in dissimilar cases": is there a case that did NOT use the tool yet was still efficient? If a counterexample exists, "tool → efficiency" is an inconclusive reason; correlation is not causation. A scalpel for the data age.
Dharmakirti (c. 600–660) is the summit of Indian epistemology after Dignaga. The Nyayabindu ("a drop of logic") is his compact digest; its complete Sanskrit survives, making it a linchpin for studying Indian logic.
To Dignaga's "perception is free of conceptual construction" Dharmakirti adds "non-erroneous"—not merely non-conceptual but undistorted (a diseased eye seeing two moons is non-conceptual yet erroneous, hence not true perception). He thus fixes: true perception grasps only the particular (svalaksana, a thing's unique here-and-now reality); inference grasps only the universal (samanyalaksana, the abstract conceptual form).
"Cognition is twofold" because "the knowable is twofold"—particular and universal. Whatever registers neither is not valid cognition. A minimal yet rigorous epistemology.
Neuroscience: "perception free of conceptual construction" maps onto the pre-conceptual stage of perception—raw input in primary sensory cortex before categorization; once "this is a cup" intervenes, we are already in inference/universal. Predictive processing's split between "bottom-up sensory signal vs top-down conceptual prior" runs strikingly parallel to particular vs universal.
Epistemology: Dharmakirti restricts "knowledge" to non-erroneous direct experience plus valid inference, holding the rest in doubt—almost isomorphic to modern science's twin pillars of empirical observation and logical inference.
Traditional: the Tibetan Gelug school ranks the Nyayabindu and Pramanavarttika among its great treatises, training monks in reasoning as groundwork for Madhyamaka and Yogacara.
Modern: separate "what you saw" from "what you assumed." A colleague frowns—"frown" is perception (particular); "he's displeased with me" is inference (universal, and often mistaken). Deliberately insert a gap: admit only the perception first, then infer with caution. Cognitive surgery for interpersonal misjudgment and anxiety.
Apoha ("exclusion of the other," rendered in Chinese as "exclusionary denotation") is the subtlest node of Dignaga's epistemology, answering how concepts and words refer. His answer is negative: language cannot positively grasp the ineffable particular; it fixes meaning only by "excluding the rest." "Cow" is simply "the exclusion of non-cow."
This yields the distinction between exclusionary and affirmative denotation. Zongmi's Preface to the Collected Chan Sources puts it best: "Exclusion rules out what a thing is not; affirmation displays what it is; exclusion screens off the rest, affirmation points directly at the thing." When scriptures call suchness "unborn, unceasing, undefiled, unpurified," this is all exclusion—layered negation—because reality is beyond words: one can only exclude, not affirm.
Hence all conceptual knowledge is a construction of universals, a product of drawing boundaries and excluding, forever one layer removed from the immediate particular. This converges completely, at the level of epistemology, with Madhyamaka's "the path of words is cut off" and Prajna's "nothing to attain."
Machine learning: apoha is virtually the philosophical prototype of discriminative classification—a machine recognizes "cat" not by positively defining the essence of cat but by learning a decision boundary that excludes "non-cat." Word embeddings are even plainer: a word's meaning is fixed by its difference/contrast with other words, not by an inner essence—exactly "meaning disclosed through exclusion."
Structural linguistics: Saussure's "in language there are only differences, without positive terms" echoes apoha across thirteen centuries—meaning is relational and negative, not substantial.
Traditional: logicians used apoha to refute the non-Buddhist "word-essence" view (that words are intrinsically bonded to real substances), showing that names are only posited universals.
Modern: when you label yourself—"I'm an introvert," "I'm bad at management"—remember these are boundary-universals drawn by exclusion, not your particular. A label only states "what you are not"; it never truly captures "who you are." Loosening your grip on labels as real is a step back from exclusion toward the wordless particular.