Anumāna: Inference in Outline
Chapter Seven
Syllabus topic Module 1, "Anumāna"
Pages 19 to 21 of 378
In one line
Anumāna is knowing something you have not observed, from something you have, by way of a general connection between the two.
In the wording you can write in an examination: Nyāya Sūtra 1.1.5 defines inference as knowledge which is preceded by perception, and states that it is of three kinds: a priori, a posteriori, and commonly seen. Inference therefore always rests on a prior observation and on an invariable connection between what was observed and what is concluded.
Why it is called "preceded by perception"
Two things are meant by that phrase and both are worth having.
You must have observed the sign. Inference does not start from nothing. Something was perceived, and the conclusion hangs off it.
You must already have observed the connection. You conclude fire from smoke only because you have previously seen smoke together with fire, many times, and never smoke without fire. That prior observation is doing most of the work, and Module II's chapter on vyāpti is about exactly it.
So an inference has three ingredients: a thing you are talking about, a mark you observed in it, and a general connection you learned earlier. Take any one away and there is no inference.
The provision
Nyāya Sūtra 1.1.5, in Vidyabhusana's translation:
Inference is knowledge which is preceded by perception, and is of three kinds, viz., a priori, a posteriori and "commonly seen."
The three kinds, with the Sūtra's own examples
| Kind | Direction | Vidyabhusana's example |
|---|---|---|
| A priori | from cause to effect | "one seeing clouds infers that there will be rain" |
| A posteriori | from effect to cause | "one seeing a river swollen infers that there was rain" |
| Commonly seen | from one thing to another regularly found with it | "one seeing a beast possessing horns, infers that it possesses also a tail", or "one seeing smoke on a hill infers that there is fire on it" |
He then records a disagreement, and it is the sort of thing an examiner likes: Vātsyāyana, the earliest commentator, takes the third kind to be "not commonly seen" instead, and reads it as knowledge of something never observed at all, as when "observing affection, aversion and other qualities one infers that there is a substance called soul."
That is not a small difference. On the first reading the third kind is the weakest of the three, a correlation. On Vātsyāyana's it is the strongest claim inference can make, reaching a thing that could never be perceived. Say which reading you are using.
The three kinds as three kinds of program
The classification is not decorative. Each kind has a different reliability profile, and a modern system meets all three.
Cause to effect is prediction. You have the cause and you want the future. It is the least reliable of the three, because a cause can be blocked. Clouds do not always bring rain.
Anumāna: Inference in Outline
Effect to cause is diagnosis. You have the symptom and you want the fault. It is unreliable in a different way: several causes can produce the same effect. A swollen river may mean rain upstream, or a dam release.
Correlation is the one that looks safest and is not. Horns and tails go together in the animals you have seen. Nothing makes them go together. The connection is a summary of observation, not a mechanism, and it fails the first time you meet an animal outside the sample.
That third one is the classical statement of the standard modern warning, and it is stated in the Sūtra's own terms: the reason is erratic when there is no universal connection between it and what you concluded. [Hetvābhāsa: The Five Fallacies of the Reason] works that out.
Worked example
A student's laptop will not boot. Three inferences follow, one of each kind.
A priori. "The battery is at two per cent and the charger is not plugged in, so it is going to shut down." Cause to effect. Blockable: someone may plug it in.
A posteriori. "It powers on, the fan spins, and nothing appears on the screen, so the display or its cable has failed." Effect to cause. Several causes fit, so the inference narrows rather than settles.
Commonly seen. "Every laptop of this model I have seen with this fault had a failed cable, so this one has a failed cable." Correlation. It carries no mechanism and no guarantee that the sample was representative.
The three conclusions are not equally good, and knowing which kind you are making is how you know how much weight to put on it.
What it does NOT mean
It does not mean inference is second class. It is a pramāṇa in its own right, on every school's list including the shortest ones after the Cārvākas'.
It does not mean an inference is a guess. An inference whose general connection is sound is knowledge. The classical machinery exists precisely to check the connection, which is what [Anumāna: Vyāpti, and the Three Kinds of Inference] is about.
It does not mean "preceded by perception" makes it about the past. The perception that precedes it may be of the sign now; what is prior is the observation of the connection.
Where Module II takes this
Everything else about inference is there, and none of it is repeated here.
- The structure of an inference in five stated members, in [The Five-Member Syllogism].
- The invariable connection itself, vyāpti, and the vocabulary of subject, mark, positive and negative instance, in [Anumāna: Vyāpti, and the Three Kinds of Inference].
- The five ways the reason can be bad, in [Hetvābhāsa: The Five Fallacies of the Reason].
- Writing the whole thing in logical notation, in [The Five Members Written in Logical Notation].
Anumāna: Inference in Outline
Quick revision
- Nyāya Sūtra 1.1.5: inference is knowledge preceded by perception, of three kinds: a priori, a posteriori, commonly seen.
- Ingredients: a subject, a mark observed in it, and a previously established general connection.
- A priori is cause to effect, prediction; a posteriori is effect to cause, diagnosis; commonly seen is correlation.
- Vātsyāyana reads the third kind as "not commonly seen", reaching what can never be perceived, which is the opposite of a correlation.
- Correlation is the kind that looks safest and is weakest, because it carries no mechanism.
Test yourself
1. Define anumāna and give its three kinds with one example each.
Knowledge preceded by perception, drawn from an observed mark by way of a general connection. A priori: clouds, so rain. A posteriori: a swollen river, so there was rain. Commonly seen: a horned beast, so it has a tail.
2. Why does the definition insist that inference is "preceded by perception"?
Because two prior observations are needed: the mark must have been perceived now, and the general connection between mark and conclusion must have been observed earlier.
3. Which of the three kinds is a correlation, and what is its characteristic weakness?
"Commonly seen." It summarises past observation without supplying a mechanism, so it fails as soon as a case outside the sample appears.
4. What does Vātsyāyana do to the third kind, and why does it matter which reading you use?
He reads it as "not commonly seen", inference to something never perceptible at all, such as the soul from its qualities. It matters because on one reading the third kind is the weakest form of inference and on the other it is the boldest.
The rest of this subject
These notes are cut from the University's printed syllabus. Open the syllabus itself for the same subject.