Gaṅgeśa on Epistemic Luck

Abstract

This essay explores a problem for Nyāya epistemologists. It concerns the notion of pramā. Roughly speaking, a pramā is a conscious mental event of knowledge-acquisition, i.e., a conscious experience or thought in undergoing which an agent learns or comes to know something. Call any event of this sort a knowledge-event. The problem is this. On the one hand, many Naiyāyikas accept what I will call the Nyāya Definition of Knowledge, the view that a conscious experience or thought is a knowledge-event just in case it is true and non-recollective. On the other hand, they are also committed to what I shall call Nyāya Infallibilism, the thesis that every knowledge-event is produced by causes that couldn’t have given rise to an error. These two commitments seem to conflict with each other in cases of epistemic luck, i.e., cases where an agent arrives a true judgement accidentally or as a matter of luck. While the Nyāya Definition of Knowledge seems to predict that these judgements are knowledge-events, Nyāya Infallibilism seems to entail that they aren’t. In this essay, I show that Gaṅgeśa Upādhyāya, the 14th century Naiyāyika, solves this problem by adopting what I call epistemic localism, the view that upstream causal factors play no epistemically significant role in the production of knowledge.

Introduction

This essay explores a problem for Nyāya epistemologists. It concerns the notion of pramā. Roughly speaking, a pramā is a conscious mental event of knowledge-acquisition or learning, i.e., a conscious experience or thought in undergoing which an agent learns or comes to know something. Suppose I see that there is a white picket fence outside my window. In undergoing this perceptual awareness, I learn or come to know that there is a white picket fence outside my window. So, this perceptual awareness is a pramā. Similarly, suppose I see that the sky is overcast and, on that basis, consciously infer that it will rain today. In making that inferential judgement, I may learn or come to know that it will rain today. If that is the case, this inferential judgement is a pramā. Call any such event of knowledge-acquisition a knowledge-event.

The problem is this. Many Naiyāyikas accept:

The Nyāya Definition of Knowledge. An awareness-event (jñāna, i.e., a conscious experience or thought) counts as a knowledge-event if and only if it is a true non-recollective awareness-event (yathārthānubhava).¹

But these Naiyāyikas are also committed to:

Nyāya Infallibilism. Any knowledge-event is produced by a causal complex (kāraṇasāmagrī or collection of causes) that couldn’t give rise to an error.

These two commitments seem to conflict with each other in cases of epistemic luck, i.e., cases where an agent undergoes an awareness-event that is true accidentally or as a matter of luck.² Consider four cases that we will discuss throughout this paper.

Mist and Fire. I look at a hill and see what looks like smoke emerging from it. So, I judge that there is smoke on the hill. I am wrong: all I see is a wisp of mist. I had previously observed (in kitchens, etc.) that smoke is always accompanied by fire. On the basis of those observations, I had judged that, wherever there is smoke, there is fire. Now, I remember that generalization. So, I conclude that there is fire on the hill. My judgement is true: there is fire on the hill.

Horns and Cows. From a distance, I see an animal with horns. Earlier, I had observed many cows with horns. On the basis of these observations, I judged that all animals with horns are cows. Now, I recall that. generalization So, I conclude that the animal is a cow. My judgement is true: the animal I see is a cow.

The Mistaken Deceiver. You think that there is no pot in the next room. You want to deceive me. So, you tell me, “There is a pot in the next room.” Since I have no reason to distrust you, I take your utterance at face value. So, I judge that there is a pot in the next room. My judgement is true: there is a pot in the next room.

The Parrot. A parrot is hidden behind a door, and it can randomly string together words to form sentences. I don’t know this. On this occasion, imitating the voice of a friend, the parrot utters the sentence, “There is a pot in the next room.” Since I think that my friend is behind the door and have no reason to distrust her, I take the utterance at face value. So, I judge that there is a pot in the next room. My judgement is true: there is a pot in the next room.

In each case, I undergo a non-recollective awareness-event that is true as a matter of accident or luck.³ Given the Nyāya Definition of Knowledge, it should count as a knowledge-event. But it’s hard to shake off the intuition that the causes of the awareness could easily have led to a mistake. If this intuition is right, then the Naiyāyikas are in trouble. For, if the Nyāya Definition of Knowledge is correct, then Nyāya Infallibilism is false in these cases. This is what I shall call the problem of epistemic luck.

In this essay, I lay out Gaṅgeśa Upādhyāya’s (14th century CE) solution to this problem in Tattvacintāmaṇi (TCM). To solve this problem, Gaṅgeśa—as well as his commentators such as Jayadeva Miśra (15th century CE)—adopted a more permissive form of Nyāya Infallibilism. This form of infallibilism involves what I will call epistemic localism, the thesis that upstream causal factors (e.g, a speaker’s epistemic standing in the case of testimony) don’t play any epistemically significant role in the production of knowledge.⁴ By downplaying the epistemic role of such factors, Gaṅgeśa and his commentators were able to treat epistemically lucky inferential and testimonial awareness-events as knowledge-events. This, in turn, allowed them to resolve the tension between the Nyāya Definition of Knowledge and Nyāya Infallibilism.

Why is this significant? First, my discussion reveals that some modern interpretations of Gaṅgeśa are simply wrong. Mukhopadhyay (1992, p. 285) and Phillips (2012, pp. 84–85) think that Gaṅgeśa does not take the judgements in The Parrot and The Mistaken Deceiver to be knowledge-events. In what follows, I show that there is little textual support for this claim. Second, if I am right, Gaṅgeśa’s treatment of epistemic luck in TCM reveals a radical shift in the attitudes of Naiyāyikas towards cases like Mist and Fire. While earlier Naiyāyikas seem to rule them out from the class of knowledge-events, later Naiyāyikas (from Gaṅgeśa onwards) certainly do not. This forces these later Naiyāyikas to revise their epistemological commitments quite radically.

This essay has five parts. I will begin by describing the stance of early Naiyāyikas on epistemically lucky awareness-events. As I will show, they accepted a version of Nyāya Infallibilism that would prevent them from recognizing such awareness-events as knowledge-events. I will then turn to Gaṅgeśa. I will consider whether his theory of inference and testimony would allow him to treat my judgements in cases like Mist and Fire, etc. as knowledge-events. The answer, I will argue, is “Yes.” Next, I will explain how Gaṅgeśa frames the problem of epistemic luck. I will then lay out the two distinct solutions that he offers to this problem. The second of these is an instance of the approach that I have called epistemic localism. I will go on to show how Gaṅgeśa’s commentator, Jayadeva, extends this approach to cases that Gaṅgeśa doesn’t address.

Background: Nyāya Infallibilism

Most of Gaṅgeśa’s Nyāya predecessors subscribed to an infallibilist conception of epistemic instruments (pramāṇa). For these writers, a knowledge-event is a true, non-recollective awareness-event, and an epistemic instrument is an instrument or means (karaṇa) by which any such awareness-event arises. An instrument by which an effect arises is the maximally efficient (sādhakatama) cause of that effect. Though there is some disagreement amongst these Naiyāyikas on what maximal efficiency actually consists in, many of them agree that a maximally efficient cause of an effect is a cause such that, when it occurs, the effect must immediately follow. In this sense, an instrument that gives rise to an effect or result (phala) is excluded from a lack of connection with its result (phalāyogavyavacchinna).⁵ So, if an epistemic instrument is a maximally efficient cause of a knowledge-event, it must be an event or an entity such that its occurrence is immediately (as a matter of necessity) followed by a knowledge-event. Udayana states this conception of an epistemic instrument succinctly in the fourth chapter (stavaka) of Nyāyakusumāñjali (NKu).

According to the view of Gotama [i.e., author of the Nyāya-sūtra], a knowledge-event is a correct discrimination (samyakparicchitti). Moreover, the status of being a knower (pramātṛtā) consists in possessing that knowledge-event, while the status of being an epistemic instrument consists in being excluded from a lack of connection with that knowledge-event (tadayogavyavaccheda).⁶

For these Nyāya writers, what distinguishes an epistemic instrument from other instruments of awareness is that it never gives rise to an error; in this sense, it doesn’t err from its object (arthāvyābhicārin).⁷ In his sub-commentary on the Nyāyasūtra (NS), Nyāyavārttikatātparyapariśuddhi (NVTP), Udayana explains the idea as follows:

An epistemic instrument is simply the instrument for a knowledge-event…A knowledge-event is a non-erroneous apprehension (aviparītopalabdhi)… Instrumenthood is maximal efficiency (sādhakatamatva). However, that is specified simply by a specific event (kriyā). Moreover, in this case, that event has the characteristic of being a knowledge-event. Therefore, being an epistemic instrument consists in not erring, which is characterised as the property of producing non-erroneous non-recollective awareness-events.⁸

If Udayana is right in his description of earlier Nyāya, then early Naiyāyikas are committed to Nyāya Infallibilism: since all knowledge-events must arise from some epistemic instrument and no epistemic instruments could fail to yield true awareness-events, the causal complex that gives rise to a knowledge-event couldn’t give rise to an error.

This commitment to infallibilism drove some early Naiyāyikas, such as Bhaṭṭa Jayanta and Udayana, towards a virtue-theoretic conception of knowledge-events.⁹ On this view, if an agent arrives at a knowledge-event, the causal complex that gives rise to her awareness-event must include certain positive factors—called epistemic virtues (guṇa)—that guarantee the truth of the resulting awareness.¹⁰ Though these early writers are reticent on which epistemic virtues are necessary for which kinds of knowledge-events, their treatment of individual epistemic instruments strongly suggests that, if an instrument of awareness is to serve as an epistemic instrument, it must possess epistemic virtues that are typically absent from cases of epistemic luck.¹¹ Let’s see why.

Start with inference (anumāna). Suppose I see that there is smoke coming out of a hill. I had judged earlier that, wherever there is smoke, there is fire. Now, I remember that. So, I conclude that there is fire on the hill. Any such inference involves three components: a target (sādhya), a reason (hetu) or an inferential mark (liṅga), and a site (pakṣa). The target is the thing that is inferred; here, it’s fire. The reason or the inferential mark serves as evidence for the target; here, it’s smoke. The site is the place where the presence of the target is inferred; here, it’s the hill. For Naiyāyikas beginning with Uddyotakara, the process by which an inferential knowledge-event arises involves three steps.¹² First, the agent becomes aware of the site as possessing the reason. For example, in this case, I simply see that there is smoke on the hill. This is called the awareness of the reason as a property of the site (pakṣadharmatājñāna). Then, the agent recalls that there is a relation of pervasion (vyāpti) between the reason and the target. In this case, I recall that, wherever there is smoke, there is fire; thus, I recall that fire pervades smoke. This step is called the recollection of pervasion (vyāptismaraṇa). Finally, the agent combines these two bits of information in a single judgement: she judges that the site contains a reason that is pervaded by the target. For example, in this case, I may judge that the hill possesses smoke that is pervaded by fire. This is called a subsumptive judgement (parāmarśa). This gives rise to the inferential knowledge-event (anumiti) that there is fire on the hill.

According to these Naiyāyikas, a good or non-defective reason (saddhetu) must have five characteristics: (1) it must be present in the site, (2) it must be present at a similar site (sapakṣa), i.e., a place where the target is observed to be present, (3) it must be absent from a dissimilar site (vipakṣa), i.e, a place where the target is absent, (4) it must be such that the relevant target isn’t already proved (by some other epistemic instrument) to be absent from the site, and (5) it must be such that there is no competing (and equally strong) inferential mark that supports the opposite thesis, i.e., that the target is absent from the site. Any inferential mark that fails to satisfy any of these conditions is said to be a pseudo-reason (hetvābhāsa).¹³ An inferential mark that is absent from the site is unproved (asiddha or sādhyasama); when it isn’t present at any similar site, it is said to be incompatible (viruddha); when it is present at a dissimilar site, it is said to be deviating (savyabhicāra); when the relevant target is proved to be absent from the site, the inferential mark is rebutted (bādhita or kālātīta); finally, when there is an equally strong competing inferential mark, the inferential mark is said to be counterbalanced (satpratipakṣa or prakaraṇasama). Notice that the inferential marks involved in Cows and Horns and Mist and Fire are pseudo-reasons. In Mist and Fire, the smoke is the inferential mark, while the fire is the target. But since the smoke is actually absent from the site, i.e., the hill, the inferential mark ends up being unproved; in particular, later Naiyāyikas call this kind of unproved reason unproved by nature (svarūpāsiddha), because the inferential mark, by its own nature, isn’t proved to be present in the site.¹⁴ In Cows and Horns, the horns serve as the inferential mark, while the target is cowhood. Since animals other than cows can have horns, the inferential mark is deviating because it is present at a dissimilar site.

Importantly, early Naiyāyikas, such as Vātsyāyana and his sub-commentators, claimed that a genuine inference—i.e., an episode of reasoning that yields an inferential knowledge-event—must involve an epistemically virtuous inferential mark, i.e., an inferential mark that satisfies (at least some of) the conditions laid out above. In particular, it cannot involve an inferential mark that is either deviating or unproved: an episode of reasoning that involves such a defective inferential mark will merely be a pseudo-inference (anumānābhāsa).¹⁵ This immediately implies that my judgements in Mist and Fire and Cows and Horns cannot be inferential knowledge-events.

Let us now turn to the case of testimony (śabda). According to NS 1.1.7, “Testimony is the teaching of a trustworthy person.” In his commentary, Vātsyāyana explains the notion of trustworthiness as follows:

Certainly, a trustworthy person is a teacher who is directly acquainted with existent objects (dharma) and is motivated (prayukta) by the desire to convey things as they have been perceived. The direct acquaintance with an object is the attainment (āpti) of the object. Since he undertakes action on the basis of that, he is trustworthy (āpta).¹⁶

On Vātsyāyana’s view, a trustworthy person must have three features. First, she must have been directly acquainted with the content that she wishes to communicate. Second, she must have compassion for other beings to whom she is communicating this content; in other words, her utterances must be motivated by a desire to help others. Third, she must want to communicate how things are exactly the way she herself has found them. Note that this notion of trustworthiness is quite strong: it seems to imply that a palaeontologist who has never encountered dinosaurs but has made lots of good inferences about them still cannot be treated as trustworthy with respect to them. While Uddyotakara seems to largely agree with this characterisation of the trustworthy speaker (NV 257.19-258.2), Vācaspati Miśra seems to relax, or reinterpret, some of these requirements (perhaps to accommodate a wider class of testimony). Instead of taking the requirement of direct acquaintance literally, he interprets this requirement as follows: “A person who is directly acquainted with, i.e., has determined by means of a firm epistemic instrument, the existent objects (dharma), i.e., the entities (padārtha) that are useful for the attainment of benefits and the avoidance of harms, is said to be so [i.e., trustworthy].”¹⁷ Thus, on Vācaspati’s view, even a palaeontologist who has never seen dinosaurs could still count as trustworthy with respect to them.

Even if we accept Vācaspati’s weakened conception of trustworthiness, this Nyāya view entails that a piece of testimony can serve as an epistemic instrument only if its speaker possesses at least two virtues: she must have gained by means of an epistemic instrument a correct awareness of the content that her utterance conveys, and she must have the desire to sincerely convey the truth. Both of these virtues are (arguably) missing in cases like The Mistaken Deceiver and The Parrot. In both cases, the speaker hasn’t determined the content of the utterance to be true by means of any epistemic instrument, and lacks the desire to convey the truth. So, the relevant linguistic utterances cannot be treated as epistemic instruments. Therefore, the resulting judgements cannot be knowledge-events.

The upshot: the early Naiyāyikas’ commitment to Nyāya Infallibilism would have prevented them from treating epistemically lucky awareness-events as knowledge-events.

This strongly suggests that the conception of knowledge-events that these Naiyāyikas were working with was closer to our contemporary notion of knowledge. According to a simple account of knowledge, a belief (or, more generally, an information-bearing state) has the status of knowledge just in case it is true but not as a matter of luck. On the early Nyāya view, a knowledge-event is simply a conscious non-recollective experience or thought in undergoing which one non-luckily or non-accidentally acquires true information. Thus, it’s plausible to think of knowledge-events as conscious mental events of learning or knowledge-acquisition. This way of connecting this conception of knowledge-events to the notion of knowledge explains at least two aspects of early Nyāya epistemology. First, it explains why these Naiyāyikas thought that the causes underlying knowledge-events couldn’t give rise to any error. Second, it also explains why they thought that knowledge-events have to be non-recollective: since reliable recollective awareness-events (typically) only help us retrieve information that we had already acquired earlier, we don’t independently acquire any true information through them. Thus, even when they are true (and reliable), they cannot be events of knowledge-acquisition.

What unifies the early Nyāya approach to knowledge-events is a form of epistemic anti-localism. According to the early Naiyāyikas, the production of inferential and testimonial knowledge depends on the transmission of knowledge from other causally upstream awareness-events that belong either to the agent herself or to some other agent. For example, in the case of inference, the production of an inferential knowledge-event depends on whether the initial steps of the relevant cognitive process—the agent’s initial awareness of the reason as a property of the site or her awareness of pervasion—are themselves knowledge-events. An inferential judgement can be a knowledge-event only if these initial awareness-events are. So, the epistemic status of these causally upstream awareness-events matters. Similarly, in the case of testimony, the epistemic virtues of the speaker play an important role: unless the speaker undergoes a knowledge-event regarding the content that she wishes to convey, the resulting testimonial awareness cannot be a knowledge-event. Once again, the production of knowledge in this case depends on the epistemic status of a causally upstream awareness-event. This form of anti-localism explains why these early Nyāya authors wouldn’t treat epistemically lucky awareness-events as knowledge-events. As we shall see later, Gaṅgeśa rejects this form of anti-localism.

Gaṅgeśa on Inference and Epistemic Luck

In Khaṇḍanakhaṇḍakhādya (KKh), Śrīharṣa showed that the early Nyāya theory of knowledge-events doesn’t handle cases of epistemic luck well. On the one hand, since Naiyāyikas like Udayana were committed to the Nyāya Definition of Knowledge, they couldn’t rule out epistemically lucky inferential judgments (like my judgements in Mist and Fire and Cows and Horns) from the class of knowledge-events. Yet, given their other commitments, these Naiyāyikas also couldn’t treat these as knowledge-events. For none of the characteristic epistemic virtues that are supposed to accompany inferential knowledge-events are present in such cases. Here, I won’t rehearse Śrīharṣa’s arguments. ¹⁸ In this section, my aim will be to show that Gaṅgeśa partially concurs with Śrīharṣa: he agrees that, if his own preferred version of Nyāya Definition of Knowledge is right, inferential judgements based on pseudo-reasons cannot be excluded from the class of knowledge-events.

In TCM, Gaṅgeśa endorses a version of the Nyāya Definition of Knowledge.

A non-recollective awareness of something at a place where it exists is a knowledge-event. Alternatively, it is an awareness that attributes a certain qualifier to something that possesses that qualifier. The awareness of something at a place where it doesn’t exist, or an awareness that attributes a qualifier to something that possesses the absence of that qualifier is not a knowledge-event (apramā).¹⁹

We can state this more precisely.

Gaṅgeśa’s Definition of Knowledge. An awareness-event is a knowledge-event if and only if

• (i) it is a non-recollective awareness, and

• (ii) if it attributes a qualifier (prakāra) x to a qualificand (viśeṣya) y by a relation R (or presents y as characterised by x in virtue of the relation R), then x is related to y by R.

To see how the definition works, consider a case where I see a banana before me as yellow. Here, the banana is the qualificand, while the yellow colour I perceptually attribute to it is the qualifier. My perceptual awareness is a knowledge-event just in case that yellow colour that I perceptually attribute to the banana actually is present in the banana.

Apply this definition to cases like Mist and Fire and Horns and Cows. In these cases, the relevant agent forms inferential judgements on the basis of pseudo-reasons. Can such an inferential judgement be a knowledge-event? Given Gaṅgeśa’s definition of knowledge-events, the answer (we might think) has to be a resounding “Yes.” In Mist and Fire, for example, if the qualificand of my inferential judgement is the hill and the qualifier is fire, then my judgement indeed is a knowledge-event by Gaṅgeśa’s lights. For there is fire on the hill. However, Gaṅgeśa’s discussion of these cases reveals that things aren’t as straightforward as they appear. Let’s see why.

Non-probativity

A good place to begin will be Gaṅgeśa’s general definition of pseudo-reasons in Anumānakhaṇḍa. Gaṅgeśa offers three distinct definitions:

In that context, the property of being a pseudo-reason (hetvābhāsatva) is (i) the property of being an intentional object of a true awareness which  is the counterpositive (pratiyogin) of an absence that serves as a cause of an inferential knowledge-event (anumiti), or (ii) the object such that an awareness of the inferential mark, which has that object as an intentional object, is an impediment (pratibandhaka) to an inferential knowledge-event, or (iii) the property of being something which, when it is being apprehended (jñāyamāna), serves as an impediment to an inferential knowledge-event.²⁰

The first definition simply says that, if an agent were to correctly judge that the inferential mark of a putative inference had the property of being a pseudo-reason, she wouldn’t be able to arrive at the relevant inferential knowledge-event. In this sense, the absence of such a correct judgement (which has as its intentional object the property of being a pseudo-reason) is a cause of the inferential knowledge-event. The second definition restates that very idea in a slightly different way: it says that, if an agent were aware of an inferential mark as possessing the property of being a pseudo-reason, then that awareness would prevent the relevant inferential knowledge-event from arising. Finally, the third definition says that the property of being a pseudo-reason, when it is being apprehended, itself prevents the relevant inferential knowledge-event from arising. Notice that all of these three definitions gesture at the same idea: namely, that if an agent were to (correctly) judge that a putative inferential mark is a merely apparent or defective reason, then she couldn’t arrive at an inferential knowledge-event on the basis of it. In other words, such a judgement serves as a source of defeating evidence, which, in turn, prevents the agent from (rationally) making the inferential judgement that she would have made otherwise. Therefore, a pseudo-reason isn’t, by definition, an inferential mark that prevents the agent from arriving at inferential knowledge-events. While this doesn’t immediately tell us whether an inferential knowledge-event can be based on a pseudo-reason, it does clear some theoretical space for saying so.

In a later section of Anumānakhaṇḍa—named “asādhakatāsādhakaprakaraṇa”—Gaṅgeśa takes up this question. Suppose that, in the context of a debate, a participant points out that her opponent’s argument relies on a pseudo-reason. This reply counts as a good response (saduttara) to that argument. Why is this so? A plausible answer: in pointing out a pseudo-reason, this participant is able to prove that the relevant inferential mark is non-probative (asādhaka), i.e., that it doesn’t prove the presence of the relevant target. This, in turn, blocks the opponent’s argument. What is non-probativity (asādhakatva)? Before presenting his preferred proposal, Gaṅgeśa considers, and rules out, a number of proposals about what non-probativity could be. We will focus on a proposal to which Gaṅgeṡa devotes the greatest amount of attention: namely, that non-probativity is simply the property of not producing a correct awareness of the site as characterised by the target (samīcīnasādhyaviśiṣṭapakṣapratyayājanakatva).

Gaṅgeṡa notes a troubling consequence of this proposal. If this view is true, and if we agree that defective reasons are non-probative, then one cannot arrive at a true inferential judgement by reasoning from a pseudo-reason. This would imply that, since the inferential mark in Mist and Fire is unproved by nature (svarūpāsiddha), I cannot arrive at a true inferential judgement in that situation. But my final judgement is indeed correct in that case! So, the proposal fails. Gańgeśa explains this idea as follows:

It [i.e. non-probativity] is also not the property of not producing a correct awareness of the site as characterised by the target. For (i) in cases involving rebutted (bādha), incompatible (viruddha) and unproved (asiddha) reasons, when the site isn’t a locus of the target, a true (satya) awareness of the target isn’t well-established, and (ii) an inferential awareness of fire (vahnyumiti) in a site that contains fire—which arises from an erroneous awareness of mist as smoke—is true (satya).²¹

Gaṅgeśa’s argument is this. At least, in cases where the site doesn’t contain the target, some pseudo-reasons, e.g., rebutted, incompatible or unproved reasons, cannot yield any true awareness of the site as characterised by the target. However, in a case like Mist and Fire, where the target is genuinely present in the site, even an inferential mark that is unproved in the site (svarūpāsiddha) can give rise to a true inferential awareness.

Gaṅgeśa considers two distinct strategies for resisting this conclusion.

Strategy 1. In Mist and Fire, the inferred target (i.e., the fire) isn’t present in the site (i.e., the hill). So, the inferential judgement is false.

Strategy 2. In Mist and Fire, the site (i.e., the hill) or the target (i.e., the fire) appears in the final inferential judgement as connected to the relevant inferential mark (i.e., smoke) in a certain way. But, since the inferential mark is absent from the site, the final judgement is false.

Both these strategies, according to Gaṅgeśa, are unsuccessful.

Strategy 1

Let’s begin with Strategy 1.

What makes this strategy attractive from a Nyāya standpoint? The Naiyāyikas are realists about intentional objects of awareness: if anything is an intentional object of a conscious thought or experience, it must exist independently of that thought or experience. This compels them to accept a misplacement theory of error (anyathākhyātivāda). According to this theory, when an agent misperceives an object o as characterised by some property F, the erroneous awareness ascribes to that object a property, i.e., F, which she has earlier veridically perceived elsewhere. So, when I misperceive a mother-of-pearl as silver, my erroneous perceptual awareness ascribes to the mother-of-pearl the property of silverhood that I have encountered elsewhere. Let’s now see how this applies to cases like Mist and Fire. In Mist and Fire, I misperceive the hill as containing smoke. This may be smoke I have seen elsewhere, e.g., in the kitchen. If this is right, one could argue that my final judgement in these cases is false. For, if the smoke that I ascribe to the hill is absent from the hill, then the fire I ascribe to the hill—insofar as it accompanies the smoke I perceptually ascribe to the hill—also cannot be present on the hill. This is precisely the conclusion that defenders of Strategy 1 support.

Partially following Śrīharṣa’s treatment of such cases in Khaṇḍanakhaṇdakhādya,²² Gaṅgeśa offers two responses to this strategy.

Moreover, it is not the case that it is simply some other fire that appears in that awareness. For, there is no evidence for this, since that fire may be recognized [later], and such an inferential awareness is possible in a case that involves just one individual [as the target].²³

According to Gaṅgeśa, there is no good reason for us to think that the inferred target in cases like Mist and Fire is in fact missing from the site. In Mist and Fire, for example, after inferentially judging that there is fire on the hill, if I walk up to the fire that is actually present on the hill, I can—at least seemingly—recognize it as the fire that I inferred. If we take this to be a genuine case of recognition, then we must concede that the fire I inferred is indeed the fire that I now see. This argument isn’t all that convincing: the opponent could simply deny that this is a genuine case of recognition. But Gaṅgeśa’s second response (which is borrowed from Śrīharṣa) is more persuasive: he points out that the proposal in question cannot succeed when it comes to an episode of reasoning which involves just one object as its target. Consider Horns and Cows: here, the target is a universal (sāmānya), i.e., cowhood.²⁴ The opponent cannot argue that the cowhood that I ascribe to the animal is distinct from the cowhood that is present in the animal that I see. For there is just one such property!

Despite these problems, the opponent might insist that a version of this strategy could still be made to work. In a case like Mist and Fire, I take the inferred fire to be identical to something that pervades the smoke that I saw. Thus, in my final inferential judgement, something that pervades the defective inferential mark (i.e., the smoke) appears as fire. But, if there is really no smoke on the hill, the fire that is present on the hill cannot pervade the defective inferential mark. Similarly, in Cows and Horns, I take cowhood to be identical to something that pervades the possession of horns. Thus, in my final inferential judgement, something that pervades the defective inferential mark (i.e., the possession of horns) appears as cowhood. But, if cowhood doesn’t really pervade the possession of horns, the cowhood that is present in the animal cannot pervade the defective inferential mark. So, in each case, the inferential judgement will end up being false.

Once again, Gaṅgeśa thinks that this strategy cannot succeed.

[The opponent:] With regard to the defective inferential mark, one is aware of the identity with something that is pervaded by fire. And, thus, with regard to fire, one is also aware of the identity with something that pervades the defective inferential mark. Otherwise, there wouldn’t be an awareness of the defective inferential mark as pervaded by fire. In this manner, something that pervades the defective inferential mark appears as fire. So, the inferential awareness of fire is simply untrue.

[Reply:] No. For, in a case where an inferential awareness [of fire] arises due to the superimposition of smoke on to light that is pervaded by fire, there would be the consequence that the inferential awareness doesn’t fail to be true despite there being a variety of unprovedness, because there is an identity between what pervades light and what pervades smoke.²⁵

Gaṅgeśa is imagining a case like this.

Light and Fire. I look at a hill and mistake what is in fact light to be smoke emerging from the hill. In fact, there is no smoke on the hill. Earlier, on numerous occasions, I had noticed, in kitchens, etc., that smoke goes hand in hand with fire. On the basis of those observations, I had judged that wherever there is smoke, there is fire. Now, I remember that. So, I conclude, “There’s fire on the hill.” My judgement is true: there is fire on the hill.

In this case, the inferential mark—i.e., smoke—is defective: it suffers from the fault of being unproved by its own nature (svarūpāsiddha), since it is absent from the site. According to the opponent, something that pervades that defective reason appears as fire in my final judgement. But since there is light on the hill, the hill contains fire, and that fire pervades the light present on the hill. Since the fire that pervades the smoke isn’t distinct from the fire that pervades the light, the final inferential judgement (which ascribes to the hill a fire that pervades the smoke) will come out true.

Strategy 2

Let’s move on to Strategy 2. According the defenders of this strategy, when I conclude that there is fire on the hill in Mist and Fire, the inferential mark somehow appears as an intentional object in my final inferential judgement. For example, the content of my judgement could be expressed in one of the following two ways:

• (1) The hill that contains smoke also contains fire.

• (2) The hill contains the fire that pervades smoke.

Both these judgements are false. (1) is false because the hill doesn’t contain any smoke; (2) is false because the fire that is present on the hill doesn’t accompany (and therefore doesn’t pervade) smoke.

Gaṅgeśa thinks that this strategy is hopeless.²⁶ First of all, he thinks that there is no evidence for thinking that the reason actually appears as an intentional object in the final inferential judgement. Second, he invokes a case like this.

Darkness. I can’t tell whether darkness is a positive entity (bhāva) like a material object or its size or colour, or a negative entity (abhāva), e.g., a mere absence of light. I have noticed that both positive and negative entities are knowable (prameya). So, despite being uncertain about whether darkness is a positive or a negative entity, I reason like this, “Is it a positive entity or a negative entity? In both cases, it’s knowable.” Thus, I reason from both those properties—positivity (bhāvatva) and negativity (abhāvatva)—to the conclusion that darkness is knowable.

On one way of reconstructing the reasoning, it involves a conjunctive inferential mark which combines these two mutually incompatible properties. If this were true, then the inferential mark would be unproved by nature (svarūpāsiddha). For nothing is both a positive entity and an absence. But the inferential judgement that darkness is knowable is undeniably true. But, if the opponent were to say that the inferential mark appears as an intentional object of this judgement, she would be forced to say that this judgement is false. That is the problem.

In response, the opponent might argue that the inferential mark in this case isn’t a conjunction of both positivity and negativity, but in fact is disjunctive, i.e., the property of being positive or negative (bhāvābhāvānyataratva). In reply, Gaṅgeśa makes two points. First, he notes that an inference like this could be made even by someone who isn’t aware of any pervasion between this disjunctive property and knowability. Moreover, he says that including “either…or…” (anyataratva) into the specification of the inferential mark is an unnecessary qualification (vyarthaviśeṣaṇa), presumably because it would not rule anything out from the scope of the inferential mark (since everything is either positive or negative).²⁷ The opponent also cannot claim that, in this case, the inferential judgement is true because it has the content, “Darkness, which is characterised by some property (either positivity or negativity) that is pervaded by knowability, is knowable.” For, the analogous judgement in Mist and Fire, “The hill, which is characterised by some property that is pervaded by fire, contains fire,” is also true.²⁸

In response to all these problems, the opponent might simply point out that there are good reasons for thinking that, in any case of reasoning, the inferential mark does appear as an intentional object of the final inferential judgement.

[The opponent:] The inferential mark is an intentional object of the inferential awareness, (i) because, as a matter of rule, it is the intentional object of any awareness of the reason as a property of the site, (ii) because it is the intentional object of [the awareness of] pervasion, and (iii) because, as a matter of rule, it is the intentional object of the awareness of the qualifier [i.e., the target] which causes the inferential awareness, just like hillhood and like the target. Moreover, [in a case where] there exists a causal complex for the awareness of some other qualifier with respect to something that has been apprehended as possessing a qualifier, there is—in that very case—an awareness of the qualification of a qualified object (viśiṣṭavaiśiṣṭyajñāna). So, with respect to that very thing [i.e., the hill] which is qualified by smoke, there is an inferential awareness of being qualified by fire.²⁹

There are two arguments here. The first argument is relatively simple. Since the inferential mark appears in every essential step of the inference (the awareness of the site as possessing it, the recollective awareness of pervasion, and the subsumptive judgement), it must also appear in the final inferential judgement. In this respect, it should be on the same footing as the delimitor of sitehood (pakṣatāvacchedaka, i.e., the property that specifies which object plays the role of the site in the inference). In Mist and Fire, the delimitor of sitehood is hillhood (parvatatva). It appears as a qualifier of the hill not only in the initial awareness of the hill as possessing smoke and in the subsumptive judgement, but also in the final inferential judgement. The same is true of the target, i.e., fire. The target appears in the recollection of pervasion as well as in the subsumptive judgement. But it also appears in the final inferential judgement. Given that these two components of the inference appear as intentional objects in the final inferential judgement, why shouldn’t the same be true of the inferential mark?

The second argument is different. When some object o is apprehended as qualified by some property F and the causal conditions for a further awareness of o as qualified by some other property G are present, then G should appear in the resulting awareness as the qualifier of an o that is already qualified by F. This is what Gaṅgeśa calls the awareness of the qualification of a qualified object. So, in Mist and Fire, if I am already aware of the hill as qualified by smoke, then, even when I infer the presence of fire on the hill, the fire should appear in my inferential judgement as a qualifier of a hill that is already qualified by smoke. If that happens, my inferential judgement will be false.

In response to these two arguments, Gańgeśa gives one final (and I think decisive) response. He appeals to a variant of Mist and Fire:

Mist and Fire Redux. I look at a hill and see what looks like smoke emerging from it. What I see is a wisp of mist. But there is smoke elsewhere on the hill. On the basis of what I see, I judge that there is smoke on the hill. Since I remember that fire always accompanies smoke, I judge, “There’s fire on the hill.” My judgement is true: there is in fact fire on the hill.³⁰

In this case, the inferential mark, i.e., smoke, is defective, but not altogether absent from the site.³¹ Thus, even if the inferential mark were to appear as an intentional object (i.e., as a qualifier of the hill) in my final inferential judgement, my final judgement wouldn’t be false. Gaṅgeśa explains the idea as follows.

Now, let this be true. Even then, when smoke is present by chance (daivāt) on that hill, how can the inferential awareness be untrue even with respect to that part? Therefore, the fire that is brought about by wet fuel is a pervader of smoke, not any other fire. Moreover, it is not the case that, since some other fire pervades smoke in virtue of firehood, that other fire is also a pervader of smoke. For smoke is present even in the absence of that other fire. In this manner, the following is also refuted: “Due to an erroneous awareness of smoke with respect to mist, a fire that pervades smoke appears [in the inferential awareness], and that fire doesn’t exist in that case. So, that inferential awareness is not true.” For, when the smoke is present by chance, the inferential awareness is true.³²

The point is this. Even if the smoke appears as an intentional object in the final inferential judgement, the judgement could be entirely true when there is in fact smoke as a matter of luck on the hill. The opponent cannot reject this conclusion by arguing that the kind of fire I infer isn’t the same kind of fire that is present on the hill. For, given that smoke is only produced due to the combustion of wet fuel, the only kind of fire that pervades smoke is the fire that is produced from wet fuel. In Mist and Fire Redux, since I take the fire that I infer to be a pervader of smoke and there is in fact smoke on the hill, the inferred fire isn’t distinct in kind from the fire that is present on the hill. So, the inferential awareness comes out true, and the opponent’s strategy fails.³³

I think these passages reveal a significant aspect of Gaṅgeśa’s approach to cases like Mist and Fire. Gaṅgeśa seems to concede that, if the Nyāya Definition of Knowledge is right, episodes of reasoning that involve apparent or defective reasons can yield true (satya) awareness-events. In fact, in this very section, when Gaṅgeśa states his own considered view (siddhānta), he accepts all the objections that he himself put forward as part of the prima facie position (pūrvapakṣa). That is why he doesn’t take a non-probative inferential mark to be something that doesn’t produce any correct awareness about the target. Rather, he defines non-probativity as follows:

We reply. Non-probativity is the property of not producing any awareness of the target in a state where there is an awareness of itself [i.e., of non-probativity].³⁴

The idea is simple. As we already know from Gaṅgeśa’s definition of pseudo-reasons, if one were to judge that an inferential mark is an apparent or defective reason, one wouldn’t (rationally) judge—on the basis of the reason—that the target is indeed present in the site. That is precisely what makes such inferential marks non-probative. In the same way, therefore, the non-probativity of an inferential mark (like the misperceived smoke in Mist and Fire) doesn’t by itself prevent a true inferential awareness from arising. However, if an agent were to recognize that the reason in question is non-probative, she (if rational) wouldn’t infer the target on the basis of it. Thus, this account leaves open the possibility that the inferential judgements that arise in cases like Mist and Fire are knowledge-events.

Gaṅgeśa on Testimony and Epistemic Luck

Gaṅgeśa’s stance on cases like The Parrot and The Mistaken Deceiver is much less clear. On the one hand, Gaṅgeśa’s own definition of knowledge-events seems to straightforwardly predict that these are knowledge-events: in The Parrot and The Mistaken Deceiver, my judgements ascribe to the next room a pot that it actually contains. How can they fail to be knowledge-events?³⁵ On the other hand, Gaṅgeśa’s own definition of testimony as an epistemic instrument doesn’t seem compatible with this verdict: “The epistemic instrument that is testimony is produced by a true awareness (tattvajñāna) about the content (artha), which serves as a cause of the utterance (prayoga).”³⁶ On a natural interpretation, this says that a linguistic utterance has the status of an epistemic instrument just in case it is produced by a true awareness of its own content. In cases like The Parrot and The Mistaken Deceiver, the linguistic utterances that produce my judgements aren’t produced by the speaker’s true awareness of its content. Therefore, Gaṅgeśa’s definition of testimony as an epistemic instrument seems to imply that these judgements aren’t produced by epistemic instruments, and therefore aren’t knowledge-events. In this section, my aim is to resolve this apparent inconsistency.

We shall focus here on a section called “śabaprāmāṇyavādaḥ” in Śabdakhaṇḍa of TCM. At the beginning of that section, Gaṅgeśa’s Vaiśeṣika opponent casts doubt on the status of testimony as an independent epistemic instrument. According to this opponent, in cases where an agent comes to know something on the basis of testimony, her knowledge-event is in fact based on an inference. Why? For both Gaṅgeśa and his Vaiśeṣika opponent, the content of a sentence (vākya) is simply a semantic relation (saṃsarga) amongst the referents of different words (pada) that are part of the sentence. The Vaiśeṣika thinks that, on hearing a linguistic utterance, a hearer can correctly infer which semantic relation the speaker intends to convey simply on the basis of certain properties of the utterance. What is the structure of that inference? The Vaiśeṣika explains:

Even then, testimony is not a distinct epistemic instrument. For the semantic relation amongst the referents of words is proved simply on the basis of an inference: namely, “Words like ‘Bring the cow with the stick’, or words in the Veda, are preceded by an awareness of a semantic relation which (i) is an intentional object of [the speaker’s] intention (tātparyaviṣaya) and (ii) holds amongst the recollected referents of those words or they are a group of words that have syntactic dependency and so on, just like [the words occurring in] ‘Bring the pot.’”³⁷

Let’s unpack this. Consider a situation where a speaker utters the sentence, “Bring the cow with the stick!” On Gaṅgeśa’s view, such a command can be true or false: it means that the act of bringing a cow can be achieved by means of effort and is a means to some desired outcome, but doesn’t bring about any pain which exceeds the pain that is necessary for bringing about that outcome.³⁸ When a hearer is exposed to a command like this, how does she become aware of its content? The Vaiśeṣika would tell the following story. This string of linguistic expressions satisfies three conditions: (i) syntactic dependency (ākānkṣā), i.e., the dependency of the expressions in virtue of which they together convey a content, (ii) contiguity (āsatti), i.e., the temporal proximity amongst the utterances of the expressions, and (iii) semantic fitness (yogyatā), i.e., the absence of rebutting knowledge-events (bādhakapramā) that show that the content of the utterance is false. When the hearer correctly judges that a string of linguistic expressions satisfies these conditions, she may infer (on that basis) that the relevant expressions were in fact produced by the relevant speaker’s awareness of a content which (i) the speaker wants to communicate, and (b) which consists in a semantic relation amongst the referents of the relevant expressions. On the basis of this inference, the hearer may understand, and judge as true, the content of the relevant utterance. In this inference, the site consists in the words that are uttered. The target is being preceded by the speaker’s awareness of a semantic relation which she intends to convey and which holds amongst referents of those expressions. The reason is the property of being a group of words that satisfy conditions like syntactic dependency, and so on.

Soon, however, Gaṅgeśa considers a Nyāya objection to this account: namely, that the reason in question deviates from the target, i.e., is present at a place where the target is absent.

[A Naiyāyika:] In the case of a sentence uttered by a deceiver, there is a deviation. For, in that case, there is no awareness of a semantic relation on the basis of the observation of some specific characteristic. One shouldn’t say, “Since it is not possible for anyone to construct a sentence without an awareness of a semantic relation, it is possible for that [deceiver] to have a suppositional awareness of a semantic relation (āhāryaṃ tasya saṃsargajñānam).” For, first of all, it is possible for one to construct sentences—just like a parrot—simply on the basis of one’s awareness of the relevant words, and the same is the rule even in other cases.³⁹

The point is this. Suppose you want to deceive me. You know that there is no pot in the next room, but you tell me that there is one. Even though the words  that you utter satisfy all three conditions mentioned above, the target won’t be present here: since you don’t judge (on the basis of any specific piece of evidence) that there is a pot in the next room, your utterance isn’t preceded by your judgement that the relevant content is true (i.e., that the semantic relation amongst the referents of the words holds). So, the reason deviates from the target. The Vaiśeṣika might offer the following response: since it’s not possible for anyone to construct a sentence without grasping its content, even the deceiver must undergo some sort of suppositional awareness about the content of the relevant sentence before she utters it. But this response fails: just as a parrot can string together expressions without understanding the content of the relevant utterance, so also can the deceiver construct utterances without grasping or reflecting on the content of her utterances.

The Vaiśeṣika replies to this objection as follows.

No, because even that deceiver has an awareness of the semantic relation, since he utters the sentence with the intention (āśaya), “This sentence will convey to this person the semantic relation amongst the referents of the words.” Moreover, [the Nyāya objection fails] because there is an absence of semantic fitness. Therefore, in the case of sentences that don’t conform to reality and are uttered in the manner of a parrot, there is no deviation. Rather, the awareness of a semantic relation, which arises from testimony, takes place due to an error regarding semantic fitness.⁴⁰

The passage contains two arguments. First of all, the Naiyāyika opponent is simply wrong in thinking that, in the case of the deceiver, the speaker lacks an awareness of the content of her utterance. For, the deceiver utters the relevant expressions precisely because she wishes to convey a certain content to the hearer, and she couldn’t have that desire without undergoing an awareness regarding that content. So, even if the reason is present in this case, the target isn’t absent. Thus, the charge of deviation is avoided. However, this reply isn’t robust. Consider a case where a parrot mechanically utters a false sentence, and an agent undergoes an awareness on the basis of it. In such a case, the kind of intention that underlies the deceiver’s utterance is missing. So, the problem of deviation will remain intact.

That is possibly why the Vaiśeṣika offers a second argument: when a sentence is false, one of the three conditions mentioned above—namely, semantic fitness—is absent. According to Gaṅgeśa, semantic fitness is the absence of rebutting knowledge-events (bādhakapramāviraha), i.e., roughly, knowledge-events that show that the relevant content is false.⁴¹ If the content of a sentence is false, there always are rebutting knowledge-events—e.g., knowledge-events belonging to Īśvara, an omniscient God-like being—which show that the relevant content is false. Thus, in this case, the expressions uttered by the deceiver lack semantic fitness. So, since the reason is absent in this case, the problem of deviation doesn’t arise. Similarly, in cases where a parrot or a child mechanically utters a false sentence, the reason put forward in the Vaiśeṣika inference doesn’t deviate from the target, because, in those cases too, semantic fitness is absent. However, in such cases, the hearer may still come to understand what the sentence means, because she mistakenly thinks that the expressions are semantically fit.

This, however, paves the way for a different worry for the Vaiśeṣika. With reference to cases like The Mistaken Deceiver and The Parrot, the Naiyāyika opponent says:

There is a deviation in the case of a fact-conforming sentence (samvādivākya) uttered by a parrot or someone else, and a sentence uttered by a mistaken deceiver, which aren’t accompanied by any awareness of a semantic relation. Moreover, how can there be a knowledge-event regarding a semantic relation, given that an inference about the awareness of the speaker is impossible?⁴²

The problem is basically the same as before. In these cases, there is no absence of semantic fitness, since the content of the sentence is true and thus there is no rebutting defeater that shows that it is false. So, the hearer can correctly judge that the relevant expressions satisfy the three conditions mentioned above. Therefore, the reason can be present in the site, i.e., the words occurring in the sentence. But the target is absent. In the parrot example, this is obvious: the parrot simply has no awareness as of there being a pot in the next room, so it couldn’t have uttered the relevant expressions on the basis of its awareness of the content of the relevant sentence. Moreover, in a version of the deceiver example where the deceiver is mistaken, if the deceiver doesn’t utter the relevant sentence on the basis of any awareness of its content (but merely on the basis of her awareness regarding the relevant words), the utterance won’t be preceded by any awareness of that content. Thus, in both cases, despite the presence of the reason, the target will be absent. So, the problem of deviation cannot be avoided.

The Naiyāyika opponent’s second remark raises a different problem. The Naiyāyika presupposes that, in The Mistaken Deceiver and The Parrot, since the sentence uttered by the parrot or the deceiver conforms to reality, the hearer’s judgement is indeed a knowledge-event. But that isn’t something that the Vaiśeṣika can easily accommodate. For the conclusion of the Vaiśeṣika inference could be false in such cases: given that neither the mistaken deceiver nor the parrot may undergo an awareness of the relevant sentential content, the hearer’s inferential judgement that the speaker undergoes such an awareness may not be true. So, testimonial knowledge-events cannot be reduced to inferential knowledge-events.

The Vaiśeṣika’s response to these problems is somewhat cryptic:

No. Moreover, it has been said [in “prāmāṇyavāde utpattivādaḥ”] that, if that awareness [which arises from the parrot’s or the mistaken deceiver’s utterance] were a knowledge-event regarding a semantic relation, the relevant sentence would be comparable to the Veda.⁴³

Mathurānātha explains the point as follows:

“It would be comparable to the Veda.” The meaning is that, just as, in the case of the Veda, the target is present due to its being preceded by a knowledge-event of Īśvara, so also is true in the relevant case.⁴⁴

We can unpack the thought as follows. If my judgements in The Parrot and The Mistaken Deceiver are knowledge-events, then the Vaiśeṣika will happily say that the relevant sentences are similar to the Veda. In the case of the Veda, the Vaiśeṣika’s inference yields a correct conclusion precisely because Īśvara, the omnipotent and omniscient God-like being, has composed the Veda with the intention of communicating its content to us. Similarly, even when the parrot or the mistaken deceiver utters a sentence, Īśvara serves as the agent of the relevant utterance. For he is a cause of every effect. So, we may argue that the relevant utterance is in fact caused by Īśvara’s true awareness of the relevant sentential content. Thus, the conclusion of the Vaiśeṣika’s inference will come out true.

Ultimately, Gaṅgeśa rejects the Vaiśeṣika’s reduction of testimony to inference. But the Vaiśeṣika’s response also contains a hint of a solution to the problem that we started out with. Recall that, for Gaṅgeśa, a piece of testimony can only serve as an epistemic instrument if it is caused by a true awareness of its content. Since, in The Parrot and The Mistaken Deceiver, the speakers needn’t have any true awareness of the relevant sentential contents, the relevant sentences cannot straightforwardly be treated as epistemic instruments. But the Vaiśeṣika’s response shows us a way out. Following the Vaiśeṣika, we could argue that, in each of these cases, the utterance of the relevant sentence is caused by Īśvara’s true awareness of its content. Thus, the sentence can indeed end up having the status of an epistemic instrument.⁴⁵ As we shall later see, Gaṅgeśa himself will endorse this solution in the section of TCM called “prāmāṇyavāde utpattivādaḥ.”

The lesson is this. If my arguments in this section and the last are sound, then Gaṅgeśa’s conception of inference and testimony as epistemic instruments don’t exclude epistemically lucky awareness-events from the class of knowledge-events. This aspect of Gaṅgeśa’s view creates trouble for him. For it cannot easily be reconciled with his commitment to Nyāya Infallibilism. This is precisely the problem that we shall now turn to.

The Problem of Epistemic Luck

Following his Nyāya predecessors like Udayana, Gaṅgeśa accepts a specific version of Nyāya Infallibilism in the section of TCM called “prāmāṇyavāde utpattivādaḥ.”⁴⁶

Virtue Infallibilism. For any kind K of knowledge-events, there is a kind E of epistemic virtue associated with that kind K, such that, if any awareness belongs to that kind K, then it is produced by an instance of E.

What does this say? For each kind of knowledge-event—perceptual, inferential, analogical, or testimonial—there is a proprietary epistemic virtue that produces knowledge-events of that sort. According to Gaṅgeśa, there is no uniform epistemic virtue that serves as the cause of all knowledge-events.

We reply. There is no uniform epistemic virtue for every knowledge-event. Rather, depending on the circumstances (yathāyatha), the contact of a sense-faculty with many parts of an object and true awareness-events regarding an inferential mark, similarity, and a sentential content serve as epistemic virtues only individually with respect to each specific [kind of] knowledge-event, because there are positive and negative correlations [between each kind of knowledge-event and each of these virtues].⁴⁷

As Gaṅgeśa goes on to explain, each specific kind of knowledge-event has a corresponding epistemic virtue that gives rise to it. In the case of perceptual knowledge-events, it is observation of specific characteristics, which—in the case of composite material objects—may be mediated by the contact of the relevant sense-faculty with a sufficiently large number of parts of the relevant object.⁴⁸ In the case of inferential knowledge-events, the epistemic virtue is a true subsumptive judgement, i.e., a correct awareness of the site as characterised by an inferential mark that is pervaded by the target. Finally, in the case of testimonial knowledge-events (including those produced by the Veda), the epistemic virtue is the (speaker’s) true awareness about the content of the relevant sentence.⁴⁹

Gaṅgeśa’s commitment to Virtue Infallibilism creates trouble for him. On the one hand, none of the epistemic virtues on the list given above are (or have to be) present in cases like Mist and Fire, Horns and Cows, The Mistaken Deceiver, and The Parrot. In Mist and Fire and Horns and Cows, my subsumptive judgement is certainly false, because the reason is either absent from the site or not pervaded by the target. In The Mistaken Deceiver and The Parrot, the basis of the speaker’s utterance needn’t be her true awareness of the content of the relevant sentence. Thus, if (following Gaṅgeśa’s Definition of Knowledge) we treat my judgements in these cases as knowledge-events, Gaṅgeśa’s version of Virtue Infallibilism will be really difficult to defend. This is simply an instance of a more general tension between The Nyāya Definition of Knowledge and Nyāya Infallibilism. Here, we will see how this problem is framed by Gaṅgeśa himself.

Let’s begin with a standard objection, offered by Bhāṭṭa Mīmāṃsakas, against Virtue Infallibilism. According to earlier Naiyāyikas like Bhaṭṭa Jayanta, Vācaspati Miśra and Udayana, the Veda has the status of an epistemic instrument precisely because its author is Īśvara, who possesses a correct awareness of the contents of Vedic sentences. The Mīmāṃsaka disputes this:

Since there is a rebutting defeater for this view in the case of Veda, e.g., the fact that no author of the Veda is recollected, and so on, therefore, even in ordinary practice, a sentence serves as the cause of a knowledge-event simply in virtue of its defectlessness. However, in the case of Veda, even though a speaker is absent, its defectlessness is determined solely on the basis of its permanence.⁵⁰

The thought is this. In the case of Veda, no author is recollected. This, according to the Bhāṭṭa Mīmāṃsaka, suggests that the Veda has no author.⁵¹ So, in order to explain the status of the Veda as an epistemic instrument, we cannot appeal to its author’s correct awareness of its contents. We can only appeal to the Veda’s lack of epistemic defects (i.e., epistemic defects that normally give rise to misleading testimony). Similarly, the Mīmāṃsaka claims, ordinary testimony also serves as an epistemic instrument because it is defectless. So, Virtue Infallibilism is false.

Gaṅgeśa’s response here is significant:

No, because the rebutting defeater will be [later] refuted extensively, and because that defectlessness is absent from (i) the sentence uttered by a mistaken deceiver and (ii) the defect-induced sentence, “There is a cloth,” that is uttered when the sentence, “There is a pot,” is to be uttered, both of which are epistemic instruments insofar as they conform to the facts (saṃvādāt). Moreover, if that Mīmāṃsaka view is right, the sentence, “One should worship a stūpa,” and a sentence uttered by a parrot, etc. by chance would also be epistemic instruments. For the defects of the speaker are absent in those cases and they are similar to the Veda in virtue of being independent of any epistemic instrument.⁵²

The passage contains two distinct arguments. First, Gaṅgeśa thinks that the Mīmāṃsaka is simply wrong in thinking that the Veda has no author; later, he will offer arguments against this position in Śabdakhaṇda.⁵³ On the one hand, if the Mīmāṃsaka is right, in a case like The Mistaken Deceiver or in a case where a true sentence is uttered instead of another false utterance due to a slip of tongue, the relevant sentence wouldn’t serve as an epistemic instrument due to the presence of epistemic defects. This is important: the fact that Gaṅgeśa takes this to be a problem for the Mīmāṃsaka clearly suggests that he takes these accidentally true awareness-events to be knowledge-events. Second, it’s obvious that a false sentence uttered by a sincere Buddhist, e.g., the sentence, “One should worship a stūpa”, or a false sentence uttered by a parrot or a child, cannot be an epistemic instrument. But, if the Mīmāṃsaka is right in thinking that any defectless sentence can be an epistemic instrument, even such a false sentence should generate knowledge-events on the Mīmāṃsaka’s view. For such an utterance isn’t accompanied by any of the typical epistemic defects, e.g., the desire to deceive, which give rise to misleading testimony.

However, the Mīmāṃsaka notices that the Naiyāyika, who is committed to Virtue Infallibilism, also faces a similar problem.

Even for you, how can a sentence uttered by a parrot, etc. or a mistaken deceiver be an epistemic instrument? For it is not produced by any epistemic virtue. Moreover, the following inference does not work: “Since such a sentence is not produced by any epistemic virtue and doesn’t have as its intentional object what is intended [by the speaker], it is not an epistemic instrument.” For, due to the conformity of the sentence to the facts, the target of this inference is rebutted.⁵⁴

As Gaṅgeśa and his predecessors acknowledge, the epistemic virtue that typically explains the status of a sentence as an epistemic instrument is the speaker’s correct awareness regarding its content. This, presumably, is not (or needn’t be) present in cases like The Mistaken Deceiver and The Parrot. So, if Gaṅgeśa and his predecessors accept Virtue Infallibilism, they cannot accept the sentences uttered in these cases to be epistemic instruments. Given that these sentences aren’t produced by the epistemic virtues of the speaker (or given that their contents don’t reflect the speaker’s intention), why can’t Gaṅgeśa and his Nyāya comrades simply accept the conclusion that these utterances aren’t epistemic instruments? As Gaṅgeśa’s opponent points out, these utterances conform to reality. So, the resulting awareness-events must be knowledge-events. This, in turn, rebuts any argument that seeks to show that the relevant utterances aren’t epistemic instruments. That’s bad news for Virtue Infallibilism.

However, a Naiyāyika could resist this conclusion in a different way. For example, she could claim that, at least according to a certain conception of semantic fitness, in cases like The Mistaken Deceiver, the relevant sentence lacks semantic fitness. Suppose we define semantic fitness not as the absence of rebutting defeaters for the content of a sentence, but rather as the absence of rebutting defeaters for the sentential content that the speaker is aware of. Since the mistaken deceiver doesn’t take the content of her sentence to be true (and, in that sense, lacks awareness of it) and utters the relevant sentence on the basis of an erroneous awareness about how things are in the world, semantic fitness is absent in this case. The Mīmāṃsaka’s response goes like this.

However, there is a view that says: “In the case of a sentence uttered by a mistaken deceiver, there is simply no semantic fitness. For semantic fitness is the absence of rebutting defeat for the sentential content that the speaker is aware of, and an intentional object of error is rebutted.” That is wrong. For, since the [deceiver’s] error—which takes the form, “There is no pot,” with respect to something that contains a pot—has a different intentional object from the sentence [uttered by the deceiver], it is not a cause of that sentence. Moreover, for reasons of parsimony, semantic fitness is the absence of rebutting defeat for the content of a sentence. And, in that scenario, the content of the sentence is unrebutted.⁵⁵

The Mīmāṃsaka’s reply has two parts. In The Mistaken Deceiver, when the deceiver says, “There’s a pot in the next room,” she incorrectly thinks that there is no pot in the next room. However, her error plays no (direct) causal role in generating her utterance. What explains her utterance is her desire to communicate a certain content to the hearer (which might in turn be explained by her error and her desire to deceive the hearer). So, even if she makes the mistake, why should that prevent an awareness-event based on the utterance from being a knowledge-event? Second, according to the Mīmāṃsaka, there is a simpler (and therefore preferable) notion of semantic fitness—namely, the absence of rebutting defeat for the content of the  uttered sentence—which allows us to show that, in this case, the sentence uttered by the deceiver is semantically fit (given that its content cannot be rebutted).

A Naiyāyika who doesn’t treat a sentence uttered by a parrot as an epistemic instrument might raise a different problem: namely, that, when a hearer is exposed to the utterance of a parrot, she simply doesn’t judge the content of the sentence to be true, but merely undergoes an awareness that takes the form, “This person [or animal] says this.” Since the hearer doesn’t judge the content of the sentence to be true, the relevant awareness can’t be regarded as a testimonial knowledge-event. Once again, the Mīmāṃsaka dismisses this worry quite quickly.

It is also not to be said, “An awareness of the content [of the sentence] simply doesn’t take place on the basis of a sentence uttered by a parrot, a child, and so on. Rather, there is an awareness of the following sort, ‘This being says this.’” This is because one cannot deny the existence of a non-recollective awareness when the causal complex for the awareness of a semantic relation—e.g., syntactic dependency, etc.—is present, and because the resulting awareness is a true non-recollective awareness in virtue of conforming to the facts.⁵⁶

The point is simple: given that all the causes necessary for testimonial awareness are present in this case, a testimonial awareness (which has as its content the content of the uttered sentence) cannot fail to arise. And that awareness-event—insofar as it conforms to reality—must also be true. Thus, it will end up being a knowledge-event.

At last, the Mīmāṃsaka opponent extends this objection to cases like Mist and Fire. As Gaṅgeśa himself says earlier, a true subsumptive judgement is the epistemic virtue that invariably precedes all inferential knowledge-events. However, in Mist and Fire, given that there is no smoke on the hill, the subsumptive judgement that the hill possesses smoke that is pervaded by fire cannot be true. So, given the absence of this epistemic virtue, the resulting awareness cannot be a knowledge-event. But, according to Gaṅgeśa’s own definition of knowledge-events, this is a knowledge-event.

If this were so, the inferential awareness of fire in a place that indeed contains fire—based on an erroneous awareness of smoke—wouldn’t be a knowledge-event. For it wouldn’t be produced by a true awareness about the inferential mark. And it is not the case that some other fire is simply the intentional object in that case. For the fire is recognized, and, in the case of an inference involving just one individual like cowhood, that [i.e., some other individual] is absent. Moreover, there is no superimposition of an identity with something else. For, even though the inferential awareness may be an error—due to a superimposition of an identity—with respect to the part that involves the superimposition (upadhāna) of the inferential mark, it would be a knowledge-event with respect to the part that concerns the target.⁵⁷

As we have already seen from Gaṅgeśa’s own discussion of such cases, a Naiyāyika cannot escape the conclusion that inferential judgements that arise in cases like Mist and Fire are knowledge-events. As the Bhāṭṭa points out, one cannot argue that the inferred fire is in fact distinct from the fire that is actually present on the hill, because one can perceptually recognize the fire on the hill as the fire that one inferred earlier, and the same strategy of response isn’t available in a case like Horns and Cows. The Naiyāyika also cannot argue that, in a case like Horns and Cows, the agent’s mistake lies in taking cowhood to be identical to some other property. For, even then, insofar as the agent correctly takes the target, i.e., cowhood, to be present in the animal, the awareness at least will be true (and therefore a knowledge-event) in relation to the part that concerns the target.

The general thrust of the Mīmāṃsaka’s objection is this. If we accept Virtue Infallibilism, then all inferential and testimonial knowledge-events must be caused by some epistemic virtue. But, in cases like Mist and Fire, The Parrot and The Mistaken Deceiver, no such virtues seem to be present. Yet, given the Nyāya Definition of Knowledge, it’s difficult to avoid the conclusion that these are knowledge-events. So, there is a tension between Virtue Infallibilism and the Nyāya Definition of Knowledge.

Gaṅgeśa’s First Solution: Appealing to Divine Awareness

Gaṅgeśa offers two distinct solutions to this problem.

The first of these isn’t original: Gaṅgeśa attributes the first solution to “those who know the tradition” (sampradāyavit), i.e., presumably other Naiyāyikas. It involves an appeal to Īśvara’s awareness as a cause of both inferential and testimonial knowledge-events.

We reply. In the case of a sentence uttered by a mistaken deceiver and a sentence uttered by a parrot, etc., since the sentences must be uttered by a trustworthy person (āpta) insofar as they are pieces of testimony that are epistemic instruments, the cause of those sentences is simply Īśvara’s true awareness regarding the relevant sentential contents just as in the case of the Veda. For he is the agent of every effect. Moreover, even someone who claims that defectlessness makes a piece of testimony an epistemic instrument must admit that the utterance of a parrot, etc. is comparable to the Veda.

[The opponent:] If that were so, pseudo-testimony (śabdābhāsa) would completely disappear, since it would also have Īśvara as its speaker.

[Reply:] No. Since the content of such a sentence is false, it is not an intentional object of the awareness of Īśvara.⁵⁸

In framing her objection to Gaṅgeśa’s Virtue Infallibilism, the Mīmāṃsaka opponent was assuming that, in The Parrot and The Mistaken Deceiver, even when my testimonial judgements are true, the utterances cannot be caused by the speaker’s true awareness about the contents of the relevant sentences. So, if the opponent is right, the kind of epistemic virtue that normally explains the truth of testimonial awareness-events are missing in these cases. Gaṅgeśa wants to deny precisely this. He starts from the simple observation that the sentences uttered by both the parrot and the mistaken deceiver are epistemic instruments insofar as they produce knowledge-events. Now, according to the Naiyāyikas, any piece of testimony that is an epistemic instrument must be produced by a trustworthy speaker. Who is the trustworthy speaker in the case of these utterances? It can’t be the deceiver or the parrot. It has to be Īśvara, given that he is implicated in every effect as an agent. So, these utterances too are produced by Īśvara’s true awareness of the relevant contents. Thus, the epistemic virtue which, for Gaṅgeśa, explains the status of testimony as an epistemic instrument—namely, the speaker’s true awareness of the sentential content—will be present even in these cases. This, however, doesn’t mean that there cannot be any misleading testimony—what Gaṅgeśa calls pseudo-testimony—that produces erroneous testimonial awareness-events. For, in cases of misleading testimony, Īśvara’s awareness of the relevant sentential content isn’t present, since he cannot undergo any false awareness. That explains why, in such cases, a knowledge-event cannot arise.

This story can be easily extended to cases like Mist and Fire. How?

In this manner, even in the case of a knowledge-event produced by a pseudo-reason, the cause is simply Īśvara’s awareness of the hill as possessing something pervaded by fire. This is what people who know the tradition say.⁵⁹

Recall Mist and Fire. In that scenario, my inferential judgement that there is fire on the hill is true but my subsumptive judgement that the hill contains smoke that is pervaded by fire is false. For there is no smoke on the hill. This, in turn, implies that the proprietary epistemic virtue that explains inferential knowledge-events goes missing in this case. Gaṅgeśa thinks this is wrong. Since Īśvara causes every effect, his awareness of the hill as containing something that is pervaded by fire can serve as the true subsumptive judgement that causes my inferential judgement. Thus, even in Mist and Fire, my inferential judgement can be caused by an epistemic virtue.

This treatment of cases like Mist and Fire seems to have been popular amongst Naiyāyikas of this period. For example, in his refutation of Śrīharṣa, Khaṇḍanoddhāra, Vācaspati Miśra II (14th century CE) appeals to an explanation like this.⁶⁰ One of Gaṅgeśa’s early commentators, Jayadeva Miśra (15th century CE), offers some insight into the motivation for this solution.

Rather, a knowledge-event [in these cases] is the virtue, and that has a knowledge-event as its result. Moreover, it is not the case that a suppositional superimposition that accidentally conforms to reality is of that sort. If this were the case, then it would follow that a distinct epistemic instrument [other than the four recognized ones] exists, since—without having a uniform and non-overextended property as its delimitor—something cannot have the status of being an epistemic instrument.⁶¹

Let’s flesh out this thought more carefully. On one way of reading Jayadeva, this solution is inspired by a certain general principle about epistemically indirect awareness-events (e.g., inferential, analogical, or testimonial knowledge-events): namely, that an epistemic virtue that gives rise to an epistemically indirect knowledge-event must itself be a knowledge-event. In The Mistaken Deceiver, even though a deceiver’s utterance may be based on an accidentally true suppositional awareness about the content of the relevant sentence, her suppositional awareness isn’t a knowledge-event. So, it cannot play the role of an epistemic virtue in relation to my testimonial awareness-event.

Why is that plausible? Jayadeva offers an argument. If the Naiyāyika were to treat a suppositional awareness as an epistemic virtue, she would have to revise her account of testimony. She would have to say that a sentence can serve as an epistemic instrument insofar as its utterance is based on either (i) the speaker’s knowledge of its content, or (ii) the speaker’s true suppositional awareness of its content. This would imply that there is no uniform or non-disjunctive (anugata) property which is shared by all sentences that serve as epistemic instruments. Why is this bad? Each epistemic instrument, according to Jayadeva, is delimited by a uniform property that doesn’t extend to other epistemic instruments. If this principle is right, then we shouldn’t treat sentences that are uttered on the basis of suppositional awareness-events as instances of the same epistemic instrument as sentences that are uttered on the basis of knowledge-events. So, we will be compelled to posit some new epistemic instruments over and above the four traditional ones. To avoid these two problems, it is better to say that testimonial knowledge-events must ultimately be based on Īśvara’s knowledge.

This solution seems somewhat ad hoc. The Naiyāyika is forced to recognize Īśvara’s true awareness as an epistemic virtue, because she doesn’t have any other way of explaining how an accidentally true awareness could be a knowledge-event.⁶² Arguably, this is why Gaṅgeśa distances himself from this proposal. In the next section, I will lay out his own solution to the problem of epistemic luck.

Gaṅgeśa’s Second Solution: Epistemic Localism

Gaṅgeśa’s second solution to the problem of epistemic luck focuses solely on the case of testimony. Against his Nyāya predecessors, he argues that the production of testimonial knowledge-events doesn’t depend on the epistemic virtues of the source of a sentence (e.g., its speaker), but rather on properties of the sentence and its hearer. This strategy is an instance of epistemic localism: by reducing epistemic virtues to local properties of a sentence and its hearer, it downplays the epistemic significance of upstream causal factors in the production of knowledge-events.

Gaṅgeśa explains his solution to the problem of epistemic luck in the following passage.

Here, we say the following. In ordinary practice, the true awareness of the speaker doesn’t serve as an epistemic virtue with respect to testimonial knowledge-events. Rather, semantic fitness and so on, or a true awareness about those conditions, does. For that is parsimonious, and these conditions are necessary for testimonial knowledge-events. There is no true awareness of semantic fitness in the case of a non-fact-conforming sentence (vākye visaṃvādini) that is produced by error, carelessness, or the desire to deceive. For the content of that sentence is rebutted. The same is true of a non-fact-conforming sentence that is uttered, due to physical indexterity (karaṇāpāṭava), regarding something when something else is to be stated. However, when a sentence conforms to the facts, it is indeed an epistemic instrument.⁶³

Gaṅgeśa’s argument is that, in the case of ordinary speech, the speaker’s true awareness of the sentential content doesn’t serve as an epistemic virtue. Rather, semantic fitness, etc., or the hearer’s true awareness regarding these factors, plays this role. Why does that matter? Two kinds of cases need to be considered. In a scenario where a parrot or child utters a false sentence, the sentence doesn’t conform to the facts. So, there is a rebutting defeater (bādhaka) for that content (i.e., a piece of evidence that would decisively show that the content is false). But, as we already know, semantic fitness involves the absence of knowledge about such defeaters. Things are different in cases like The Parrot and The Mistaken Deceiver. In such cases, the content of the relevant sentence conforms to the facts. Hence, there is no such rebutting defeater. As a result, semantic fitness, or a true awareness regarding it, may indeed be present. So, these sentences  can serve as epistemic instruments.⁶⁴

This second solution to the problem of epistemic luck runs into an objection. Gaṅgeśa claims elsewhere in Śabdakhaṇḍa that “the status of testimony as an epistemic instrument is dependent on [the underlying] intention (tātparya).”⁶⁵ According to Gaṅgeśa, the intention underlying an utterance is its having a certain aim (tatprayojanakatva), which could be either an awareness that the speaker wants the hearer to undergo, or an action that the speaker wishes the hearer to perform.⁶⁶ So, what does it mean for the status of testimony as an epistemic instrument to be dependent on the underlying intention? According to the commentator, Mathurānātha, this means that whether a piece of testimony produces a non-recollective awareness depends on the hearer’s awareness of the speaker’s intention. Gaṅgeśa’s second solution contradicts this principle. According to this solution, a correct (or incorrect) awareness of the speaker’s intention (tātparya) isn’t a necessary condition for any testimonial knowledge-event. But this means that an agent can undergo a testimonial knowledge-event even when she misconstrues the intention underlying an utterance or simply has no clue about what it is. Mathurānātha offers a concrete example of this, which we can spell out as follows.⁶⁷

Hari. You are an avowed atheist. Your friend, who believes in the many Hindu gods, wants to dispel what she takes to be your illusion. So, in order to convey the idea that Viṣṇu exists, she says, “Hari exists” (harir asti). But you misconstrue her intention., since “hari” also stands for lions. So, you come to judge that a lion exists. That, of course, is true.

According to Gaṅgeśa’s second solution, since you have a correct awareness of semantic fitness, etc., your judgement can indeed be a knowledge-event, and therefore the relevant sentence should be treated as an epistemic instrument. But this contradicts the view that an awareness of the speaker’s intention is necessary for testimonial knowledge-events.

Gaṅgeśa addresses this objection as follows.

[The opponent:] If this is so, then, in the presence of a true awareness of semantic fitness, etc., even a fact-conforming awareness regarding some other semantic referent (śakya), which arises from a polysemous expression that is meant to convey something else, would be a knowledge-event, either when there is an error with respect to the [underlying] intention or without that error. So, even in that case, that sentence would be an epistemic instrument.

[Reply:] No. For this is accepted. Moreover, since that sentence doesn’t produce a knowledge-event with respect to the intentional object of the [underlying] intention, it is not an epistemic instrument [with respect to that object]. In this manner, even in the case of the Veda, only a true awareness of semantic fitness is the epistemic virtue. So, Īśvara isn’t proved on the basis of the fact that Veda-based knowledge-events are produced by epistemic virtues.⁶⁸

Gaṅgeśa’s response is subtle. First, he thinks that this result—namely, that a true judgement that is based on a misunderstanding about the speaker’s intention can be a knowledge-event—is acceptable. However, he acknowledges also that there is a sense in which the sentence in question isn’t an epistemic instrument. Since it fails to produce a knowledge-event with respect to the content that the speaker in fact intends to convey, it cannot be treated as an epistemic instrument with respect to that content.

In effect, Gaṅgeśa rejects an argument for Īśvara that earlier Naiyāyikas like Udayana put forward.⁶⁹ The argument goes like this. The Veda gives rise to testimonial knowledge-events. In the case of all testimonial knowledge-events, the epistemic virtue is (a) the speaker’s true awareness of the content of the relevant sentence and (b) her desire to convey that content to the hearer (i.e., the intention). So, the author of the Veda too must be someone who has a true awareness of the content of the Veda and has the desire to convey that content to the hearer. But such an agent cannot be someone like us, since we don’t have any independent access to the truths that the Veda communicates. So, the author of the Veda must be Īśvara. Gaṅgeśa thinks that this argument is unsound. For the epistemic virtue in the case of testimonial knowledge-events is neither the speaker’s awareness of the relevant sentential content nor her intention. It is simply the hearer’s true awareness regarding semantic fitness, etc. So, any attempt to prove the existence of Īśvara by appealing to the epistemic virtues underlying testimonial knowledge-events must fail.

What are Gaṅgeśa’s reasons for rejecting the view that the speaker’s true awareness of the sentential content (or her intention) is an epistemic virtue? Gaṅgeśa states these reasons more carefully in response to a Naiyāyika opponent who holds that view.

[The opponent:] Let this be true. In the case of the Veda, the speaker’s true awareness of the sentential content is also an epistemic virtue. For, in ordinary practice, that kind of awareness is a cause of any testimony that serves as an epistemic instrument. That is why you infer the speaker’s awareness in ordinary practice. And, in the same way, the Veda is composed by an autonomous person who possesses a true awareness that has as its intentional object the contents of the relevant Vedic sentences. For it is a piece of testimony that is an epistemic instrument, just like the sentence, “Bring the cow.” So, Īśvara is proved.

[Reply:] Don’t say this. The status of being a piece of testimony that serves as an epistemic instrument is possible even without being preceded by any true awareness of the sentential content, due to the presence of necessary conditions such as semantic fitness and so on, or due to a true awareness regarding those conditions. This has been explained. So, this condition [namely, the speaker’s true awareness regarding the sentential content] isn’t conducive (prayojaka) [to the status of testimony as an epistemic instrument].

[The opponent:] Even then, in ordinary practice, [the utterance of] a sentence that is an epistemic instrument is thought to be caused by an awareness of the sentential content. So, how can such a sentence occur without that awareness?

[Reply:] No. For a sentence is uttered for the sake of practical undertakings, etc. on the basis of the awareness, “From such words, he will become aware of the sentential content,” with the aim of producing an awareness of the sentential content in a person to be motivated (prayojya). Thus, the speaker’s initial awareness of the sentential content is causally superfluous (anyathāsiddha). In fact, it doesn’t cause the application of the string of words. For, just as in the case of a parrot, that collection of words is produced (upapatteḥ) simply by the causes of the individual words themselves.⁷⁰

Gaṅgeśa denies that the status of any sentence as an epistemic instrument depends on the speaker’s true awareness of its content. The last paragraph explains why. Suppose a farmer tells her farmhand, “Bring the cow.” The farmhand immediately judges that she is supposed to bring the cow, and complies with the command. In this case, according to Gaṅgeśa, the speaker utters the sentence with the explicit aim of motivating the farmhand to act. But the utterance isn’t produced directly by the speaker’s true awareness of the sentential content, but rather by her awareness, “From such words, the farmhand will become aware of the content of the sentence.” So, under Gaṅgeśa’s account, it is simply an awareness regarding the relevant words, which causes the speaker to utter those expressions. Therefore, her own awareness of the sentential content is causally superfluous (anyathāsiddha) with respect to the utterance: even if she didn’t have the relevant awareness, she could still make the utterance on the basis of a similar awareness regarding the individual words. Cases like The Parrot can be explained in exactly the same way: the parrot obviously has no clue about the content of the sentence it utters, but it utters the relevant expressions on the basis of some awareness about each of them. If this is right, then a speaker’s true awareness of the sentential content cannot be regarded as a cause of her utterance. Hence, it cannot be treated as an epistemic virtue that causes the resulting knowledge-event.

Note that this argument (which is only intended to show that the speaker’s true awareness of the sentential content isn’t an epistemic virtue) generalizes to the case of intention. For, in the parrot case, the parrot might have no desire to convey any particular content to the hearer; it might only utter the relevant expressions out of a desire to simply utter those expressions. Since such a desire cannot explain the truth of the resulting testimonial awareness, it cannot be treated as an epistemic virtue.

Suppose this is right. Even then, couldn’t Gaṅgeśa’s Naiyāyika opponent still argue that the Veda functions as an epistemic instrument with respect to a content that its speaker intends to convey? And, if that is right, wouldn’t that be enough to show that Īśvara exists? The opponent states her argument as follows.

The Veda is an epistemic instrument with respect to the intentional object of [the underlying] intention (tātparya). And the intention [underlying the Veda] is its being uttered due to a desire to produce the awareness of that intentional object. Moreover, without the Veda, people like us can’t undergo any awareness about the imperceptible content of the Veda, in virtue of which people like us could utter he Veda out of a desire to produce an awareness of that content. Furthermore, it is not the case that such an awareness arises from the Veda itself. For this would lead to mutual dependence (anyonyāśraya). Therefore, that Veda—which is uttered out of a desire to produce an awareness of a certain content by someone who perceives the entire content of Veda—is an epistemic instrument with respect to that content. So, such a desire is simply the epistemic virtue. Any knowledge-event regarding the content of the Veda is produced by that. Thus, an autonomous being, who is the greatest of all persons, is proved to be the locus of that virtue.⁷¹

The argument starts out from the assumption that the Veda serves as an epistemic instrument with respect to some content that its speaker intends to convey. Who is this speaker? It cannot be a person like us. First, independently of the Veda, people like us aren’t aware of the imperceptible truths about dharma that the Veda conveys. Second, the speaker who composes the Veda couldn’t become aware of the content of the Veda by means of the Veda itself, since that would lead to a problem of mutual dependence: the existence of the Veda would depend on that person’s awareness of its content, and the person’s awareness of its content would depend on the existence of the Veda. So, the speaker of the Veda must be someone who (unlike us) has direct epistemic access to its content and who utters the Veda with the desire to produce in us an awareness of that content. This agent is none other than Īśvara. And his desire, according to the opponent, is the epistemic virtue that explains the status of the Veda as an epistemic instrument.

Gańgeśa’s response is cautious. On the one hand, he doesn’t want to say that the testimonial knowledge-events that arise from the Veda are as arbitrary, or as unconstrained by the speaker’s intention, as the knowledge-events that arise from sentences uttered by a parrot. On the other hand, he stands by his previous argument, namely that the status of the Veda as an epistemic instrument cannot help us prove the existence of Īśvara. That is why he now argues that, though the testimonial knowledge that arises from the Veda is indeed constrained by the intention of some speaker, this speaker needn’t necessarily be Īśvara.

[Reply:] Don’t say this. Since a preceptor (adhyāpaka)—who is aware of the contents of Vedic sentences with the help of all the auxiliary disciplines (aṅga) like Mīmāṃsā, etc.—utters the Veda out of a desire to produce an awareness regarding those specific contents, someone who is truly aware of the content of the Veda indeed has an intention to convey those specific contents. Thus, having become aware of this intention (tatparatva), i.e., the Veda’s being uttered by such earlier preceptors out of a desire to produce an awareness regarding those specific contents, future generations (uttarottareṣām) become aware of the content of the Veda. Thus, there is a beginningless sequence of intentions. What is the point of Īśvara?

[The opponent:] In that case, since a preceptor who doesn’t know the content of the Veda doesn’t utter the Veda out of a desire to produce an awareness of those specific contents, the Veda [when uttered by such a preceptor] isn’t an epistemic instrument due to the absence of intention. Neither is there an ascertainment of the content of Veda on the basis of that.

[Reply:] No. In the beginningless cycle of rebirth, that Veda has been uttered with the desire to produce the awareness of those specific contents at some time by someone who is aware of the content of the Veda on the basis of Mīmāṃsā, etc. The intention [underlying the Veda] obtains simply to that extent.⁷²

According to Gaṅgeśa, some sentences that are epistemic instruments don’t have that status in virtue of producing knowledge-events with respect to the contents that their speakers intend to communicate. Rather, they have that status in virtue of producing true awareness-events about contents that their speaker didn’t intend to communicate. For they are uttered either by a speaker who intends to communicate some other content (e.g., a person who uses a polysemous expression but is misunderstood), or by a speaker who has no such intention (e.g., a parrot or a babbling child). In such cases, the speaker’s intention plays no role in explaining the status of the relevant sentence as an epistemic instrument. However, the Veda isn’t like this: it serves as an epistemic instrument insofar as it produces awareness-events about certain fixed contents that it is intended to convey. But the agent to whom that intention belongs isn’t Īśvara; it is just a preceptor who has understood the contents of Vedic sentences by means of exegetical tools like Mīmāṃsā, etc.

For our purposes, the important question is this. When the Veda produces knowledge-events about the contents that it is intended to communicate, does a true awareness of the speaker’s intention play any role in generating the relevant knowledge-event? Gańgeśa’s response to the objection raised in the passage quoted above suggests that the answer is “No.” Suppose my teacher doesn’t quite understand the Veda, but I am able to grasp its intended content (perhaps because I am better at Vedic exegesis than my teacher). In that case, even though my teacher may utter the Veda to convey some other content, my awareness of the true content of the Veda—which may be based on a miscontrual of my teacher’s intention—would still count as a knowledge-event. Thus, even though I won’t be getting my teacher’s intention right, I would still gain an awareness of the intended content of the Veda in a looser sense, i.e., in the sense that, at some point in the beginningless cycle of rebirth, someone who correctly understood the Veda uttered the Veda precisely with the intention of communicating that content. Even in this case, therefore, grasping the immediate speaker’s intention isn’t necessary for a knowledge-event to arise. The hearer’s true awareness of semantic fitness, etc. should suffice.

This view is an instance of epistemic localism. For Gaṅgeśa, testimonial knowledge-events are produced not due to the transmission of knowledge from a trustworthy speaker, but rather due to the truth-conducive properties of a sentence (its semantic fitness, etc.) and the hearer’s true awareness of those truth-conducive properties. Therefore, the epistemically significant factors which explain testimonial knowledge-events are local to the sentence and the hearer and don’t belong to the source of the sentence (i.e., the speaker).

Now, we might worry that this localist strategy doesn’t fit well with Gaṅgeśa’s own remark that the status of testimony as an epistemic instrument depends on the underlying intention. Gaṅgeśa’s commentator, Rucidatta, addresses this point with reference to cases like The Parrot.

[Objection:] Since, in the case of a sentence uttered by  a parrot, etc., there is a testimonial awareness even though the absence of intention is ascertained, there is a deviation [from the rule that the status of testimony as an epistemic instrument depends on the hearer’s awareness of the intention underlying the utterance]. It is not to be said that, since Īśvara’s intention is present, this is not so. For it is impossible to apprehend that intention, since there is no means of apprehending it, given that the sentence uttered by a parrot and so on is unconstrained by exegetical rules, etc.

[Reply:] No. For, in that case, it is possible to apprehend that intention even by means of semantic fitness and so on, because, without an apprehension of those factors, an awareness of the semantic relation isn’t established. This is what some say. According to others, that awareness of the underlying intention is said to be a cause only in other cases [i.e., in cases that don’t involve a sentence uttered by a parrot, etc.].⁷³

Rucidatta sketches two solutions on behalf of Gaṅgeśa. First of all, it is possible to argue that, in these cases too, Īśvara’s intention brings about the relevant utterance, and it is possible for the hearer to grasp that intention on the basis of various features of the sentence such as semantic fitness, etc. The other strategy is to give up the idea that an awareness of the speaker’s intention is actually necessary in cases like this. In other words, the claim that an awareness of the speaker’s intention is a cause of testimonial knowledge-events needs to be qualified so that it applies only to some (but not all) testimonial knowledge-events.

Let’s sum up. Contrary to some modern interpretations, Gaṅgeśa clearly thinks that the testimonial awareness-events that arise in cases like The Parrot and The Mistaken Deceiver are knowledge-events.⁷⁴ According to Gaṅgeśa’s preferred proposal, the epistemic virtues that explain the epistemic status of those awareness-events are local to the sentence or to the hearer: it is either semantic fitness, etc., or the hearer’s true awareness of semantic fitness, etc. This localist approach to epistemic virtues became influential amongst later Naiyāyikas. For example, in his Kārikāvalī, Viśvanātha Nyāyapañcānana says, “In the case of testimonial awareness, the epistemic virtue should be a knowledge-event either regarding semantic fitness or the speaker’s intention.”⁷⁵ Similarly, in Nyāyakaustubha, Mahādeva Puṇatāmakāra (17th century CE) straightforwardly endorses Gańgeśa’s second proposal: “In this manner, in the case of a testimonial knowledge-event, a knowledge-event regarding semantic fitness is the epistemic virtue.”⁷⁶ Thus, the second solution that Gaṅgeśa offers to the problem of epistemic luck seems to have been widely accepted in later Nyāya.

Jayadeva’s Extension of Localism

Gaṅgeśa’s localist solution to the problem of epistemic luck focuses on testimony. Can the story be extended to inference? Gaṅgeśa’s commentator, Jayadeva, says, “Yes.” In his commentary Āloka, Jayadeva says:

It is to be understood that, even in the case of inference, a true awareness of the absence of rebutting defeat (bādha) serves as the epistemic virtue.⁷⁷

Just as the hearer’s true awareness about semantic fitness, i.e., the absence of rebutting knowledge-events (bādhakapramā), serves as the epistemic virtue in the case of testimony, so also does a correct awareness about the absence of rebutting knowledge-events serve as the epistemic virtue in the case of inference. Thus, we end up with a perfectly systematic solution to the tension between the Nyāya Definition of Knowledge and Nyāya Infallibilism.

However, Jayadeva thinks that this solution doesn’t quite work. The problem is expressed in the voice of a Naiyāyika who thinks that Gaṅgeśa’s solution works for testimonial knowledge-events, but not for inferential ones.

[The opponent:] The absence of rebutting defeat, simply insofar as it exists in itself, is a constituent element (aṅga) of inference, but not insofar it is an object of awareness. For the former is parsimonious. This because, [if things were otherwise], given that the absence of the absence of a target boils down to the target, it would follow that an awareness of the target is the cause of an awareness of the target. And this is not possible, since the fault of proving that which has been proved is an impediment to inferential knowledge-events.⁷⁸

The problem is this. If we take a correct awareness about the absence of rebutting defeat to be the cause of inferential knowledge-events, then the agent can only correctly infer a target if she already correctly judges that the absence of the target isn’t present in the site. Now, according to a principle widely accepted in Nyāya, the absence of an absence of a positive entity (bhāva), e.g., fire, is simply that entity itself.⁷⁹ So, this would mean that an agent’s inferential knowledge-event about a target, e.g., fire, must be preceded by her own true awareness regarding the presence of that target in the site. Thus, the agent would not be inferring anything new. Such an inference would suffer from the fault of proving what has been proved (siddhasādhana), which (under normal circumstances) prevents inferential knowledge-events from arising.

Jayadeva entertains a possible solution to this problem: the relevant epistemic virtue doesn’t have to be a true judgement about the absence of rebutting defeat; it only has to be an awareness of a lack of incompatibility between the target and the relevant delimitor of sitehood (pakṣatāvacchedaka).⁸⁰ The delimitor of sitehood is the property that specifies which object plays the role of the site in an inference. When I undergo an inferential awareness, “The hill contains fire,” the delimitor of sitehood may be hillhood (parvatatva). An awareness of there being no incompatibility between the delimitor of sitehood and the target is necessary for this inferential awareness to arise. If I were to think that the property of being a hill is incompatible with the presence of fire, then I wouldn’t conclude that the hill contains fire.

Jayadeva isn’t happy with this solution. For the solution commits us to the idea that an awareness of an incompatibility between the delimitor of sitehood and the target serves as an impediment to inferential awareness. Given Gaṅgeśa’s definition of pseudo-reasons, this would mean that this sort of incompatibility is a separate kind of pseudo-reason. So, this view will violate the traditional five-fold classification of pseudo-reasons.⁸¹ Jayadeva’s own commentator, Maheśa Ṭhakkura, seems to think that this concern is unfounded. For we can subsume this sort of incompatibility under a recognized variety of pseudo-reasons, namely rebutted reasons (bādhitahetu). Still, Maheśa thinks Jayadeva ignores this solution because his own solution is much clearer.⁸²

Jayadeva’s solution is based on an account of epistemic virtues and defects. ⁸³ He rejects the conception of epistemic virtues and defects that motivated Gaṅgeśa’s first solution to the problem of epistemic luck: the claim that only knowledge-events can serve as the epistemic virtues in cases of inference. Jayadeva thinks that this is wrong. Even though the causes of an awareness may include a knowledge-event, that alone cannot guarantee that the awareness is true. For example, suppose I perceptually recognize Pierre outside a café. When I enter the café, this earlier perceptual awareness may cause me to infer that the café doesn’t contain Pierre. But this inference of Pierre’s absence, though based on a knowledge-event, may be erroneous (e.g., if Pierre has snuck into the café unbeknownst to me). Analogously, if I correctly infer the absence of Pierre from the café after undergoing an illusion as of him being outside the café, my inferential judgment may constitute a knowledge-event about Pierre’s absence from the café (e.g., if Pierre is in fact absent from the café). This shows that, even if knowledge-events (or errors) are included amongst the causes of our awareness-events, they needn’t function as epistemic virtues (or as epistemic defects). According to Jayadeva’s own proposal, epistemic virtues simply are positive factors (i.e., not mere absences of epistemic defects) which bring about knowledge-events, while epistemic defects simply are positive factors (i.e., not mere absences of epistemic virtues) which bring about instances of error.

What does this tell us about cases like Mist and Fire? Jayadeva’s final explanation is relatively simple. In any inference, if an agent arrives at a true inferential judgement, her final judgement must be based on a correct judgement that the site contains something that is pervaded by the target. Consider Mist and Fire: even though I am wrong to think that the hill contains smoke, my subsumptive judgement that the hill contains smoke that is pervaded by fire is still partially true. For, after all, the hill does contain something that is pervaded by fire! That, in turn, explains why I arrive at a true judgement. Therefore, the epistemic virtue that explains the truth of my inferential judgement (and its status as a knowledge-event) is my correct awareness of the hill as characterised by something that is pervaded by fire. Hence, Jayadeva writes:

Therefore, in this case, the awareness of a pervaded object with respect to something that possesses a pervaded object and the awareness of a pervaded object with respect to something that doesn’t possess a pervaded object are the epistemic virtue and the epistemic defect respectively. For that is parsimonious. However, [the epistemic virtue] isn’t the awareness of a specific pervaded object with respect to something that possesses that very pervaded object, and so on. For that is not parsimonious.

And, thus, since something that is pervaded [by fire] must necessarily be present at a place that actually contains fire even when one isn’t aware of that specific pervaded object, how can the status of an inferential awareness as a knowledge-event be ruled out even in a case where the smoke, etc. which is apprehended as pervaded by fire is absent from that place? This is because, when there is an awareness of some other pervaded object [i.e., smoke] with regard to a site that actually contains a pervaded object, it is possible for one to have an awareness of the site as possessing a pervaded object. For it is not possible that the awareness of a site only as possessing that very pervaded object which is present in it is the cause of inferential awareness-events, since it has already been said that this won’t be parsimonious.

Therefore, this [i.e., the status of a true inferential awareness based on a pseudo-reason as a knowledge-event] can be accommodated by appealing to the awareness-events of people like us. What then is the point of admitting the awareness of Īśvara in order to account for that? This is the direction the reader should go in.⁸⁴

In a nutshell, Jayadeva proposes the following revision to the Nyāya theory of epistemic virtues (which Gaṅgeśa himself had mentioned earlier). Instead of treating a true subsumptive judgement as the epistemic virtue that gives rise to inferential knowledge-events, we should treat the agent’s true awareness of the site’s possessing something that is pervaded by the target as the epistemic virtue. Not only is this proposal parsimonious, but it also allows us to explain the truth of such awareness-events without appealing to any kind of divine awareness.

Jayadeva’s proposal is an instance of the same kind of epistemic localism that Gaṅgeśa himself endorses. Not only did the early Naiyāyikas take the production of testimonial knowledge-events to be dependent on the transmission of knowledge from a speaker to a hearer, but they also took the production of inferential knowledge-events to be dependent on the transmission of knowledge from certain initial awareness-events—e.g., the agent’s awareness of the reason as a property of the site and her awareness of pervasion—to a final cognitive state, i.e., her inferential judgement. According to them, if these initial awareness-events weren’t knowledge-events, the final inferential judgement couldn’t be a knowledge-event either. Following Gaṅgeśa, Jayadeva rejects this picture. For him, the epistemic status of these initial awareness-events doesn’t really affect the epistemic status of an inferential judgement. What matters is whether the subsumptive judgement (which is based on these initial awareness-events) is partially true, i.e., whether the agent correctly judges that the site contains something that is pervaded by the target. And this partial truth of the subsumptive judgement is independent of the epistemic status of those initial awareness-events. As long as the subsumptive judgement is partially true in the relevant respect, the final inferential judgement is guaranteed to be true. So, a subsumptive judgement that is correct in this way constitutes the epistemic virtue that produces inferential knowledge-events. This approach to inference makes the relevant epistemic virtue local to the last step of the cognitive process, and therefore downplays the epistemic importance of the initial steps of the process. Thus, this is an instance of epistemic localism.

How influential was Jayadeva’s solution? A version of this solution seems to have been widely accepted by later Naiyāyikas. In his Kārikāvalī, Viśvanātha Nyāyapañcāna says, “Moreover, in the case of inferential knowledge-events, the epistemic virtue should be a subsumptive judgement with respect to a site that possesses the target.”⁸⁵ He explains in his commentary, “In the case of inferential knowledge-events, the awareness of being characterised by something pervaded by the target with respect to a site that possesses the target is the epistemic virtue.”⁸⁶ Similarly, Mahādeva Puṇatāmakāra says:

In the case of inferential knowledge-events, a good subsumptive judgement is the epistemic virtue. And a good subsumptive judgement isn’t a subsumptive judgement that, by nature, is a knowledge-event. For an inferential knowledge-event is produced even by means of a subsumptive judgement that apprehends something which isn’t pervaded or isn’t a property of the site as such [i.e., either as pervaded or as a property of the site]. Rather, it is a subsumptive judgement that has as its qualificand [i.e., the site] something that possesses the target. And, thus, in virtue of being a subsumptive judgement which has as its qualificand something that possesses fire, etc., such a subsumptive judgement serves as the cause of an awareness which is delimited by the property of being an inferential knowledge-event that has fire, etc. as its qualifier.⁸⁷

For both Viśvanātha and Mahādeva, if a subsumptive judgement is to bring about an inferential knowledge-event, it doesn’t itself have to be a knowledge-event. It only has to be partially true: it must correctly ascribe to the relevant site the property of possessing something that is pervaded by the target. And this will be true just in case the site in question possesses the target. This strongly suggests that Jayadeva’s solution came to be the standard way of reconciling the Nyāya Definition of Knowledge with Gaṅgeśa’s Virtue Infallibilism in cases of inference like Mist and Fire.

Conclusion

Let’s take stock. Gaṅgeśa resolves the tension between the Nyāya Definition of Knowledge and Nyāya Infallibilism by appealing to a form of epistemic localism, i.e., the view that upstream causal factors play no epistemically significant role in the production of knowledge-events. What forced him to adopt this view?

The early Naiyāyikas reject epistemic localism. They accept a view according to which the production of inferential and testimonial knowledge-events depends on the epistemic status of causally upstream awareness-events, e.g., the agent’s initial awareness of the reason as present in the site or the speaker’s awareness of the sentential content. They think that, if these awareness-events weren’t knowledge-events, the resulting testimonial or inferential awareness-events couldn’t be knowledge-events either. This commits these Naiyāyikas to a theory that excludes epistemically lucky awareness-events from the class of knowledge-events. But this commitment is problematic. For it is in tension with their view that any true non-recollective awareness can be a knowledge-event.

As Gaṅgeśa’s first solution to this problem shows, some Naiyāyikas try to solve this problem without embracing epistemic localism. To account for the production of knowledge-events in cases like Mist and Fire, they appeal to Īśvara’s knowledge. This solution seems ad hoc. Gaṅgeśa’s preferred solution avoids this disadvantage. He argues that the production of testimonial knowledge-events doesn’t depend on the epistemic status of the speaker’s awareness, but rather depends on the truth-conducive properties of the relevant sentences and the hearer’s true awareness regarding those properties. These are the epistemic virtues that produce such knowledge-events. By restricting epistemic virtues to downstream causal factors in this manner, Gaṅgeśa adopts a robust form of epistemic localism. His commentator, Jayadeva, extends this localist approach to the case of inference. Therefore, if the arguments of Gaṅgeśa and Jayadeva succeed, they will have shown that epistemic localism can help us resolve the conflict between the Nyāya Deffinition of Knowledge and Nyāya Infallibilism.