Certainty and Assertion

It is widely held that assertions are partially governed by an epistemic norm. But what is the epistemic condition set out in the norm? Is it knowledge, truth, belief, or something else? In this paper, I defend a view similar to that of Stanley (2008), according to which the relevant epistemic condition is epistemic certainty, where epistemic certainty (but not knowledge) is context-sensitive. I start by distinguishing epistemic certainty, subjective certainty, and knowledge. Then, I explain why it's much more plausible to think that "certain", rather than "know", is context-sensitive. After that, I respond to an important worry raised by Pritchard, according to which the proposed view is too strong to accommodate our current practice of assertion. I then show that the main linguistic and conversational data advanced in the recent literature in favour of the knowledge condition are best explained by the certainty view. Finally, I offer two principled considerations: the certainty view is the only one compatible with three independently plausible claims and it fits very well with the common thought that knowledge does not entail certainty.

In this paper, my aim is to introduce and defend a view similar to that of Stanley (2008), according to which the relevant epistemic condition is epistemic certainty, where "certainty" is understood in a context-sensitive way. In section 1, I introduce the certainty view by clarifying the relations between epistemic certainty, subjective certainty and knowledge. I explain why I think it is epistemic certainty, rather than subjective certainty, which primarily matters for assertion. In section 2, I show that, in contrast to "know", there are strong linguistic and conversational reasons to think that "certain" is context-sensitive. In section 3, I respond to Pritchard's worry that the proposed account is too strong to accommodate our current practice of assertion. Section 4 shows that the certainty view easily explains all the conversational and linguistic data recently put forth in favour of the knowledge view, while also explaining data recalcitrant to the knowledge view. Finally, I adduce two principled considerations in favour of the certainty view in section 5. 2 3

Epistemic Certainty, Subjective Certainty, and Knowledge
According to the proposal defended in this paper, the epistemic condition relevant for appropriate assertion is epistemic certainty: 2 As we will see, I often follow Stanley (2008), but I consider data and responses to objections that he does not consider, as well as new reasons to think that the certainty norm is superior to its competitors. I also make claims and arguments that Stanley might want to reject. For example, I suggest that being in a position to know is often sufficient for epistemic certainty, a claim which is inconsistent with the testimonial argument he proposes at p. 52. Some other differences will be noted in due course. 3 While the present paper was under review in this journal, two papers were published defending views congenial to the one proposed here (Petersen 2019;Beddor 2020). There is some overlap between these papers and the present one, but also important differences. Let me mention the most striking. Beddor (2020) mainly defends a certainty norm for practical reasoning and only briefly mentions some basic data in favour of a similar norm for assertion. Petersen (2019) defends a certainty norm for assertion, but the way in which this norm is understood carries assumptions that one might want to reject. For example, Petersen's arguments often rely on the assumption that the certainty norm is additional to the knowledge norm. Petersen's certainty norm includes a belief component which, on the view defended in the present paper, should be merely derivative.
In contrast to Petersen's certainty norm, the certainty norm defended in this paper is overtly gradable and not binary. It must also be noted that the ways in which this paper and Petersen's deal with Pritchard, Williamson and Turri's objections do not coincide, but complement each other. Finally, while Petersen offers arguments which are not considered here (e.g. the isolated second-hand knowledge argument and the concessive knowledge attributions argument), the present paper discusses more data, establishes the bad prospects of competing views, and suggests two further motivations for the certainty view (section 5).

CN-E. S (epistemically) ought to assert that only if is epistemically certain for S.
Some basic clarifications are required in order to get the proposal right. First, we can think of epistemic certainty as a high degree of epistemic justification (many would say the highest degree) and of subjective certainty as a high degree of confidence (many would say the highest degree). Second, in ordinary and philosophical contexts, we often use the expression "S is certain/sure that " to refer to subjective certainty. But while we mainly use "It is certain for S that " to refer to epistemic certainty in philosophical contexts, this expression is not frequently used in ordinary contexts. 4 However, the notion of epistemic certainty is not technical. Suppose there is a televised poker tournament where the broadcasters and audience can see the hands but the players cannot. One player bets confidently at a point when it is not certain, given the information available to her, that her hand is the best, though the broadcasters and audience can see that it is. It seems perfectly natural to say something like "She can't be sure that her hand is the best" or "She should not be so sure that her hand is the best." These are statements about lack of epistemic warrant for subjective certainty, i.e about lack of epistemic certainty. A third clarification concerns the relation between knowledge and epistemic certainty. It's natural to think that epistemic certainty entails knowledgelevel justification (or being in a position to know). Perhaps more surprisingly, the view defended in this paper also takes it that knowledge or knowledgelevel justification does not entail epistemic certainty. 5 At first sight, this seems 4 As an anonymous referee pointed out to me, we find slightly different expressions, like "It is certain that for S" (e.g. "It is certain that there will be punishment for the prisoners"), typically used to express the speaker's certainty that a proposition being about a subject S is true. Another example is "It is certain, for S, that " (e.g. "It is certain, for the prisoners, that there will be punishment"), typically used to express that S is certain that . The expression "It is certain for S that " when used in ordinary language sometimes seems to be equivalent to the further expression "S takes for certain/granted that ." For example, "It is certain for John that it will rain" seems equivalent to "John takes for certain that it will rain", and this seems to express John's subjective certainty. Still, "It is certain for S that " may also be taken as equivalent to "According to S, it is certain that ." For example, "According to John, it is certain that it will rain." Here, we seem to express John's belief that the proposition that it will rain is certain, which is to express epistemic certainty. 5 According to Stanley (2008, 35), to say that is epistemically certain for S is to say that S "knows that (or is in a position to know that ) on the basis of evidence that gives one the highest degree of justification for one's belief that ." This definition implies that certainty is an absolute notion. As we will see below, however, we can distinguish certainty (a contextually-influenced doi: 10.48106/dial.v74.i3.02 to clash with infallibilist approaches to knowledge. On closer inspection, however, it should be clear that, on pain of scepticism, everyone should grant that knowledge does not require satisfying absolutely maximal epistemic standards (e.g. Cartesian certainty). 6 In addition, and following for example Williamson's influential non-sceptical infallibilist view, it's natural to think that there is some epistemic space between knowledge-level standards and absolutely maximal epistemic standards. 7 This fact is reflected in ordinary language. The expression "I know that with certainty (/for sure)" does not appear redundant and expresses something stronger than "I know that " (although not necessarily something as strong as the satisfaction of Cartesian standards). Thus, the ordinary notion of certainty seems to capture a degree of justification (or confidence) between knowledge-level justification (or confidence) and absolutely maximal certainty. This notion of certainty is the notion invoked by the view defended in this paper. Let me add two further clarifications. In section 4 below, we will see that expressions of subjective certainty and uncertainty are also highly relevant for assertions. On this basis, one might think that what really matters is subjective certainty: 8 CN-S. S (epistemically) ought to assert that only if S is subjectively certain that . degree of justification, potentially above knowledge-level justification) and absolute certainty. I do not take the certainty view of assertion to demand absolute certainty. It is worth noting that Stanley also distinguishes certainty from absolute certainty later in his paper (see 2008, 54). 6 See Brown (2011Brown ( , 2018. Of course, if it is also true that certainty does not require the satisfaction of absolutely maximal epistemic standards, as I will argue below, this point is not enough to show that knowledge does not require certainty. But, at least, it is enough to show that knowledge does not have to require certainty (at least in some sense) and to suggest that there is an epistemic space between knowledge and absolute certainty. Thanks to a reviewer for raising this point. 7 Williamson's probability one infallibilism has it that you know that only if has probability one given your evidence (i.e. only if has the highest degree of evidential probability). In this sense, knowledge requires epistemic certainty. But Williamson also grants that our epistemic position with respect to can be improved further by knowing that we know (that we know…) that . As he writes (2009,339): "Thus some propositions with evidential probability 1 will have epistemic advantages over other propositions with evidential probability 1." In addition, although Williamson assumes that knowledge requires some kind of subjective certainty, namely outright belief, he also admits that there are weaker and stronger forms of outright beliefs (see 2000,99). See also Wedgwood (2012) and Gao (2019). 8 Stanley favours the epistemic norm over the subjective norm only "for the sake of discussion" (2008, 52) and does not argue for its superiority, as I will do.
Further, I will take the data concerning subjective certainty as favouring CN-E and not CN-S. This may seem problematic. Some reflection on the relation between epistemic and subjective certainty can dispel these worries, though. To begin with, a natural thought is that epistemic certainty is the epistemic norm of subjective certainty: 9 ECNSC. If is epistemically certain for S (and S considers whether ) then S (epistemically) ought to be subjectively certain that , and if is not epistemically certain for S, then S (epistemically) ought not to be subjectively certain that .
If so, if we accept either CN-E or CN-S, there will be something inappropriate if a subject asserts that while lacking epistemic certainty or subjective certainty (see Stanley 2008, [51-52]). Indeed, suppose we accept CN-E and suppose that S asserts that while lacking epistemic certainty. It follows that S violates the supposed norm of assertion (CN-E). Assume that S asserts that while lacking subjective certainty. S violates either the norm of subjective certainty (ECNSC) or the supposed norm of assertion (CN-E). Alternatively, suppose we accept CN-S. If S asserts that while lacking epistemic certainty, then S violates either the norm of subjective certainty (ECNSC) or the supposed norm of assertion (CN-S). If S asserts that while lacking subjective certainty, then S violates the supposed norm of assertion (CN-S). In brief, given ECNSC and the subject's adherence to either CN-E or CN-S, if it's appropriate for S to assert that , then S does not lack epistemic and subjective certainty with respect to .
Still, why should we prefer CN-E over CN-S? The main reason is the following. Suppose the norm of assertion is CN-S. It follows that if, after reading his favoured guru's book, John is certain or completely convinced that he is a cabbage, then John can appropriately assert "I am a cabbage." But it is far from clear, to say the least, that John's assertion is epistemically warranted. Further, in general, we should not allow normative reasons (and permissions) to be generated merely by the adoption of unjustified attitudes. This point has been largely and forcefully made in the literature with regard to normative requirements in general and there is no reason to think that similar considerations do not apply here. 10 Yet, according to CN-S, if John wants to assert that he is a cabbage while he is not certain that he is a cabbage, he can warrant his assertion by adopting the attitude of certainty. CN-S has implausible consequences. 11 It's important to stress, though, that in this framework the apparent plausibility of CN-S can be explained: necessarily, if S satisfies ECNSC and CN-E, then either S is subjectively certain or she does not assert. But, again, that does not imply that there is a norm of assertion such as CN-S.

The Context-Sensitivity of Epistemic Certainty
An additional claim essential to the proposed view is that "certain" is contextsensitive. To illustrate, consider the following cases (inspired by Huemer 2007): airport. Mary is picking up Sam from the airport, but she is a little late, so she calls Sam on his cell phone.
Mary: Where are you? Sam: I'm on the ground; we've just landed. Mary: Is it possible that you're still in the air? Sam: No, it's certain that I'm on the ground. I can see it through the window. epistemology class. John is teaching a class about philosophical scepticism. After reviewing Descartes' sceptical scenarios in the First Meditation, John gestures at the table at the front of the room and asks the class: "So, is it certain that there's is table here?" A student replies: "No, it isn't certain. We might be the victim of an evil demon." A visual experience that seems to be sufficient for an appropriate assertion of "It is certain that " in airport, but insufficient for an appropriate assertion of 11 Here are some further reasons to think that the norm is CN-E and not CN-S. First, CN-S is incompatible with the plausible claim that "selfless assertions" (where speakers assert in accordance with their evidence but against their beliefs) can be epistemically appropriate (see Lackey 2007, 608). Second, when challenged, we defend our assertions by putting forth reasons to think that they are true, not by citing our strong convictions that they are true. Third, suppose the evidence is sufficient, but one irrationally does not believe that and, as a result, one does not assert that . The mere fact that one is not convinced that (while it's clear that the evidence sufficiently indicates that ) does not seem to justify -although it can explain-the fact that one refrains from asserting that . "It is certain that " in epistemology class. This suggests that the epistemic standards that must be satisfied for the truth of sentences such as "It is certain that " and "It is uncertain that " shift with the context (i.e. the aim of the discussion, the interests of the speakers, etc.). 12 Similar pairs of cases were initially offered in favour of contextualism about "know." 13 However, a potential problem for contextualism about "know" is that, prima facie, there are no clear linguistic reasons to think that "know" is context-sensitive and some linguistic reasons to think otherwise. 14 In addition, we seem to be ignorant of the alleged context-relativity of "know" whereas we are not ignorant of the context-sensitivity of other uncontroversial contextsensitive terms, like indexicals, gradable adjectives, etc (see Schiffer 1996). Crucially, things are radically different with "certain". From a linguistic point of view, it is highly plausible that "certain" is context-sensitive. For it is uncontroversial that "certain" (contrary to "know", and like "tall") is gradable. It can take the comparative form (more… than…), it can be modified by degree modifiers such as adverbs (very, extremely, totally, absolutely…), and it can be combined with "how" to form questions. We can say that John is more certain than Bill that the bank is open, that John is absolutely certain that the bank is open, and we may ask how certain John is. (Similar examples can be provided involving an epistemic sense of "certain".) If so, there is a scale associated with this adjective. Like "tall", it's very plausible to think that the degree on the scale which picks out the appropriate threshold for "certain" is context-sensitive.
12 Since these cases are first-person cases, they are not sufficient to rule out that the variation is due to the subject's situation, rather than the speaker's (thanks to a reviewer for raising this issue). Still, the linguistic considerations below lend further independent support to the suggestion that "certain" is context-sensitive. We must also keep in mind that adopting a contextualist semantics for "certain" is not incompatible with holding a view like pragmatic encroachment about certainty, that is, a view in which truth-irrelevant factors of the subject's situation (such as the rationality for the subject of acting on the proposition) are part of the truth-conditions of "certainty" attributions (see e.g. Stanley 2005 for a defence of pragmatic encroachment about knowledge and evidence). 13 See DeRose (2009), Cohen (1999), Blome-Tillmann (2014) and Ichikawa (2017). For an alternative explanation of these pairs of cases appealing to a norm of assertion which has features compatible with those of the certainty norm defended here, see Vollet (2020). 14 See Stanley (2005) and Blome-Tillmann (2014) for possible replies. My point here is just that it is much easier to accept that "certain" is context-sensitive.
Someone may object that "certain" is an absolute maximum-standard gradable adjective, which always picks out the highest degree on the scale. 15 Even so, recent linguistic theories have suggested that absolute adjectives may also be context-sensitive (what counts as "the highest degree" on the scale may be context-sensitive). 16 Further, it's common (and useful) to distinguish a relative (or non absolutely maximal) sense of "certain" from an absolute sense of "certain", so that we can distinguish "certainty" from "absolute certainty", and various degrees of certainty in between. 17 For example, it does not seem weird at all to say "I'm sure that I have hands, but I am even more certain that I exist." This makes sense of the idea that non-sceptical infallibilists about knowledge can grant that we can know that with "more or less certainty", where "less certainty" does not imply that the proposition is uncertain or that we are not certain of it in some important sense. 18 Thus, even if "absolute certainty" were not context-sensitive, it remains that the threshold for "certain" could 15 On the distinction between the different kinds of gradable adjectives (relative vs absolute and minimum-standard vs maximum-standard) see Kennedy and McNally (2005). The claim that "certain" is an absolute maximum-standard adjective, so that certainty implies absolute certainty, is defended by Unger (1975). 16 See Cruse (1980); McNally (2011). One way of developing a context-sensitive semantics for "certain", understood as a maximum-standard absolute adjective, is suggested by Lewis (1979, 353-54), according to whom there are context-sensitive standards of precision for the correct use of absolute terms (like "flat"). Some differences (say, some bumps) may be irrelevant in some contexts or given the object under consideration (e.g. a pavement, a table). Beddor (2016, ch. 3) develops this suggestion by assuming that certainty is the highest grade on the scale and that the granularity of the scale is context-dependent. On this model, propositions which do not count as equally certain on a fine-grained scale can count as equally certain on a less fine-grained scale. This view aims to reconcile the idea that "certainty" is a maximum-standard adjective with the intuition that we can truly ascribe "certainty" to many things in many contexts. 17 Like Stanley, I'm inclined to think that "certain" is not a maximum-standard adjective and that "[t]he semantic function of 'absolutely' is to raise the degree on the scale above that for 'certain' " (2008,54). See also Popper: "it is not impossible to improve even on the most certain of certainties" (1972,79). Many other authors reject the absolutely maximalist view of "certain" by relativizing the standards of certainty (see, among others, Firth 1967;Ayer 1973, 232;Miller 1978;Williamson 2000, 213, 254). Klein (1981, 181-89) distinguishes two concepts of absolute certainty: absolute certainty in the actual world and absolute certainty in all the possible worlds. 18 To illustrate, consider subjective certainty. Assume Williamson's view that outright belief comes in degrees. Then, even if "outright believing" at the first degree may count as being certain (for Williamson, it involves being willing to rely on , at least in some situations), it does not count as the highest grade of certainty (for we may not be willing to act on in any situation). Or assume Wedgwood's view that to outright believe is to be disposed to act on in normal situations. Then you can outright believe , and in this sense be certain that , even if you do not have the highest degree of certainty, for you could have a stronger disposition to act on in normal and in (some further) abnormal situations.
be context-sensitive. In contrast to "know", there is no linguistic objection to the idea that "certain" (or "absolutely certain") is context-sensitive and some good reasons to think that it is. It is also important to emphasize that we are not ignorant of the contextsensitivity of "certain". Consider the following dialogue: tall. John: Robert is tall.
Paul: But is Robert taller than six feet? John: No. Paul: So Robert is not tall, right? John: I did not mean he is that tall.
Clearly, John's last answer is perfectly understandable, for "tall" is contextsensitive.
Consider now the following dialogue, with "know": traffic jam 1. John: I know that Robert will be here at 10 a.m. Paul: But can you rule out that he will be late due to an exceptional traffic jam?
John: No. Paul: So you don't know that Robert will come at 10 a.m, right? John: I didn't mean that knowledge (/I did not mean "I know" or "knowledge" in that sense).
John's last utterance is rather puzzling and this is easily explained if "know" is not context-sensitive (see Stanley 2005, [52-53]). John should either grant that he does not know, or else challenge the relevance of the traffic jam possibility.
Consider now a similar dialogue with "certain": traffic jam 2. John: It's certain that Robert will be here at 10 a.m. Paul: But can you rule out the possibility that he will be late due to an exceptional traffic jam?
John: No. Paul: So, it is not certain that Robert will come at 10 a.m., right? John: I did not mean it is that certain (/I was not considering such a level of certainty/I didn't mean it is absolutely certain).
John's last answer is much less puzzling than John's last answer in traffic jam 1, and as acceptable as that of John in tall. 19 The claim that "certain" is context-sensitive, like "tall", is immensely plausible -certainly much more plausible than the claim that "know" is context-sensitive.
In sum, linguistic and conversational considerations give us good reasons to adopt a context-sensitive view of "certain", and if we adopt a shifty view at all, it's much more natural to adopt a shifty view about "certain" than about "know".
If "certain" is context-sensitive, we must determine the context which is relevant to assess the epistemic appropriateness of assertions. For this purpose, it seems quite natural to invoke the speaker's context and relativise the certainty norm of assertion as follows: CN.R. S (epistemically) ought to assert that in C only if S satisfies the epistemic standards of epistemic certainty which are operative in C. 20 19 An anonymous referee pointed out to me that traffic jam 2 seems no more natural than the following dialogue involving "know": traffic jam 3. John: I know that Robert will be here at 10 a.m. Paul: But can you rule out that he will be late due to an exceptional traffic jam? John: No. Paul: So, you don't know that Robert will come at 10 a.m., right? John: I did not mean it is that certain (/I was not considering such a level of certainty/I didn't mean it is absolutely certain). I agree that John's last utterance seems equally acceptable in traffic jam 3. But this is what we should expect. If "certain" is context-sensitive, "know with certainty" is context-sensitive, and to assert "I know that " does not necessarily mean that one knows that with the level of certainty suggested by Paul. Therefore, John's last answer is easily understood as meaning "I did not mean that I knew that with this level of certainty, but only that I knew that ." We can make sense of Paul's challenge if we assume that Paul is mistaken about the level of certainty associated with John's knowledge. Compare: basketball. John: Robert is a basketball player. Paul: But Robert is not taller than five feet! John: Yes. Paul: So Robert can't be a basketball player, right? John: I did not mean he is not a very short basketball player. Even if we expect a basketball player to be taller than five feet, that's not a necessary condition for being a basketball player. Similarly, even if Paul expects John's knowledge to be associated with a fairly high degree of certainty (presumably due to the context of John's assertion), it remains the case that knowledge does not entail this degree of certainty, as John rightly notes. 20 See DeRose (2009, 99) for a similarly relativised knowledge norm. As it is assumed that the degrees of certainty depend on the degrees of justification, we can see CN.R as belonging to the family of gradable justification or warrant norms for assertion (and action), such as those With this relativised formulation in mind, we are now in a position to assess the main objection to the proposed view, namely, that it is too demanding to accommodate our current practice of assertion.

Pritchard's Worry
The certainty view of assertion does not have many advocates nowadays. It is often thought that this view is too demanding. 21 In particular, following Pritchard (2008), we might worry that the epistemic standards of "certain" cannot be sufficiently low to accommodate our current practice of assertion. More precisely, according to Pritchard (2008, 60), "prima facie at any rate, one would expect that the threshold for 'certain' should be fairly high in all contexts." If so, we have some reason to expect that these standards will not be met often enough to warrant most of our assertions. However, while we can agree that the (contextually-influenced) standards of absolute certainty should be fairly high in all contexts, we must distinguish certainty and absolute certainty. Further, even if we grant that the threshold for "certain" should be at least as strong as knowledge-level justification in all contexts and stronger at least in some contexts, we need not assume that the epistemic standards for knowledge are particularly high. On some accounts, they are even as weak as a true belief. 22 If the standards for knowledge are that weak, standards stronger than knowledge-level standards need not be that high. Of course, the weaker defended by Brown (2010), McKinnon (2015), Locke (2015) or Gerken (2017). However, there are substantial differences between these views and the one proposed here. According to these views, truth is not required and there are contexts in which a warrant or justification insufficient for knowledge can be sufficient for epistemic assertability (or actionability). An apparent advantage of such views is that they can easily handle cases in which the asserted belief is false, or not justified, but in which, intuitively, it would not be reasonable to criticise the assertion (see e.g. Gerken 2017). Yet, these potentially problematic cases can also be handled by proponents of the certainty view if they interpret them as cases where the assertion is epistemically inappropriate but excusable (see, among many others, Kelp and Simion 2017 for a similar move in defence of the knowledge view). In addition, a disadvantage of these alternative views is that they have to explain why the linguistic and conversational data involving "knowledge" are invariant across contexts (see section 5). Finally, I should add that although in this paper I focus on the certainty norm for assertion, I think good arguments can be marshalled in favour of a similar certainty norm for action (see Vollet 2017; Beddor 2020). One such argument might use a commonality assumption between the epistemic norms of assertion and action (see Gerken and Petersen 2020, sect. 3 for a very good overview of this "commonality" issue). Thanks to a anonymous reviewer for raising these issues. 21 See Kvanvig (2009, 143), Turri (2010), Benton (2020), Gerken (2017, 138). 22 See Sartwell (1991Sartwell ( , 1992 and DeRose's discussion about the "floor" of "knows" (2009,(13)(14)(15)(16)(17)(18). the knowledge-level standards are supposed to be, the less plausible is the thought that knowledge is (in many or some contexts) sufficient for certainty. But then, the more plausible is the thought that certainty does not demand very high standards. Conversely, the higher the knowledge-level standards are supposed to be, the more plausible is the claim that in many contexts they are sufficient for certainty, so that in many contexts certainty can be reached.
In addition, it's natural to assume that knowledge is the "floor" of certainty, so that in low-standards contexts "S has knowledge-level justification for if and only if is certain for S" is true, and that these low-standards contexts are fairly frequent. This assumption explains why we might be naturally inclined to think that knowledge entails certainty. If this assumption is correct, indeed, in many contexts it would be false to say "S knows that but S is not certain that (/ is not certain)." We can combine this natural assumption with the consideration that, with regard to many propositions, our epistemic position is stronger than knowledge-level. For many propositions, we have testimonies coming from various sources, repeated visual and tactile experiences, knowledge that we know the relevant proposition, etc. 23 It follows that even if we grant that certainty requires fairly high standards in all contexts, this is not a compelling reason to think that certainty is not often reached. Pritchard (2008, 60-61) also proposes the following case: Jenny. Jenny steps off the train in an unfamiliar town and asks the first person that she meets for directions. This person is indeed knowledgeable about the area and communicates this knowledge to Jenny, who promptly heads off to where she needs to go.
According to Pritchard, "it is hard to see why Jenny (or anyone else for that matter) would regard her as being certain of what she believes, whether the certainty in question is of the subjective or epistemic variety." Thus, champions of the certainty view would encounter the following dilemma: either they must say that Jenny herself cannot appropriately flat-out assert the relevant proposition about the direction which, according to Pritchard, The claim that one's epistemic position is often stronger than knowledge-level justification can be accepted even if one thinks that knowledge-level justification requires evidential probability one (see footnote 7). Importantly, we may think that safety or reliability also matters for the strength of one's epistemic position. Non-sceptical views of knowledge grant that knowledge does not require maximal safety or reliability. For more on this issue, see Brown (2011, 161-62) and Schulz (2017). dict our intuitions and amount to conceding that "far more of our assertions are improper than we typically suppose" (2008, 61); or else they must grant that Jenny's assertion would be appropriate, and the standards of certainty met, which is implausible, for "no-one would surely describe Jenny as certain of what she knows" (2008,61). Let me start with the second horn of the dilemma. Why can't we see Jenny or the relevant proposition as certain in this context? Jenny is described as promptly acting on what she has been told, which suggests that she acts unhesitatingly, and therefore that she is in some sense certain in this context. Further, the person who communicates the information to Jenny is supposed to be knowledgeable about the area. On this basis, we may think that what she says is certain. For example, were Jenny to raise a doubt by asking, "But is it certain?", it would seem perfectly appropriate for this person to reply, "Yes it is! I've been living here for 10 years!" All that Pritchard says in favour of the fact that Jenny is uncertain is that she has a low degree of confidence. But "low" is context-sensitive. We may think that, in Jenny's context, this degree of belief counts as sufficiently high for certainty.
Consider the first horn of the dilemma. Suppose we accept that Jenny is uncertain and conclude that she herself cannot appropriately flat-out assert the proposition regarding the direction. First, contrary to what Pritchard suggests, this would not lead us to concede that much, for most of the propositions we assert in ordinary life are more warranted for us than this proposition is for Jenny. Second, it seems to me that if we suppose that Jenny is uncertain, the idea that her flat-out assertion would be inappropriate is rather intuitive, in particular if that supposition is fleshed out in more detail. Suppose, for example, that we say that Jenny is uncertain because she feels doubtful about the truth of the proposition or because she hesitates to act on it. Now, imagine that someone asks Jenny for the direction, and she unhesitatingly flat-out asserts the proposition in question. It seems that we should regard her flat-out assertion as inappropriate. We would expect her to hedge the assertion by saying something like, "This is the right direction, I believe." Alternatively, it would be very natural for her to say, "I was told that this is the right direction." Lastly, it pays to note that the possible intuition that unconfident Jenny can warrantedly assert the target proposition can be explained by invoking the notion of conditional assertion. When we make assertions, we often speak loosely. We leave aspects of the asserted content implicit. In particular, as Bach (2010, 122-25) notes, many assertions seem to involve an implicit assumption of normality or an implicit ceteris paribus clause. For example, if unconfident Jenny asserts "This is the right direction", we may argue that she asserts loosely and merely commits herself to the conditional "Provided things are normal (i.e. if what I've been told is true), this is the right direction." Understood in this way, Jenny's flat-out assertion does not constitute a counterexample to the certainty view of assertion.
Finally, Pritchard (2008, 63) notes that, "it is in fact very easy to get people to concede that they are not certain of something that they believe, even when no additional practical considerations are being raised." If that is true, that is problematic for the certainty view, for the most plausible explanation why people easily concede that they are not certain is precisely that, in fact, they are not certain. Pritchard proposes the following case to support his claim: car park. My wife and I are heading out of the shopping arcade and I stride purposively towards the part of the car park where I believe our car to be. Nothing in my behaviour indicates any doubt on my part on this score and, indeed, I do know that my car is parked at the relevant location. My wife asks me whether I'm certain that it is parked there, whether there is any possibility that I could be wrong.
According to Pritchard (2008, 63-64), "I would be unlikely to say 'yes'." I must say that I do not share Pritchard's intuition here. At least, we should note that it is also far from clear that an ordinary subject (in a normal situation with low stakes) would be likely to say "no". In contrast, it is clear that "Yes, I'm fairly/pretty/reasonably certain that the car is parked there" would be a very natural reply. 24 By using "fairly/pretty/reasonably certain" in this way, it does not seem that the subject retracts, qualifies or hedges his or her assertion (compare: "No, but I think/Probably it's parked there"). Rather, in doing so the subject seems to maintain the assertion, which suggests that he or she thinks he or she satisfies the epistemic norm. If CN.R is true, that is what we should expect, for mentioning that we are reasonably certain (that is, certain although less than absolutely certain) indicates that we satisfy the certainty norm of assertion.
Perhaps the case is underdescribed. For example, it is unclear with what degree of clarity the subject is supposed to remember where he has parked the car or to what extent he has paid attention when he parked. Undoubtedly, there are cases where we are unsure, for example when we do not remember very well. But in many ordinary cases with low stakes, it appears to me that the most natural reply would be that we have no doubt or that we are certain enough. On this point, I can just encourage the reader to check whether people are willing to insist that they are (fairly) certain of where they have parked their car, what they have eaten at lunch, what job they have, where they live, etc.

Linguistic and Conversational Data
Let us now turn to the consideration of conversational and linguistic data. We can show that appealing to a certainty condition allows us to explain all the data put forth in favour of the knowledge account of assertion while also explaining data recalcitrant to this account. 25 26

Moore's Paradoxical Sentences
It is common to appeal to Moore's paradoxical sentences to defend one or another view about the epistemic condition required for appropriate assertion. Moore (1942, 543) notes that it sounds incoherent to assert " but I do not believe that ." Yet, it is clear that this sentence does not express a semantic contradiction, for it may well be true that and that the subject who utters this sentence does not believe . So, how are we to explain the fact that asserting this sentence sounds incoherent?
A popular explanation appeals to the norm of assertion. Suppose, for example, that you should not assert what you do not believe. Suppose that you are seen as following this norm. Then, if you assert that , you are seen as 25 If we maintain that knowledge entails certainty and that "know" is context-sensitive, the certainty account will collapse into the knowledge account. The defender of the knowledge norm will then be in a position to explain the data involving "certain" (see Ichikawa 2017, 185-86). However, the claims that knowledge entails certainty and that "know" is context-sensitive are far from trivial and rather controversial. 26 As anticipated in section 1, these data involve indifferently epistemic and subjective certainty, but that is not problematic. As explained, if we assume ECNSC these data can be accounted for either by CN-S or CN-E, and I have offered independent reasons to think that CN-E is superior to CN-S. Also, I assume below that "it is certain that p" expresses epistemic certainty, which might be doubted given the considerations of footnote 4. However, I think that in the present context this expression is naturally understood as meaning, "In my opinion, it is certain that p", which I think expresses the assertor's belief that is epistemically certain.
believing that (or at least as taking yourself to believe that ). In other words, in virtue of the supposed belief norm of assertion, by asserting that you represent yourself as believing that . In the second half of your assertion, though, you say that you do not believe that . Thus, such an assertion sounds incoherent because it represents the assertor as believing and not believing that . Consider the following sentences: (1) but I do not believe that (2) but I do not know that (3) but I am not certain that (4) but it is not certain that Assertions of these sentences sound incoherent. 27 As Stanley notes, if we embrace the certainty view of assertion, we can easily explain in a unified way why assertions of (1)-(4) sound incoherent. Suppose you follow the certainty norm of assertion. When you assert that , you take to be epistemically certain. Given the plausible bridge principle (ECNSC) according to which if is epistemically certain for you, then you should be subjectively certain that (at least if you consider whether ), if you assert that . Subjective certainty rightly based on epistemic certainty entails knowledge.
27 See also Williamson (2000), Kvanvig (2009), Turri (2010 and Gerken (2017). Some philosophers might think that (3) does not always sound incoherent. Consider Radford's case (1966): In a quiz, Albert is asked when Queen Elisabeth died. Albert thinks he does not know the answer. Yet, he reliably answers "Elisabeth died in 1603." Albert does not trust his answer, and hence, according to Radford, Albert does not believe that Elisabeth died in 1603. Still, according to Radford, he knows the answer. We might think that this case illustrates a situation where a subject can assert without infelicity " but I am not certain that ." If so, assertions of (3) are not always infelicitous. But first, suppose we accept that Albert can felicitously assert (3). Then, it must also be granted that Albert can without incoherence assert (1) and (2), for Albert does not believe that Elisabeth died in 1603, and yet he says "Elisabeth died in 1603"; also, Albert believes that he does not know that Elisabeth died in 1603, and yet he says "Elisabeth died in 1603." If Radford's case shows that asserting (3) is not always incoherent, then it shows that the strategy consisting in appealing to Moore's paradoxical sentences is misguided in the first place. Second, note that it is far from clear that if Albert utters "Elisabeth died in 1603 but I am not certain," he is really asserting the first half of this sentence. Albert is participating in a quiz. As Radford stresses, Albert takes his answer to be a mere guess. So, arguably, Albert's utterance of "Elisabeth died in 1603" is not an assertion, but a guess.
So, when you assert that , you represent as epistemically certain for you, and you represent yourself as believing with certainty that , and hence as knowing that . But when you assert the second half of sentences (1)-(4), you deny a necessary condition for epistemic and/or subjective certainty (namely, you deny that you believe that , that you know that , that is certain, and that you are certain that ). You represent yourself as having incoherent attitudes.

Rejection of Two Explanations Consistent with the Knowledge Norm
Advocates of epistemic conditions weaker than certainty-such as knowledge-must propose a specific explanation for the infelicity of (3) and (4). A first explanation proposed by knowledge normers appeals to a supposed "contextual" connection between the epistemic standards of knowledge and those of certainty. When considering sentences similar to (3) and (4), Williamson (2000, 254) writes: What seems to be at work here is a reluctance to allow the contextually set standards for knowledge and certainty to diverge. Many people are not very happy to say things like "She knew that A, but she could not be certain that A." According to this proposal, we can explain the infelicities of asserting (3) and (4) in the following way. By saying that , you represent yourself as knowing that (assumption of the knowledge norm). In virtue of a general reluctance to dissociate the standards of knowledge from those of certainty, we expect you to be reluctant to say "I cannot be certain that ," for this would amount to representing yourself as endorsing (5) I know that but I am not certain that (/it is not certain that ). Now, if we assume that knowledge is the norm of assertion and that we are reluctant to make assertions like (5), this may explain why we are reluctant to assert (3) or (4). However, while it can be granted that an assertion of (5) sounds incoherent, Williamson's explanation ultimately relies on the claim that people are reluc-tant to assert sentences like, "She knew that A, but she could not be certain that A." But Williamson does not explain why people are so reluctant. 28 The main problem for Williamson's approach, however, is that it is far from clear that we are reluctant to assert such sentences. As Stanley (2008) notes, asserting the following sentence does not sound incoherent: (6) S knows that , but being a cautious fellow, she is not certain that . 29 We may reinforce this line of thought by noting that assertions of the following sentence do not sound incoherent: (7) S knows that but she does not know that she knows that . That's why she is not certain that . 30 28 Williamson's proposal seems to be that in many (if not most) contexts, the context-sensitive epistemic standards of "certain" and the invariant epistemic standards of knowledge are identical.
In these contexts, we can explain the infelicity of (3) and (4) with the knowledge norm, for in such contexts if you do not satisfy the epistemic standards of certainty (in the relevant sense) you do not satisfy the epistemic standards of knowledge. Williamson adds that in contexts in which "certain" is used with higher standards, like when we use "absolutely certain"-so that the epistemic standards of knowledge and certainty diverge-, "assertability goes with knowledge, not with the highest possible standards of certainty" (2000,254). I agree, but as Turri (2010, 458-59) notes, when we use "certain" we generally invoke "ordinary standards of certainty and assuredness, not the highest possible standards." Therefore, since these standards can be stronger than those required by knowledge, although weaker than standards of absolute certainty, Williamson's solution "does not speak to this problem" (Turri 2010, 459). 29 Of course, we are reluctant to assert the epistemic version of (6): (6*) S knows that , but since she is a cautious fellow, it is not certain that for S. This can be explained by the fact that the epistemic certainty of a proposition relative to a subject does not depend on whether this subject is cautious but on his evidence. 30 An anonymous referee pointed out to me that (7) sounds unnatural as an utterance of ordinary language. I agree. The point is just that this sentence does not sound incoherent, not that we commonly say that kind of things. This referee also notes that the following sentences, involving epistemic certainty, do not sound particularly good, with (7**) sounding particularly bad: (7*) S knows that but she does not know that she knows that . That's why it is not certain (for her) that .
(7**) We knew that but we did not know that we knew that . That's why it was not certain for us that . Here too, I agree that we do not often say this, but I feel no contradiction in asserting these sentences. Perhaps modifying these sentences with "certain enough" would make them more acceptable: (7***) S knows that but she does not know that she knows that . That's why it is not certain enough (for her) that .
(7****) We knew that but we did not know that we knew that . That's why it was not certain enough (for us) that .
If we are not always reluctant to make third-person "knowledge" ascriptions while denying third-person certainty, it is hard to see how Williamson's approach can explain the infelicity of (3), (4) and (5).
A second possible explanation of the infelicity of (3), (4) and (5) consistent with the knowledge norm for assertion could appeal to the distinction between warranted assertion and knowledge that the assertion is warranted, and to the claim that certainty is necessary for knowledge of knowledge.
The distinction between warranted assertion and knowledge that the assertion is warranted is sometimes used by proponents of the knowledge norm to explain why assertions of the following sentence sound infelicitous: (8) but I do not know whether I know that .
According to proponents of the knowledge norm, indeed, knowledge that one knows is not required for epistemically appropriate assertion. So, as Sosa (2009) emphasizes, if they are right, why is it that asserting (8) sound infelicitous?
In reply, advocates of the knowledge norm sometimes appeal to the thought that there is something bad in doing something appropriate without knowing that it is appropriate; and they say that an assertion of (8) sounds infelicitous because the subject represents herself as not knowing that the assertion is warranted (i.e. known), which is somehow wrong (see Benton 2013). Now, suppose we understand certainty in terms of knowledge that one knows, as some writers suggest (e.g. Turri 2010, 459). The kind of explanation offered for the infelicity of (8) can be used by the knowledge normer to explain the infelicity of (3), (4) and (5). Someone asserting " but I am not certain that (/it is not certain that )" would represent herself as knowing that (in virtue of the supposed knowledge norm of assertion), but as not knowing that she knows that . Hence, she would represent herself as not knowing that the assertion is warranted. And someone asserting "I know that but I cannot be certain that " would represent herself as being certain that she knows that (in virtue of the knowledge norm of assertion), but as not being certain that . Hence the infelicity.
Consider also: (7*****) We began to learn that , but it wasn't certain enough yet. The fact that it's not incoherent to use the expression "it is not certain enough that " alongside "know that " in this way shows that we allow the epistemic standards for knowledge and those for certainty (in the relevant sense of "certainty") to diverge.
While interesting, there are several reasons to think that this explanation is ultimately unsuccessful. A first point to note is that it is far from clear that the problem with someone who asserts " " while she cannot be certain that has to do with the fact that she cannot know that she knows that . Indeed, asserting (9) seems equally infelicitous: (9) , but it is not certain that I know that .
Yet it may well be true that the subject knows that she knows that , without knowing that she knows that she knows that . But according to the proposal under examination, not knowing that one knows that one knows that (or, in other words, not knowing that is certain, or not being certain that one knows that ) should make no difference for an appropriate assertion that p, since what is required is merely warrant (i.e. knowledge) and knowledge that one has warrant (i.e. knowledge of knowledge that or certainty that ).
Secondly, it is unclear that certainty is always necessary for knowing that one knows. Indeed, it seems that you can know that you know that even if is still uncertain. Suppose it is still uncertain that you know that you know that . Your epistemic position with respect to could be better and, if so, we may think that is not certain. More generally, if we think that knowledge does not entail certainty, it's clear that you can know that you know that even if is not certain.
Thirdly, the proposal under examination crucially relies on the assumption that, in some sense, it is always bad to do something for which one has warrant if one does not know that one has warrant for doing it. Suppose that this assumption is true. It is then very plausible to think that the strength of the (alleged) requirement to know that one has warrant varies with the importance of being warranted. 31 But if so, in contexts in which it is not at all important to make a warranted assertion, it should be possible to assert without obvious infelicity (3), (4), (5) and (8). Yet, assertions of these sentences always sound infelicitous.
In addition, it is hard to see why such a requirement should always be in force, in particular when it comes to epistemic warrant. Suppose that our epistemic position with respect to the target proposition is good enough for assertion, given the norm of assertion. Suppose that, for whatever general reason, a further relevant second-level epistemic position with respect to the first-level epistemic warrant is always required. Now, presumably, the same kind of reason should lead us to think that a relevant third-level epistemic position with respect to the second-level epistemic position is also always required. And so on. This obviously leads to an infinite regress. Since we have to stop somewhere, it is natural to stop at the first-level.
Fourthly, if the approach under examination is correct, it undermines a fundamental argument for the knowledge account. This argument is based on the fact that it is appropriate to challenge an assertion by using "know" (I consider this argument below). But if the present approach is correct, an advocate of a norm of truth, for example, could argue that when we require knowledge of the asserted proposition, we require knowledge that the assertion is warranted, i.e. we require that it is known that the assertion is true. 32 If these considerations are correct, there is little hope for the proponent of the knowledge norm in appealing to a distinction between warrant and knowledge of warrant. We cannot exclude that another proposal could be offered by the advocate of the knowledge norm or some other weaker condition. However, let me point out that, in contrast to rival views, the certainty view provides a unified and very straightforward explanation of the infelicity of asserting Moorean sentences. 33

Appropriate Challenges
In support of the knowledge account, Williamson (2000, 252-53) emphasizes that it is natural to challenge an assertion by asking "How do you know that ?" or "Do you know that ?" (see also Unger 1975, [263-264]). Of course, these data suggest that knowledge is necessary, but they do not show that knowledge is sufficient. Therefore, they do not speak against the certainty view, for according to this view being in a position to know is necessary for certainty, and if you are in a position to know that and you consider whether , you should know that . In addition, it's crucial to note that we 32 Similarly, Weiner (2005, 235-36) argues that truth is the epistemic condition for proper assertions but that secondary property requires a reasonable belief that one satisfies this condition, i.e. a reasonable belief that the proposition is true. Pagin (2015, 19) considers the possibility of reinforcing the notion of secondary property by appealing to knowledge instead of a reasonable belief. 33 A possible objection here is that the certainty view does not seem to be able to explain the infelicity of asserting "p but it is not certain that is certain." However, this infelicity can be explained by the assumption that in mentioning certainty of certainty in the second half of the assertion, the assertor raises the epistemic standards for appropriate assertion to higher-order certainty.
can challenge an assertion by invoking certainty (e.g. "Are your sure?") (see Stanley 2008, [51]). If knowledge is the norm of assertion, this is a surprising fact. An explanation consistent with the knowledge account, relying on the claims that we ought to know that we have warrant and that certainty is necessary for knowledge of knowledge, is proposed by Turri (2010). However, to repeat, it's far from clear that certainty that is necessary for knowledge that one knows that . Second, it is hard to see why a requirement that one knows that one has epistemic warrant should always be in force. Third, this strategy undermines the argument for the knowledge norm based on the appropriateness of knowledge-based challenges.
Another suggestion would be that we must distinguish assertions and guarantees. This would explain why "certain" is used in some challenges. However, it seems that any kind of assertion can be properly challenged by using "certain" or "sure". Further, asserting that clearly seems to be a way of guaranteeing that . It would be very odd to say, " but I do not guarantee that is true." Finally, if we still want to distinguish assertions and guarantees, given the distinction between certainty and absolute certainty, it seems more natural to associate guarantees with absolute certainty. Benton (2011) points out that a satisfying account of assertion should be able to explain in a unified way the Moorean data and the appropriateness of challenges. Indeed, appropriate challenges "can elicit a de facto Moorean paradox within a conversational context" (2011,686 Benton (2011, 686) argues that this favours the knowledge account because this account explains the relation between Moorean data and challenges in terms of knowledge.

Unified Explanation of Moorean Sentences and Appropriate Challenges
However, the certainty view of assertion fares equally well in explaining why challenges in terms of knowledge may elicit de facto Moorean paradoxes. According to this view, the same thing explains why utterances of Moorean paradoxical sentences sound paradoxical and why we can challenge an assertion by using "know": a necessary condition for an appropriate assertion that you know that is that you are in a position to know that you are in a position to know that , and hence, by factivity of knowledge, that you are in a position to know that . If you are in a position to know that you know that and you consider whether you know that , you should know that you know that , and by factivity of knowledge you should know that .
Further, it is not difficult to imagine a conversation with "certain" eliciting a de facto Moorean paradox in terms of certainty: A: It is snowing.
B: Is it certain/Are you sure? A: Oh, no. B: Huh? A: Still, it's snowing.
The certainty view has a simple and unified explanation for all these data, by appealing to the epistemic certainty norm for assertion (CNA-E) and the epistemic certainty norm for subjective certainty (ECNSC), whereas the knowledge account does not.

Parenthetical Uses
Other linguistic data which have been produced have to do with parenthetical uses. First, consider parenthetical uses expressing the assertor's mental state. When we want to express a mere belief in what we assert, we can use "believe" in a parenthetical position: (9) It is, I believe, raining.
As Benton (2011) notes, it is striking that we cannot (or do not) use "know" parenthetically in the same way. Consider: (12) It is raining, I know.
This use of "know" in parenthetical position seems redundant. Benton argues that this constitutes a further argument in favour of the knowledge account. Indeed, on this account, in asserting that you already represent yourself as knowing that . But if so, it is redundant to parenthetically use "know" to express your knowledge that when you are already asserting that . However, Benton fails to note that, like "know", "certain" cannot be (or is not) used in this way without redundancy. Consider: (13) It is, it's certain, raining (14) It is raining, it's certain.
The certainty view of assertion can explain why these parenthetical uses of "certain" and "know" are redundant. The knowledge account of assertion merely explains why this use of "know" is redundant.
A possible worry is that one might think that if certainty is the norm of assertion, then, as knowledge is typically weaker than certainty, the parenthetical use of "know" should have the same effect as the parenthetical use of "believe" in hedging the assertion. However, the use of "know" in parenthetical position does not hedge the assertion. As Blaauw (2012) notes, it can even have a reinforcing effect. Consider the following case (see 2012, 106): lazy. John is having a fight with his wife Jill. Apparently, as Jill brings up repeatedly during their heated conversation, John is very lazy; a point that Jill supports with ample evidence. At one point, exasperated, John asserts, (15) I am very lazy, I know! What John says sounds natural, and he could also have said: (16) I am, I know, very lazy! Of course, it is striking that the use of "believe" in parenthetical position does not have this reinforcing effect, but, rather, typically hedges the assertion. Blaauw takes these considerations to favour the knowledge account. If you already express your knowledge that in asserting that p, then by saying that you know that p, you can reinforce what is already expressed. We might think that if certainty, rather than knowledge, is the epistemic norm of assertion, it is unclear why the parenthetical use of "know" can have this reinforcing effect whereas the parenthetical use of "believe" has a diminishing effect.
There is a reply, though. By asserting that you believe that , you represent as certain the proposition that you believe that , but you do not represent as certain the proposition that . Thus, you do not represent yourself as satisfying the epistemic norm for asserting , which is why the parenthetical use of "believe" can hedge the assertion. In contrast, by asserting that you know that , you represent as certain the proposition that you know that , and by factivity of knowledge, you represent as certain the proposition that . As a result, you represent yourself as satisfying the epistemic norm with respect to . This is why the parenthetical use of "know" does not hedge the assertion. In sum, the set of data having to do with parenthetical uses are more easily explained by the certainty view than by the knowledge account. Benton (2020) notes that "a standard response when one feels not wellpositioned to assert, in reply to a prompt like 'Is it the case that ?', is to answer 'I don't know'." According to him, this speaks in favour of the knowledge account. Indeed, "the query was about , not about whether one knows that ", and thus the fact that it is appropriate to answer "I do not know that " has to be explained. The knowledge account explains why this answer is appropriate: it is appropriate because it is appropriate to decline a prompt to assert by saying that one does not meet the epistemic norm for assertion.

Responses to Prompts to Assert
Still, this line of argument also favours the certainty view of assertion. Suppose that is not certain. A perfectly correct response to " ?" is "I'm not sure." Further, note that the certainty view explains why "I do not know" is always an appropriate answer, because being in a position to know is a necessary condition for epistemic certainty. In contrast, the knowledge account cannot explain why "I'm not certain" is always an appropriate answer, since, according to this view, certainty is not required for assertion. As a consequence, considerations concerning typical ways of declining prompts to assert favour the certainty view over the knowledge account.

Two Principled Considerations
The certainty view of assertion appears to provide a straightforward and unified explanation of all the main linguistic and conversational data. To reinforce the case for this account, I shall now present two arguments based on principled considerations.
The first principled consideration is the following. It is hard to deny that the warrant required for appropriate assertion varies with the context (e.g. with the audience). Few would deny, for example, that in normal circumstances you can assert that on the basis of a mere testimony that , whereas when it comes to testify that before a court you should have first-hand knowledge. But assuming this variability, it is difficult to maintain the knowledge norm for assertion, while explaining the data involving "know", without endorsing a shifty semantics about "know". 34 Indeed, these data are invariant in the sense that, in all contexts, it's infelicitous to say " and I do not know that "; in all contexts we can challenge an assertion by asking "How do you know that ?"; etc. Yet, as explained above, from a linguistic perspective it's not prima facie plausible to think that "know" is context-sensitive. If we can propose an account compatible with an invariantist semantics for "know", capable of explaining the invariant character of the data involving "know" and compatible with the claim that the epistemic warrant for assertion is variable, this account will certainly have an advantage. Therefore, it is an advantage of the certainty view that it respects the three following assumptions: A The epistemic standards that we must satisfy for appropriate assertions shift with the context. B The linguistic and conversational data (infelicity of asserting Moorean sentences, appropriate challenges, etc.) about "certain", "know" and "believe" are invariant across contexts. C The epistemic standards of "know" are invariant across contexts.
To illustrate, consider the knowledge account and the warrant account to see how they fail to respect these three assumptions. Assuming the knowledge 34 See Benton (2020). DeRose (2009) proposes an argument in favour of contextualism about "know" from the claims that the warrant for assertion varies across contexts and that knowledge is the norm of assertion. norm, we can (partially) explain B. But if we embrace C we must reject A, for we assume that the epistemic standard of appropriate assertion is knowledge and that the epistemic standard of knowledge is invariant across contexts. If we accept A instead, we must then reject C. Suppose now that we adopt Gerken's warrant account, according to which the warrant for epistemically appropriate assertion shifts with the context, in the sense that the warrant is sometimes weaker and sometimes stronger than knowledge-level warrant (see Gerken 2017). We can accept A and C, but it is unclear how to account for B. For example, in situations in which the warrant is supposed to be weaker than knowledge, we cannot appeal to the supposed norm to explain why it sounds infelicitous to assert " but I do not know that ." 35 In contrast, consider the certainty view. We can adopt A, for the epistemic standard of certainty is taken to shift with the context. What degree of justification counts as good enough for certainty is partially influenced by the context. We can also respect B: the norm of assertion always requires at least knowledge-level justification and, given ECNSC, if you assert that , you consider whether , and you should be subjectively certain that . Therefore, you should know that . Finally, the certainty view respects the claim that the epistemic standards of knowledge are invariant across contexts.
The fact that claims A, B and C, which are independently plausible, are fully compatible with the certainty view, whereas together they raise problems for rival weaker accounts, provides a further argument for the certainty view. Of course, this argument is limited, for the claim that "certain" is contextsensitive could be disputed. However, given the considerations developed in section 2, it appears that it is highly plausible to think that "certain" is context-sensitive. In this regard, "certain" strongly differs from "know".
The second principled consideration is that many philosophers find it plausible that knowledge does not entail (epistemic) certainty or the highest grade of justification. But if that is correct, first, we have to explain why asserting (5) sounds infelicitous: (5) I know that but I am not certain that (/it is not certain that ).
As Stanley (2008) emphasizes, the certainty view provides a pragmatic explanation of the infelicity. Again, given the certainty norm of assertion, by asserting that you know that , you represent yourself as being certain that you know that , and by the factivity of knowledge, as being certain that . But this contradicts the second half of your assertion. 36 Second, if knowledge does not entail epistemic certainty or the highest grade of justification, it is obscure why knowledge is bound to always be sufficient for appropriate assertion. If knowledge does not require certainty or the highest grade of justification then, in principle, a situation can arise where the difference between knowledge and certainty could matter for appropriate assertion. In the absence of reasons to think that knowledge is bound to always be sufficient, it is more natural to think that certainty, rather than knowledge, is always sufficient for assertion. 37 In sum, it is plausible that the degree of justification required for warranted assertions shifts across contexts and that knowledge does not entail certainty. These two claims fit nicely with the certainty view of assertion whereas they are in tension with (many) rival weaker accounts.

Conclusion
The claim that certainty is the norm of assertion is often dismissed as implausible. In this paper, I've responded to the main objection that this view is too strong. I've also considered the main conversational and linguistic data advanced in the recent literature, and I have shown that the certainty view provides us with a straightforward and unified explanation of them. I have also argued that this account fits nicely with the plausible claims that the degree of justification required for appropriate assertion is variable and that knowledge does not entail certainty. I thereby hope to have shown that, on more careful reflection, the certainty account of assertion is a viable and respectable account.* Jacques-Henri Vollet University of Geneva jacquesvollet@yahoo.fr