Absolute gradable adjectives and loose talk

Abstract

Kennedy (Linguist Philos 30:1–45, 2007) forcefully proposes what is now a widely assumed semantics for absolute gradable adjectives. On this semantics, maximum standard adjectives like “straight” and “dry” ascribe a maximal degree of the underlying quantity. Meanwhile, minimum standard adjectives like “bent” and “wet” merely ascribe a non-zero, non-minimal degree of the underlying quantity. This theory clashes with the ordinary intuition that sentences like “The stick is straight” are frequently true while sentences like “The stick is bent” are frequently informative, and fans of the indicated theory of absolute gradable adjectives appeal to loose talk in response. One goal of this paper is to show that all extant theories of loose talk are inconsistent with this response strategy. Another goal is to offer a revised version of Hoek’s (Philos Rev 127:151–196, 2018, in: Proceedings of the 22nd Amsterdam Colloquium, 2019) recent theory of loose talk that accommodates absolute gradable adjectives after all, while being defensible against a range of important concerns.

1 Introduction

Kennedy (2007) forcefully proposes what is now a widely assumed semantics for absolute gradable adjectives (henceforth, absolute adjectives). On this semantics, so-called maximum standard absolute adjectives such as “flat”, “dry” and “straight” (henceforth, maximum adjectives) ascribe a maximal degree of the underlying quantity.

• (1) The table is flat.

• (2) The stick is straight.

• (3) The socks are dry.

Sentence (2), for instance, supposedly says that the stick is perfectly straight; and sentence (3) that the socks contain not a single drop of liquid. Meanwhile, so-called minimum standard absolute adjectives (henceforth, minimum adjectives) such as “bumpy”, “wet” and “bent” merely ascribe a non-zero, non-minimal degree of the underlying quantity.

• (4) The table is bumpy.¹

• (5) The stick is bent.

• (6) The socks are wet.

Sentence (5), for instance, supposedly says that the stick has a non-zero degree of bentness, i.e., that it has some curvature, however slight, and sentence (6) supposedly says that the socks have a non-zero degree of wetness, i.e., that they contain some amount of liquid, however little.

While the described position has much going for it (see, e.g., Kennedy, 2007), it faces an immediate challenge. Normally, the suggested propositions are either obviously false or obviously true. For instance, it is normally obvious that a given stick has some curvature if we only look closely enough. This makes (2) obviously false and (5) obviously true. Similarly, a normal sock obviously contains some amount of water, e.g., due to the humidity of the surrounding air. This makes (3) obviously false and (6) obviously true. Other things being equal, assertions of (1) to (6) should normally be objectionable, because it is objectionable to assert obvious truths or falsities. In fact, however, we use these sentences all the time and appropriately so.

The go-to response to this challenge appeals to loose talk (Kennedy & McNally, 2005: 357; Kennedy, 2007: 23–25). Strictly speaking, utterances of (1) to (6) are obviously false or obviously true, but they have a loose reading, which is true and informative. Sentence (2), for instance, expresses the obviously false proposition that the stick is perfectly straight, while on a loose reading, it suggests the informative truth that the stick is somewhat straight. Similarly, sentence (5), for instance, expresses the obvious truth that the stick has some curvature, while loosely conveying the informative truth that it is substantially curved (see similarly, e.g., Lasersohn, 1999: 532; Klecha, 2018: 88; Hoek, 2019: 179; Carter, 2019: 172, who all mention absolute adjectives as instantiating loose talk).

Let’s call the presented semantics for absolute adjectives along with the indicated appeal to loose talk the Endpoint View, because on this view, maximum and minimum adjectives semantically refer to the endpoints on the underlying scale. One goal of this paper is to show that if the Endpoint View is right, then every extant theory of loose talk is false (§2). Another goal is to propose a revision of Hoek’s (2018, 2019) recent theory of loose talk that accommodates the Endpoint View after all (§3). Finally, I defend the resulting theory against several previously unaddressed concerns (§4). The upshot will be a defense of the Endpoint View and a powerful theory of loose talk.

2 The Endpoint View and extant theories of loose talk

In this section, I argue that if the Endpoint View is right, then every extant theory of loose talk is wrong. Semantic theories of loose talk are obviously wrong if the Endpoint View is right. On these views, the semantic value of loosely used expressions varies with the context of utterance (e.g., Krifka, 2007; Sauerland & Stateva, 2007; DeRose, 2012: 714–717; Solt, 2014). On the Endpoint View, though, maximum and minimum adjectives are used loosely, and yet their semantic value does not vary with the context of utterance. They always refer to the maximum or, respectively, minimum standard (see Kennedy, 2007: 21–22).² In the following, I discuss three major pragmatic theories of loose talk to show that they can’t be squared with the Endpoint View either.

Notice that discussions of loose talk typically focus on number and measurement expressions as they occur in the following sentences.

• (7) Hannah is 6 foot 2.

• (8) It’s 3pm.

• (9) We have 500 reservations.

As in the case of absolute adjectives, the challenge is to explain why these sentences can be used appropriately when Hannah is only roughly 6 feet 2, when it’s around 3pm and when we merely have approximately 500 reservations. Absolute adjectives are frequently mentioned as instances of loose talk too (e.g., Lasersohn, 1999: 532; Klecha, 2018: 88; Hoek, 2019: 179; Carter, 2019: 172), but they are rarely discussed in detail. In the following, I show that, in the major theoretical frameworks on offer, it is impossible to fill these details in.

2.1 Lasersohn’s pragmatic halos

Lasersohn (1999) presents the first substantial pragmatic account of loose talk. On his view, loosely used sentences semantically express one proposition and pragmatically convey another. Lasersohn further describes a method to determine the pragmatically conveyed proposition in cases of loose talk. The starting idea is to associate each atomic expression in a given sentence with a so-called pragmatic halo. The pragmatic halo of an expression comprises any object that differs from the semantic value at most in “pragmatically ignorable” (526) ways while being of the same “logical type” (526). For instance, the halo of a term like “3pm” will comprise points in time that are sufficiently close to 3pm. The halo of entire sentences is compositionally derived from the halos of their atomic expressions. It comprises one proposition for each combination of elements in the latter halos. In a sentence like (8), for instance, we get one proposition for each point in time in the halo of “3pm”. The proposition conveyed by a given sentence is then determined as follows.

$$ \mathrm{The\; proposition\; conveyed\; by\; sentence\; S\; in\; context\; C }=\cup {\mathrm{H}}_{\mathrm{C}}\left(\mathrm{S}\right) $$

H_C gives you the halo of an expression in a context C, and propositions are modeled as sets of possible worlds here and throughout this paper. The proposition conveyed by a given sentence in a given context thus corresponds to the union of the propositions in its halo.

If the Endpoint View is right, this theory must be wrong. Consider minimum adjectives. According to the Endpoint View, sentence (5), for instance, is an instance of loose talk where the expressed proposition is that the stick has some curvature while the conveyed proposition is that the stick is substantially curved. Thus, the conveyed proposition is stronger than the expressed proposition (in the sense of asymmetrically entailing it). On Lasersohn’s theory, however, this should be impossible. The “difference” between the semantic value of any given expression and itself is always pragmatically ignorable (see also Lasersohn, 1999: 548). Thus, the halo of a sentence always comprises the proposition expressed. Since we obtain the proposition conveyed by unioning this proposition with other propositions, the proposition conveyed can never be stronger than the proposition expressed.

These brief considerations should suffice since the underlying point is famil-iar enough. Hoek (2018: 158) and Carter (2019: 177) already observe that, on Lasersohn’s theory, the proposition conveyed in cases of loose talk can never be stronger than the proposition expressed.³ They appeal to loosely used expressions in downward monotonic environments like negation (e.g., “Lena didn’t arrive at 3pm”) to show that this prediction cannot be right. If the Endpoint View holds, minimum adjectives are just another case in point; though, notably, one that doesn’t feature any obvious embedding.

2.2 Klecha’s optimality account

Klecha (2018) proposes another pragmatic account of loose talk on which loosely used sentences convey another proposition than they express. As we’ll see, this view too has problems with minimum adjectives on the Endpoint View. This time, though, the problem is not that the theory cannot predict stronger readings than the proposition expressed. Rather the theory predicts too strong readings in the relevant cases, namely, contradictory readings.

Klecha assumes an optimality theoretic framework to determine the proposition conveyed in cases of loose talk (see also Krifka, 2002). The basic idea is that a speaker conveys what she must believe for the sentence uttered to be an optimal choice. Klecha offers a set of ranked constraints which, taken together, determine when a sentence counts as optimal. These constraints rely on a notion of the coarsening of a proposition, which in turn relies on a certain way of partitioning the set of possible worlds. I describe this partition first, before I turn to coarsening and then, finally, to the indicated constraints.

The partition in question, call it π_C, is determined by “what distinctions are relevant to [the domain goals]” (Klecha, 2018: 104), i.e., to “the set of goals that interlocutors have when conversing” (Klecha, 2018: 96). To regiment this a bit, let’s think of these goals as shared goals of the conversational participants, i.e., goals they have somehow coordinated on and jointly pursue in their conversation. These goals can be seen as determining a set of questions that one needs to answer to achieve these goals. The question determined by a practical goal like going to the movies, for instance, may include questions like when the movie starts, which movies are on, how one can get to the movie theatre, etc. The questions determined by a theoretical goal like figuring out whether the movie was good may include questions like whether the movie was good but also sub-question like what the plot was, whether the acting was good, etc. We can then say that the partition cells of π_C comprise possible worlds such that, whichever of these worlds we occupy, the answers to all these questions are the same.

Consider a context HIKE, where we are looking for a walking stick on a hike. This does not require answering the question of whether a given stick is perfectly straight but only whether the stick is roughly straight. Consequently, possible worlds differing only in whether the stick is perfectly straight or only sufficiently close to being perfectly straight occupy the same partition cell in π_HIKE. If, for simplicity, we ignore all other relevant questions, π_HIKE will look as follows.

$$ {\pi }_{\mathrm{HIKE}}=\left\{\matrix{\left\{w |\mathrm{ At }\;w,\mathrm{the\; stick\; is\; at\; least\; roughly\; straight}\right\},\cr \left\{w |\mathrm{ At }\;w,\mathrm{ the\; stick\; is\; not\; at\; least\; roughly\; straight}\right\}}<!end array> \right\} $$

On to the coarsening of a proposition. Intuitively, coarsening makes a proposition less precise, so that it speaks only to the relevant questions as represented in π_C. Formally, the coarsening ${K}_{C}$ of a proposition p in context C gives you the union of the partition cells in π_C that overlap with p.

$$ {K}_{C}\left(p\right)=\cup \left\{p^{\prime}\in {\pi }_{C} | p^{\prime}\cap p\ne \varnothing \right\} $$

Assume the Endpoint View, according to which (2) expresses that the stick is perfectly straight. This proposition overlaps with the partition cell of π_HIKE in which the stick is at least roughly straight but not with the partition cell in which the stick isn’t at least roughly straight. Thus, K_HIKE maps the proposition expressed by (2)—that the stick is perfectly straight—onto the proposition that the stick is at least roughly straight.

With this idea of coarsening at hand, Klecha suggests that the proposition conveyed by a sentence S in context C is determined as follows (simplifying a bit; see Klecha, 2018: 108).

$$ \mathrm{The\; proposition\; conveyed\; by\; S\; in\; C}=\cup \left\{{K}_{C}\left(B\right) \right| S\in \mathrm{opt}\left({\mathcal{S}}_{C}|B\right)\} $$

Here B ranges over possible belief states on behalf of the speaker, where these are modeled as the set of possible worlds consistent with the believed propositions. ${\mathcal{S}}_{\mathrm{C}}$ denotes the set of “utterance alternatives,” i.e., roughly, the set of sentences that the speaker can relevantly utter in the context at hand, including the actually uttered sentence, S (see Klecha, 2018: 108). The set $\mathrm{opt}\left({\mathcal{S}}_{\mathrm{C}}|\mathrm{B}\right)$ is the set of optimal sentences among ${\mathcal{S}}_{\mathrm{C}}$ on the assumption that the speaker is in belief state B. The proposition conveyed thus corresponds to the union of the coarsenings of the possible belief states on behalf of the speaker relative to which the actually uttered sentence is optimal compared to its alternatives.

What is missing is a definition of the set $\mathrm{opt}\left({\mathcal{S}}_{\mathrm{C}}|\mathrm{B}\right)$, the set of optimal sentences relative to a belief state. Klecha offers three constraints that together determine this set. One constraint is Faithfulness, where sentences rank highly roughly when the speaker believes the coarsening of the proposition expressed (see Klecha, 2018: 109). Another constraint is Manner, where sentences rank highly when they are perspicuous in a relevant sense (see Klecha, 2018: 98–99, 109–110; see also Krifka, 2002). For our purposes, we can focus exclusively on the third constraint called Informativity, which ranks sentences as follows.

$$ {\hbox{Informativity}}\left( S \right) \ge {\hbox{Informativity}}\left( {S^{\prime }} \right) \hbox{iff}\; K_{C} \left[\kern-0.15em\left[ S \right]\kern-0.15em\right] \subseteq K_{C} \left[\kern-0.15em\left[ {S^{\prime }} \right]\kern-0.15em\right] $$

A sentence is more informative than another, and thus pro tanto more optimal, when the coarsening of the proposition expressed by the former sentence is stronger than the coarsening of the proposition expressed by the latter.

We can now see why Klecha’s theory is incompatible with the Endpoint View. If the Endpoint View is right, sentences featuring minimum adjectives are instances of loose talk. Qua theory of loose talk, Klecha’s theory should make decent predictions about the propositions these sentences convey, but it doesn’t. Let’s grant that we find ourselves in a context governed by π_HIKE and thus that we exclusively care about whether a given stick is roughly straight. Now look at the result of coarsening the proposition expressed by (5). On the Endpoint View, this sentence expresses the obviously true proposition that the stick has a non-zero degree of bentness. Coarsening this proposition leads to an even more obviously true proposition, now comprising the entire set of possible worlds. This is because having a non-zero degree of bentness is compatible with both (and hence all) partition cells in π_HIKE. A stick can be at least roughly straight while still having a non-zero degree of bentness, and any stick that fails to be at least roughly straight has a non-zero degree of bentness too. Sentence (5) thus comes out as maximally uninformative by the above metric since the coarsening of the proposition expressed is entailed by every proposition. Such utterly uninformative sentences shouldn’t be optimal relative to any belief state we are normally prepared to ascribe.⁴ This means that, when we derive the proposition conveyed by the principle above, i.e., by unioning the belief states relative to which the uttered sentence is optimal, we end up with the empty set. Klecha’s theory predicts that sentence (5) conveys the equivalent of a contradiction, but surely, this cannot be right.⁵

2.3 Hoek’s conversational exculpatures

Hoek (2018, 2019) provides another pragmatic account of loose talk, where the proposition expressed and the proposition conveyed come apart, and he also provides a mechanism by which we can derive the proposition conveyed. On his view, this mechanism is a subtype of a more general mechanism of “conversational exculpature”. I will present this general mechanism first before I turn to loose talk. As before, the resulting view is incompatible with the Endpoint View though this time problems arise equally for maximum and minimum adjectives.

Hoek’s view relies, first, on the notion of a question under discussion, or “conversational subject matter” (Hoek, 2018: 153). This question under discussion is represented as a partition on the set of possible worlds that Hoek preliminarily defines as follows.

Roughly speaking, the subject matter of a conversation is the partition S such that w∼_Sv if and only if the differences between w and v are ignored for the purpose of the conversation. (169)

One might worry that the relation of ignoring the difference between w and v cannot induce a partition. Small differences can add up to large ones, and so this relation doesn’t appear transitive. Let’s instead identify the question under discussion with the already introduced partition π_C from Klecha’s account, which represents the questions determined by the shared goals of the conversational participants.

Second, Hoek relies on the following notion of a proposition being wholly relevant to a question.

A proposition ${\hbox{p}} \subseteq {\hbox{W}}$ is wholly relevant to a question Q if and only if no partition cell of Q contains worlds where p is false and also worlds where p is true. (Hoek, 2019: 173)

Take the question “How many cows are there on this ranch?” This question induces one partition cell for each answer of the form “There are exactly n cows on this ranch,” containing just those worlds where this answer is true. Both the response “There are exactly three cows on this ranch” and the response “There are between three and five cows on this ranch” are wholly relevant to this question, being true throughout some partition cells while being false throughout all others.

Finally, we need the notion of a pragmatic presupposition, which, following Stalnaker (1978: 321), we can preliminarily define as follows.

A proposition is [pragmatically, AD] presupposed if the speaker is disposed to act as if he assumes or believes that the proposition is true, and as if he assumes or believes that his audience assumes or believes that it is true as well.

While one may dispute this exact characterization, the important point is that one can be disposed to act as if one believed something, and hence pragmatically presuppose it, even if one does not actually believe it. And this can be rational too, e.g., if it helps to simplify the conversation (see, similarly, Stalnaker, 2014: 25 on the idea that propositions can be common ground when the conversational participants accept them whether or not they actually believe them).⁶

Based on these definitions, Hoek (2019: 174) suggests that the proposition conveyed in cases of conversational exculpature is determined as follows.

The proposition conveyed by S in C = the unique proposition, p, such that

• (i) p is wholly relevant to the question under discussion (henceforth, QUD), and

• (ii) p is equivalent to the proposition expressed by S in C conditional on the speaker’s pragmatic presuppositions.⁷

Turning to loose talk, Hoek suggests that loose talk is a sub-type of conversational exculpature, where the pragmatic presuppositions involved are scale presuppositions, which in turn suggest a fitting QUD. Scale presuppositions are presuppositions that arise when one uses an expression that belongs to a conventionally associated scale. The content of a scale presupposition is that a relevant subject falls under one of the expressions on the given scale. This is best illustrated with number and measurement expressions. The expression “6 foot 2”, for instance, occurs on the following scale, which is conventionally used to describe personal heights.

{… “5 foot 11”, “6 feet”, “6 foot 1”, “6 foot 2” …}⁸

Correspondingly, an utterance of, e.g., (7) triggers the scale presupposition that Hannah falls under one of the expressions on the above scale, i.e., that she is an exact number of inches tall. The corresponding QUD is how tall Hannah is to the nearest inch, which induces partition cells containing worlds where Hannah is n ± ½ inches tall. The scale presupposition is obviously false (or unjustified at least), but as we’ve seen, it can be rational to presuppose falsities, and that’s what we do in the case at hand, according to Hoek.

Here is how the indicated scale presupposition and the corresponding QUD can be used to predict that sentence (7) conveys the loose content that Hannah is 6 foot 2 ± ½. The latter proposition is wholly relevant to the QUD thereby satisfying (i). After all, it is true throughout the partition cell for “6 foot 2 ± ½” and nowhere else. This proposition also satisfies (ii), being conditionally equivalent to the proposition expressed, i.e., the proposition that Hannah is exactly 6 foot 2. If Hannah is exactly 6 foot 2, then she is 6 foot 2 ± ½ independently of any additional presuppositions. Meanwhile, if she is 6 foot 2 ± ½ and we assume, via the scale presupposition, that she is an exact number of inches tall, then she must be exactly 6 foot 2. No other proposition satisfies conditions (i) and (ii). Hence, the proposition that Hannah is 6 foot 2 ± ½ is conveyed.

Promising as it may seem, this theory of loose talk cannot be true either if the Endpoint View is. On Hoek’s account, loose talk arises against the background of a scale presupposition, which, in turn, depends on a scale of expressions that encodes conventional ways to describe, e.g., the height of a person. To apply this account to, e.g., (1) and (4), we need such a scale for “flat” and “bumpy”, containing expressions conventionally used to describe an object’s degree of flatness. No scale suits our purposes, however. Hoek (2019: 173) suggests the following scale.

{“flat”, “bumpy”}

But assuming the Endpoint View, this scale will not do. It would induce the scale presupposition that the table either falls under the predicate “flat” or under the predicate “bumpy”. On the Endpoint View, this means that the table has either a maximum degree of flatness or a non-zero degree of bumpiness.⁹ Since the zero degree of bumpiness is the maximum degree of flatness, this presupposition is entirely trivial, saying merely that the table is either perfectly flat or not. As such, it cannot generate any additional entailments, and hence it will not establish the conditional equivalence between, e.g., the proposition that the table is perfectly flat and the proposition that it is roughly flat.

To be sure, Hoek suggests that, in the above scale, “‘bumpy’ means containing a lot of bumps” (179). If “bumpy” meant that, the scale could be fine. But on the Endpoint View, it does not. It means having a non-zero degree of bumpiness, which leads to the problems described. In response, one could propose to replace “bumpy” by “contains a lot of bumps” in the above scale. But, first, this looks ad hoc. The expression “contains a lot of bumps” is decidedly unconventional. Second, any reason to add this expression would presumably be a reason to also add “contains some bumps”. But those attracted to the Endpoint View will likely hold that this latter expression is semantically equivalent to “bumpy”. Consequently, we end up with the trivial scale presupposition from above, which gets us nowhere.

Hoek’s account of slack regulators (see below) exacerbates this problem. Slack regulators supposedly shift the governing scale to a more fine-grained one. Even if the above scale were defensible, we would thus need further, more fine-grained scales, and it would remain unclear what they should look like.

3 A revised theory of loose talk

If the Endpoint View is correct, then we must abandon all extant theories of loose talk. This might seem like a reductio ad absurdum of the Endpoint View, but in what follows, I argue that the situation is not as bleak as it appears, because a modification of Hoek’s account makes the right predictions after all. I suggest that while scale presuppositions can feature in loose talk, they are not essential to loose talk. Once we abandon the idea of tying loose talk to scale presuppositions, this opens up a wider range of presuppositions that we can use to derive loose contents for absolute adjectives via conversational exculpature. Indeed, there are natural candidates that help to defend the Endpoint View, as I will explain in this section.¹⁰

Consider contexts where absolute adjectives are understood loosely and focus on (2) (“The stick is straight”) and (5) (“The stick is bent”) for concreteness. We might be on a hike looking for a walking stick, we might be looking for a stick to fry our marshmallows, or we might be in the garden looking for a stick to support a loose branch. Suppose we are looking for a walking stick on a hike. If we find a stick, this shared goal determines several questions such as how thick the stick is, whether it is still attached to a tree and whether it has thorns, and all these questions will be represented in the QUD. It does not matter though whether a stick is perfectly straight, and as far as straightness is concerned, the only relevant question is whether the stick is straight enough for our purposes, i.e., straight enough to be used as a walking stick. For short, all that matters is whether the stick is at least roughly straight. Now, contexts in which, straightness-wise, only rough straightness matters naturally come with certain pragmatic presuppositions. In such contexts, we ignore insignificant curvatures, i.e., curvatures compatible with being roughly straight. We pragmatically presuppose that these curvatures do not exist, even if, in fact, they do. Given these assumptions about the context, Hoek’s theory yields the desired outcomes.

Consider maximum adjectives and keep focusing on (2). According to the Endpoint View, this sentence expresses the proposition that the stick is perfectly straight, and Hoek’s modified theory nicely predicts that it instead conveys that the stick is at least roughly straight. This latter proposition is wholly relevant to the QUD, thereby satisfying (i). No partition cell of the QUD contains possible worlds where the stick is at least roughly straight and possible worlds where it is not, as these worlds determine different answers to our question of whether the stick is at least roughly straight. Moreover, the indicated proposition is conditionally equivalent to the proposition expressed thereby satisfying (ii). If the stick is perfectly straight, then it is at least roughly straight, independently of any pragmatic presuppositions. Meanwhile, if the stick is at least roughly straight, it does not have any significant curvatures. Given the pragmatic presupposition that it does not have any insignificant curvatures either, it follows that it has no curvature whatsoever. And this means that it is perfectly straight. Since we do not care about perfect straightness, no other proposition should satisfy (i) and (ii). Take, for instance, the expressed proposition that the stick is perfectly straight. This proposition fails to be wholly relevant because possible worlds that differ only in whether the stick is perfectly straight rather than having some minute curvature go into the same partition cell. These possible worlds determine different answers to the question of whether the stick is perfectly straight, but since this question is conversationally irrelevant, they will be lumped together.

This time, we can tell a parallel story about minimum adjectives. On the Endpoint View, sentence (5), for instance, expresses the proposition that the stick has a non-zero degree of bentness, and Hoek’s modified theory neatly predicts that it instead conveys that the stick is substantially bent. Again, this proposition is wholly relevant to the QUD, satisfying (i). This is because the stick is substantially bent iff it is not at least roughly straight, and because, as indicated, no partition cell of the QUD contains possible worlds where the stick is at least roughly straight and possible worlds where it is not. Furthermore, the indicated proposition is conditionally equivalent to the proposition expressed, thereby satisfying (ii). If the stick is substantially bent, it follows trivially that it has a non-zero degree of bentness. Meanwhile, assume that the stick has a non-zero degree of bentness. On the presupposition that it does not have any insignificant curvatures, this can be the case only if it has significant curvatures. And this means that it is substantially bent. Given that we do not care about perfect straightness, no other proposition satisfies (i) and (ii). Take, for instance, the expressed proposition that the stick has a non-zero degree of bentness. This proposition fails to be wholly relevant because possible worlds where the stick is perfectly straight and otherwise identical worlds where the stick is minimally bent go into the same partition cell. By assumption, we don’t care about perfect straightness, and so these possible worlds determine the same answers to all conversationally relevant questions.

It is important to note here that both the QUD and the associated pragmatic presuppositions are subject to negotiation and accommodation and that they vary as conversational participants decide to drop or adopt shared goals for their conversation or to act, or stop acting, as if some proposition was true. This goes particularly for the (false) pragmatic presuppositions introduced above, which only serve to simplify the conversation (contrast presuppositions actually deemed true). Consider, for instance, the following sentence.

• (10) The stick is straight, but it isn’t perfectly straight.

This sentence standardly conveys that the stick in question has insignificant curvatures, and this clashes with the previously assumed pragmatic presupposition that such curvatures do not exist. The sentence is still fine because this presupposition is in place only when the first conjunct of (10) is interpreted. On pain of infelicity, it is dropped once the second conjunct is interpreted. Or consider a case where we sort sticks into three categories: perfectly straight ones, roughly straight ones and ones that aren’t even roughly straight. This goal presumably entails the question of whether the stick is perfectly straight, but we can still use (10) to convey that the stick is roughly straight. The reason here is that, if the conversational participants hold on to the indicated goal, (10) gets a contradictory reading. An utterance of (10) therefore pushes them to coordinate on a different goal, e.g., the goal of figuring out whether the stick goes into the roughly-straight category. Relative to this goal, (10) gets the desired reading in the familiar way (see also the considerations on slack regulators below).¹¹

4 Some challenges

Proponents of the Endpoint View can appeal to the proposed modified version of Hoek’s account to defend their view. It remains to be seen whether this theory is independently plausible. Hoek (2018, 2019) covers a lot of ground already. In the following, I address residual concerns, including concerns from non-detachability, cancelability and calculability (§4.1), concerns from slack regulators (§4.2) and a concern from unidirectionality (§4.3).

4.1 Non-detachability, cancelability, calculability

Hoek’s account is supposed to be a pragmatic account of loose talk. More specifically, the loose interpretations that this account predicts are supposed to be conversational implicatures in the sense of Grice (1989).¹² Thus, these contents should pass familiar tests for conversational implicatures including the non-detachability test, the cancelability test and the calculability test. Each of these tests raises concerns, which I will address in the following.

Consider non-detachability. Conversational implicatures are non-detachable in the sense that truth-conditionally equivalent sentences carry the same conversational implicatures when used in otherwise identical circumstances (e.g., Grice, 1989: 43). It is somewhat difficult to show that a given conversational implicature is non-detachable because we would have to go through all truth-conditionally equivalent sentences and see if the implicature remains. This is not feasible. Still, the implicatures in cases of loose talk remain in sentences that readily come to mind, such as “The stick is straight/unbent/linear” or “The table is flat/even/plane”. This is good evidence for non-detachability.

Carter (2019: 10–11) raises a challenge to non-detachability based on slack regulators. I discuss slack regulators below.¹³ Independently of slack regulators, one may worry that there are contexts in which the following, truth-conditionally equivalent sentences receive differently strict readings.

• (11) The stick is bent.

• (12) The stick has a non-zero degree of curvature.

If this was so, loose talk would be detachable after all. But the supposed datum is questionable. Consider a context where (11) would be understood loosely. As indi-cated, we might be on a hike searching for a walking stick. Now suppose that in this context, one utters (12). Since talk of “non-zero degrees of curvature” is at least semi-technical, it’s hard to judge what people would make of such an odd utterance. Maybe hearers would conclude that, since the speaker is using technical vocabulary, she is interested in whether the stick is technically straight. This might lead them to a strict, non-loose interpretation. But this is unproblematic. For we have now moved to a relevantly different context featuring a different standard of precision, and to test detachability, we must hold the context fixed.

One could further suggest that there are contexts in which the following sentences are truth-conditionally equivalent while (13) receives a looser interpretation than (14).

• (13) The socks are dry.

• (14) The socks couldn’t be drier.

Again, this would show that loose talk is detachable, but again, the supposed datum is questionable. Normally, a sentence like (14) is interpreted in terms of circum-stantial modality, and on such an interpretation, it is not truth-conditionally equivalent to (13). Even if we don’t possess the means to dry our socks further (say because we are travelling through the rain forest), it does not follow that the socks contain no liquid whatsoever. The sentences are truth-conditionally equivalent only if we interpret (14) as featuring something like metaphysical modality, but this requires quite peculiar contexts, and it is difficult to say how (13) would be interpreted there. I suspect that, in such contexts, (13) is interpreted strictly, in line with non-detachability. For instance, we might have put our socks into a vacuum dryer that has been running for some time, such that the socks really couldn’t be any dryer. In such a context, (14) might receive a metaphysical reading, but sentence (13) would presumably receive a likewise strict interpretation.

More generally, these putative counter-examples either use non-standard vocabulary or non-standard interpretations of standard vocabulary. Either way, this makes intuitions unstable, and it creates noise, which makes it hard to draw semantic or pragmatic conclusions from whatever intuitions we may have.

Consider cancelability. Conversational implicatures are contextually and explicitly cancelable. Contextual cancelability means that we can find contexts in which the implicature disappears. Explicit cancellability means that we can felicitously conjoin the target sentence, s, with a denial of the implicature, t, as in “s, but I don’t mean to imply t’’ or “s, but not-t” (e.g., Grice, 1989: 44; Zakkou, 2018). The implicatures in the case of loose talk are contextually cancelable. When we talk in the mathematical domain, for instance, we may use “straight” and “bent” to convey their strictly literal meaning. Meanwhile, maximum adjectives fail the explicit cancelability test, as sentences of the following kind attest.

• (15) # The stick is straight, but it isn’t even roughly straight.

This may seem problematic, but it is not. We are dealing with implicatures that happen to be entailed by what is said. Being perfectly straight entails being at least roughly straight. Entailed implicatures are generally acknowledged exceptions to the explicit cancelability test (see, e.g., Zakkou, 2018: 3–4 for brief discussion and references). Correspondingly, the non-entailed implicatures in the case of minimum adjectives are cancelable.

• (16) The stick is bent, but I don’t mean to suggest that it is substantially curved. (You see this only under the microscope.)

Overall, the assumed implicatures are just as cancelable as you would expect.

Consider calculability. There are many different ways to spell out this notion, but to a first approximation, an implicature is calculable only if hearers can retrieve it despite the fact that it is not semantically encoded (see Grice, 1989: 30–31 for seminal discussion). How can hearers retrieve loose interpretations?

To begin with, it’s not enough for them to rely on, e.g., familiar conversational maxims like Quality, Quantity, Relation and Manner (Grice, 1989: 26–27). Hearers might notice that, on a strict interpretation, sentences ascribing maximum adjectives are obviously false and therefore violate Quality, while sentences ascribing minimum adjectives literally state trivialities, contrary to Relation and Quantity. Hearers may thus conclude that the speaker must mean something other than what she literally says. They may assume further that what the speaker means must obey the maxims by being true, relevant, etc. But this still leaves many possibilities open, and it doesn’t get them to a loose interpretation rather than, e.g., a metaphorical interpretation or some entirely independent interpretation that happens to be true, relevant, etc.

With Hoek’s theory at hand, however, we can tell a simple story about how hearers derive loose interpretations. Hearers not only assume that speakers tend to obey (something like) the indicated maxims. They also start with a defeasible default assumption that speakers speak loosely and thus mean to convey the loose interpretation of the sentences they use.¹⁴ What is the loose interpretation of a sentence? In my view, Hoek’s theory is best construed as a conceptual analysis of this very notion.

The loose interpretation of a sentence S in context C is the unique proposition, p, such that

• (i) p is wholly relevant to the QUD, and

• (ii) p is equivalent to the proposition expressed by S in C conditional on the speaker’s pragmatic presuppositions.

Hearers obviously cannot explicate this analysis, but this doesn’t mean that they can’t tell whether a given proposition is the loose interpretation of a given utterance. Likewise, people may be at a loss to define pornography while still being able to “know it when they see it”. That’s the kind of ability they use when interpreting each other loosely.¹⁵

4.2 Slack regulators

Let us turn to slack regulators. Expressions that lend themselves to loose talk characteristically come with an associated set of slack regulators. Slack regulators include expressions like “exactly” and “precisely” as they occur in the following sentences involving number and measurement expressions.

• (17) Hannah is exactly 6 foot 2.

• (18) It’s precisely 3pm.

• (19) We had exactly 500 reservations.

The effect of the slack regulators in the previous sentences is to strengthen the proposition conveyed. For the proposition conveyed by an utterance of e.g., (17) to be true, Hannah must be closer to being 6 foot 2 than if the slack regulator “exactly” were missing. Maximum adjectives likewise come with slack regulators including “perfectly” and “completely”.

• (20) The table is perfectly flat.

• (21) The stick is perfectly straight.

• (22) The socks are completely dry.

Again, these expressions strengthen the conveyed proposition in the sentences above. The proposition conveyed by an utterance of e.g., (20) is true only if the table is particularly close to the maximum standard of flatness, closer than it would have to be if “perfectly” were missing.¹⁶

On the face of it, proponents of the Endpoint View who appeal to Hoek’s modified account of loose talk cannot accommodate the described effects of slack regulators (see, similarly, Carter, 2019: 180–181.). Given the Endpoint View, adding a slack regulator like “perfectly” as in e.g., (20) has no truth-conditional effect. The proposition expressed by “The table is perfectly flat” presumably is the proposition that the table is perfectly flat, and according to the Endpoint View, “The table is flat” expresses that very proposition.¹⁷ Given Hoek’s modified account of loose talk, it seems to follow that slack regulators don’t affect the proposition conveyed either, contrary to the data just described. After all, and as indicated, conversational implicatures, which cover loose talk on Hoek’s view, are non-detachable, meaning that truth-conditionally inert material does not affect them.

Hoek (2019: 178) offers an account of slack regulators that may evade this conundrum. Unfortunately, the basic idea is that slack regulators shift the governing scale to a more fine-grained one, thereby shifting the corresponding scale presupposition, thereby affecting the proposition conveyed. Since as we've seen, scale presuppositions play no role in the interpretation of absolute adjectives, this proposal isn’t general enough.

Here is a more promising account. The mentioned slack regulators belong to a broader class of expressions that semantically express meta-linguistic intentions. We could call them interpretation regulators. Interpretation regulators include expressions like “literally”, “technically” and “in the just defined sense”. The standard view of the core meaning of “literally”, for instance, has it that “literally adds no content to a sentence, but simply regulates its interpretation. […] Literally says, in effect, ‘I mean what I’m saying: to understand me correctly you need add nothing to the meaning of my words’.” (Israel, 2002: 425) In the same way, the use of, e.g., “perfectly” in a sentence like (20) semantically expresses the intention not to use the subsequent maximum adjective loosely, or at least to use it less loosely than the context would otherwise suggest.

This explains the observed effects of slack regulators. Take sentence (20). Were “perfectly” missing, hearers would calculate an implicature based on the default assumption that speakers speak loosely. As indicated, though, this default assumption is defeasible, and one evident way to defeat it is to express one’s intention not to speak loosely, as per the slack regulator “perfectly”. This leads to a more literal interpretation and hence to the more demanding interpretation we get.

4.3 Unidirectionality

Loose talk supposedly shows unidirectionality, in the sense that it is easier to reduce the admissible amount of slack than to widen it. Klecha (2018: 92–93) following Lewis (1979: 351–354) presents this supposed datum as a major motivation for his account, and so it is imperative to explain how the suggested, alternative account can deal with this.

The way to address unidirectionality, I think, is to reject the datum. Loose talk is not unidirectional in the suggested way, as the following examples bring out. Consider two carpenters assessing antique tables. The question is whether they must continue working on their tops or whether the tables are ready for sale. One carpenter, Carla, assesses the tables, the other carpenter, Carl, is supposed to take notes, but constantly interferes. The following dialogue transpires.

(23) Carla: The table top of table A is flat. Ready to sell.

Carl: No, it’s not! Look at this dent here. Our customers are more demanding than that.

Carla: OK, OK. The table top of table A isn’t flat. We need to work on that one. The table top of table B isn’t flat either…

The admissible amount of slack has been reduced in this dialogue, in that “flat” now allows fewer dents (assume that Carla is aware of the mentioned dent throughout). Consider the following dialogue next.

(24) Carla: The table top of table A isn’t flat. We still need to work on this one.

Carl: Sure it is! Just because of this little dent here? No one will notice. We’ll sell that one right away.

Carla: OK, OK. The table top of table A is flat. Ready to sell. The table top of table B is flat too…

The admissible amount of slack has been widened here, with just the same amount of effort with which it was raised in the previous case. There is no asymmetry, and I think the same goes for all functioning conversations where standards of precision shift.

Any residual sense of unidirectionality results from a very general, easily explicable phenomenon, not specifically tied to loose talk. I call it literalism. People sometimes interpret utterances overly literally. This can happen for a variety of reasons. They may have missed the pragmatic message; they may intend to appear witty or funny (Kennedy & McNally, 2005: 359); or they may try to suggest that you should speak literally, because of confusion about how orderly communication works or specific contextual demands (e.g., in court, when we do not want to rely too much on the vagaries of pragmatic interpretation). Literalism can target any type of non-literality, including loose talk. The following dialogues bring this out.

(25) Metaphor:

• A: I got tomatoes coming out of my ears.

• B: Tomatoes don’t grow in human ears.

• A: Yes, technically, …

(26) Irony:

• A: She’s a fine friend.

• B: Didn’t you just say that she betrayed you? Then she’s not a fine friend.

• A: Well, yes, …

(27) Loose talk:

• A: The stick is straight.

• B: Nothing is truly straight.

• A: Right, but …

When you face a literalist, you must speak literally unless you want to continuously engage with complaints about false or otherwise defective literal contents. In the specific case of loose talk, you thus begin to act as if no slack was admissible, hedging everything in appropriate ways (you say, e.g., “The stick is roughly straight”). In this way, the availability of literalism makes it easy to reduce the admissible amount of slack, indeed, to eliminate it entirely, and a sense of unidirectionality results. Notice that while a phenomenon of anti-literalism, where hearers complain whenever the speaker speaks literally, is conceivable, it is at least very uncommon.¹⁸

5 Conclusion

The Endpoint View of absolute adjectives is at odds with ordinary usage by positing extremely demanding semantic values for maximum adjectives and extremely undemanding semantic values for minimum adjectives. Loose talk supposedly explains this, but this response strategy clashes with extant theories of loose talk. I have defended the Endpoint View against this challenge by offering a revised version of Hoek’s recent theory of loose talk. Taken together, the Endpoint View and the proposed theory of loose talk offer a compelling account of absolute adjectives. Furthermore, I have shown that the proposed theory of loose talk accommodates recalcitrant data while being defensible against several important concerns. In this way, I have not only defended the Endpoint View, I have also offered what I hope is strong evidence for the proposed theory of loose talk.¹⁹