Thomasson’s defence of easy arguments

Abstract

According to Amie Thomasson, most of the existence questions that concern ontologists can be easily answered using trivial arguments from uncontroversial premises. In this paper, I examine Thomasson’s main argument for this significant and striking thesis, focusing on the crucial premise that sortal terms have application conditions. I argue that Thomasson’s defence of this premise fails to support it. I also argue that it faces a serious objection.

1 Introduction

According to Amie Thomasson:

[…] existence questions can be answered easily—often by trivial arguments from uncontroversial premises, and always by invoking nothing more than competence with language and reasoning and straightforward empirical knowledge.

(Thomasson, 2015, p. 318)

Thomasson has developed, and defended, this view, across many papers and books (2007, 2008, 2009a, 2009b, 2010, 2015, 2019a, 2019b) and it is the central topic of her 2015 book Ontology Made Easy (henceforth, OME). In OME, Thomasson offers the following examples of the way in which existence questions can be easily answered:

(1):

a. Snow is white

b. Therefore, that snow is white is true.

c. Therefore, there is a proposition (namely, that snow is white).¹

(2):

a. Beyoncé’s dress is red

b. Therefore, the dress has the property of redness.

c. Therefore, there is a property (namely, the property of being red).²

(3):

a. The cups and the saucers are equinumerous.

b. Therefore, the number of cups = the number of saucers.

c. Therefore, there is a number (namely, the number that is the number of cups and the number of saucers).³

(4):

a. There are particles arranged tablewise.

b. Therefore, there is a table.⁴

(5):

a. May was born on Monday.

b. Therefore, May’s birth occurred on a Monday.

c. Therefore, there is an event (namely, May’s birth).⁵

(6):

a. Arthur Conan Doyle used the name ‘Sherlock Holmes’ fictionally in writing a number of stories.

b. Therefore, there is a fictional character (namely, Sherlock Holmes).⁶

(7):

a. Bob and Ben said the proper vows in the right context.

b. Therefore, there is a marriage (namely, the marriage of Bob and Ben).⁷

This list of easy arguments is only intended as a suggestive sample. In fact, Thomasson defends the following thesis:

THE EASY ONTOLOGY THESIS: There are easy arguments for the existence of most of the controversial entities that ontologists debate and these arguments are sound.

The easy ontology thesis is significant and striking. It is significant because if it is true then propositions, properties, numbers, and many other controversial kinds exist. It is striking because it suggests that coming to know this is no great feat. For example: if easy arguments like (2) are sound, then ancient Mesopotamian farmers, observing that their dates are ripe, may be in a position to know that their dates have the property of being ripe and that properties exist. So, easy arguments appear to lead to easy knowledge.

The easy ontology thesis may be striking for another (closely related) reason. According to Thomasson, the easy ontology thesis strongly supports the view “that something is wrong with many of the hard ontological debates that have been earnestly engaged in over the past fifty years or so” (2015, p. 158). Thomasson calls this view metaontological deflationism. Does the easy ontology thesis support metaontological deflationism, as Thomasson suggests? This is an interesting question, but not the topic of this paper which focuses on Thomasson’s defence of the easy ontology thesis.

The easy ontology thesis is a conjunction. The first conjunct says that there are easy arguments for the existence of most of the controversial entities that ontologists debate. Thomasson presents this as a legitimate generalization from a set of cases. The cases are the easy arguments listed above (and a few others that I’ve omitted for simplicity). The second conjunct says that these arguments—the easy arguments described by the first conjunct—are sound; i.e., their premises are true and they are valid.

According to Thomasson, the initial premises of (1)–(7) are “undisputed,” “uncontroversial,” or “uncontested” so she invests most of her energy in defending the claim that easy arguments are valid. Her argument is roughly this. Words are associated with rules of inference that we must know in order to be competent speakers. But these rules, together with our general reasoning capacity, licence us to draw the inferences in easy arguments like (1)–(7).

I begin, in Sect. 2, by examining the claim that the premises of easy arguments are “undisputed,” “uncontroversial,” or “uncontested”. In Sect. 3, I present Thomasson’s argument for the claim that easy arguments are valid, which I call the competence argument. In Sects. 2 and 3, we will see that Thomasson’s defence of the easy ontology thesis has a lot of moving parts. That’s not a criticism. After all, we can hardly expect a simple defence of such a significant and striking thesis. The main aim of Sects. 2 and 3 is to show how these moving parts fit together.

In Sects. 4 and 5, I will focus on the crucial second premise of the competence argument, which I call the sortal thesis. The sortal thesis is the claim that sortal terms—terms like ‘proposition’, ‘property’, ‘number’, ‘table’, ‘event’, ‘fictional character’ and ‘marriage’—are associated with informative application conditions, which are inference rules of a certain kind. In Sect. 4, I present, and reject, Thomasson’s arguments for the sortal thesis. And in Sect. 5, I argue that the sortal thesis faces a serious objection.

2 True premises?

Do easy arguments have true premises? There are two questions here: a narrow question and a broad question. The narrow question is whether the initial premises of (1)–(7) are true. The broad question is whether this generalizes to a sufficiently large set of easy arguments to make it the case that there are easy arguments, with true premises, for the existence of most of the controversial entities that ontologists debate. I will focus on the narrow question.

According to Thomasson, the initial premises of (1)–(7) are “undisputed,” “uncontroversial,” or “uncontested”. Unfortunately, none of these arguments have premises that are undisputed (etc.). For example: (1a), the claim that snow is white, is disputed by philosophers who deny the existence of inanimate composite objects (e.g., Merricks, 2001; van Inwagen, 1990).

However, although the initial premises of (1)–(7) are not undisputed in the usual sense, they may be undisputed in an extended sense. On this extended sense, the premises of an argument are undisputed iff for every philosopher there is a version of the argument from premises that they believe. Consider (1a) again and notice that (1) is an argument of a certain form:

(8):

a. p.

b. Therefore, that p is true.

c. Therefore, there is a proposition (namely, that p).

Further, it’s plausible that any simple instance of this form is an easy argument. But since the initial premise, (8a), can be any simple proposition whatsoever, for every philosopher there is a version of this argument from a premise that they believe. So, (1a), the claim that snow is white, is undisputed in the extended sense. Plausibly, (2a) and (3a) are also undisputed in this sense because it’s plausible that every philosopher believes at least one simple instance of ‘a is F’ and ‘the Fs and the Gs are equinumerous’.

What about the initial premises of (4)–(7)? Are they undisputed in the extended sense? Consider (4a), the claim that there are particles arranged tablewise. Thomasson acknowledges that (4a) is not undisputed in the narrow sense when she admits that “[s]ome table-eliminativists […] would accept neither that there are particles nor that there are tables” (Thomasson, 2015, n. 15, p. 142). However, eliminativists about ordinary objects, like tables, “need a non-committal paraphrase strategy to distinguish their position from ‘the madman’s’, and [they] need to capture the sense in which we may still say something true, or ‘nearly as good as true’ in our talk of ordinary objects” (Thomasson, 2019b, p. 260).

Further, Thomasson believes that if an adequate paraphrase can be found, we can use it as the (true) premise of an easy argument for the existence of a table. So: any philosopher who wants to deny that there are tables must offer an adequate paraphrase of the sentence ‘there are tables’ and once they do Thomasson will offer an easy argument from this paraphrase back to the claim that there are tables. Whether this argument is valid is another question, but that’s not the question we are addressing in this section. Presumably, Thomasson can make a similar claim about the initial premises of (5)–(7) as she makes about the initial premise of (4). So, by her lights, (4)–(7) are undisputed in the extended sense.

3 Valid inferences?

Are easy arguments valid? Again, there is a narrow question and a broad one. The narrow question is whether (1)–(7) are valid. The broad question is whether this generalizes to a sufficiently large set of easy arguments to make it the case that there are valid easy arguments for the existence of most of the controversial entities that ontologists debate. Again, I will focus on the narrow question.

Thomasson argues that (1)–(7) are analytically valid, in an epistemic sense. There are two contrasts here. First, there is a contrast between formal validity and analytic validity. An argument is formally valid iff it is necessarily truth-preserving in virtue of its logical form. In contrast, an argument from p to q is analytically valid iff the conditional ‘if p then q’ is analytically true.

Second, there is a contrast between a metaphysical notion of analyticity and an epistemic one. A sentence is analytically true in a metaphysical sense iff it is true in virtue of its meaning alone. In contrast, a sentence is analytically true in an epistemic sense iff its truth can be known by competent speakers merely by drawing on their linguistic competence (and, if need be, their general reasoning skills).⁸ Here is Thomasson’s argument that easy arguments are analytically valid in an epistemic sense:

THE COMPETENCE ARGUMENT

P1.:

The first applied thesis: (1)–(7) each has a conclusion of the form ‘there is a K,’ where ‘K’ is a sortal term.

P2.:

The sortal thesis: Every sortal term is associated with informative application conditions, which are rules of inference that provide sufficient conditions for determining whether the sortal term applies—i.e., whether it (successfully) refers, or denotes.

P3.:

The second applied thesis: If the premises of (1)–(7) are true, then the application conditions for each of the relevant sortal terms is fulfilled—i.e., the sufficient conditions for their application are met; hence, they apply to something.

P4.:

The competence thesis: Every speaker must (at least tacitly) know the rules that are associated with a term in order to be competent with that term.

C.:

Competent speakers can come to know that if the premises of (1)–(7) are true, then their conclusions are true, merely by drawing on their linguistic competence (and, if need be, their general reasoning skills). In other words: (1)–(7) are analytically valid in an epistemic sense.

I think that the competence argument (or a nearby version of it) is valid, although its validity is not obvious. In any case, I will assume its validity and focus on whether its premises are true. In the rest of this section, I will discuss each premise in turn. However, I cannot evaluate all these premises in a single paper, so in Sects. 4 and 5. I will focus on the sortal thesis.

3.1 The first applied thesis (P1)

According to the first applied thesis (P1), (1)–(7) each has a conclusion of the form ‘there is a K,’ where ‘K’ is a sortal term. In the literature, there is no standard definition of a sortal (and Thomasson doesn’t offer one). However, as far as I can see, the competence argument only relies on the claim that a sortal is a common count noun that functions as a referring (or denoting) term. Further, the relevant terms in (1)–(7) are common count nouns that function in this way.⁹ So, if we adopt this definition, then the first applied thesis is true.

3.2 The sortal thesis (P2)

There are two distinct sortal theses in OME: a linguistic sortal thesis (LST) and a conceptual sortal thesis (CST). LST is the thesis that appears as the second premise (P2) of the competence argument. However, LST and CST are connected and must be evaluated together. In this section, I will introduce LST, leaving CST for Sect. 5. According to LST, every sortal term is associated with informative application conditions. On Thomasson’s account, an application condition is a rule of inference that provides sufficient conditions for determining whether a sortal term applies. This rule has the following form:

(AC) From p, you are licenced to infer that there is something that ‘K’ applies to. (Alternatively, from p, you are licenced to infer that the sentence ‘there is a K’ is true).

However, (AC) only provides sufficient conditions for the application of ‘K’ if it is backed by a true conditional of the form:

(I) If p, then there is something that ‘K’ applies to.

For simplicity, Thomasson often treats application conditions as though they were true instances of (I), and I will follow suit. Further, application conditions can be informative (substantive) or uninformative (trivial). Consider the following examples:

• (i) If there is a triangle, then there is something that the term ‘triangle’ applies to.

• (ii) If there is a plane shape with three angles and three sides then there is something that the term ‘triangle’ applies to.

(ii) is informative; (i) is not. (ii) is informative because one can understand the antecedent of (ii), taken as a sentence, without understanding the meaning of the word ‘triangle’. Alternatively, one can grasp the antecedent of (ii), taken as a thought, without possessing the concept triangle.¹⁰

A thought is (roughly) the fine-grained meaning of a sentence. For example: the thought that I express when I say ‘I am hungry’ is distinct from the thought that Bob expresses when he addresses me and says ‘you are hungry’. (Note: throughout this paper, I will use small caps to refer to concepts and thoughts). In contrast to (ii), (i) is uninformative because one cannot understand, or grasp, the antecedent of (i), without understanding the meaning of the word ‘triangle’ or possessing the concept triangle.

Finally, an application condition is fulfilled if its antecedent is true. For example: if (ii) is one of the application conditions for the term ‘triangle’, then it is fulfilled if there is a plane shape with three angles and three sides. However, if (ii) is fulfilled, then there is something that the term ‘triangle’ applies to. Further, it’s uncontroversial that if there is something that the term ‘triangle’ applies to, then there is a triangle and (fairly) uncontroversial that every competent speaker knows this.¹¹

Noting this, I will further simplify the way that I talk about application conditions by treating them as true conditionals of the form ‘if p, then there is a K’. Thus, (ii) becomes:

(iii) If there is a plane shape with three angles and three sides, then there is a triangle.

3.3 The second applied thesis (P3)

According to the second applied thesis (P3), if the premises of (1)–(7) are true, then the application conditions for each of the relevant sortal terms is fulfilled. Consider (7):

(7):

a. Bob and Ben said the proper vows in the right context.

b. Therefore, there is a marriage (namely, the marriage of Bob and Ben).

The relevant sortal in (7) is ‘marriage’. According to P2, ‘marriage’ is associated with informative application conditions and, according to Thomasson, the following is among them: if x and y say the proper vows in the right context, then there is a marriage.¹² Further, if (7a) is true, then the application conditions for ‘marriage’ are fulfilled.

A similar point can be made about (4) and (6), but complications arise for (1)–(3) and (5). For example: the inference from (2a)—Beyoncé’s dress is red—to (2c)—there is a property—could be directly licenced by an application condition, but then (2b)—the dress has the property of redness—would be redundant. On the other hand, neither the inference from (2a) to (2b) nor the inference from (2b) to (2c) has the right form to be licensed by an application condition.

This issue can be addressed by deleting (2b) or by arguing that (AC) is not the only form that an application condition can take. Here, it makes no difference which option we choose. So, for simplicity, I will suppose that (2b) is redundant and that the following is one of the application conditions that Thomasson associates with the sortal ‘property’: if x is red, then there is a property. Given this, the application conditions for ‘property’ are fulfilled, if (2a) is true. Further, the complications that arise in (1), (3), and (5) are similar to the complications that arise in (2) and they can be resolved in a similar manner.

This argument for P3 only succeeds if the relevant sortal terms in (1)–(7) are associated with the right application conditions—specifically, application conditions that are fulfilled if the premises of (1)–(7) are true. Of course, you could reject these specific application conditions but if LST is true, then you must provide alternate application conditions. What are the plausible alternatives? According to Thomasson, every plausible alternative adds an object condition. Consider:

(iv):

A. If x is red, then there is a property.

B. If x is red because it instantiates an object, then there is a property.¹³

(v):

A. If the Fs and the Gs are equinumerous, then there is a number.

B. If the Fs and the Gs are equinumerous because there is an object that numbers them, then there is a number.

(vi):

A. If there are Fs arranged tablewise, then there is a table.

B. If there is an object composed by some Fs arranged tablewise, then there is a table.

(vii):

A. If x and y say the proper vows in the right context, then there is a marriage.

B. If x and y say the proper vows in the right context and, in so doing, they constitute a new (social) object, then there is a marriage.

The (A) examples are the application conditions that Thomasson proposes. The (B) examples are the alternatives. At this point, Thomasson argues that there are only two senses of the word ‘object’ (or ‘thing’): a sortal sense and a ‘covering’ sense. Further, she argues that neither sense can appear in the antecedent of the application conditions associated with the relevant sortals in (1)–(7) (2007, p. 157–8, 2015, p. 108–11).

On the sortal sense, an object is (roughly) a coherent chunk of matter bounded in space. On this reading, the (B) sentences are clearly false, with one exception. The following may be true: if there is a sortal object composed by some particles arranged tablewise, then there is a table. However, someone who denies the existence of tables will also want to deny the existence of sortal objects, so this alternate application condition is not available to them.

On the covering sense, the word ‘object’ is “a dummy sortal” (2015, p. 109) with the following application condition: if there is a K, then there is an object in the covering sense, (where ‘K’ is any sortal except ‘object’ in the covering sense). According to Thomasson, the covering notion of an object cannot figure in the antecedent of an application condition without making it uninformative. So, again, the (B) sentences cannot provide alternate application conditions.

One objection to this argument is that there is a third sense of ‘object’. Schaffer (2009) and Korman (2019) both reject Thomasson’s argument on these grounds. They argue that ontologists appeal to a perfectly legitimate sense of ‘object’ when they attempt to quantify over absolutely everything whatsoever. In response, Thomasson argues that the ontologist’s notion of an object is irredeemably defective (2009, p. 460–2). I am inclined to agree with Schaffer and Korman. In any case, if you want to place pressure on the competence argument, P3 is a good place to press. However, any defence of P3 presupposes P2 (i.e., LST), so P2 is an even better place to press.

3.4 The competence thesis (P4)

According to the competence thesis, every speaker must (at least tacitly) know the rules that are associated with a term in order to be competent with that term. This assumes that there is a common set of truths that every competent speaker must know in order to be competent—an assumption that has been challenged by Williamson (2007) in his discussion of epistemic analyticity.

Williamson argues that even if competent speakers must know a set of truths to count as competent, there is no common set of truths that every speaker must know. That is, different speakers might know different sets of truths that don’t completely overlap (or don’t overlap at all). Thomasson (2015) addresses Williamson’s arguments, but I won’t evaluate her success (or the general issue) here.

This brings Sects. 2 and 3 to a close. As we have seen, Thomasson’s defence of the easy ontology thesis has a lot of moving parts. Again, that’s not a criticism. Further, I hope to have shown how these moving parts fit together. In the next two sections (Sects. 4, 5), we turn our attention to the sortal thesis, which is a crucial premise in Thomasson’s defence of the easy ontology thesis.

4 Thomasson’s defence of the sortal thesis

Thomasson asks: “Why think that our terms have application conditions […]?” (2015, p. 95). Her answer is the metasemantic argument:

The usual grounds given for thinking that our terms have no application conditions […] come from those who are inclined to purely causal theories of reference, which would hold that our terms and concepts derive their entire content from the things and kinds they pick out. I have argued extensively elsewhere […], however, that causal factors alone cannot sufficiently disambiguate whether a term refers and if so to what. For purely causal theories of reference notoriously face the so-called ‘qua problem’. In any situation, a speaker who would ground the reference of a term is apparently related to a great many things: a person, her clothes, the color of her hair, her current temporal part, the air between them, and so forth. Something more is needed to help disambiguate what is referred to (e.g., that the name is introduced as a person-term). Moreover, a similar problem arises in determining whether our terms refer at all. For speakers are always in causal contact with a great many things, so how could attempts to ground reference ever fail, and a term fail to refer? As I have argued elsewhere […], in order to allow that terms may sometimes fail to refer, we need to allow that our terms also come with rules in the form of something like application conditions, which play the needed role in determining whether or not a term succeeds in referring at all (or if its uses end in a block).

(Thomasson, 2015, footnotes removed, p. 95)

This passage contains two distinct arguments. The first is aimed at someone who rejects a purely causal theory of reference and goes like this:

METASEMANTIC ARGUMENT 1 (MA1)

P1.:

If purely causal theories of reference are false, then LST (the linguistic sortal thesis) is plausible.

P2.:

Purely causal theories of reference are false.

C.:

LST is plausible.

The reason for believing P2 is that purely causal theories of reference face the qua problem and the reference-failure problem (which Thomasson doesn’t name but discusses in the passage above). Of course, you may wish to defend purely causal theories of reference against these problems but MA1 is aimed at someone who rejects purely causal theories of reference, so its success turns on the truth of P1.

Thomasson offers no explicit reason for believing P1. Perhaps she thinks that LST is initially plausible—the view that one would hold, if there were no reason to believe in a causal theory of reference. I don’t find LST initially plausible. In any case, in Sect. 5. I show that it faces a serious objection, so even if it has some initial plausibility, that is not enough to recommend it.

Another possibility: according to Thomasson, the main rivals to a causal theory are views that support LST. So, if causal theories of reference are false and there are no serious objections to LST, then LST is plausible. However, this cannot be assumed and requires a discussion of other metasemantic theories, which Thomasson doesn’t offer. Further, as we will see below, it doesn’t follow from metasemantic inferentialism, which Thomasson mentions and provisionally endorses in a footnote. I conclude that MA1 is not successful (pending the discussion in Sect. 5).

The second metasemantic argument is aimed at someone who endorses (or is tempted to endorse) a purely causal theory of reference and goes like this:

METASEMANTIC ARGUMENT 2 (MA2)

P1.:

Purely causal theories of reference face the qua problem and the reference-failure problem.

P2.:

The most plausible solutions to these problems support LST.

C.:

It is likely that LST is true.

P1 is clearly true. Thomasson argues that P2 is true, but on inspection her reasons appear to support a weaker claim only. According to Thomasson, the qua problem and the reference-failure problem can only be solved by assuming that referring terms are associated with descriptive information about the category that the referent belongs to, if it exists. Discussing the qua problem, she says: “[s]omething more is needed to help disambiguate what is referred to (e.g., that the name is introduced as a person-term)” (2015, my italics, p. 95). In the following passage, she is even more explicit:

[…] the qua problem […] may be resolved by moderating pure causal theories, holding that reference is only unambiguously established to the extent that our nominative terms are associated with a high-level conceptual content establishing what category of entity is to be referred to by the term, if it refers at all.

(Thomasson, 2007, p. 38)

In this passage, Thomasson argues that a nominative term (i.e., a noun) must be associated with descriptive information about the category that its referent, or referents, belong to. In other words, nouns are associated with conditionals like the following:

• (viii)
  

  If ‘Sally’ (successfully) refers, then Sally is a person.

• (ix) If ‘tiger’ (successfully) denotes, then tigers are animals.

But neither (viii) or (ix) is an application condition in the relevant sense—they provide necessary conditions for the application of a term, rather than sufficient conditions.¹⁴ So, nothing Thomasson says gives us a reason to believe that the best solution to the qua problem supports LST.¹⁵ What about the reference-failure problem? Thomasson says that “in order to allow that terms may sometimes fail to refer, we need to allow that our terms also come with rules in the form of something like application conditions” (2015, p. 95). In the following passage, she spells this out with an example:

[…] if I attempt to ground the name ‘Orky’ as the name for an animal (swimming near my boat), my attempt to ground the reference may fail if all that has perturbed the water near my boat is a large clump of seaweed, or a strange event in the ocean current causing an unusual wave.

(Thomasson, 2007, p. 39).

The idea is this: if I attempt to fix the referent of the name ‘Orky’ by pointing and saying ‘that is Orky’ I must intend to name an entity of a certain sort (in this case an animal). If ‘Orky’ refers to whatever caused me to say ‘that is Orky’ then this name cannot fail to refer. For example: if a clump of seaweed caused me to say ‘that is Orky’ then that is what ‘Orky’ refers to. If we accept this argument, then we will conclude that ‘Orky’ is associated with conditionals like the following:

(x) If ‘Orky’ refers, then Orky is an animal.

But (x) is not an application condition in the relevant sense—it provides a necessary condition for the application of the name ‘Orky’, rather than a sufficient condition. So, although Thomasson says that the best solution to the reference-failure problem is one on which referring terms “come with rules in the form of something like application conditions”, her discussion of the solution does not support this claim. I conclude that MA2 is not successful.

In a footnote, Thomasson (2015) hints at a third metasemantic argument. Here she admits:

I am now […] more inclined to an inferentialist or use theory of meaning (see Brandom, 1994; Horwich, 1999) […] If one takes this general approach, application conditions for general nouns can be treated as among the introduction rules licensing us to apply a certain term or concept (and partially constitutive of its meaning).

(Thomasson, 2015, n. 13, p. 95)

If this is intended as an argument, the relevant argument might be drawn out like this:

METASEMANTIC ARGUMENT 3 (MA3)

 P1.:

If inferentialism is true, then sortal terms have informative application conditions.

P2.:

Inferentialism is true.

C.:

Sortal terms have informative application conditions.

Inferentialism is a package of semantic or metasemantic views, or the analogue of these views for mental content. Semantic inferentialism is the view that the meaning of a linguistic expression is identical to its inferential role (i.e., its meaning is the property of having a certain inferential role). For example: according to a version of this view, the meaning of ‘and’ is its inferential role, which can be specified by the standard introduction and elimination rules for conjunction.¹⁶

Metasemantic inferentialism is usually a view about meaning determination or semantic competence. Inferentialism about meaning determination is the view that the meaning of a linguistic expression is determined by its inferential role. Put differently, an expression has the meaning it does in virtue of the inferential role that it plays. For example: according to a version of this view, the meaning of ‘and’ is determined by its inferential role in the following way: ‘and’ has whatever meaning it must have to make the standard introduction and elimination rules for conjunction come out as valid.

Inferentialism about semantic competence is the view that a speaker must know the inferential role that a linguistic expression plays in order to be competent with this expression. For example: according to a version of this view, a speaker is competent with ‘and’ iff they know the standard introduction and elimination rules for conjunction.

The main problem with MA3 is that its first premise (P1) is false. There are versions of inferentialism that don’t require sortal terms to have informative application conditions. For example: logical inferentialism is limited to logical constants (like ‘and’) and says nothing about sortals (or nouns more broadly). However, in the passage above, Thomasson approvingly cites Brandom (1994) and Horwich (1999). So, perhaps, her argument is this: if Brandom or Horwich’s inferentialism is true, then sortal terms have informative application conditions; either Brandom or Horwich’s inferentialism is true; therefore, sortal terms have informative application conditions. But again, the first premise of this argument is false.

I will focus on Brandom (but similar remarks apply to Horwich). According to Brandom (1994), every meaningful term is associated with an inferential role that is characterized by a set of inference rules.^{17, 18} However, many of these rules have a ‘non-deductive’ character (I use this term loosely). For example: the word ‘red’ might be associated with a rule backed by the following conditional: if I see an object that appears red, and there is no countervailing evidence, then it is likely that there is a red object.¹⁹

So, if Brandom’s inferentialism is true, then every sortal term is associated with an inferential role, but it doesn’t follow that every sortal term is associated with an informative application condition that is backed by a true conditional of the form: if p, then there is a K. For example: the sortal ‘table’ might be associated with a rule that is backed by the conditional: if I see something with such-and-such an appearance, and there is no countervailing evidence, then it is likely that there is a table.²⁰

5 An objection to the sortal thesis

According to LST, every sortal term is associated with informative application conditions. For example, (iii) might be one of the application conditions for the sortal ‘triangle’:

(iii) If there is a plane shape with three angles and three sides, then there is a triangle.

It is worth noting that ‘shape’, ‘angle’, and ‘side’ are all sortals (i.e., they are common count nouns that function as referring, or denoting, terms—see Sect. 3.1). However, the fact that (iii) is one of the application conditions for ‘triangle’ in a certain language (if it is) does not entail that this language has the words ‘shape’, ‘angle’, and ‘side’ in its vocabulary. Or as Thomasson puts it: “for a term to have application conditions does not require that those conditions be statable at all” (Thomasson, 2015, p. 92). The idea is that (iii) directly connects the word ‘triangle’ to the concepts shape, angle, and side, and only indirectly (if at all) connects it to words that express these concepts. This brings us to the conceptual sortal thesis (CST).

As I understand it, a sortal concept is a concept that could be expressed by a sortal term in some possible language. According to CST, every sortal concept, k, is associated with informative application conditions. In this context, an application condition is a rule of inference that is backed by a true conditional of the form: if p, then there is something that the concept k applies to; or more simply: if p then there is a K.

Thomasson treats LST and CST as interchangeable. This suggests that she endorses the following biconditional: LST is true iff CST is true. I will argue (below) that CST is tangled in the horns of a trilemma, which gives us a reason to believe that it’s false. However, if CST is false and the biconditional connecting LST and CST is true, then LST is false as well.

Can Thomasson evade this argument by rejecting the biconditional? Unlikely. The relevant part of this biconditional, the left-to-right direction, is very plausible (and I doubt that Thomasson would want to deny it in any case). The left-to-right direction states: if LST is true, then CST is true. This is logically equivalent to the statement: if CST is false, then LST is false. So, suppose that CST is false. In other words, suppose that you possess a sortal concept that does not have informative application conditions. On the face of it, there is nothing to stop you from having, or introducing, a word that expresses this concept. However, if such a word exists, then LST is false, because there is a sortal term—a term that expresses a sortal concept—that does not have informative application conditions. Hence, if CST is false (as I will argue), then LST is false as well.

Like LST, CST is paired with a competence thesis. According to the conceptual competence thesis, every cognizer must (at least tacitly) know the rules that are associated with a concept in order to possess that concept. Further, when easy arguments are framed in conceptual terms, Thomasson claims that they are conceptually valid in an epistemic sense.

An argument from p to q is conceptually valid iff the conditional if p then q is conceptually true. And a thought is conceptually true in an epistemic sense iff its truth can be known by a cognizer merely by drawing on their conceptual competence (and, if need be, their general reasoning skills). Now, suppose that k is a sortal concept and that the following is one of its application conditions:

• (xi) If p, then there is a K.

I will say that a thought t contains a concept c iff we cannot grasp t without grasping c.²¹ Does p contain any sortal concepts? If it does, then according to CST each of these sortal concepts has informative application conditions. Let’s suppose that p contains some sortal concepts, including the concept k*, which has the following as one of its application conditions:

(xii) If q, then there is a K*.

Does q contain any sortal concepts? If it does, then according to CST each of these sortal concepts has informative application conditions. Presumably, this can’t go on forever. At very least, it can’t go on forever if the conceptual competence thesis is true because anyone who possesses the concept k (e.g., the concept triangle) would have to (at least tacitly) know an infinite series of informative application conditions, which is implausible.

Further, the application conditions of our sortal concepts cannot be jointly circular. For example: if (iii) is among the application conditions for triangle, then (xiii) cannot be among the application conditions for shape:

(xiii) If there is either a triangle, or a square, or a circle, … , then there is a shape.

(iii) is informative because we can grasp its antecedent without possessing the concept triangle. However, if (xiii) is among the application conditions for the concept shape, then we can’t grasp the antecedent of (iii) without possessing the concept triangle. Hence, (iii) is not informative. This leaves only one option: the hierarchy of informative application conditions must bottom out. For example:

(xiv) If p(K*), then there is a K.

(xv) If q(K**), then there is a K*.

(xvi) If r(K***), then there is a K**.

     ⋮

(#)  If z, then there is a K*** ^….

In this diagram, p(K*) expresses the thought that p, which contains the concept k*; q(K**) expresses the thought that q, which contains the concept k**; and so on. In contrast, z expresses a thought that doesn’t contain any sortal concepts whatsoever.

Here’s the rub: it’s not obvious that we possess any sortal-free thoughts, even though we possess sentences that don’t contain sortal terms. For example: the sentence ‘it is raining’ doesn’t contain any sortal terms, but arguably we can’t grasp its meaning without grasping sortal concepts like raindrop, raincloud, sky, ground, and location. If this is correct, then it is raining is not a sortal-free thought. Further, although ‘raining’ is not a sortal term (because it’s a verb, not a noun) it doesn’t follow that raining is not a sortal concept. Could the concept raining be expressed by a sortal term in some possible language? The answer depends on how we understand this concept. If the concept raining describes events of a certain kind, the raining kind, then arguably it could be expressed by a sortal term; so, it’s a sortal concept. But if raining is a sortal concept, then it is raining is not a sortal-free thought.

Consider another case: the sentence ‘this [pointing] exists’ does not contain any sortal terms, but arguably we can’t grasp the meaning of this sentence unless we know what sort of thing ‘this’ refers to—e.g., this table, this person, this event, or this object. However, table, person, and event are sortal concepts. And, according to Thomasson, there are only two senses of the word ‘object’: a sortal sense and a covering sense.

Obviously, if ‘this’ refers to this object, in the sortal sense, then the thought that this exists is not sortal-free. On the other hand, if ‘this’ refers to this object, in the covering sense, then Thomasson forbids it from appearing in the antecedent of an application condition (see Sect. 3.3). So, the hierarchy of informative application conditions can’t bottom out in the thought this exists.

The objection can be summed up like this: CST is tangled in the horns of a trilemma. Either the hierarchy of sortal application conditions is infinite, or it is circular, or it bottoms out in application conditions that have sortal-free antecedents. The first and second horn are implausible.²² The third horn is far from obvious and shouldn’t be adopted unless we can find a strong reason in its favour. This gives us a reason to believe that CST is false. But if CST is false, then (as we have seen) LST is false as well.

In the remainder of this section, I will discuss two other versions of this trilemma that have appeared in the literature. The first version is due to Brenner (2018); the second is due to Raab (2021). The arguments that these authors offer against the first and second horn of the trilemma are roughly the same as the arguments that I gave above. We differ in our assessment of the third horn.

Brenner seems to grant that we possess some sortal-free mental states²³ and offers two candidates: feature-placing thoughts (e.g., it is raining) and perceptual experiences or thoughts that describe our perceptual experience. His worry is that for most sortals the hierarchy of application conditions cannot bottom out in an application condition with either of these candidates in the antecedent. We don’t need to delve into the details of this argument in order to see that it differs from the argument that I offer.

I think that Brenner misses the main thrust of the third horn, but part of his argument is worth rehearsing. At various points in OME, Thomasson suggests that some application conditions licence an inference from a perceptual experience (or perception-describing thought) to the application of a term (or concept). According to Thomasson, for example, a competent speaker is “entitled, using her conceptual competence, to infer ‘there is a table’ from veridically observing the inside of a restaurant” (2015, p. 143). Elsewhere, she suggests that some application conditions are “perceptual input analyzers” that take perceptual experience as their input and deliver, as their output, the verdict that a certain concept applies, or doesn’t apply (e.g., 2015, n. 5: p. 90, and pp. 93, 110–1).

However, as Brenner notes, it is “clearly false that our having any sorts of perceptual inputs is a sufficient condition for any sort of external object to exist” (2018, p. 612). In other words, there is no true conditional of the form: if I perceive such-and-such, then there is a triangle, table, and so on for any sortal that applies to objects in the external world.

Brenner adds that if Thomasson endorses conditionals of this form, then she is committed to a kind of phenomenalism or idealism and she doesn’t appear to endorse either. (I would add: it’s not obvious that we can have perceptual experiences without possessing some sortal concepts. For example: it is not obvious that I can perceive a triangle without possessing the sortal concept triangle).

Raab focuses on LST, rather than CST. According to Raab, if Thomasson takes the third horn of the trilemma, she undermines her reason for introducing LST—to solve the qua problem. He concludes that Thomasson can’t take the third horn. However, some of Thomasson’s arguments for LST are not about the qua problem. For example: MA3 is about inferentialism.

Further, the qua problem is a problem for sortal terms and it’s not obvious that there is an analogue of this problem for sortal concepts. So, the qua problem is not a reason to believe CST in any case. Raab might object that he is offering a critique of LST, not CST. Fair enough. But in that case, Thomasson can dodge his critique simply by leaning on CST, rather than LST, in her discussion of easy arguments.

I conclude that LST is troubled. We don’t appear to have any good reason to believe LST (Sect. 4) and we have good reasons to deny it (Sect. 5). Further, LST is crucial to Thomasson’s defence of the easy ontology thesis (Sects. 1, 2, 3). So, if the easy ontology thesis is true, we may need to look elsewhere for reasons to believe it.²⁴