Metalinguistic apophaticism

Peter van Elswyk

University of Wisconsin–Milwaukee

A conviction had by many Christians over many centuries is that natural

language is inadequate for describing God. This is the doctrine of divine
ineffability. Apophaticism understands divine ineffability as it being justi-
fied or proper to negate statements that describe God. This paper develops
and defends a version of apophaticism in which the negation involved is
metalinguistic. The interest of this metalinguistic apophaticism is two-
fold. First, it provides a philosophical model of historical apophaticisms
that shows their rational coherence. Second, metalinguistic apophaticism
enables a minimal understanding of ineffability that is independently
plausible given its minor commitments.

1 Divine ineffability

A conviction had by many Christians over many centuries is that natural

language is inadequate for describing God. For example, in Paradise, Dante
Alighieri starts his poem by informing the reader that what was seen in heaven
were things “as man nor knows / Nor skills to tell.” And the poem ends, when
Dante is finally in the presence of God, with the claim that he has reached a place
“where speech is vanquished and must lag behind.” Let’s call this conviction
divine ineffability. To understand it, we should distinguish two separate but
closely related questions.
explication question acceptance question
What does it mean to say that Why should one accept divine inef-
natural language is inadequate for fability?
describing God?

This paper provides an answer to the first question that is compatible with nearly
all, if not all, answers to the second question.
Such a focus might seem amusingly unambitious. In most settings, it would
be. Matters are different here. It is hard to intelligibly explicate what it
means to say that language is inadequate for describing God. Two immediate
challenges face any attempt to do so. The first is the statibility problem. Many
explications of divine ineffability are self-refuting or at least self-undermining
(Alston 1956; Plantinga 2000). The second challenge I call the uniqueness
problem (Kenny 2006; Howard-Snyder 2017). A difference between theists
in contrast to agnostics and atheists is that theists accept propositions about
God that the others do not. Chief among them is that God exists, but the
list for Christian theists will also include the doctrinal contents of Creeds and
Councils. Divine ineffability threatens to erase this doxastic difference. If
natural language is inadequate for describing God, then doctrinal statements
such as God is triune or ones that distinguish God as a perfect being like God
is omniscient run the risk of being inadequate. Explications of divine ineffability
must therefore explain how natural language is inadequate while leaving room
for certain propositions to be expressed or possibly true such that they can be
rationally accepted by Christian theists. Otherwise the explication of divine
ineffability is not particularly Christian.
My answer to the explication question will be apophatic by giving a starring
role to negation. In particular, ineffability is understood as it being justified
or proper to negate all statements that describe God. Apophaticisms, though,
have extra difficulty. In maintaining it is proper to accept negations about God,
the apophaticist is committed to statements like (2) being acceptable. However,
atheists will accept at least one interpretation of (2) as well. They concur that
God is not omniscient because God does not exist. To avoid running into
the uniqueness problem, the apophaticist will often accept (1) too. That price
appears steep.
(1) God is omniscient.
(2) God is not omniscient.
At least on the surface, the conjunction of (1) and (2) constitutes a contradiction.
For apophatic attempts to answer the explication question, then, the statibility
and uniqueness problems act as constraints that it is difficult to find a way through
without terminating in contradiction.
So merely answering the explication question is not unambitious. It is
not without interest either. Both of the problems noted above have epistemic
consequences. Self-refuting views cannot be known because they are false, and
they are difficult to believe rationally. Likewise, it is difficult to imagine how
a Christian could rationally adopt a view that erases the doxastic difference
between theists and non-theists. And yet, as contemporary work has stressed,
divine ineffability was important to the early church (Scott and Citron 2016;
Keller 2018; Hewitt 2020). One can find the conviction that natural language is
inadequate in the writings of Anselm of Canterbury, Gregory of Nyssa, John
Chrysostom, Psuedo-Dionyisus, Maximus the Confessor, Thomas Aquinas,
Meister Eckhardt, and in Councils like the Fourth Lateran. So it is worth
investigating whether divine ineffability, especially apophatic varieties, can be
explicated in a manner that is rational. The charitable assumption is that it can
be, and to theorize from that starting point.
In proposing an apophaticism, I am up to what McDaniel (2015, 644) calls
philosophical modeling:

2
 A philosophical model is a worked out philosophical position that has
features that are analogous to the historical position of which it is supposed
to be a model. But the philosophical model might have features that are
‘mere artifacts’ of the model, i.e., not part of the fully faithful account of the
meaning of some statement of a historically important philosophical view.
But still the philosophical model can be useful if the following conditions
obtain. First, it is clear that the philosophical model is itself a coherent
philosophical position, and second, when taken as a model for the historical
position, it shows a way in which that historical position could be coherently
maintained. This is especially important when the position in question is
one whose coherence has often been doubted. . . And finally, there must be
no elements in the model that are explicitly contradicted by elements in
the theory that is being modeled; I take the importance of this to be self-
explanatory.

The apophaticism I provide will fulfill these conditions. It offers an apophatic

model of ineffability that is coherent by solving both the statibility and unique-
ness problems and compatible with historical precursors. However, one im-
portant caveat is that I will not be modeling the apophaticism of a particular
historical figure. My focus instead is on explicating apophaticism in a broad way
that can model most, if not all, earlier apophaticisms. Since it will be compatible
with most answers to the acceptance question, the apophaticism is flexible in
how substantive or minimal it is.
But the paper’s aim is not merely modeling. I will also show that model
enables the development of a minimal apophaticism that can be widely en-
dorsed. This minimal approach takes the form of an answer to the explication
question without an answer to the acceptance question. This paper is therefore
of two-fold interest. It advances a philosophical model of apophaticism that
demonstrates its coherence and that model enables a minimal version of an
apophatic view of divine ineffability that is independently plausible given its very
minimal commitments.
The essay proceeds as follows. In §2, I develop metalinguistic apophaticism
and explain how it answers the explication question while solving the two
problems that typically afflict attempts at doing so. Then §3 compares my
answer to a number of recent attempts to explicate divine ineffability. I argue
that metalinguistic apophaticism outperforms alternatives because either an
alternative does not solve both the statibility and uniqueness problems, or it
has other problems that my apophaticism lacks. I conclude in §4 with a brief
discussion about describing God.

2 Metalinguistic apophaticism

My explication of apophaticism gives a starring role to metalinguistic negation.

In §2.1, I introduce metalinguistic negation. Then §2.2 explains how it enables
a new explication of apophaticism. In §2.3, I defend that metalinguistic

3
apophaticism provides a probative philosophical model of historical apophati-
cisms. I discuss in §2.4 how it is compatible with most, if not all, answers
to the acceptance question and present the minimal version of metalinguistic
apophaticism.

2.1 Metalinguistic negation

The familiar meaning of not is truth-conditional.1 It has the simple task of

flipping truth-conditions. For a proposition p, not p is true iff p is false. But that
is not the whole story about not. Once we step outside of logic to consider how
competent speakers use negation in natural language, another use is noticeable.
This use is for expressing disapproval. Consider the following examples from
Horn (1989, 370-371).

(3) James is not a bastard—he is a nice guy!

(4) I didn’t manage to trap two mongeese—I managed to trap two
mongooses.
(5) Suzie didn’t call the [pólice]. She called the [políce].

Another pair of examples is owed to Almotahari (2014, 488).

(6) I’m not a secretary; I’m an administrative assistant.

(7) For a pessimist like him, the cup isn’t half full.

These six examples demonstrate what is now widely known as metalinguistic

negation. In each example, the not or negation clitic n’t is denying that
something subsequent is correct or appropriate. What exactly is the target of
this denial varies considerably.
Consider them top to bottom. In (3), the not targets the noun bastard. It
negates to reject or protest the use of the offensive term. In (4), the clitic n’t
targets the morphology of the term mongeese to indict it as the incorrect term
for referring to more than one mongoose. In (5), the negation clitic corrects
the pronunciation of police. In (6), not targets and objects to the connotation
that being a secretary is a diminutive position (in contrast to an administrative
assistant). And, finally, the clitic n’t in (7) targets the whole phrase the cup is half
full to object to it for being an inaccurate description of a pessimist’s attitude.
1
By familiar I also mean classical and static. One can find other meanings for negation in
paraconsistent logics and dynamic semantic theories that are not metalinguistic and yet still
different from the familiar meaning. Since my aim in this paper is exploring how metalinguistic
negation enables us to understand apophaticism, I overlook other non-metalinguistic meanings.
For an explication of ineffability in Buddhist thought that relies on non-classical logic, see Priest
(2015).

4
So our non-exhaustive list of targets includes: the meaning of, morphology
of, pronunciation of, connotations of, and applicability of an uttered term or
sentence. That variety of targets is then why metalinguistic negation is not the
familiar meaning. The truth-conditional not can only target the meanings of
other expressions. Morphological or phonological properties, connotations of,
or applicability for an expression are not part of its semantic meaning. They are
ineligible targets for the usual negation.
So what is metalinguistic negation? I proceed on the assumption it is
a pragmatic ambiguity (Horn 1985; Burton-Roberts 1989; Pitts 2011). In
particular, it is an atypical use of a negation term that, as opposed to flipping a
term’s truth-conditions, expresses the speaker’s disapproval of an utterance or an
aspect thereof. Though pragmatic, we can model metalinguistic negation with an
operator M( ⋅ ). We can loosely think of M( ⋅ ) as taking scope over an utterance
to express disapproval with it or one of its parts. The narrowness of intonational
stress often determines the exact target. From here, M( ⋅ ) expresses that the
speaker disapproves of some property of the target. Recall example (4). The
stress on geese in mongeese targets morphology as opposed to a bigger constituent
of the sentence. The speaker then disapproves of geese as the plural morphology
for mongoose.
Often metalinguistic negation is followed by a correction that illustrates what
the proper or appropriate meanings, morphology, pronunciation, connotations,
and so forth are. Each of our examples illustrated as much. The correction
provides a reason for the speaker’s disapproval. It rights what is wrong with
the target utterance. But the correction is not mandatory for metalinguistic
negation to be felicitous (Horn 1989; Kay 2004). It may be clear from
context what the speaker’s reason for disapproval is or, as Martins (2020, 350)
observes, metalinguistic negation can be used by speakers to merely express
disapproval. In other words, correcting is not the purpose of metalinguistic
negation. Expressing disapproval is. In describing M( ⋅ ) as expressing
disapproval, I proceed on the assumption that M( ⋅ ) does not express a
propositional content. It expresses an attitude in the way that phrases like What
a game! or terms like Damn! do. The attitude of disapproval is exclaimed as
opposed to asserted by M( ⋅ ).2
A consequence of metalinguistic negation merely expressing disapproval is
2
A referee wonders whether positing metalinguistic negation requires one to confront any of
the usual problems confronted by expressivism about moral terms. It does not. Problems
like the Frege–Geach problem concern how moral terms that are supposed to express attitudes
semantically compositionally integrate into complex sentences that do not express attitudes (e.g.
negation, conditionals, attitude verbs). Understood as a pragmatic ambiguity, metalinguistic
negation is owed to a post-semantic reinterpretation of not that speakers impose on a sentence as
opposed to a semantic interpretation that is compositionally derived. So such problems do not
arise.

5
that discourses that might appear to be contradictions are not. Consider (5)
without the extra intonation: Suzie didn’t call the police—Suzie called the police.
It looks like a contradictory pair. But when it is clear via the difference in
pronunciation—[pólice] versus [políce]—that the negation is metalinguistic, the
contradiction disappears. To put it schematically, the speaker isn’t asserting not
p via an utterance u and then asserting p via u⋆ . The speaker is disapproving of
u via negation in u, and then asserting p with u⋆ partly to explain the basis of the
prior disapproval.

2.2 Metalinguistic apophaticism

Apophaticisms maintain that divine ineffability consists in it being justified or

proper to negate all statements that describe God. I propose to understand the
negation involved as being metalinguistic. Accordingly, apophaticism expressed
with metalinguistic negation can be understood as the following claim about
utterances.

metalinguistic apophaticism (ml)

For every utterance u describing God in a context c, M(u) is proper
in c.

By proper, I mean that there is an explanation for how the disapproval expressed
by M( ⋅ ) is sincere. So ml states that there is no description of God uttered that
can be uttered that cannot also be metalinguistically negated. Every statement
about God merits disapproval.
There are a variety of reasons one might use metalinguistic negation to
express disapproval about an utterance. Likewise, there are a variety of reasons
one might have to express disapproval with an utterance about God. This is
why metalinguistic apophaticism provides an answer to the explication question
but not the acceptance question. The view does not require elaborating why
metalinguistic negation can be proper; it just claims that it can be for any
description of God we may utter.
A selling point of ml is that it solves or avoids each of the problems that tend
to afflict answers to the explication question. Start with the statibility problem.
The theory is not self-undermining in at least two important ways. First, ml
is defined only over utterances that describe God. It itself as a statement is
not about God; it is about a type of utterance. So ml cannot self-undermine
by negating itself because it does not apply to itself. Second, even if ml
did apply to itself, nothing contradictory or epistemically amiss would result.
Since metalinguistic negation only expresses disapproval, the result would be
presenting a thesis that one disapproves of. As long as the reason for disapproval
is not that ml is false, statibility is secured.

6
 Take the uniqueness problem next. As noted earlier, apophaticists risk
committing to a contradiction when they state or accept any description about
God. But a pair like (1) and (2) only constitutes a contradiction if the not is
truth-conditional.

(1) God is omniscient.

(2) God is not omniscient.

When it is metalinguistic, no contradiction appears. (1) utters a sentence to

assert that God is omniscient, and (2) expresses disapproval of this utterance
type. What part of the utterance is the exact target for disapproval, and what
reason motivates or justifies the disapproval will vary between contexts. But still
no contradiction will be produced. Nothing irrational will be required either
as long as the reason for disapproval is not that (1) is false. So ml enables an
apophatic theist to rationally believe and state every description of God found
in the Creeds and Councils. It just requires that such descriptions be sincerely
disapproved by the apophaticist.
Another selling point of ml is that it is actually pretty hard to articulate an
apophatic claim without using metalinguistic negation. Horn (1989) notes that,
unlike truth-conditional negation, metalinguistic negation cannot be expressed
with a negation prefix like -im or -un. Once incorporated, negation remains
truth-conditional.
(8) It is not possible—it is actual.
(9) It is impossible—it is actual.
Compare (8) and (9). The negation in (8) is metalinguistic, but (9), which
attempts to provide the same correction with negation incorporated through the
prefix -im, expresses a contradiction. When attempting to make an apophatic
claim with a negation prefix, the same limitation is expressed. (10) is an
intelligible claim an apophaticist would make about describing God with the
adjective just.

(10) God is not just–God is beyond that!

(11) God is unjust—God is beyond that!

The same cannot be said for (11). It sounds confused, and is very difficult
to recover an intended meaning from.3 Explicating apophaticism as the view
3
As an apophaticism, ml does not try to make special sense of predicates like ineffable or
unsayable that are especially paradoxical. A natural proposal continuous with ml is to treat these
as metalinguistic predicates that describe talk about God. For example, see Gäb (2017). But the
prefix data causes some trouble for this approach. It rules out the motivating argument that these
predicates are metalinguistic because the negation involved is. I explore a different explanation
in van Elswyk (2021).

7
all descriptions of God are eligible for metalinguistic denial captures this fact.
Apophatic claims require metalinguistic negation because apophaticism is a
metalinguistic view.

2.3 Metalinguistic apophaticism as a philosophical model

Let’s turn to how ml models apophaticism as a historical view of divine ineffabil-

ity. We have already seen in §2.2 that ml is coherent. But it is worth showing that
it can explain other apophatic theses. Though my aim is to offer an explication of
divine ineffability that broadly models a range of apophaticisms, I will not walk
through many particular theses. Instead, I will just focus on how ml explains
the three-staged notion of apophatic ascent that is found in Gregory of Nyssa,
Psuedo-Dionyisus, and others.
Metalinguistic negation is importantly iterable. The operator M( ⋅ ) can be
applied to an utterance that involves either metalinguistic or truth-conditional
negation. As a result, we can stack disapproval. We can disapprove of what we
disapproved when we disapproved. Though speakers rarely have cause to iterate
metalinguistic negation, its iterability provides a way to understand ascent. We
begin with a catophatic description like God is omniscient. Then we ascend to
the second stage by metalinguistically denying that description. The sentence
God is not omniscient is endorsed to express disapproval of describing God with
the natural language predicate is omniscient. Finally, we arrive at the third stage
by denying that denial by uttering God is not not omniscient. Here the double
not serves double-duty. It first clarifies that the not used in the second stage was
not truth-conditional. In denying that God omniscient, a negation is not being
asserted. Disapproval is being expressed. But the double not also enables second-
order disapproval to be conveyed. The disapproval expressed at the second stage
is disapproved of at the final stage.
Disapproving of what one disapproves of may be an uncomfortable position
to be in cognitively. But it is neither contradictory nor epistemically defective.
Here are a few examples. Suppose a person disapproves of drinking coffee
with milk. It hides the notes of the underlying bean, and they are bothered
whenever they see people doing so. But suppose they also recognize that this
disapproval is snobby; they wish they weren’t so judgmental. In this case,
they disapprove of what they disapprove. Another example is moral. Suppose
someone long ago internalized that eating meat is wrong. The person has since
surprisingly changed their mind, but a byproduct of that internalization is that
they disapprove whenever they see someone eating meat. The person knows
they should no longer feel this way. They disapprove of what they disapprove.
What remains to be seen is whether there are elements in ml that contradict
elements in historical apophaticisms. I will consider two objections that it does.

8
In their brief discussion of metalinguistic negation as a concept which might be
useful in understanding historical apophaticisms, Scott and Citron (2016) argue
that the resulting apophaticism is too thin. Many early apophatic figures were
not merely committed to descriptions of God being worthy of disapproval. They
maintained more. For such figures, ml is inadequate as a model. However, this
objection misses the mark. ml answers the explication question. It does not
answer the acceptance question. At worse, then, ml is incomplete as a model
of historical apophaticisms. But it can always be supplemented with an answer
to the acceptance question to be a complete model. I shortly discuss how
ml is compatible with various answers in §2.4. So ml carries no commitment
that contradicts the commitment of a historical apophaticism. By only being an
answer to the explication question, it might just lack a commitment that is key
to modeling a particular apophaticism.
The next objection is more on-target. Scott and Citron (2016) argue that
historical apophaticists would reject understanding their view with metalinguistic
negation because they regarded God as being beyond negation. To motivate this
claim, they provide two excerpts that I repeat below. The first is from Dionysius
(1987, 141):
It is beyond assertion and denial. We make assertions and denials of
what is next to it, but never of it, for it is both beyond every assertion,
being the perfect and unique cause of all things, and, by virtue of its
preeminently simple and absolute nature, free of every limitation, beyond
every limitation; it is also beyond every denial.

The second is from Palamas (1983, 57):

The excellence of Him Who surpasses all things is not only beyond all
affirmation, but also beyond all negation; it exceeds all excellence that is
attainable by the mind.

I submit that neither excerpt proves that ml contradicts their respective apophati-
cisms. In fact, a closer read shows its compatibility. Consider the line we make
assertions and denials of what is next to it, but never of it. As stated, ml is not about
God. It is about utterances. It therefore does not assert or deny anything about
God. Instead it ascends to state that every assertion or denial about God is
eligible for disapproval via metalinguistic negation. In this way, it accommodates
the Palamas line that God is beyond all negation. If we understand the negation in
question to be truth-conditional, ml states that such negation is to be disapproved
of. Or, if we understand such negation to be metalinguistic, ml states that such
disapproval is to be disapproved of. Though there may be other passages from
historical apophatic thinkers that are harder to square with ml, these two excerpts
are easily accommodated by ml.
As the earlier quote from McDaniel (2015) noted, models may introduce
artifacts, or elements that are not present in the historical theories or statements

9
being modeled. Metalinguistic negation is that artifact in the model I have
proposed. Metalinguistic negation as a pragmatic ambiguity was first theorized
by Horn (1985). Historical figures would not have understood themselves as
deploying an operator on utterances.
But the natural language phenomenon Horn identified is by no means new
or recent. It is a robust, cross-linguistic phenomenon that has been present for
a long while (Martins 2020). A fitting historical illustration is that metalinguistic
negation can be found in Scripture. Consider the emphasized portion of Mark
9:37 esv: “Whoever receives one such child in my name receives me, and
whoever receives me, receives not me but him who sent me.” When the negation
is interpreted descriptively, the passage expresses a contradiction. One receives
and does not receive Jesus. But a better interpretation is produced if the negation
is read metalinguistically. One use of metalinguistic negation is blocking scalar
inferences.

(12) Dean didn’t choose some of the records to play—he chose all of
them.

An illustration is provided by (12). Uttering the sentence Dean chose some of the
records to play invites the inference that Dean did not choose all of the records
to play. But the metalinguistic negation in (12) cancels this inference. The
subsequent correction he chose all of them strengthens what is asserted. Following
Craigmiles (2016, 264-270), the negation in Mark 9:37 should be understood
similarly. It blocks the scalar inference that only Jesus will be received as opposed
to both Jesus and whoever sent Jesus.
It is therefore plausible that historical apophaticists used negation met-
alinguistically, especially given that, as argued earlier, it is very difficult to
express apophatic denial of a particular description without using metalinguistic
negation. To claim as much is not to claim that they did so intentionally. They
might not have been consciously aware that the were exploiting a pragmatic
ambiguity in expressing their conviction.
One might be puzzled by the claim that a historical figure could have used
negation metalinguistically without being aware of doing so. Another example
helpfully illustrates.

(13) One shouldn’t make resolutions; one should have goals.

To explain the significance of this metalinguistic negation, Almotahari (2014,

499) is worth quoting in full:
I encountered [(13)] while watching CNN shortly after New Year’s Day. A
commentator was invited to talk about lifestyle choices for the new year.
She began by uttering [(13]. She then tried to explain that the contrast
is substantive. Her explanation failed. Insofar as there is a contrast, it’s a

10
 contrast at the level of connotation. Perhaps ‘goal’ connotes something a
little more flexible, something that needn’t constrain one’s behavior daily.
It might be that ‘resolution’ lacks this connotation. The speaker simply
wanted to emphasize that it’s ok to occasionally indulge one’s bad habit;
it will improve one’s chances to actually stick with one’s new lifestyle
choice in the long-run. . . What’s interesting about the example is that the
speaker took herself to be drawing a substantive distinction, not merely a
terminological contrast. This suggests that ordinary speakers aren’t always
sensitive to the difference between metalinguistic and descriptive negation.
That’s not surprising, since it took some amount of theorizing to appreciate
the difference. So competent language users might detect that it would be
unacceptable to utter a particular sentence, choose to reject the sentence
by exploiting metalinguistic negation, fail to recognize that the negated
sentence is actually true, and thereby draw unwarranted conclusions.

Almotahari’s key insight is that speakers can use metalinguistic negation and fail
to recognize they are doing so. With (13), the speaker was detecting something
unacceptable with resolutions as the term for what people should make for the new
year. This compelled her use of metalinguistic negation. Since the speaker then
started reasoning as if her negation was descriptive as opposed to metalinguistic,
it shows her lack of awareness.
Since it is plausible that historical apophaticists used negation metalinguis-
tically, ml is an especially probative model. It coherently models an apophatic
approach to divine ineffability without resorting to concepts or frameworks that
would have been entirely foreign to historical figures. With respect to the project
of philosophical modeling, ml is therefore preferable to other models that answer
the explication question.

2.4 Reasons for apophaticism

Many reasons have been given that motivate why natural language is inadequate
for describing God. Here is a casual reconstruction of five that are loosely
inspired by recent work (Keller 2018; Hewitt 2020; Fakhri forthcoming). I
do not lay them at the feet of anyone, nor assess their merits. They are just
illustrations of answers to the acceptance question.

argument from simplicity

God is simple. Simple things don’t have properties. Descriptions
like God is omniscient ascribe the property being omniscient. So
descriptions in natural language are inadequate.
argument from otherness
God is wholly other. So we are not acquainted with God. De-
scription requires acquaintance with what is being described. So
descriptions in natural language are inadequate.

11
 argument from infinitude
God is infinite. Infinite things cannot be comprehend by a finite
mind. What cannot be comprehended cannot be described. So
descriptions in natural language are inadequate.
argument from perfection
God is perfect. Perfect things cannot be described with terms created
for describing imperfect things. Terms in natural language were
created for describing imperfect things. So descriptions in natural
language are inadequate.
argument from containment
God cannot be contained. So God is not in the extension of
any predicate nor a constituent of any structured proposition. So
descriptions in natural language are inadequate.

Perhaps when formally rendered, the differences between the arguments disap-
pear and they merge into a smaller number. But, as is, they highlight reasons
for accepting that language is inadequate.
I advertised at the outset that ml would be compatible with most, if not all,
reasons for accepting divine ineffability. We can now see why. What ml requires
is that the disapproval expressed by metalinguistic negation be sincere. One
leading way that disapproval can be sincere is because it is based in one or more
of these arguments. For example, one might disapprove of an utterance of God
is omniscient because one is wary of describing God with the adjective omniscient.
Perhaps the meaning of that adjective, as per the argument from perfection, is
inadequate because it is an imperfect term. Or, perhaps one disapproves, as per
the argument from containment, of limiting God to what omniscience ascribes.
Either way, these reasons as answers to the acceptance question can underwrite
the disapproval required by ml. Accordingly, ml can be supplemented as an
answer to the explication question with any of the reasons above to fully model
a historical apophaticism. Together, they provide a philosophical of model what
an apophatic conception of divine ineffability consists in, and why someone
would accept such a conception.
In defending earlier that ml models historical apophaticisms, I considered
the objection that it was too thin to provide a suitable philosophical model. My
response was that ml answers only the explication question. To model the full
commitments of a particular figure, ml would need to be supplemented with a
particular answer to acceptance question. The arguments above illustrate such
answers. However, one may want to press the objection again by reasoning as
follows.4 Even with one of the arguments, ml remains too thin. The reason why
4
I am indebted to a referee for pressing this objection.

12
is that the arguments support a stronger conclusion than the conclusion that
descriptions about God meriting disapproval. For example, one understanding
of the argument from infinitude supports the conclusion that describing God
is not intelligible. So ml fails to model apophaticisms committed to stronger
conclusions about language’s inadequacy.
My response remains the same. ml answers the explication question. Its aim
is to provide a model for what apophaticism is that does not have the statibility
and uniqueness problem. This explicative aim is prior to and independent
of any reasons for accepting apophaticism. Any stronger conclusions about
language’s inadequacy are modeled by answers to the acceptance question. As
just discussed, the arguments sketched are fully compatible with ml. They are
what makes the disapproval sincere.
A notable consequence of distinguishing the explication question and accep-
tance question cleanly is that variants of the statibility and uniqueness prob-
lems may surface for the latter question even if solved for the former. In other
words, the reasons for accepting apophaticism may still be self-undermining
or erase the difference between theists and non-theists. As a philosophical
model, ml does not therefore show that the reasons or arguments for historical
apophaticisms are coherent. It shows that apophaticism generally as a particular
explication of divine ineffability is coherent.
What about us contemporaries? The substantivity of ml as an explication
of apophaticism depends on the explanation why the disapproval expressed by
metalinguistic negation is sincere. Pairing ml with any of the above gives us
a substantive theory. The reason explains the disapproval, and accepting that
reason will yield implications that ml lacks on its own. Importantly, though,
ml does not require or presuppose an answer to the acceptance question.
Disapproval can be sincere without having to endorse an argument like the above
to explain that disapproval. In other words, we can adopt ml while letting the
acceptance question go unanswered.
To illustrate sincere disapproval without a reason that constitutes an answer
to the acceptance question, we should turn to other topics that are hard to
describe. I will focus on art objects and emotion. For both subject matters,
we often experience what I will call descriptive reluctance. Here is an aesthetic
example of such reluctance. Suppose a person visited the Metropolitan Museum
of Modern Art and viewed Dalí’s The Persistence of Memory (1931). Then suppose
they were describing the artwork to a friend. There is much they could say about
the surrealist painting with its melting clocks and dripping watches. But either at
the end of the description, or mid-way through describing, it would be natural
to use metalinguistic negation to deny what was said. For example, they might
describe it as being like a hallucination and follow that with And yet, it is not like
a hallucination. They may even follow every description with a metalinguistic

13
negation of that description.
Here now is an emotional example of descriptive reluctance. Suppose
someone is processing with a therapist their complicated divorce. It has taken
months of sessions just to lay the foundation for talking about how they are
currently feeling. There is much they could say about what transpired between
them and their ex-spouse. But either at the end of the description, or mid-way
through describing, it would be natural to use metalinguistic negation to deny
what they said. For example, they might describe themselves as feeling relieved
and follow that with And yet, it is not that I am relieved. One may even follow
every description with a metalinguistic negation.
I think both examples of descriptive reluctance are cogent and ordinary. Two
further observations are worthwhile. First, the metalinguistic denials in these
examples are not retractions. The speaker is not taking back the description of
the Dalí painting as being like a hallucination. They are merely disapproving
of that description to express reluctance. What they want to communicate
to their addressee is that the painting was like a hallucination and that they
are uncomfortable with that very description. Likewise, the speaker denying
their self-ascription of feeling relief is not retracting that ascription. What they
want their therapist to know is that they feel relief and yet are uneasy about
saying as much. Second, the metalinguistic denials in these examples are not
hedges. Following Benton and van Elswyk (2020), hedges are expressions like
the parenthetical verb I think that can be used to weaken the epistemic position
a speaker represents themselves as occupying. In contrast to It is raining, the
sentence It is raining, I think represents the speaker as thinking as opposed to
knowing that it is raining. The denials in our examples do not weaken the
epistemic position of the speaker. Though the person describing the painting
or their feelings about the divorce are qualifying their descriptions by further
conveying disapproval, they are not hedging.
Descriptive reluctance is thus a distinct state of mind. It explains why
metalinguistic negation is sincere in the examples. It can also explain why
metalinguistic negations concerning God can be sincere. One may think that
every description of God is eligible for disapproval because one is experiencing
reluctance. What explains and underwrites the disapproval is not that one is
signed up for one of our five arguments for divine ineffability. One may even
believe that there is no answer to the acceptance question because there are no
good reasons for divine ineffability. It is just that, when it comes to describing
God, one encounters unease or discomfort. Metalinguistic negation is how that
reluctance is expressed as disapproval.
Another basis for sincere disapproval that is not an answer to the acceptance
question is wonder. Here I follow Yadav (2016). Yadav argues that apophaticism
is a type of mystical theology. Accordingly, an explication of apophaticism

14
must account for how apophaticism plays a regulative role in the life of the
apophaticist. In part, an explication needs to specify what attitude an apophatic
theist takes. His proposal is that the attitude of wonder can play this regulative
role. To wonder at something is to possess the enduring feeling that one lacks the
ability to understand that thing. In this manner, wonder contrasts with curiosity.
Curiosity is an attitude directed at a question that involves a desire to know the
answer to that question (Whitcomb 2010). Once an answer is known, curiosity
disappears because the desire is satisfied. Wonder persists as one continues to be
amazed or astonished by the thing.
Like descriptive reluctance, wonder directed at what is being described
can make the disapproval expressed by metalingusitic negation sincere. What
underwrites the disapproval is not that one accepts one of the arguments for
divine ineffability. It is just that, when it comes to describing God, one
experiences an awe or felt inability to understand God. Metalinguistic negation
is one way of voicing this wonder. It enables one to affirm statements like God is
triune while disapproving of that description as a way of sincerely conveying that
one is in a state of wonder at God.
The apophaticism that results from pairing ml without an answer to the
acceptance question but with descriptive reluctance or wonder is very minimal.
It is committed to every description of God meriting disapproval. But what
justifies that disapproval are facts about the speaker’s psychology as opposed
to facts about God. Nothing more. Let us call this apo-minimalism. Apo-
minimalism is clearly not what many historical apophatic figures endorsed. As
discussed, they are committed to some argument for divine ineffability like one
of the five I sketched. However, apo-minimalism does strike me as a view that
most contemporaries can accept. It requires metalinguistic negation, a well-
attested pragmatic phenomenon in natural language that one should already
acknowledge, and experiencing descriptive reluctance or wonder when it comes
to God.

3 Alternative explications

I have advanced ml as both a model of historical apophatic views of divine inef-

fability and to motivate apo-minimalism, an apophaticism where disapproval
is based in a speaker’s attitudes. This section compares ml to other recent
explications of divine ineffability.
My discussion is not exhaustive. First, I do not discuss general theories of
religious language. For instance, theories that treat statements about God as
metaphorical or analogical are overlooked. Since analogies and metaphors are
still used to express propositions, I do not regard these theories as explicating
how language is inadequate as opposed to defending that language is different

15
when used to describe God. Note that such views of religious language are
argued to still fail to solve the statibility and uniqueness problems (Howard-
Snyder 2017). Second, I do not discuss theories that are broadly incompatible
with model-theoretic semantics in the generative grammar tradition.5 Though
I lack space to defend as much, I regard such compatibility as a necessary
condition on plausible explications. In what follows, I focus on four recent
accounts of divine ineffability owed to Jacobs (2015), Keller (2018), Lebens
(2014, 2017), and Pruss (forthcoming).

3.1 Ineffability and fundamentality

Jacobs (2015) takes an approach to ineffability that distinguishes between two

ways a proposition can be true. A proposition can be true or fundamentally
true. Where F( ⋅ ) is an operator that applies to a proposition when it
is fundamentally true and P is the set of all propositions describing God
intrinsically, we can state his view thusly.

fundamental ineffability (fi)

For all propositions p ∈ P, not(F(p)) and not(F(not-p)).

Note that fi has the consequence that there are no propositions about God
whatsoever that are fundamentally true. If there are any true propositions about
God, they are non-fundamentally true.
Jacobs’s theory solves both problems afflicting answers to the explication
question. Start with the statibility problem. fi is about whether certain
propositions are fundamentally true or not. As such, it cannot undermine itself
because it does not apply to itself. Next consider the uniqueness problem. The
propositions contained in the Creeds and Councils are true. They just are not
fundamentally true. What grounds the truth of these propositions is God, full
stop. So these truths are not made true by nor do they supervene on a base of
additional fundamental truths about God.
The differences I want to highlight between ml and fi are their metaphysical
commitments. By itself, ml has none. Commitments can be incurred depending
on what answer to the acceptance question one adopts. But ml does not require
any as an explication of divine ineffability. In contrast, fi does. It requires a
fundamentality-theoretic approach to metaphysics, and a particular application
of that approach that some find objectionable because of how it leaves certain
truths about God ungrounded (Lebens 2014; Fakhri forthcoming). For this
5
For this reason, I do not engage with Hewitt (2020). Hewitt expresses sympathy for
inferentialism as an alternative approach to syntax and semantics. But inferentialism is well-
known to have serious failures like the inability to explain compositionality. See Fodor and
Lepore (2001, 2007).

16
reason, ml is preferable as an explication. It can explain what divine ineffability
consists in without taking on controversial commitments. As a result, ml cleanly
distinguishes the question of whether divine ineffability is an intelligible doctrine
from the further question of whether one should accept or endorse divine
ineffability.

3.2 Ineffability and semantic incompleteness

Keller (2018) offers an explication of divine ineffability that turns on subtle

issues at the interface between semantics and pragmatics. She starts with radical
minimalism, a controversial view about meaning owed to Bach (2006). Radical
minimalism holds that some declarative sentences fail to have propositions
as meanings because they contain a semantically incomplete term. Putative
examples include ready.

(14) Pim is ready.

Assessed in isolation, (14) is difficult to interpret. Ready for what? is the

natural follow-up. Adding context eases interpretation. When (14) is an
answer to the question Is Pim ready for dinner?, a proposition is communicated.
However, incomplete terms are not context-sensitive. Even relative to a context,
declaratives like (14) are incomplete. What the addition of context facilitates
is world knowledge that conversational participants can draw on to construct a
proposition. To get to a proposition from a sentence like (14) therefore requires
a post-semantic pragmatic process.
Call this enrichment. For Keller, a necessary condition for enrichment is
that the speaker uttering an incomplete sentence intends to convey a proposition
expressed by a complete counterpart of that sentence. So an utterance of Pim is
ready is enriched into a proposition only when the speaker intends to convey the
content of a counterpart like Pim is ready for dinner. Without such an intention,
some declaratives remain incomplete.
With these two assumptions in place, we can appreciate what Keller calls the
Franciscan-Thomist view (ft). It maintains that sentences ascribing positive,
intrinsic properties to God do not express propositions. To the extent that any
sentences about God express propositions, they ascribe extrinsic and/or negative
properties. To elaborate, consider a sentence ascribing a positive, intrinsic
property to God.

(15) God is good.

This sentence contains the gradable adjective good. For Keller, such adjectives
are incomplete. As a result, (15) fails to have a proposition for a meaning as

17
per radical minimalism. More than that, speakers uttering (15) cannot intend to
express the proposition of a counterpart sentence because there is no counterpart
sentence about God that speakers can grasp.6 So a sentence like (15) cannot be
enriched into a proposition either. For contrast, compare example (15) with (16)
below.

(16) God is maximally/perfectly good.

Keller holds that such a sentence does express a proposition. But she adds
two qualifiers. The first is that sentences like (16) ascribe a negative property.
To be maximally or perfectly good is to deny or negate that God is limited
in goodness. The second qualifier is that, to the extent that we can grasp the
proposition expressed, we understand it dimly because of non-propositional,
personal knowledge we have of God.
Keller’s view solves the statibility problem straightforwardly. Insofar as
sentences used to state ft involve incomplete terms, we can enrich those
sentences to express propositions. The failure to enrich only happens with
sentences about God because speakers cannot form intentions about what they
cannot grasp. ft solves the uniqueness problem by allowing propositions to still
be expressed about God. What differentiates the theist from the non-theist
is that the theist believes propositions that ascribe extrinsic and/or negative
properties God that the non-theist does not.
However, ft is not without problems given the assumptions about semantics
and pragmatics that Keller adopts. I am not a radical minimalist. Indeed, few
are. But ft requires it. To illustrate, consider a minimalism that is not radical
(Borg 2004; Cappelen and Lepore 2005; Schoubye and Stokke 2016). On
this alternative, sentences with incomplete terms like ready express a minimal
proposition. But then ineffability cannot result. Sentences ascribing positive,
intrinsic properties like (15) would express minimal propositions. Even if those
propositions are never enriched into perhaps more determinate propositions,
sentences about God would still have propositions as meanings. Or, consider
a contextualism where alleged incomplete terms are context-sensitive. That is,
the terms contribute a variable to the logical form of the sentences in which
they appear that needs to be saturated in a context before the sentence has a
proposition for a meaning (Stanley 2000; King and Stanley 2005). Then there
is a further question about how that variable gets saturated. But an increasingly
popular view is that, even if the saturation process is itself incomplete, sentences
hosting context-sensitive terms express a range of propositions (von Fintel
6
Note here that Keller adopts something akin to what I called the argument from infinitude
in §2.4. Accordingly, her answer to the explication question requires a particular answer to the
acceptance question.

18
and Gillies 2009; King 2017). The idea is that each proposition in the
range corresponds to a particular resolution of the variable. With cloudy
contextualism, ineffability cannot result. Instead of one proposition about God
being expressed by a sentence like (15), a cloud of them are for each resolution
of the context-sensitivity of good.
Turn next to Keller’s discussion of gradable adjectives like good. She
takes sentences like (16) to express negative properties as opposed to positive
properties. Since the only difference between (16) and (15) is the addition
of an adverb like maximally or perfectly, Keller is committed to a semantics for
these adverbs on which they convert positive properties into negative ones. This
commitment is a highly controversial one.
To illustrate, let us consider the leading semantics for gradable adjectives. it
is a degree semantics on which adjectives denote relations between individuals
and degrees (Cresswell 1977; von Stechow 1984; Heim 2000; Kennedy 2007).
On this semantics, Nick is tall states that the degree of Nick’s height is above a
certain standard along a scale that orders comparable objects. The comparison
class is context-sensitive. Nick is tall may be true when the comparison class is
professors at state universities but false when the comparison class is professional
basketball players who play the center position.
One important way in which gradable adjectives differ is with respect to the
structure of the scale. Scales are either open or closed. Open scales have no
maximal or minimal element. For each object along the scale, there is a higher
and lower object. In contrast, closed scales have a maximum or minimum.
Compare full with tall.

(17) The glass is full.

(18) The glass is tall.

A closed scale is evoked by full whereas an open scale is evoked by tall. For full,
there is a maximal degree of fullness, a degree at which the glass cannot be more
full. Not so for tall. There is not a degree of height at which the glass cannot
be taller. Gradable adjectives with open scales are relative adjectives (RAs)
whereas gradable adjectives with closed scales are absolute adjectives (AAs).
Since they differ in the associated scale structure, RAs and AAs also differ in how
the standard is determined. The scale is context-sensitive for RAs but not for
AAs. The standard for tall regularly shifts in a context with comparison classes
while the standard for full does not.
Within a degree semantics for gradable adjectives, many adverbs receive
a semantics as degree modifiers. They specify where along the scale evoked
by an adjective an object falls. Nick is almost tall, for example, indicates that
the degree of Nick’s height falls on the scale below the standard for tallness

19
given the comparison class. Some degree modifiers are what Kennedy and
McNally (2005) call maximality modifiers. They compose exclusively with
AAs to indicate that an object has the property denoted to a maximal degree.
Examples include totally, maximally, and perfectly.
Let’s circle back to (15) and (16). Good is an AA with a closed scale. God
is good therefore states that the degree of God’s goodness is at or above the
standard for goodness. The addition of maximally or perfectly in (16) does not
change anything about the property ascribed. It clarifies that the degree of
goodness is at the top of the scale. Accordingly, the degree semantics is at
odds with Keller’s claim that is good ascribes a positive property but is perfectly
good somehow ascribes a negative property. The modifier does not change the
underlying property ascribed nor involve negation or denial in its meaning.7
Since a degree semantics is the leading semantics, accepting ft requires us to
abandon the leading semantics. In the absence of an alternative semantics that is
equally explanatory and which explains how modifiers convert a property from
positive to negative as the outcome of semantic composition, ft is a very costly
explication of ineffability to adopt.
So far, the problems pressed have concerned ft’s consequences for how we
theorize about natural language. But there are more theory-neutral concerns
too. ft is at odds with linguistic judgment to maintain that (15) does not express a
proposition. Consider entailment data. It appears that God is good entails Someone
is good. But Keller cannot accommodate this intuition. Since the former sentence
fails to express a propositional content and content is what enters into entailment
relations, the former cannot be understood as entailing anything. The intuition
must be denied. Or, consider data involving propositional anaphora. Sentences
that express propositions license the use of dedicated expressions in subsequent
sentences that refer to those propositions (van Elswyk 2019). For example,
consider the reply in (19b).
(19) (a) Pim is ready for dinner.
(b) I think so too.
It features the anaphor so. This expression refers to the proposition that Pim is
ready for dinner that is expressed by (19b). As a result, what (19b) states is that
7
Notice that the modifier will not convert a positive property to a negative one even with a
different semantics. For example, one might deny that good is an AA. Then there’s a question
about what modifiers like perfectly could mean. The natural answer suggested in Kennedy and
McNally (2005, 354) is that they are akin to other modifiers like very. This semantics would not
place God at the top of the scale with God is perfectly good because there is no top. But it would
still place God high along a scale. Similarly, one might pursue a mereological interpretation of
perfectly on which it ascribes a property to all the parts of the grammatical subject (Burnett 2014).
Applied to God is perfectly good, it would state that all parts of God possess goodness to the degree
that is at or above the standard. But this again would still place God along a scale just in multiple
ways.

20
the speaker thinks that Pim is ready for dinner. Statements describing God also
license anaphors.

(20) (a) God is good.

(b) I think so too.

Like (19b), it appears that the so in (20b) refers to the proposition expressed
by the prior sentence. What (20b) states is that the speaker thinks that God is
good. Again, Keller cannot accommodate this intuition. Her proposal predicts
that (20b) also fails to express a proposition because so fails to refer. Notice
that the need to deny these and related linguistic intuitions are not problems
with merely radical minimalism. Bach can respond that God is good is enriched
into a proposition that entails or which provides the antecedent for downstream
anaphors, but Keller with ft cannot.
I conclude that the costs of ft are not worth paying. In comparison, adopting
ml as an answer to the explication question does not require any stance on
the boundary between semantics and pragmatics, the compositional semantics
of gradable adjectives, or when declaratives enter into entailment or license
anaphors.

3.3 Ineffability and illuminating falsehoods

The next theory I discuss is owed to Lebens (2014, 2017). Leben’s proposal is
that all statements of divine ineffability are false. To state that God is ineffable
or transcendent or not omniscient is to state a falsehood. And yet, they are
illuminating falsehoods. No definition or criterion is given of this category. But
Lebens motivates the category with examples. The basic idea is that to entertain
a statement of ineffability is to do something more than represent the world as
being a certain way. It is to highlight something about God, the world, and one’s
place in the world in a way that is beneficial.
One benefit noted by Lebens is epistemic. Entertaining a false yet illumi-
nating statement like God is ineffable need not compel one to believe as much
because doing so would quickly be self-refuting. Nevertheless, entertaining
it may compel one to lower one’s credence in the proposition expressed by
statements like God is just out of discomfort or unease that the predicate involved
applies to God. Or perhaps entertaining such a statement does not direct one’s
credences downward but it does prompt one to be reminded to talk or theorize
about God with increased epistemic humility.
Leben’s view effortlessly answers the explication question. It explicates
ineffability in a manner that is not self-refuting because statements of ineffability
are not taken as true. They are explicitly identified as false. So the statibility

21
problem is solved immediately. Likewise, his view imperils the truth or intel-
ligibility of zero propositions about God. The uniqueness problem is therefore
solved because there is an abundance of propositions that theists can believe to
distinguish themselves from non-theists.
His view and the one I have developed are also importantly similar. I
argued that ml could be coupled with what I called apo-minimalism. Apo-
minimalism differs from more robust versions of metalinguistic apophaticism by
not offering an answer to the acceptance question to explain disapproval. What
makes the disapproval expressed sincere or proper is just descriptive reluctance.
The benefits of expressing merely disapproval are presumably the same as
entertaining an illuminating falsehood. For example, one might maintain that
every utterance about God is eligible for disapproval in order to remind oneself
to speak about God with humility.
Still, I think there are two reasons for preferring ml. First, it succeeds
better as a philosophical model of historical apophaticisms broadly construed.
Consider the earlier excerpts from Pseudo-Dionysius and Gregory of Palamas
on which God is described as beyond assertion and denial and beyond all negation.
Lebens treats apophatic statements like God is not omniscient as truth-evaluable.
That is how they can be false. But this does not move beyond assertion or
denial like ml does by evaluating statements only according to whether they are
worthy or not of approval. One way to see as much is to reconsider the three-
staged apophatic ascent. Lebens’s view cannot accommodate the third stage. For
example, to deny that God is not omniscient is so state something true because the
statement of ineffability at the second stage is false. But to state something true
is to crash back down to the first stage where one makes catophatic statements.
ml stays at the third stage of ascent by treating the relevant instance of negation
as metalinguistic as opposed to descriptive.
The second reason for preferring ml is that it is a more flexible explication of
divine ineffability. Though ml enables apo-minimalism which is akin to Lebens’s
view as a minimal approach to divine ineffability, ml does not mandate deflation.
It is compatible with most, if not all, answers to the acceptance question. Once
paired with one of these answers—like the arguments in §2.4—ml becomes
a more substantive theory of ineffability. Lebens’s view does have this same
flexibility. It is committed to the falsity of all statements expressing divine
ineffability.

3.4 Ineffability and misleading

Linguistic deception comes in two forms: lying and misleading. Both involve
the communication of a proposition that is false or at least not believed by the
speaker. The exact boundary between them is a point of controversy (Saul 2012;

22
Stokke 2016; van Elswyk 2020). But generally the lying/misleading distinction
tracks whether the disbelieved proposition is what the speaker said with a
sentence, or another proposition communicated by the sentence’s utterance. A
classic example is a scalar implicature. Some words order on scales. When they
do, the use of a word that is lower on the relevant scale typically implicates that a
sentence featuring stronger words on the scale are false. An example is provided
by (21).

(21) Dean chose some of the records to play.

Since all > some, (21) carries the scalar implicature that Dean did not choose all of
the records to play. To turn this into example of misleading, imagine that Dean
did choose all of the records to play and the speaker uttering (21) knows as much.
Then what is said by (21) is not a lie; it is a truth known by the speaker. But the
associated scalar implicature is false and known to be false by the speaker. As a
result, they mislead by uttering (21).
Pruss (forthcoming) proposes that all descriptions about God are misleading.
This proposal solves both the statibility problem and the uniqueness problem.
It solves the former in at least two ways. First, it is defined over utterances of
descriptions about God as opposed to being a description of God. So it does
not apply to itself. Second, even if it did apply to itself, its truth would not be
threatened. Statements expressing truths can mislead. His proposal solves the
uniqueness problem in a similar manner. One can affirm whatever propositions
about God that one likes. The qualification that one’s affirming utterances are
misleading just needs to be added.
Pruss’s proposal is akin my own in that it is pragmatic. It does not tinker with
the semantic content of terms used to describe God. Such tinkering quickly
leads to problems, as I argued when discussing ft. Our views are also very
complementary. One might appeal to the generalization that all uttered de-
scriptions about God are misleading to explain why all such utterances merit the
disapproval expressible with metalinguistic negation. The result of combining
our views would plausibly constitute another form of apo-minimalism because
one would not need to embrace a substantive answer to the acceptance question
like the answers sketched in §2.4.
Nevertheless, I am skeptical that all uttered descriptions about God are
misleading. Since an utterance is misleading only if it communicates a false
or disbelieved proposition, Pruss’s proposal is committed to all utterances
about God communicating a secondary proposition that is false or disbelieved.
But it is not clear that they do. Reconsider scalar implicatures as a kind
of secondary proposition. Whether an utterance carries a scalar implicature
is a very contingent matter. It depends on whether scalar terms appear in

23
the sentence uttered, and the structure of the broader discourse in which
the sentence appears (van Kuppevelt 1996; Asher 2013). On the traditional
explanation owed to Grice (1989), the implicature also has to be calculated, or
worked out by the hearer. Other secondary propositions that are communicated
in a non-semantic manner are similar. So contingent features about every
utterance describing God would need to align just right for them to be misleading
utterances.8

4 Conclusion

This paper defended metalinguistic apophaticism or ml as an explication of

divine ineffability. What ml offers is a way of modeling historical apophaticisms
within the Christian tradition that shows their coherence. Of special note is
that ml is compatible with a range of different arguments that conclude God is
ineffable. Indeed, it is also compatible with the denial of these arguments. The
minimal view that results is apo-minimalism, a view that is of broader interest
than just philosophical modeling. Yadav (2016, 17) notes that apophaticism
“tends to make theologians and philosophers of religion with analytic sensibilities
queasy.” Apo-minimalism need not have the same effect. In concluding, I want
to highlight a feature of apo-minimalism.
A separate conviction from divine ineffability is that describing God with
natural language is somehow different. When we predicate goodness with God
is good that statement perhaps involves metaphor or pretense or the predicating
is analogical as opposed to ordinary. This conviction constitutes the denial of
what Scott (2017) calls the face-value theory. The face-value theory holds that
language used to describe God can be understood the same as language used to
describe other persons and objects. Importantly, the theory is not committed
to all descriptions about God being literal. Metaphor, pretense, and analogy are
used to describe God because they are part of everyday language use too. The
core of the theory is that a distinctive semantic and pragmatics are not required
to understand talk about God.
Most denials of the face-value theory treat denial as a cost of possessing either
the conviction that God is different or that God is beyond language. But ml
8
Pruss provides a number of examples where what is said by a sentence prompts a hearer to
draw an extra conclusion. God is a Trinity of persons illustrates. He describes it as being apt
to mislead hearers into falsely believing that God is not one. But the proposition that God is
not one fails to qualify as a secondary proposition communicated by an utterance. This false
proposition is one the hearer arrives at independently after adding what was said by an utterance
to their existing beliefs. Accordingly, such examples do not involve misleading utterances. They
are examples of mistaken inference. Since any utterance whatsoever has the power to prompt a
mistaken inference, appealing to such a general effect is not enough to distinctively answer the
explication question.

24
shows that divine ineffability is compatible with the face-value theory. When
ineffability consists in the eligibility of all utterances about God for metalinguistic
disapproval, one can still proceed to analyze those utterances at face-value. To
do so, one needs to refrain from endorsing an answer to the acceptance question
that is itself incompatible with the face-value theory. That can be done by
endorsing apo-minimalism. What is facilitated by apo-minimalism is therefore
a way to reconciled two claims: that God is ineffable and that describing God is
done ordinarily.9

References

Almotahari, Mahrad. 2014. “Metalinguis- Cresswell, Max. 1977. “The semantics of

tic negation and metaphysical affirmation.” degree.” In Barbara Partee (ed.), Montague
Philosophical Studies 167:497–517. Grammar, 261–292. Academic Press.
Alston, William P. 1956. “Ineffability.” Dionysius. 1987. The Mystical Theology. New
Philosophical Review 65:506–522. Jersey: Paulist Press.

Asher, Nicholas. 2013. “Implicatures and Fakhri, Omar. forthcoming. “The ineffability
discourse structure.” Lingua 132:13–28. of God.” International Journal for Philosophy of
Religion .
Bach, Kent. 2006. “The excluded mid-
Fodor, Jerry and Lepore, Ernie. 2001.
dle: Semantic minimalism without minimal
“Brandom’s burdens: compositionality and
propositions.” Philosophy and Phenomenological
inferentialism.” Philosophy and Phenomenolog-
Research 73:435–442.
ical Research 63:465–481.
Benton, Mathew and van Elswyk, Peter. —. 2007. “Brandom beleaguered.” Philosophy
2020. “Hedged Assertion.” In Sandy and Phenomenological Research 74:677–691.
Goldberg (ed.), Oxford Handbook of Assertion.
Gäb, Sebastian. 2017. “The paradox of
Oxford University Press.
ineffability.” International Journal of Philosophy
Borg, Emma. 2004. Minimal semantics. and Theology 78:1–12.
Oxford University Press.
Grice, Paul. 1989. Studies in the Way of Words.
Burnett, Heather. 2014. “A delineation Harvard University Press.
solution to the puzzles of absolute adjectives.” Heim, Irene. 2000. “Degree operators and
Linguistics and Philosophy 37:1–39. scope.” SALT 10:40–64.
Burton-Roberts, Noel. 1989. “On Horn’s Hewitt, Simon. 2020. Negative theology and
dilemma: presupposition and negation.” philosophical analysis. Palgrave Macmillan.
Journal of Linguistics 25:95–125. Horn, Larry. 1985. “Metalinguistic negation
Cappelen, Herman and Lepore, Ernest. and pragmatic ambiguity.” Language 61:121–
2005. Insensitive Semantics. Oxford: 174.
Blackwell. —. 1989. A Natural History of Negation.
Craigmiles, Shawn. 2016. The pragmatic University of Chicago Press.
constraints of ÁΛΛÁ in the Synoptic Gospels. Howard-Snyder, Daniel. 2017. “Pan-
Ph.D. thesis, Asbury Theological Seminary. metaphoricism.” Religious Studies 53:25–49.
9
I am indebted to Marilyn McCord Adams, Jonathan Jacobs, Wes Siscoe, Lorraine Keller,
Sameer Yadav, Matthew Benton, Dean Zimmerman, Lara Buchak, and anonymous referees for
OSPR for helpful comments.

25
Jacobs, Jonathan. 2015. “The ineffable, Palamas, Gregory. 1983. Gregory Palamas:
inconceivable, incomprehensible God: Fun- The Triads. New Jersey: Paulist Press.
damentality and apophatic theology.” Oxford
Pitts, Alyson. 2011. “Exploring a ’Pragmatic
Studies in Philosophy of Religion 6:158–176.
ambiguity’ of negation.” Language 87:346–
Kay, Paul. 2004. “Pragmatic aspects of 368.
grammatical constructions.” In Laurence
Horn (ed.), The Handbook of Pragmatics, 675– Plantinga, Alvin. 2000. Warranted Christian
700. Oxford: Blackwell. belief. Oxford University Press.

Keller, Lorraine. 2018. “Divine ineffability Priest, Graham. 2015. “Speaking of the
and Franciscan knowledge.” Res Philosophica Ineffable, East and West.” European Journal
95:347–370. of Analytic Philosophy 11:6–20.

Kennedy, Christopher. 2007. “Vagueness Pruss, Alexander. forthcoming. “Ineffability

and grammar: The semantics of relative and and creation.” Oxford Studies in Philosophy of
absolute gradable adjectives.” Linguistics and Religion 10.
Philosophy 30:1–45.
Saul, Jennifer. 2012. Lying, misleading, and
Kennedy, Christopher and McNally, Louise. what is said: An exploration in philosophy of
2005. “Scale structure and the semantic language and in ethics. Oxford: Oxford
typology of gradable predicates.” Language University Press.
81:345–381.
Schoubye, Anders and Stokke, Andreas.
Kenny, Anthony. 2006. “Worshipping an 2016. “What is said?” Nous 50:759–793.
unknown God.” Ratio 441–453.
Scott, Michael. 2017. “Religious language.”
King, Jeffrey C. 2017. “Strong contex- In Edward N. Zalta (ed.), Stanford Encyclope-
tual felicity and felicitous underspecification.” dia of Philosophy.
Philosophy and Phenomenological Research .
Scott, Michael and Citron, Gabriel. 2016.
King, Jeffrey C. and Stanley, Jason. 2005. “What is apophaticism? Ways of talking
“Semantics, Pragmatics and the Role of about an ineffable God.” European Journal for
Semantic Content.” In Zoltan Szabó (ed.), Philosophy of Religion 8:23–49.
Semantics versus Pragmatics. Oxford: Oxford
University Press. Stanley, Jason. 2000. “Context and Logical
Form.” Linguistics and Philosophy 23:391–
Lebens, Samuel. 2014. “Why so negative
434.
about negative theology?” International
Journal for Philosophy of Religion 76:259–275. Stokke, Andreas. 2016. “Lying and Mis-
leading in Discourse.” Philosophical Review
—. 2017. “Negative theology as illumi-
125:83–134.
nating/therapeutic falsehood.” In Michael
Fagenblat (ed.), Negative theology as Jewish van Elswyk, Peter. 2019. “Propositional
modernity. Indiana University Press. anaphors.” Philosophical Studies 176:1055–
Martins, Ana Maria. 2020. “Metalinguistic 1075.
negation.” In Viviane Deprez and Teresa Es- —. 2020. “Deceiving without answering.”
pinal (eds.), The Oxford Handbook of Negation. Philosophical Studies 177:1157–1173.
Oxford University Press.
—. 2021. “Saying the unsayable.”
McDaniel, Kris. 2015. “A Philosophi-
cal Model of the Relation between Things van Kuppevelt, Jan. 1996. “Inferring from
in Themselves and Appearances.” Nous topics.” Linguistics and Philosophy 19:394–
49:643–664. 443.

26
von Fintel, Kai and Gillies, Thony. 2009. Semantics 3:1–77.
“‘Might’ Made Right.” In Andy Egan and
Whitcomb, Dennis. 2010. “Curiosity was
Brian Weatherson (eds.), Epistemic Modality,
framed.” Philosophy and Phenomenological
chapter ‘Might’ Made Right. Oxford: Oxford
Research 81:664–687.
University Press.
Yadav, Sameer. 2016. “Mystical experience
von Stechow, Arnim. 1984. “Comparing and the apophatic attitude.” Journal of Analytic
semantic theories of comparison.” Journal of Theology 4:17–43.

27