236–267

https://doi.org/10.1093/philmat/nkad006
Philosophia Mathematica
Advance Access Publication on April 12, 2023

The Caesar Problem —

A Piecemeal Solution

Downloaded from https://academic.oup.com/philmat/article/31/2/236/7116333 by guest on 06 September 2024

J.P. Studd∗
Faculty of Philosophy, University of Oxford, Oxford OX2 6GG, U.K.

A BS T R AC T
The Caesar problem arises for abstractionist views, which seek to secure refer-
ence for terms such as ‘the number of Xs’ or #X by stipulating the content of
‘unmixed’ identity contexts like ‘#X = #Y ’. Frege objects that this stipulation
says nothing about ‘mixed’ contexts such as ‘#X = Julius Caesar’. This arti-
cle defends a neglected response to the Caesar problem: the content of mixed
contexts is just as open to stipulation as that of unmixed contexts.

1. THE CAESAR PROBLEM

In his Grundlagen der Arithmetik,1 Frege confronts a strange question:

(1) Is Julius Caesar a cardinal number?

Might the last dictator of the Roman Republic moonlight as a mathematical

object? Frege apologises for asking a question that ‘looks nonsensical’ (§66). All
the same, Caesar has given his name to a long-standing problem facing Fregean
approaches to abstract objects.
The Caesar problem arises in Frege’s discussion of how numbers are ‘given to
us’ (§62). The answer he floats contains the kernel of what is now called abstrac-
tionism. Central to this view is a ‘top-down’ approach to the metasemantics of
abstract terms.2 Even if a community lack an ostensive or descriptive means
to baptise numbers, they may accord reference to a term such as ‘the number
of Xs’ (in symbols: #X)3 by determining the content of a suitable range of

∗
Orcid.org/0000-0003-4345-3168. E-mail: [email protected].
1
[Frege, 1980]. Section references refer to this work unless indicated otherwise.
2
Warren [2017] uses the label ‘top-down metasemantics’ in a similar way. Williams
[2007] dubs this a ‘two-step’ approach to metasemantics.
3
When there is no risk of confusion, quotes are sometimes omitted from formal
expressions.

Philosophia Mathematica (III) Vol. 31 No. 2  c The Author [2023]. Published by Oxford University
Press. This is an Open Access article distributed under the terms of the Creative Commons Attribution
License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution,
and reproduction in any medium, provided the original work is properly cited.
• 236
 The Caesar Problem • 237

whole sentences involving it. Frege, and most abstractionists after him, focus
on identity contexts of the form ‘#X = #Y ’. According to what I will call
his abstractionist proposal, moreover, their content may be stipulated with a
‘definition’ (§63) that has come to be known as hp:

hp For any classes X and Y, the following are equivalent:

#X = #Y ; X and Y are equinumerous.

Downloaded from https://academic.oup.com/philmat/article/31/2/236/7116333 by guest on 06 September 2024

Construed in this way, hp stipulates that the identity context on its left-hand
side ‘is to mean the same as’ (§65) the sentence on its right-hand side.
Frege’s interest in hp stems from his desire to give a purely logical foundation
for arithmetic. As is now well known, the system that results from adding
hp, construed as an axiom, to a standard formulation of second-order logic
interprets Peano Arithmetic. In other words, Frege’s definitions of arithmetical
expressions permit us to rewrite the theorems of arithmetic using no non-logical
terms save for # so that their #-based translations emerge as theorems of the
hp-based system.4
Does this provide a purely logic- and definition-based foundation of
arithmetic? Frege objects that his abstractionist proposal faces the Caesar
problem:

. . . our proposed definition . . . does not provide for all cases. It will not,
for instance, decide for us whether [Julius Caesar] is the same as [the
number of Xs] . . . Naturally no one is going to confuse [Caesar] with [the
number of Xs]; but that is no thanks to our definition of [number]. That
says nothing as to whether the proposition [‘#X = q’] should be affirmed
or denied, except for the one case where q is given in the form of [#Y ].
(§66, example changed.5 )

The problem needs some unpicking. But abstractionists generally follow

Frege in maintaining that a stipulation does not succeed in conferring reference
on terms such as #X unless, at least in some cases, it does ‘decide’ questions
such as (1). Frege rapidly concludes that hp’s stipulation does not settle Caesar
questions, as I will call questions of this kind, and abandons his abstractionist
proposal. He opts instead for his well-known extension-based account of num-
ber, leading him, in Grundgesetze der Arithmetik, to Basic Law V and Russell’s
paradox.
Frege’s abrupt change of tack may seem like an overreaction. Might his
abstractionist proposal, or something like it, instead solve the Caesar prob-
lem? Nowadays, the dominant abstractionist response contends, contrary to

4
This result (Frege’s Theorem) is noted in [Parsons, 1964] and proved in [Wright, 1983].
5
§66 considers ‘England is the same as the direction of the Earth’s axis’. The now-
standard example (Caesar) appears in an objection to a competing account of number
terms in §56.
238 • Studd

Frege, that hp does decide Caesar questions. According to what I will call
wholesale responses, number–Roman identity conditions come bundled whole-
sale with number–number identity conditions. Less metaphorically, although
hp only overtly stipulates the content of ‘unmixed’ identity contexts — those
of the form #X = #Y — this stipulation also serves covertly to determine the
content of a range of other contexts, including ‘mixed’ identity contexts such
as #X = q, in which q is not a #-term. Consequently — given how the world
is6 — hp decides the corresponding Caesar questions.

Downloaded from https://academic.oup.com/philmat/article/31/2/236/7116333 by guest on 06 September 2024

This has been the guiding approach behind the various iterations of Bob Hale
and Crispin Wright’s influential response to the Caesar problem. According to
Hale and Wright, the stipulation of hp fixes the meaning of the #-operator
so as to introduce a sortal concept: cardinal number. Moreover, the concept’s
application-condition, which settles whether something (e.g., Caesar) is a car-
dinal number is, as they put it, ‘extractable from’ the identity conditions for
cardinal numbers, directly stipulated by hp [2001b, p. 369]. Hale and Wright’s
most recent proposal appeals to a category-based ontology to effect this extrac-
tion. Competing wholesale responses have deployed different assumptions about
the ambient metaphysics or metasemantics to bridge the gap between the con-
tent hp explicitly stipulates and the content that, on a wholesale view, hp
determines by some other means.7
In my view, wholesale responses grossly overestimate what is achieved just
by laying down hp. On the face of it, however, there is a much easier way to
solve the Caesar problem. According to my favoured response, the content of
a mixed context such as ‘#X = Caesar’ is just as open to stipulation as the
content of an unmixed context such as ‘#X = #Y ’. On this view, then, Caesar
questions may be decided piecemeal, via further stipulations governing unmixed
contexts laid down alongside hp. This is the main idea behind the piecemeal
response to the Caesar problem.
Suggestions along these lines occasionally surface in the extensive debate
surrounding this issue. Frege, in Grundgesetze, makes short work of the Caesar
questions that arise in the context of his formal system by stipulating, in effect,
that each truth value is identical to an extension, namely its singleton [2013,
§10]. In a similar spirit, Michael Dummett notes in passing that it would be
‘straightforward’ to attain negative answers to the Caesar questions that arise
in the Grundlagen by ‘direct stipulation’ [1978, p. 111]. More recently, Øystein
Linnebo [2018, p. 160] has offered a more sustained defence of the view that

6
A sentence’s content determines its truth conditions (in a context); its truth value
also depends on the world. Henceforth I sometimes follow Frege in leaving the world’s
contribution tacit.
7
For the category-based approach, see [Wright, 2020, pp. 309–315] and [Hale and
Wright, 2001b, p. 369]. Earlier neologicist responses are offered in [Wright, 1983, ch. 3]
and [Hale, 1987, ch. 8]. For a competing approach, based on ‘real definitions’, see [Rosen
and Yablo, 2020].
 The Caesar Problem • 239

answers to Caesar questions often turn on ‘conceptual decisions’ and not just
‘factual discoveries’.8
Despite this, the piecemeal response remains underdeveloped. One reason for
this, I suspect, is that a would-be stipulative solution to the Caesar problem is
widely taken to be hopeless. Richard Kimberly Heck, for example, offers this
verdict on the stipulative identifications in Grundgesetze:

Downloaded from https://academic.oup.com/philmat/article/31/2/236/7116333 by guest on 06 September 2024

Plainly, Frege is not here offering a solution to the Caesar problem: A
piecemeal ‘solution’ is not a solution to the problem but a recipe for
side-stepping it. [2005, n. 17]

Heck does not elaborate on why, in the general case, a piecemeal response falls
short of a fully fledged solution. Fortunately, others have been more explicit
about their misgivings. Fraser MacBride objects that piecemeal stipulations
may conflict with ‘antecedent facts’ [2006a, p. 193]:

Suppose that Caesar leads a double life. Suppose that in addition to

leading his material existence Caesar is also a number. In that case the
stipulation that sentences that say Caesar is a number are all false cannot
succeed. For some of these sentences will be true and true sentences cannot
be stipulated to be false. . . . Stipulation cannot suffice as a basis for
determining that Caesar is no number. [2006a, p. 192]

Another worry stems from Frege’s observation that his Grundgesetze stipula-
tion cannot be coherently generalized to identify each item — and, in particular,
each extension — with its singleton, on pain of conflict with the identity con-
ditions laid down for extensions [2013, §10].9 Hale and Wright object that
piecemeal stipulations risk incoherence:

. . . before we can safely stipulate that some object . . . is a certain exten-

sion, we need an assurance that it is not (behind our back, as it were)
some other extension — else our new stipulation might conflict with the
original stipulation of identity-conditions . . . A solution to the Caesar
Problem is thus presupposed, and cannot be provided, by generalizing
the kind of stipulation Frege envisages for truth-values. [2001b, n. 8]

Both worries merit closer examination. In the end, though, I will argue that
neither poses a serious threat to the response I defend. Moreover, when prop-
erly developed, the piecemeal response not only withstands these objections

8
A view in a similar spirit is sometimes defended in the context of mathematical struc-
turalism. Shapiro [1997, p. 81] maintains that cross-structural identifications are ‘matters
of decision, based on convenience, not matters of discovery’.
9
The incoherence here is separable from Basic Law V’s inconsistency. The piecemeal
stipulation also conflicts with consistent versions of this axiom, such as New V (stated in
2.2).
240 • Studd

but, in important respects, improves on its wholesale competitors. First, we

need a better grasp of the underlying problem. The remainder of this section
elaborates on what I take to be the metasemantic core of abstractionism (1.1)
and the challenge to this metasemantics posed by Caesar (1.2). Section 2 then
argues that the problems facing the wholesale view are worse than has so far
been recognized. Most damagingly, a wide range of wholesale views fall foul
of Benacerraf’s multiple-reduction problem and unduly constrain mathemati-

Downloaded from https://academic.oup.com/philmat/article/31/2/236/7116333 by guest on 06 September 2024

cal freedom. Section 3 then develops a piecemeal version of abstractionism that
avoids these problems and has a robust reply to the MacBride and Hale–Wright
worries.

1.1.
Frege’s abstractionist proposal may be developed in various ways. The best
known stops just short of full-blown logicism. According to Hale and Wright’s
neologicist version of abstractionism, hp does not provide a foundation for
arithmetic that is based purely on logic and definitions, strictly conceived; but
hp is a definition-like truth, in a more liberal sense.10
To extend their programme to other branches of mathematics, neologicists
have sought other axioms that enjoy a similar status. An abstraction principle
is standardly taken to have an hp-like form:11

ap For any x and y, the following are equivalent:

σx = σy; x ∼σ:σ y.

In the axiom, x and y are first- or higher-order variables that range over entities
of the corresponding type (individuals, classes of individuals, etc.); x ∼σ:σ y
expresses an equivalence relation on these entities; and σ is a singular-term-
forming operator. Let us call a term of the form σx an abstract- or σ-term,
and its referent (if any) for a suitable value of x a σ-abstract, or simply, an
abstract. The axiom ap thus states identity conditions for σ-abstracts in terms
of a unity relation ∼σ:σ holding between their specifications. I also speak of
#-abstracts as #-cardinals, and so on, when an abstraction principle seeks to
introduce mathematical objects of a familiar kind.12
Two comments are in order. First, in cases where x and y are higher-order
variables, the usual range of options for interpreting them is available. For con-
creteness, I call the value of a first-order variable — x, y, etc. — an individual
(object, item) and the value of a monadic second-order variable — X, Y , etc.
— a class of individuals. But an abstractionist may equally adopt a Fregean

10
See, for instance, [Hale and Wright, 2001a, p. 4].
11
Formally, an instance of the ap-schema is canonically taken to be the universal closure
of a biconditional: σx = σy ↔ x ∼σ:σ y. See, for instance, [Hale, 2000, p. 100; Hale and
Wright, 2001a, p. 16; Ebert and Rossberg, 2016, p. 3].
12
My terminology is intended to be neutral on whether #-cardinals (items introduced
by hp) are cardinals (familiar mathematical objects). I return to this question in 1.2.
 The Caesar Problem • 241

or plural interpretation of second-order variables. In this case, my ‘class’-talk

below should be seen as elliptical for discourse about Fregean concepts or plu-
ralities. In each case, the class-, concept-, or plurality-interpretation may be
extended to polyadic and higher-order variables in the standard way.13
Second, although official characterizations of abstraction principles in the ap-
mould are widespread, abstractionists often rely on a more liberal conception of
the distinguished class of axioms. For example, Hale [2000] proposes to obtain

Downloaded from https://academic.oup.com/philmat/article/31/2/236/7116333 by guest on 06 September 2024

positive reals from a suitable domain of quantities, and then to identify reals,
positive or non-positive, with differences of positive reals:14

d For any reals x, y, z, w > 0, the following are equivalent:

diff(x, y) = diff(z, w); x + w = y + z.

Hale’s axiom fails to fit the ap-mould unless it is liberalized in two ways: (i)
to permit polyadic σ-terms of the form σx for x = x1 , . . . , xn and (ii) to
allow for the specification variables to be restricted to a domain, Dσ .15 Since
Hale intends d to ‘work in essentially the same way as paradigm abstractions’
(p. 107),16 I can see no principled reason for abstractionists not to liberalize
their characterization, as per (i) and (ii). In deference to tradition, however,
I use the term ‘abstraction principle’ in its standard sense and use unmixed
postulate for both strict instances of ap and their liberalized counterparts.
Terminology aside, a central neologicist contention is that, in a wide range
of cases, we may secure reference to σ-abstracts, and gain a priori knowledge of
their identity conditions, by laying down the relevant postulate as an ‘implicit
definition’.17 Here I propose to bracket much-debated questions surrounding
the epistemic status of abstraction principles and focus on what I take to be
the metasemantic core of Frege’s abstractionist proposal, shared in its essentials
by neologicism and other variants of abstractionism.
For simplicity, let us focus on an idealized linguistic community who initially
speak a higher-order language. The language need not contain abstract-term-
forming operators; but it is convenient to suppose that it does contain a singular
term ‘Caesar’ that refers to Caesar.18 Suppose further that the language’s initial
interpretation is encoded as a model-theoretic interpretation of the standard

13
For the plural case, see, for instance, [Linnebo and Rayo, 2012, apps. A, B.2].
14
Strictly, Hale abstracts d-reals from the abstracts he identifies with positive reals
(namely, ratios of quantities). Similar remarks apply to cp and cp below.
15
Boldface x and y are henceforth used in this way for a sequence of one or more
variables, which may differ in type. Officially, the domain Dσ is specified by a condition
δσ (x) associated with σ, and comprises the entities that satisfy this condition under the
pre-abstraction interpretation.
16
Shapiro [2000, p. 338] makes similar remarks about a polyadic postulate.
17
See, for instance, [Wright, 1997, pp. 278–280].
18
This account of abstraction is readily refined to permit the initial language to include
higher-order predicates and function symbols other than σ-operators, which I omit for ease
of exposition.
242 • Studd

kind — an MT-interpretation — which assigns semantic values to its expres-

sions and determines truth values for its sentences in a broadly Tarskian way.19
In outline, an attempt to confer reference on σ-terms via abstraction takes place
in three stages:20

• First, if necessary, the community may expand their lexicon with one
or more abstract-term-forming operators. This purely syntactic addition
leaves them with an initial interpretation of the expanded lexicon which

Downloaded from https://academic.oup.com/philmat/article/31/2/236/7116333 by guest on 06 September 2024

accords no reference to the newly available σ-terms.21
• Second, the community stipulate the content of a range of sentences
framed in the expanded lexicon. To begin with, let us assume, in line
with the abstractionist mainstream, that an abstraction attempt is based
on one or more unmixed postulates that each stipulates the content of
the relevant identity context in terms of an antecedently expressible unity
relation.22
• Third, semantic values for expressions in the expanded lexicon (and thus
referents for σ-terms) are selected subject, at least, to the constraint that
these compositionally determine the stipulated sentential content. If the
abstraction attempt is successful, the expanded lexicon is furnished with a
new interpretation that extends the initial one in line with this constraint
(leaving the referent of ‘Caesar’ unchanged). In this case, let’s say that the
post-abstraction interpretation extends the initial interpretation according
to the stipulated postulates.23

Abstractionists may flesh out this bare-bones sketch depending on their pre-
ferred view of stipulation, sentential content, semantic values, and so on. But

19
Officially, an MT-interpretation is a pair M, I comprising a non-empty set M as its
domain and a function I that assigns a suitable referent/extension to each expression in
the lexicon. I say ‘broadly Tarskian’ since the semantics allows for empty abstract terms.
For an n-ary operator σ, I(σ) is a partial function from n-tuples of appropriately typed
entities based on M (either members of M or classes built from members of M ) to members
of M . To assign truth values to atomic contexts which contain undefined σ-terms, I adopt
a ‘negative free’ semantics that rules them false, and otherwise assigns truth values in the
usual Tarskian way.
20
Compare [Hale, 1997, p. 98; Wright, 1997, pp. 276–279; Heck, 2011a, pp. 51–52;
Studd, 2016, pp. 583–590, 597–599]; and [Linnebo, 2018, ch. 8].
21
A community may make successive abstraction attempts. In this case, an attempt’s
initial interpretation may accord reference to σ-terms introduced by earlier attempts.
22
The notion of sentential content here must be unstructured, permitting the same
content to be attached to sentences with different syntactic structures.
23
Formally, when the initial and post-abstraction interpretations — I and J — are
MT-interpretations (of the expanded lexicon), M, I and N, J, then J extends I accord-
ing to some stipulated unmixed postulates if two conditions are met: (i) J is an extension of
I, in the model-theoretic sense that M ⊆ N , I(R) = J(R) ∩ M n , I(σ) = J(σ)dom(I(σ)),
and I(c) = J(c), for each n-ary predicate R, operator σ, and constant c in the expanded
lexicon; (ii) for each stipulated postulate, when x and y are assigned to members of Dσ ,
the terms σ(x) and σ(y) are defined under J , and the truth value of the postulate’s
left-hand side under J coincides with the truth value of its right-hand side under I.
 The Caesar Problem • 243

here, I propose to stay as neutral as possible on the details and focus on the
core account of how — to indulge in some suggestive metaphors — σ-abstracts
are ‘abstracted from their specifications’, ‘introduced’, or ‘given to us’.

1.2.
Frege’s complaint that hp fails to decide Caesar questions poses a significant
challenge to the proposed metasemantics.24 Consider the following question fac-

Downloaded from https://academic.oup.com/philmat/article/31/2/236/7116333 by guest on 06 September 2024

ing a community who make an abstraction attempt based solely on hp (writing
Λ for the empty class):

(2) Is #Λ = Caesar?

The Caesar problem then emerges in the following inconsistent triad:

C1 The hp-abstraction attempt determines a unique referent for #Λ (and

leaves the referent of ‘Caesar’ unchanged): post-abstraction, under the
new interpretation, ‘Caesar’ is a singular term that refers to Caesar and
#Λ is a singular term that refers to a unique item b.
C2 The hp-abstraction attempt confers the standard syntax and semantics
on the identity predicate: post-abstraction, if s and t are singular terms
that refer to a and b, then ‘s = t’ is a well-formed identity context that
is true if a is identical to b and false if a is not identical to b.
C3 The hp-abstraction attempt settles no determinate — yes/no — answer
for (2): post-abstraction, ‘#Λ = Caesar’ is neither true nor false.

A solution to this version of the Caesar problem calls for a well-motivated denial
of one of C1–C3.
Before I come to the wholesale response, two less popular responses merit
attention. The first rejects the first assumption in the triad. Of course, the
assumption C1 would hold if the hp-abstraction attempt singled out a unique
model-theoretic interpretation as the new interpretation. But, on reflection, how
could it do so? If there are any MT-interpretations that extend the initial one
according to hp, there are infinitely many isomorphic MT-interpretations that
extend it according to hp. Some take #Λ to refer to a non-Roman; others to a
Roman. How, then, could the hp-abstraction attempt determine a unique, and
presumably non-Roman, referent for #Λ?
Rather than attempt to answer this question, advocates of the first response
reject C1 and maintain that the hp-abstraction attempt, even if successful,

24
Compare, for instance, [Heck, 1997a, pp. 277–278; MacBride, 2006a, pp. 186–189;
Hale and Wright, 2001b, pp. 341–342]. Heck and MacBride also emphasize an epistemo-
logical aspect to the problem. The main text continues to focus on the metasemantic
issue. But let me add that, on the piecemeal view I defend, answers to Caesar questions
are settled by postulates that plausibly enjoy a similar epistemic status to abstraction
principles.
244 • Studd

leaves the reference of #-terms indeterminate. The usual range of options is

available to account for indeterminacy. One straightforward way is to deploy
a supervaluationist semantics.25 On this view, the post-abstraction interpreta-
tion corresponds to a class of MT-interpretations, encoding admissible ways of
selecting determinate semantic values for the extended lexicon. A sentence is
then true (false) under the post-abstraction interpretation if it is true (false)
under each admissible MT-interpretation; otherwise, it lacks a determinate

Downloaded from https://academic.oup.com/philmat/article/31/2/236/7116333 by guest on 06 September 2024

truth value. In the absence of further constraints on the reference of #-terms,
the abstractionist might adopt a radical version of the indeterminacy response
that deems admissible every MT-interpretation that extends the initial one
according to hp. According to the radical indeterminacy response, #Λ may be
admissibly interpreted to refer to anything whatsoever, with the result that
‘#Λ = Caesar’ lacks a determinate truth value.
A second response shares something of the quasi-structuralist feel of the
indeterminacy view. This response maintains that (2) embodies a category mis-
take.26 Consequently, C2, which accords the standard syntax and semantics to
the problematic identity context, is denied on the grounds that ‘#Λ = Caesar’
is either syntactically or semantically defective. This response also admits of a
radical version, which takes every Caesar question to be defective.
Might the indeterminacy or category-mistake response be developed into
a satisfying solution to the Caesar problem? Not, I think, in their radical ver-
sions. Caesar questions are much more commonplace than ‘nonsensical’-looking
examples like (1) and (2) may suggest. Mixed identity contexts also give rise
to questions like the following:

(3) Is #Λ = 0N ?

Here, ‘0N ’ is not presumed to be a #-term: instead it is a disambiguated version

of the ordinary English numeral that refers to a familiar natural number.27 (I
adopt a similar convention with subscripts below, reserving ‘N’, ‘Q’, and ‘R’ for
the familiar natural, rational, and real numbers.) More generally, the abstrac-
tionist also needs to account for atomic contexts of the form R(t1 , . . . , tn ) where
one or more of t1 , . . . , tn are σ-terms. These give rise to a further stock of Caesar
questions:28

(4) Is #Λ Roman?

25
Compare Boccuni and Woods’s [2020] supervaluationist treatment of what they take
to be the arbitrary reference of #-terms. MacBride [2006a, pp. 190–193] critically discusses
this option.
26
Compare [Benacerraf, 1965, pp. 64–67] and [Heck, 1997a, pp. 280–281; Heck, 2011b,
pp. 17–19].
27
More cautiously, ‘0N ’ purports to refer to a number. Since ontology is not at issue
here, I omit this caveat, and take for granted the existence of familiar mathematicalia.
28
Questions such as (3) and (4) respectively exemplify what have been dubbed the
Counter-Caesar problem [MacBride, 2003] and the Roman problem [Fine, 2002]. In my
view, the various issues are really different aspects of a single problem.
 The Caesar Problem • 245

(5) Is #Λ non-concrete?
(6) Is #Λ ∈ N?

As in the case of (2), there is a straightforward sense in which hp ‘says

nothing’ to answer questions such as (3)–(6): provided the abstract terms are
undefined pre-abstraction, if any MT-interpretations extend the initial inter-
pretation according to hp, some of them render the embedded context true
and others render it false.29 According to the radical indeterminacy response,

Downloaded from https://academic.oup.com/philmat/article/31/2/236/7116333 by guest on 06 September 2024

therefore, the Caesar questions (2)–(6) all lack determinate answers, as is also
the case according to the radical version of the category-mistake response.
Abstractionists, however, cannot leave (3) and (6) unanswered — at least,
not if they wish to explain how the natural numbers are given to us. For,
unless these numbers themselves are found among the #-abstracts introduced
by hp, abstraction at best introduces ersatz copies of the natural numbers. If,
on the other hand, each natural number is identical to the corresponding #-
abstract, this requires an affirmative answer to (3) and (6).30 A similar point
can be made about (5). This time, an affirmative answer is required by the
broader metaphysical outlook adopted by neologicist abstractionists like Hale
and Wright.31
Abstractionists, then, have independent reasons to think that (3), (5), and
(6) have determinate answers. Semantically ascending, each of the embedded
contexts has a determinate truth value and, therefore, is not semantically or

29
The ‘says nothing’ point may be established with a permutation argument. More
generally, suppose that J is an MT-interpretation N, J that extends I according to one
or more unmixed postulates. Consider a Caesar question that embeds a context of the form
R(c1 , . . . , cm , σ1 (x1 ), . . . , σn (xn )) whose σi -terms, for i = 1, . . . , n, are formed using dif-
ferent operators and undefined under I for all assignments. Then, except in a trivial case,
whatever truth value the R-context receives under J (when its free variables are respec-
tively assigned to a1 , . . . , an ), there are also MT-interpretations J1 and J2 that extend I
according to the same unmixed postulates, and differ from J at most on the reference of
σ1 , . . . , σn , that respectively render the R-context true and false (under this assignment).
The trivial case is when the n-ary relation on N defined by R(c1 , . . . , cm , x1 , . . . , xn )
under J is permutation invariant. In this case, the Caesar question is insensitive to which
members of N are the referents of σi -terms. Otherwise, there is a permutation π on N such
that a1 , . . . , an  stands in the defined relation and πa1 , . . . , πan  does not. In this non-
trivial case, for i = 1, . . . , n, let πi transpose (J(σi ))(ai ) and ai . Then J1 and J2 may be
defined to be just like J , except that J1 (σi ) = πi ◦ (J(σi )) and J2 (σi ) = π ◦ (J1 (σi )). By
construction, J1 and J2 differ on the Caesar question, and adapting the reasoning of Fine’s
switching lemma [2002, p. 110] both J1 and J2 extend I in accordance with the relevant
postulates. The same argument may be applied, mutatis mutandis, when the embedded
context has the form σx = c. In this case, the non-triviality condition requires only that
N contain more than one item. For discussion of whether Frege anticipates permutation
arguments of this kind see [Wehmeier and Schroeder-Heister, 2005].
30
MacBride [2006a, p. 134] emphasizes this point. How far ersatz copies would serve
Frege’s logicist aims is open to question: see [Benacerraf, 1981; Weiner, 1984], and
[Blanchette, 2012, ch. 4]. When it comes to abstractionism, however, the interest of the
view is significantly diminished if it cannot explain reference to familiar mathematical
objects. Wright [1999, p. 322] is clear that his target is ‘genuine arithmetic’.
31
See, for instance, [Hale and Wright, 2001a, p. 7].
246 • Studd

syntactically defective. This rules out the radical versions of the indeterminacy
and category-mistake responses. But it is compatible with moderate versions
of these responses that accord a determinate answer to (3), (5), and (6) and
accord no determinate answer to other Caesar questions, perhaps including
(2) and (4). Developed in this way, however, the indeterminacy and category-
mistake responses at best provide a partial solution to the Caesar problem. For
no progress has yet been made in explaining how abstraction settles answers in

Downloaded from https://academic.oup.com/philmat/article/31/2/236/7116333 by guest on 06 September 2024

the determinate, non-defective cases.
This brings me to the wholesale response. On this view, at least in some
cases, abstraction does decide Caesar questions. In the case of Caesar himself,
it is open to a wholesale abstractionist to follow Hale and Wright in maintaining
that hp does decide (2), contrary to C3.

2. WHOLESALE RESPONSES
According to wholesale versions of abstractionism, an abstraction attempt not
only stipulates the content of unmixed identity contexts, it also determines the
content of mixed identity contexts or other atomic contexts and thereby —
given how the world is — settles answers for the corresponding Caesar ques-
tions. The view encompasses a wide range of responses to the Caesar problem
that differ on their accounts of how abstraction decides Caesar questions.32
Rather than attempting to pick these off one by one, my aim in this section is
to raise three objections against this general style of response: (i) a broad range
of wholesale responses conflict with various intuitive identity judgements; (ii)
they give rise to a version of Benacerraf’s multiple-reduction problem; and (iii)
they unduly constrain mathematical freedom.

2.1.
The first objection arises in connection with Caesar questions concerning
the identity of abstracts. Suppose that σ- and ρ-abstracts are introduced by
different unmixed postulates. Under what conditions are they identical?33
Two wholesale answers have dominated the discussion. According to the
first, σ- and ρ-abstracts are distinguished in a fine-grained way:34

fg For any x ∈ Dσ and y ∈ Dρ , the following are equivalent:

σx = ρy; ∼σ:σ and ∼ρ:ρ express the same relation and x ∼σ:σ y.

Despite its superficial similarity with an abstraction principle, fg is no

stipulation. Instead, the wholesale proposal is intended as a substantive meta-
physical thesis.35 This particular proposal — which automatically distinguishes

32
Some of the options are listed in fn. 7.
33
Cook and Ebert [2005] dub this version of the Caesar problem the C-R problem.
34
[Fine, 2002, p. 48] outlines this fine-grained response.
35
Compare, for instance, [Hale and Wright, 2001b, p. 370].
 The Caesar Problem • 247

abstracts associated with different unity relations — may be motivated by Hale

and Wright’s suggestion that ontology divides into disjoint categories:36

Within a category, all distinctions between objects are accountable by

reference to the criterion of identity distinctive of it, while across cate-
gories, objects are distinguished by just that — the fact that they belong
to different categories. [Hale and Wright, 2001b, p. 389]

Downloaded from https://academic.oup.com/philmat/article/31/2/236/7116333 by guest on 06 September 2024

Whatever the merits of a category-based ontology, Roy Cook and Philip
Ebert [2005, p. 125] maintain that fg is subject to a counterexample, deriving
from [Fine, 2002].37 Consider a variant of hp they dub fhp:38

fhp For any classes X and Y , the following are equivalent:

nX = nY ; X and Y are equinumerous or both infinite.

Although they come apart in infinite cases, hp and fhp agree on the identity
conditions for the finite #- and n-cardinals which provide candidate natural
numbers according to accounts of arithmetic based on these postulates:39

0# =df #Λ (i.e. #{x : x = x}); 0n =df nΛ;

(n + 1)# =df #{m# : m# ≤ n# }. (n + 1)n =df n{mn : mn ≤ nn }.

Cook and Ebert intuit that corresponding finite #- and n-cardinals are identi-
cal: in general, n# = nn . According to fg, however, there is no overlap between
#- and n-cardinals (since hp and fhp deploy different unity relations). If Cook
and Ebert are right, the wholesale proposal incorrectly distinguishes identical
abstracts: fg is sometimes too fine-grained.

2.2.
Cook and Ebert’s intuition is not beyond question; but let it stand for now.40
Might a wholesale abstractionist accommodate this intuition? Fine proposes a
more coarse-grained view [2002, pp. 47–49]:

cg For any x ∈ Dσ and y ∈ Dρ , the following are equivalent:

σx = ρy; {z : z ∼σ:σ x} = {z : z ∼ρ:ρ y}.

36
Compare [Linnebo, 2005; Linnebo, 2018, pp. 167–169].
37
Fine offers a more cautious verdict on the cases discussed in 2.1–2.2. See 2.4.
38
A closely related axiom, hpf, is investigated by Heck [1997c].
39
Officially, I take ‘n# ’ and ‘nn ’ as shorthand for the corresponding #- and n-terms.
Similar remarks apply to ‘∅† ’ and ‘0sup ’.
40
I return to the various intuitions in play in 2.4.
248 • Studd

On this view — which, again, is not intended as a stipulation — σ- and ρ-

abstracts are identical when their specifications determine the same equivalence
class under their respective unity relations; equivalently — dispensing with the
‘class’-talk — when the same specifications stand in the σ-abstract’s unity
relation to its specification as stand in the ρ-abstract’s unity relation to its
specification.41
It is not so clear what underlying metaphysics might be used to motivate cg.
In its favour, it does deliver Cook and Ebert’s intuitive judgment that n# = nn .

Downloaded from https://academic.oup.com/philmat/article/31/2/236/7116333 by guest on 06 September 2024

Putative counterexamples to cg, however, have also been forthcoming. Here is
another case deriving from [Fine, 2002]. Neologicist attempts to recover set
theory often deploy a variant of Basic Law V to introduce what I will call
†-sets:42

new v For any X and Y, the following are equivalent:

†X = †Y ; X and Y are coextensive or both universe-sized.

The second disjunct in new v’s right-hand side implements a ‘limitation-of-size’

fix to Russell’s paradox, which would emerge without it. This time the difficulty
centres on the empty †-set: ∅† =df †Λ. According to cg, 0# = ∅† (since they
are both associated with the same equivalence class). My intuitive judgment,
however, is that there is no overlap between natural numbers and sets. Assum-
ing that #-naturals and †-sets are identical with their familiar counterparts,
it follows that 0# = ∅† . If this is right, Fine’s proposal incorrectly identifies
distinct abstracts: cg is sometimes too coarse-grained.

2.3.
The counterexamples levelled against the fg- and cg-proposals may prompt
the wholesale abstractionist to seek an intuitive Goldilocks zone somewhere
between the two — thus a medium-grained proposal:43

mg For any x ∈ Dσ and y ∈ Dρ , the following are equivalent:

σx = ρy; x ∼σ:ρ y.

41
The right-hand side of cg may be formalized ∀z(z ∼σ:σ x ↔ z ∼ρ:ρ y), following
[Cook and Ebert, 2005, p. 136]. Compare [Fine, 2002, p. 167].
42
See, for instance, [Wright, 1997] and [Boolos, 1998], which dubs the axiom ‘new v’. It
is important to note the type distinction between a †-set (a value of a first-order variable)
and the corresponding class (a value of a second-order variable). I write ∅ for the empty
set and Λ for the empty class.
43
Compare [Fine, 2002, p. 54]. The ‘mixed’ unity relation ∼σ:ρ giving identity con-
ditions for σ- and ρ-abstracts may depend on the ‘unmixed’ unity relations that govern
these abstracts (as in the cases of fg and cg).
 The Caesar Problem • 249

Really, this is a proposal schema. The relation ∼σ:ρ may be any ‘medium-
grained’ unity relation, no finer than fg’s and no coarser than cg’s. The mg-
proposal consequently shares two assumptions with fg and cg:

nff If ∼σ:σ and ∼ρ:ρ express the same relation and x ∼σ:σ y, then σx = ρy.
ncc If σx = ρy, then {z : z ∼σ:σ x} = {z : z ∼ρ:ρ y}.

Downloaded from https://academic.oup.com/philmat/article/31/2/236/7116333 by guest on 06 September 2024

The no-finer-than-fg assumption — nff — is hard to resist if the whole-
sale abstractionist follows Hale and Wright in maintaining that the meaning
of a σ-operator is fixed solely by the identity conditions laid down in the rel-
evant unmixed postulate.44 On this view, successful attempts to introduce σ-
and ρ-abstracts using notational variants of the same unmixed postulate, each
deploying the same unity relation, accord the same meaning to σ and ρ. Con-
sequently, given a modest amount of compositionality, ‘σx = ρy’ is equivalent
to ‘σx = σy’. And this is enough to secure nff on the basis of the unmixed
postulates.45
The no-coarser-than-cg assumption — ncc — is less central to the wholesale
view and potentially more controversial.46 As Fine observes, however, if we deny
ncc, ‘it is hard to see what reasonably systematic view could associate the same
abstract with two different equivalence classes’ [2002, p. 47].
What I sometimes loosely call ‘the’ mg-proposal consequently encompasses
a broad class of wholesale responses, including, as limiting cases, the fg- and
cg-proposal. Different wholesale abstractionists may choose an instance appro-
priate to their preferred account of how abstraction decides Caesar questions.
Once again, however, any mg-proposal must contend with prima facie coun-
terexamples. This time, take Shapiro’s [2000] proposal to abstract reals from
classes of rationals:47

cp For any X,Y ⊆ Q, the following are equivalent:

44
See, for instance, [Wright, 2020, p. 304].
45
Assuming the antecedent of nff, ‘x ∼σ:σ y’ holds and, by the unmixed postulates,
is equivalent to ‘σx = σy’; by the argument in the text, this is equivalent to ‘σx = ρy’,
which consequently also holds, as per the consequent of nff.
46
Linnebo and Uzquiano [2009, p. 248] note that ncc ‘figures as an uncontroversial
minimal assumption’ in all the ‘technically precise’ discussions they were then aware of. In
the case of abstracts associated with ‘coarsenings of hp’, such as fhp, however, an identity
axiom since put forward by Ebels-Duggan [2021] runs contrary to this assumption: his
structural identity principle identifies n(N) and #(N) even in cases when these abstracts are
associated with different equivalence classes. Without restriction, however, Ebels-Duggan’s
proposal also runs contrary to nff, by giving a negative answer to Caesar questions of the
form ‘Is #1 X = #2 X?’ in cases where both operators are introduced by duplicates of
hp. To avoid this, he restricts the structural identity principle to render it silent on such
questions. Whether this proposal may be developed into a ‘reasonably systematic’ view,
then, depends on whether there is a non-ad hoc way to lift this restriction and to extend
the account beyond coarsenings of hp.
47
Shapiro calls the relevant abstracts cuts.
250 • Studd

sup X = sup Y ; X and Y have the same rational upper bounds.

Reals are identified with the sup-abstracts specified by non-empty classes with
upper bounds and 0sup =df sup{r ∈ Q : r < 0Q }. Now, according to ncc,
0sup = 0# (since the abstracts are associated with different equivalence classes).
The more intuitive view, it seems to me, however, is that there is just one
number zero, both a natural and a real (and a rational, etc.). The assumption
ncc conflicts with the view that 0N and 0R are the same number if we also

Downloaded from https://academic.oup.com/philmat/article/31/2/236/7116333 by guest on 06 September 2024

suppose that the familiar zero is identical with its #- and sup-counterparts. If
this is right, any instance of mg (including fg and cg) incorrectly distinguishes
identical abstracts.

2.4.
But is this right? Let us no longer delay confronting the obvious rejoinder to
the supposed counterexamples. The argument against fg, cg, and mg relies on
two groups of assumptions. First, there are the various intuitions:
(i1) n# = nn ; (i2) 0N = ∅; (i3) 0N = 0 R .
Cook and Ebert’s intuition, (i1), concerning abstracts is in immediate conflict
with fg. My intuitions, (i2) and (i3), concerning familiar sets and numbers
conflict with cg and mg if it is assumed that these mathematical objects —
the natural 0N , real 0R , and set ∅ — are identical to the corresponding abstracts:
(r1) 0# = 0N ; (r2) 0sup = 0R ; (r3) ∅† = ∅.
The importance of reduction theses such as (r1)–(r3) has already been noted.
The abstractionist’s interpretation of arithmetic aspires to provide a faithful
reduction in the sense that the intended meaning of arithmetical expressions is
captured by their #-based translations. This requires, in particular, that ‘0# ’
is accorded the same meaning as ‘0N ’ and thus that (r1) is true.48 Similarly, a
faithful reduction of real analysis or set theory via the standard cp- or new v-
based interpretation requires the truth of (r2) or (r3).
A proponent of fg, cg, or mg may find it easier, however, to reject intuitions
like (i1)–(i3). According to Fine [2002, p. 73], for example, it ‘is not clearly
incorrect’ to reject (i1). In the case of (i2), Fine maintains that if we ‘conceive of’
sets as abstracts, it is again ‘not clear’ that the number 0 is to be distinguished
from the empty set (p. 54).49 Putative ‘cross-sort’ identities, such as (i3), are
especially vexed. For example, prior to his more recent stipulative proposal,
Linnebo [2005, p. 219] contends that intuitions of this kind present a ‘mis-
diagnosis’ of the fact that numerals like ‘0’ are ambiguous, denoting natural
numbers on some disambiguations, and distinct reals on others.50

48
This follows by compositionality again given that ‘0N = 0N ’ is true.
49
Fine himself is questioning (i2) here only if the items we ‘conceive of’ as set-like
abstracts are indeed sets (and 0 is 0N ).
50
See also [MacBride, 2003, pp. 129–130] and [Shapiro, 2006, pp. 128–129].
 The Caesar Problem • 251

What becomes of the case against fg, cg, and mg if a wholesale abstrac-
tionist bites the bullet? Well, for what it is worth, I stand by my intuitive
judgements about numbers and sets (although, unlike Cook and Ebert, I do
not have strong intuitions about #- and n-abstracts). Fortunately, though, I
need place no weight on these divisive claims. Wholesale abstractionism faces
two further objections that do not turn on intuitions such as (i1)–(i3).

Downloaded from https://academic.oup.com/philmat/article/31/2/236/7116333 by guest on 06 September 2024

2.5.
One objection is that the wholesale proposals lead to a version of Benac-
erraf’s multiple-reduction problem. The original version concerns would-be
set-theoretic reductions of arithmetic. Recall that, among other options, we
may interpret arithmetic in set theory either by identifying natural numbers
with z-ordinals (following Zermelo) or vn-ordinals (following von Neumann):

0z =df ∅; 0vn =df ∅;

(n + 1)z =df {nz }. (n + 1)vn =df {0vn , 1vn , . . . , nvn }.

The problem then comes out in a trilemma:

• ALL: each of the candidate interpretations is a faithful reduction, so that

its identifications are genuine identities. When the candidate interpreta-
tions are Zermelo’s and von Neumann’s, this implies that, for each natural
number n, n = nz and n = nvn . This horn leads to absurdity. A modicum
of set theory demonstrates that 2z = 2vn .
• SOME: some but not all of the candidate interpretations are faithful
reductions. In this case, what reason have we to think 2N = 2z , as opposed
to 2N = 2vn , or vice versa? After all, the two interpretations agree on the
arithmetical properties of the natural numbers. In the absence of a ‘cogent
reason’ to think that naturals are z-ordinals rather than vn-ordinals, or
vice versa, Benacerraf rejects as ‘hardly tenable’ the position that the
number–set identity facts are unknowable truths [1965, p. 284].
• NONE: none of the candidate interpretations are faithful reductions. Hav-
ing rejected ALL and SOME, Benacerraf concludes that natural numbers
are neither z- nor vn-ordinals.

In its original form, the trilemma poses no immediate threat to abstraction-

ism. Notwithstanding Frege’s later position, the core abstractionist view carries
no commitment to reduce familiar mathematicalia to sets or extensions. But
what about would-be reductions to abstracts? In the case of real analysis, for
example, Shapiro’s sup-based interpretation has a natural dual that takes reals
to be the inf-abstracts specified by non-empty classes of rationals with lower
bounds:

cp For any X, Y ⊆ Q, the following are equivalent:

inf X = inf Y ; X and Y have the same rational lower bounds.

252 • Studd

Which, then, of the inf- and sup-reals provide the faithful reduction? The ALL
option is unavailable to a proponent of fg, cg, or mg. According to these views
— specifically ncc — a sup-abstract identified with a real is never identical to
an inf-abstract (since the sup- and inf-abstracts in question are associated with
different equivalence classes).51 Since the two interpretations agree on questions
posed in the language of real analysis, the SOME option is no more appealing
than in Benacerraf’s original trilemma. This leaves the NONE option, which is
to deny that real numbers are sup- or inf-abstracts.52

Downloaded from https://academic.oup.com/philmat/article/31/2/236/7116333 by guest on 06 September 2024

2.6.
Another objection is that the wholesale proposals unduly constrain mathemat-
ical freedom. Suppose a community seek to introduce two different notations
for ordered pairs, x≺y and x y, pronounced ‘x-before-y’ and ‘x-after-y’, and a
projection predicate π(z, i, p) — ‘z is the ith coordinate of p’ — subject to the
following stipulations:53

≺-pair For any x, y, z, and w, the following are equivalent:

x≺y = z≺w; x = z and y = w.

-pair For any x, y, z, and w, the following are equivalent:

x y = z w; x = z and y = w.

≺-proj For any x, y, z, and any i ∈ {1, 2}, the following are equivalent:

51
According to cg, sup Λ = inf Λ, but this ‘dummy’ abstract is not identified with a
real.
52
Might a wholesale abstractionist reply by rejecting cp and cp , and instead use Hale’s
d, or another postulate to obtain the reals? The difficulty with this reply is that the objec-
tion from multiple reduction is apt to generalize to other abstraction postulates. Indeed,
in many cases, the resulting trilemma prompts Caesar questions which the wholesale pro-
posals leave unanswered. This is because fg, cg, and mg only apply to contexts of the form
σx = ρy in a special case, namely when the types of x and y match, in the sense that x
and y are either variables of the same type or sequences of variables of the same length
whose respective members match. As a result, fg, cg, and mg are silent on ‘unmatched’
Caesar questions, such as ‘diffp (u) = diff(x, y)?’ or ‘#X = #̂Y ?’. In the first case, given
a pairing operation that maps positive reals x and y to a single individual x, y, diffp
is a monadic operator, subject to a trivial variant of d which identifies the diffp -reals
specified by x, y and z, w iff x + w = y + z. In the second case, Y is a third-order
variable, and #̂-cardinals are introduced by a higher-level analogue of hp that identifies
the #̂-cardinals of classes (values of third-order variables) iff the classes are equinumerous.
(Cook [2009] dubs this postulate Upper Hume.) The only half-way principled way I can
think of to extend the wholesale proposals to these cases is by adding an ‘extremal clause’
that gives a negative answer to all unmatched Caesar questions. Once again, this rules out
the ALL option in these cases: a real number cannot be both a diffp -real and a diff-real
and a natural number cannot be both a finite #-cardinal and a finite #̂-cardinal.
53
The pronunciation guide for ≺, , and π is intended to be suggestive but should not
be taken to constrain the interpretation that abstraction may accord these expressions.
 The Caesar Problem • 253

π(z, i, x≺y); i = 1 and z = x OR i = 2 and z = y.

-proj For any x, y, z, and any i ∈ {1, 2}, the following are equivalent:

π(z, i, x y); i = 1 and z = y OR i = 2 and z = x.

By the lights of wholesale abstractionism, if all goes well, ≺-pair and -pair
effectively function as unmixed postulates that introduce ≺-pairs and -pairs,

Downloaded from https://academic.oup.com/philmat/article/31/2/236/7116333 by guest on 06 September 2024

and ≺-proj and -proj define the projection predicate π.
The trouble is that, according to fg, cg, and mg, at least one of these
stipulations must fail. This time the objection relies only on the no-finer-than-
fg assumption. Assuming nff, 4≺5 and 4 5 are identical (since they are both
abstracted from the same specifications — x has value 4 and y value 5 —
according to the same unity relation). By ≺-proj and -proj, however, 4≺5
and 4 5 are distinct (since they differ in their first coordinate).54
Of course, once the community realize their mistake, it is open to them
to sanitize their practice to bring it in line with nff. But I am reluctant to
think that mathematics is thus metaphysically constrained. By mathematical
standards, the community’s stipulations — as they stand — provide a perfectly
coherent set of axioms. Unless nff is rejected, the abstractionist has no choice
but to condemn the community’s practice as flawed.55
To take stock briefly, a broad range of wholesale proposals are subject to
three objections: fg, cg, and mg face the objection from intuition, from mul-
tiple reduction, and from mathematical freedom. Even if no importance is
attached to the first objection, the other two provide plenty of motivation to
consider alternatives.

3. A PIECEMEAL SOLUTION
The basic idea behind the piecemeal response to the Caesar problem is straight-
forward: Caesar questions left unanswered by hp or other abstraction principles
may be decided via further stipulations.

54
More fully, π(4, 1, 4≺5) holds and π(4, 1, 4 5) fails because when x, y, z, and i
respectively take the values 4, 5, 4, and 1, the first disjunct of ≺-proj holds and both
disjuncts of -proj fail.
55
Might a wholesale abstractionist reply by restricting nff? Perhaps abstraction
attempts based solely on unmixed postulates determine the content of mixed contexts,
in accordance with nff, but nff may be overridden by additional stipulations. I find this
view at least half-way congenial. This is because it goes a good way towards the piece-
meal view I wish to defend. If the wholesale mechanism which ensures that ‘4≺5 = 4 5’
expresses a truth can be bypassed in this way, then the content of mixed contexts may be
settled by additional stipulations alongside unmixed postulates, just as the piecemeal view
maintains. Indeed, what bar remains to overriding nff by directly stipulating the content
of mixed contexts? My main reservation is that this leaves the wholesale component of
the view poorly motivated: if Caesar questions may be decided via stipulation, what need
has the abstractionist for a category-based ontology or other wholesale means to answer
them.
254 • Studd

But how is this response to be developed? Linnebo is one of few abstraction-

ists to take a piecemeal response seriously:

When our ancestors first confronted Caesar-style questions, they had a

choice which way to go; and this choice played a role in shaping the
concepts that they thereby forged. [2018, p. 160]

In his view, although speakers ‘tend to operate’ according to fg (p. 168), ‘excep-

Downloaded from https://academic.oup.com/philmat/article/31/2/236/7116333 by guest on 06 September 2024

tions are certainly possible and very likely even actual’ (p. 160). For example,
Linnebo maintains that it is ‘consistent’ to identify #- and n-cardinals, in line
with Cook and Ebert’s intuition (p. 166).
Linnebo, however, does not present a general account of what can and cannot
be successfully stipulated in a piecemeal fashion.56 This leaves some important
questions unanswered. How much choice did our ancestors have? Could they
also have chosen to identify #-cardinals with †-sets? Could they have chosen to
identify numbers with Romans? What about MacBride’s worry about conflict
with ‘antecedent facts’ or the Hale–Wright worry about incoherent stipulations?
What is wanted, if the piecemeal response is to emerge as a serious candidate
solution, is a systematic account of how, in the context of an abstractionist
metasemantics, Caesar questions may be decided via stipulation. My aim in
this section is to provide such an account. I argue that the piecemeal view that
emerges has a robust response to the concerns from MacBride and Hale and
Wright and that this view improves on its wholesale competitors when it comes
to the objections from intuition, multiple reduction, and mathematical freedom.
To begin with, let us continue to focus on Caesar questions concerning dif-
ferent kinds of abstract.57 Pace Hale and Wright, there is no metaphysical or
metasemantic law that permits us to extract the content of mixed contexts from
postulates governing unmixed ones. Instead, according to piecemeal abstrac-
tionism, a community is just as free to stipulate the content of mixed contexts
as they are to stipulate the content of unmixed contexts. The three-step account
of abstraction outlined in 1.1 is unchanged save that, in addition to unmixed
postulates, an abstraction attempt may also be based on mixed postulates of
the following form:

σ:ρ For any x ∈ Dσ and y ∈ Dρ , the following are equivalent:

σx = ρy; x ∼σ:ρ y.

The postulate stipulates the content of a mixed context in terms of an

antecedently expressible unity relation ∼σ:ρ . As before, semantic values for the

56
One general constraint can be gleaned from Linnebo’s account [2018, p. 167]: his
discussion of the axiom he labels G makes clear that he endorses nff (the converse of G).
This is a noteworthy difference from the piecemeal account I wish to recommend, which
permits stipulations that run contrary to nff and ncc.
57
I return to Caesar himself in 3.3.
 The Caesar Problem • 255

expanded lexicon are then selected subject to the constraint that they compo-
sitionally recover the sentential content stipulated by the attempt’s postulates.

3.1.
Does the piecemeal view really improve on its wholesale competitors? Let me
first return to the intuitions reported in Section 2. Assuming Caesar questions
may be settled via stipulation, it is straightforward to give answers in line

Downloaded from https://academic.oup.com/philmat/article/31/2/236/7116333 by guest on 06 September 2024

with the intuitions (i1)–(i3) while maintaining the reduction theses (r1)–(r3).
Suppose a community — Community 1 — successfully lay down the following
alongside the corresponding unmixed postulates:58

n:#-1 For any classes X and Y , the following are equivalent:

#X = nY ; X and Y are equinumerous and both finite.

#:†-1 For any classes X and Y , the following are equivalent:

#X = †Y ; ⊥.

#:sup-1 For any class X and any Y ⊆ Q, the following are equivalent:

#X = sup Y ; Y has the same rational upper bounds as a class

of integral rationals {0Q , . . . , nQ }, nQ ≥ 0Q , and X is
equinumerous with {0Q , . . . , nQ }\{0Q }.

According to n:#-1, each finite #-cardinal n# is identical to its n-counterpart

nn , as per (i1). The postulates #:†-1 and #:sup-1 ensure that #-naturals are
distinct from †-sets but identical to the corresponding sup-reals. Given the
desired identification with their marketplace mathematical counterparts, as per
(r1)–(r3), it follows that 0N = ∅ and 0N = 0R , as per (i2) and (i3).
I concede (once again) that the intuitions here are open to question. But the
key point is not that piecemeal abstractionism can accommodate my intuitions
(or Cook and Ebert’s). Instead this approach is flexible enough to mirror the
identity profile of natural and real numbers, whatever this may be. Suppose, for
instance, that there is no overlap between natural and real numbers, contrary
to (i3). Then a second community — Community 2 — may seek to introduce
disjoint #- and sup-abstracts, by supplementing hp and cp with a different
#:sup-postulate:

#:sup-2 For any class X and any Y ⊆ Q, the following are equivalent:

#X = sup Y ; ⊥.

58
As usual, ⊥ stands for a trivial contradiction.
256 • Studd

More generally, whatever the natural–real identity facts may be, it is open to a
piecemeal abstractionist to contend that the abstract realm is rich and varied
enough to contain abstracts with the same identity profile.

3.2.
All this free and easy stipulation brings me back to a MacBride-style objection:

Downloaded from https://academic.oup.com/philmat/article/31/2/236/7116333 by guest on 06 September 2024

Suppose that zero leads a double life. Suppose that in addition to leading
its arithmetical existence 0N is also the real number 0R . In that case
the stipulation that a sentence that says that 0N is identical to 0R is
false cannot succeed. For this sentence will be true and a true sentence
cannot be stipulated to be false. Stipulation cannot suffice as a basis for
determining that 0N is distinct from 0R .

Needless to say, stipulation cannot render identical things distinct. What is

subject to stipulation is the meaning of expressions. It stretches the meaning
of ‘stipulate’ to speak of a sentence’s truth value being stipulated. However,
MacBride’s locution is harmless enough provided it is understood in an explic-
itly extended sense: a sentence is stipulated true if, as a matter of stipulation,
the sentence says that p and, as a matter of fact, it is the case that p; similarly
for ‘stipulated false’.59
Would, then, zero’s leading a double life undermine Community 2’s ability
to stipulate ‘0# = 0sup ’ false, in this extended sense? The objection leaves the
ambient interpretation unspecified. Clearly, if it is to threaten Community 2’s
stipulation, the occurrences of ‘false’ in the objection must mean ‘false under
i2 ’ where i2 is the interpretation resulting from Community 2’s abstraction
attempt. What about ‘true’ and ‘says’ ? To avoid an obvious non-sequitur, their
occurrences must be understood relative to the same interpretation — but is
this i2 or another interpretation? Let me consider the two options in turn.
Option 1 : ‘True’ and ‘says’ mean truth-under-i2 and says-under-i2 . The key
passage then reads as follows:

. . . the stipulation that a sentence that says under i2 that 0N is identical to

0R is false under i2 cannot succeed. For this sentence will be true under i2
and a true-under-i2 sentence cannot be stipulated to be false under i2 . . .

In response, a piecemeal abstractionist should simply deny that their abstrac-

tion attempt renders ‘0# = 0sup ’ true under i2 . After all, #:sup-2 stipulates
that this sentence coincides in content with a trivial falsehood.
Option 2 : ‘True’ and ‘says’ mean truth-under-j and says-under-j, for some
interpretation j other than i2 :

59
Analogous remarks apply to talk of ‘deciding’ Caesar questions via stipulation.
 The Caesar Problem • 257

. . . the stipulation that a sentence that says under j that 0N is identical

to 0R is false under i2 cannot succeed. For this sentence will be true under
j and a true-under-j sentence cannot be stipulated to be false under i2 . . .

In this case, the sentence in question — ‘0# = 0sup ’ — may well be true under
j. But there is no reason to think that a sentence that is true in one lan-
guage cannot be stipulated false in another. Speakers of the two languages may
respectively render the sentence true under j and false under i2 by interpreting

Downloaded from https://academic.oup.com/philmat/article/31/2/236/7116333 by guest on 06 September 2024

its abstract terms differently.
Either way, then, the identity of 0N and 0R is no obstacle to the success of
Community 2’s abstraction attempt. Nonetheless, there is an important moral
to be extracted from MacBride’s objection. Consider again Option 1: if 0N and
0R are identical, then to deny that ‘0# = 0sup ’ is true under i2 , the piecemeal
abstractionist must also deny that this sentence says that 0N = 0R under i2 .
Indeed, on the piecemeal metasemantics, if Community 2’s abstraction attempt
succeeds, semantic values are selected so that the compositionally determined
content of ‘0# = 0sup ’ is the content of a trivial falsehood (as per #:sup-2).
Assuming that 0N = 0R , this is achieved only if the relevant #- and sup-terms
do not respectively refer to 0N and 0R .60 Consequently, on the proposed metase-
mantics, if zero leads a double life, Community 2’s stipulations ensure that they
do not achieve a faithful reduction of both arithmetic and real analysis.61
Similarly, the non-identity of 0N and 0R would be no bar to the success of
Community 1’s stipulation. But in this case they cannot hope both to stipulate
‘0# = 0sup ’ true and to hit upon the familiar naturals and reals as the referents
of the corresponding #- and sup-terms. The moral here is that it is reduction
rather than success that is hostage to ‘antecedent facts’. Assuming all goes
well, both communities succeed in introducing natural- and real-like abstracts,
which underwrite interpretations of arithmetic and real analysis. But it does
not follow that they introduce the same abstracts. The different abstraction
attempts accord different meanings to their abstract terms, so that at most
one of the communities achieves a faithful reduction in both cases.

3.3.
The discussion in the last section raises the spectre of multiple reductions once
more. But before I come to that objection, it is helpful to return to Caesar
himself. The piecemeal account is readily extended to other atomic contexts,
such as σx = q or R(σx). In addition to the unmixed and mixed postulates
introduced so far, a piecemeal abstraction attempt may also include mixed
postulates of the following forms:

σ:q For any x ∈ Dσ and q ∈ Dq , the following are equivalent:

60
This is because, assuming the standard semantics for identity, ‘0# = 0sup ’ is true if its
terms both refer to the same number. Similarly, ‘0# = 0sup ’ is true on the supervaluationist
semantics if ‘0# ’ and ‘0sup ’ refer to 0N (= 0R ) under each admissible interpretation.
61
Community 2, however, may yet go on to achieve one faithful reduction. See fn. 65.
258 • Studd

σx = q; x ∼σ:q q.

R For any x ∈ Dσ , the following are equivalent:

R(σx); IσR (x).

In the first case, q is a first-order variable whose associated domain Dq com-

Downloaded from https://academic.oup.com/philmat/article/31/2/236/7116333 by guest on 06 September 2024

prises some or all of the items quantified over pre-abstraction.62 In the second
case, IσR (x) expresses what I will call an instantiation relation. The postu-
lates stipulate the content of a mixed identity or atomic context in terms of an
antecedently expressible unity or instantiation relation. Their relata are either
specifications of the relevant abstracts, as usual, or — in the case of members
of Dq — the very items in the initial domain whose identity relations are in
question. In the case of a polyadic atomic context, a polyadic postulate sup-
plants the monadic one stated above.63 As ever, semantic values are selected
for the expanded lexicon that compositionally recover the stipulated sentential
content.
The basic piecemeal response to the Caesar problem is that Caesar questions
are answered via stipulations of this kind. Community 1 may ensure that their
#-terms do not refer to Romans by supplementing their earlier stipulations
with postulates such as the following:

#:q-1 For any class X and Roman q, the following are equivalent:

#X = q; ⊥.

Roman# -1 For any class X, the following are equivalent:

#X is Roman; ⊥.

The #-cardinals introduced by piecemeal abstraction may consequently have

a rich nature extending beyond arithmetical properties. Community 1 ensure

62
Officially, I assume that each first-order variable q is associated with a domain Dq
that is specified by a condition δq (x) and comprises members of the initial domain that
satisfy this condition under the initial interpretation. For technical convenience, I assume
that at least one Dq is the whole initial domain.
63
The polyadic postulate takes the following form:
R For any x1 ∈ Dt1 ,. . . , xn ∈ Dtn , the following are equivalent:
R
R(t1 , . . . , tn ); It1 ,...,tn (x1 , . . . , xn ).

In the axiom, each term ti takes one of two forms (i = 1, . . . , n): either (i) ti is an abstract
term σi (xi ) and Dti is the domain associated with σi ; or (ii) ti and xi are the same first-
order variable and Dti is the domain associated with this variable. Allowing for R to be
the identity predicate, all the abstraction postulates considered in the text take this form
(modulo infix notation).
 The Caesar Problem • 259

that the referents of their #-terms are distinct from Caesar and non-Roman.
Further postulates may constrain them to be non-concrete, and so on.
Of course, to say a community may give stipulative answers to Caesar ques-
tions is not to say that they must. A piecemeal abstraction attempt, even if it
answers a wide range of Caesar questions, may still say nothing about others.
In these cases, it is reasonable to expect languages interpreted by piecemeal
abstraction to manifest some indeterminacy. Different abstractionists may pre-

Downloaded from https://academic.oup.com/philmat/article/31/2/236/7116333 by guest on 06 September 2024

fer different accounts of indeterminacy. But, to fix ideas, let me continue to
operate with the supervaluationist approach outlined in 1.2.
On this approach, following a successful piecemeal abstraction attempt, the
admissible MT-interpretations are those that extend the initial one according
to the attempt’s postulates. In the limiting case, a community who introduce
# with no postulates other than hp leave the reference of #-terms radically
indeterminate, as before. In my view, however, if in fact many interesting
Caesar questions are answered, it is because our use of number terms goes
far beyond a commitment to hp. Community 1 take a step in this direction
with postulates that render ‘#Λ = Caesar’ false. Further postulates, providing
an idealized reconstruction of more of our number-term practice, may further
reduce indeterminacy, and decide further Caesar questions.
Of course, our use of number terms is just one way to go. On the piecemeal
view I wish to recommend, there is nothing to stop Community 2 identi-
fying some of their #-abstracts with Romans with a postulate such as the
following:

#:q-2 For any class X and Roman q, the following are equivalent:

#X = q; q is a dictator of the Roman Republic such that the

class of dictators succeeding q is equinumerous with X.

It bears repeating, contrary to MacBride’s objection, that the success of this

stipulation in not hostage to whether the familiar numbers — the referents of
our number terms — are Roman. But a different concern calls for brief com-
ment. Assuming Community 2’s attempt succeeds, can the abstracts introduced
really include flesh-and-blood Romans?
Part of this concern is easily dealt with. The term ‘abstract’ is notoriously
polysemous. But my present use is in no way opposed to concrete or spa-
tiotemporal. I have been using ‘#-abstract’ to apply to the putative referents
of #-terms that result from a (salient) abstraction attempt. I take it that there
is nothing to stop a linguistic community using #-terms, or any other syntacti-
cally singular terms, to refer to Romans. The only remaining question is whether
this can be achieved by the top-down metasemantics posited by abstractionism.
But why should it not? In fact, the mooted cases of abstract–Roman overlap
are unusually favourable ones for abstraction to confer reference on #-terms.
In these cases, there is no doubt that there is a unique dictator in the initial
domain who stands in the postulate’s unity relation to the class assigned to
X. Consequently #:q-2, in effect, stipulates that the reference of #X for this
260 • Studd

assignment is to be selected so as to render ‘#X = q’ true of the specified

Roman. But this is little different from the clearly unproblematic stipulation
that the referent of #X for this assignment is to be the specified Roman. Even
if the ability of abstraction to secure reference is deemed questionable in other
cases, the means by which #:q-2 ensures that #-terms refer to Romans is akin
to comparatively straightforward kinds of reference fixing.

Downloaded from https://academic.oup.com/philmat/article/31/2/236/7116333 by guest on 06 September 2024

3.4.
Turn now to the objection from multiple reductions. On the piecemeal view, it
is straightforward to identify sup- and inf-reals in the natural way. Community
1, for instance, may supplement cp and cp with the following:

sup:inf-1 For any X, Y ⊆ Q, the following are equivalent:

sup X = inf Y ; {x − y : x ∈ X, y ∈ Y } has least upper bound 0Q .

On its own, however, this observation is of limited help since another commu-
nity is equally free to distinguish their analogues of these abstracts. Witness
Community 2’s addition to cp and cp :

sup:inf-2 For any X, Y ⊆ Q, the following are equivalent:

sup X = inf Y ; ⊥.

The multiplicity of would-be reductions leads back to Benacerraf’s trilemma.

To avoid confusion, I will use subscripts to distinguish Community 1’s over-
lapping sup1 - and inf 1 -reals and Community 2’s disjoint sup2 - and inf 2 -reals.
Which, then, of the sup1 -/inf 1 -reals, the sup2 -reals, and the inf 2 -reals are the
familiar reals? The ALL horn leads swiftly to absurdity. In the absence of a
reason to privilege one of the interpretations, the SOME horn is as unpalatable
as ever. This leaves the NONE horn, which amounts to denying that either
community achieves a faithful reduction.
Without further addition, this is indeed the case. The postulates listed so far
say nothing about real–sup-real identity. On the supervaluationist semantics, a
successful abstraction attempt based on Community 1’s postulates accords no
determinate truth value to reduction theses such as ‘0sup = 0R ’.64 This indeter-
minacy, however, may be relieved with further stipulations. Once Community
1 is equipped with quantifiers ranging over R — the familiar reals — they may
constrain the reference of their sup-terms so as to ensure they obtain a faithful
reduction with a suitable σ:q-postulate:

sup:y-1 For any X ⊆ Q and y ∈ R, the following are equivalent:

64
I assume here that ‘0R ’ is part of the lexicon prior to the cp-abstraction attempt.
 The Caesar Problem • 261

sup X = y; X has the same rational upper bounds

as {r ∈ Q : r < y}.

If successful, an attempt that includes this postulate, in addition to cp, cp ,

and sup:inf-1, ensures that each sup- and inf-term refers to the corresponding
real, so that, in particular, ‘0sup = 0R ’ comes out true. This gives the piecemeal
abstractionist the means to blunt the SOME horn of the trilemma. Benacerraf’s
sought-for ‘cogent reason’ to think that Community 1’s sup-terms refer to real

Downloaded from https://academic.oup.com/philmat/article/31/2/236/7116333 by guest on 06 September 2024

numbers is that this follows from their stipulation on the proposed piecemeal
metasemantics.65
It should not go unnoticed that a stipulation such as sup:y-1 is only avail-
able to a community equipped with quantification over the familiar reals.
But I want to insist that this is unproblematic. If we come to quantify over
real numbers by some means other than abstraction, sup:y-1 is analogous to
#:q-2. The latter postulate quantifies over Romans to settle that some #-terms
refer to Romans, in a way that has already been argued to be unproblematic.
The more interesting case arises if abstraction provides our most fundamen-
tal way to acquire quantification over real numbers. But, in this case, there
is nothing metaphysically special about the ‘familiar’ reals compared with the
many other fields of real-like objects that may be introduced by abstraction.
The familiar reals are simply the real-like abstracts that are familiar to us: the
ones we happened to introduce and to associate with our terms, such as ‘real
number’ and ‘0R ’. If these terms were ultimately interpreted via abstraction,
then a faithful reduction requires only that sup-terms should pick out the right
abstracts among those we have already introduced. And once we have come
to quantify over these abstracts — whether they are introduced via cp , d, or
any other abstraction attempt — we can deploy a postulate such as sup:y-1 to
constrain sup-terms to refer to the very same abstracts. The same goes, mutatis
mutandis, for natural numbers, sets, and so on.

3.5.
Two loose ends remain: the objection from mathematical freedom and the
Hale–Wright worry about incoherent stipulations. On the wholesale view, the
community who attempt to introduce ≺-pairs and -pairs fail in some of their
stipulations, which conflict with the no-finer-than-fg assumption, nff. The
basic piecemeal response to the objection from mathematical freedom is sim-
ply to reject nff. On the piecemeal view, laying down notational variants of
the same postulate to govern unmixed identity contexts does not prevent the

65
The situation is not so different for Community 2. But they face a choice. They
may stipulate a verbatim copy of sup:y-1 to constrain their sup-terms to refer to the
corresponding reals. Alternatively, they may lay down the dual of this postulate to ensure
the same for their inf-terms. Given sup:inf-2, either stipulation constrains the dual terms
to refer to real-like abstracts distinct from the familiar reals. Similarly, notwithstanding
#:sup-2, Community 2 may lay down postulates to obtain a faithful reduction of either
arithmetic or real analysis even if 0N = 0R .
262 • Studd

community from coherently distinguishing pairs such as 4≺5 and 4 5 via their
stipulations governing other contexts.
This response, however, may provoke a Hale–Wright-type worry about inco-
herent stipulations. Even if the piecemeal account carries no commitment to
nff, why think it avoids other equally objectionable constraints on success-
ful abstraction? After all, it is not hard to multiply examples of incoherent
abstraction attempts. Suppose, for instance, that a community supplement hp

Downloaded from https://academic.oup.com/philmat/article/31/2/236/7116333 by guest on 06 September 2024

and fhp with the following postulate:

n:#-! For any X and Y , the following are equivalent:

nX = #Y ; X and Y are equinumerous.

This postulate conflicts with the transitivity of identity: n(R) = n(N) (by fhp)
and n(N) = #(N) (by n:#-!) but n(R) = #(N) (by n:#-!).
The risk of incoherence is not a new problem for abstractionism, of course.
In the case of an abstraction attempt based on a single unmixed postulate,
there is a straightforward necessary condition for its success: its unity relation
must be an equivalence relation. There is an analogous necessary condition for
a piecemeal abstraction attempt to succeed in extending the initial interpreta-
tion according to its postulates.66 In this case, the attempt’s postulates provide
a patchwork of sort-specific unity relations. These in turn induce one or more
global unity relations. A global unity relation ∼ serves to identify and distin-
guish abstracts and individuals, of any sort, in accordance with the attempt’s
unity relations and the identity relations that hold between individuals in the
initial domain. Analogously, for each predicate R, the patchwork of sort-specific
instantiation relations induces one or more global instantiation relations I R .67

66
The postulates take the form indicated in fn. 63. Formally, an MT-interpretation J
may then be said to extend another I according to some stipulated postulates if it meets
the conditions stated in fn. 23 (with the second suitably generalized): (i) J is an extension
of I, and (ii) for each stipulated postulate, when x1 , . . . , xn are respectively assigned to
members of Dt1 , . . . , Dtn , the terms t1 , . . . , tn are defined under J , and the truth value
of the postulate’s left-hand side under J coincides with the truth value of its right-hand
side under I.
67
Officially, given an initial MT-interpretation I = M, I, the relations ∼ and I R
are relations on the set — henceforth, M — comprising pairs of the form t, a, where
t is either a first-order variable q or an abstract term σx drawn from the expanded lex-
icon and a belongs to the corresponding domain, Dq or Dσ . Given an n-ary predicate
R in the expanded lexicon, a relation I R is a global instantiation relation induced by
the attempt if it meets two conditions: (i) for each postulate deployed in the attempt
(of the form described in fn. 63), I R holds of a sequence of pairs drawn from M ,
t1 , a1 , . . . , tn , an  iff the postulate’s right-hand side holds of a1 ,. . . ,an under I; (ii) for
terms t1 , . . . , tn that are already defined under I (when their free variables are respectively
assigned to a1 , . . . , an ), I R holds of a sequence of members of M , t1 , a1 , . . . , tn , an 
iff R(t1 , . . . , tn ) holds of a1 , . . . , an under I. In the special case when R is the identity
predicate, a relation I R meeting (i) and (ii) is a global unity relation induced by the
attempt.
 The Caesar Problem • 263

A necessary condition for the attempt’s success is then given by what I will
call the congruence condition:68 the attempt must induce a family of relations
— a global unity relation ∼, together with a global instantiation relation I R
for each predicate in the expanded lexicon — that is congruent, in the sense
that ∼ is an equivalence relation that is respected by each I R .69 The attempt
to lay down n:#-! alongside hp and fhp fails because it violates this condition:
any global unity relation it induces fails to be transitive.

Downloaded from https://academic.oup.com/philmat/article/31/2/236/7116333 by guest on 06 September 2024

Is there an equally straightforward sufficient condition for success? Neologi-
cist abstractionists long ago gave up hope that anything like the congruence
condition might give this success condition. For this condition is met by
attempts that include the postulate widely blamed for Russell’s paradox:

blv For any X and Y, the following are equivalent:

†X = †Y ; X and Y are coextensive.

In response to what is known as the bad company problem posed by blv and
other problematic abstraction principles, abstractionists have typically sought
to defend a more demanding condition for an abstraction attempt to succeed.70
There is nothing in principle to stop a piecemeal abstractionist adopting the
same strategy. But a danger of this approach, for either kind of abstractionist,
is that an over-demanding success condition may undermine other aspects of
their programme. In the piecemeal case, my responses to the objections from
intuition, multiple reduction, and mathematical freedom ultimately turn on the
success of the communities’ stipulations. They are free to introduce abstracts
whose identity profiles accord with the reported intuitions, or which sustain
the desired reductions, or which conform to the stipulated axioms, only if their
attempts succeed.
The bad company problem, however, is another place where the abstraction-
ist may do better to depart from neologicist orthodoxy. Even in the ur-bad case,
there is no obstacle to extending an interpretation in accordance with blv unless
we assume — in line with the standard, impredicative, treatment of abstraction
— that any abstracts introduced must fall within the domain of the pre-
abstraction interpretation.71 An alternative is to defend a predicative account of
abstraction. On this view, the items introduced by abstraction are not assumed

68
Proof sketch: Suppose that there is an MT-interpretation J that extends I according
to the attempt’s postulates. Then the congruence condition is witnessed by a family of rela-
tions read off J : for pairs in M (see fn. 67), I R is defined to hold of t1 , a1 , . . . , tn , an 
iff R(t1 , . . . , tn ) holds of a1 , . . . , an under J ; ∼ is defined analogously.
69
Recall that ∼ is respected by I R if whenever I R holds of a sequence a1 , . . . , an it
also holds of any sequence b1 , . . . , bn with bi ∼ ai , for i = 1, . . . , n.
70
See, for instance, [Wright, 1997; Cook, 2012].
71
Wright [1998] defends this kind of impredicativity by appealing to the (pre-
abstraction) availability of quantification over absolutely everything (including every
abstract).
264 • Studd

to belong to the initial domain, so that abstraction may (iteratedly) intro-

duce ‘new’ items. This kind of dynamic abstraction rehabilitates abstraction
based on blv.72 More generally, for any piecemeal abstraction attempt that
meets the congruence condition, assuming that the pre-abstraction interpre-
tation is encoded as an MT-interpretation, there is also an MT-interpretation
that extends the initial one according to the attempt’s postulates.73 This opens
the way for a piecemeal abstractionist to adopt the maximally liberal view,

Downloaded from https://academic.oup.com/philmat/article/31/2/236/7116333 by guest on 06 September 2024

according to which the congruence condition is sufficient for an attempt to
succeed.
If this is right, the orthodox neologicist version of abstractionism is doubly
over-restrictive. An alternative version of abstractionism, dynamic and piece-
meal, takes a more liberal view, both on the range of unity relations that may
be successfully abstracted upon and on the variety of contexts whose content is
open to direct stipulation. When it comes to the Caesar problem, I have argued
that this provides a viable candidate solution, which has a robust response to
the MacBride and Hale–Wright worries, and which improves on its wholesale
rivals when it comes to the objections from intuition, multiple reduction, and
mathematical freedom.

4. ACKNOWLEDGMENTS
For helpful comments and discussion, thanks to Roy Cook, Richard Kimberly
Heck, Nick Jones, Graham Leach-Krouse, Øystein Linnebo, Alex Paseau, Alex
Roberts, Zeynep Soysal, and the anonymous referees.

72
This response to the bad company problem is defended by Studd [2016] and Linnebo
[2018, ch. 3], who distinguish the domains associated with the new and old interpretations
by working in a many-sorted or modal setting. The requirement imposed by Studd and
Linnebo that the unity relation be ‘stable’ follows from the congruence condition. Many-
sorted approaches in a similar spirit are also discussed by Heck [1997a]; but, in the case
of arithmetic, Heck ultimately favours a two-sorted system that includes an impredicative
formulation of hp.
73
Proof sketch: Suppose that the attempt is based on an initial MT interpretation,
I = M, I, that induces a congruent family of unity/instantiation relations ∼/I R on
the set M defined in fn. 67. When a is a suitably typed entity or sequence of entities
based on M , and t is either a first-order variable x or an abstract term σx, write tI (a)
for the reference of this term, if defined, under I, when a is assigned as the value of
x or x (i.e., tI (a) = a or (I(σ))(a)). It is then a routine exercise in model theory to
verify that the following interpretation J = N, J is a well-defined MT-interpretation
that extends I according to the attempt’s postulates. The interpretation function J is
defined as follows for each constant c, operator σ, and predicate R in the expanded lexicon
(picking ∗ outside the transitive closure of I): (i) J(c) = I(c); (ii) (J(σ))(a) is either
(I(σ))(a) if this is defined under I or tI (b) if this is defined for t, b ∈ M with t, b ∼
σx, a; otherwise, if σx, a ∈ M , (J(σ))(a) = ∗, {t, b ∈ M : t, b ∼ σx, a}
and, in all other cases, (J(σ))(a) is undefined; (iii) J(R) holds of tJ J
1 (a1 ), . . . , tn (an )
R
iff I (t1 , a1 , . . . , tn , an ) and t1 , a1 , . . . , tn , an  ∈ M (and holds of no other n-
tuples). In the last clause, tJ (a) is defined as before, with J(σ) replacing I(σ). Finally,
the domain of J is defined as follows: N =df M ∪ {tJ (a) : t, a ∈ M }.
 The Caesar Problem • 265

REFERENCES
Benacerraf, Paul [1965]: ‘What numbers could not be’, The Philosophical Review 74,
47–73. DOI: https://doi.org/10.2307/2183530.
[1981]: ‘Frege: The last logicist’, Midwest Studies in Philosophy 6, 17–36.
DOI: https://doi.org/10.1111/j.1475-4975.1981.tb00426.x.
Blanchette, Patricia A. [2012]: Frege’s Conception of Logic. Oxford University Press.
Boccuni, Francesca, and Jack Woods [2020]: ‘Structuralist neologicism’, Philosophia
Mathematica (3) 28, 296–316. DOI: https://doi.org/10.1093/philmat/nky017.

Downloaded from https://academic.oup.com/philmat/article/31/2/236/7116333 by guest on 06 September 2024

Boolos, George [1998]: Logic, Logic, and Logic. Cambridge, Mass.: Harvard Univesity
Press.
Cook, Roy T. [2009]: ‘Hume’s big brother: Counting concepts and the bad company
objection’, Synthese 170, 349–369.
DOI: https://doi.org/10.1007/s11229-007-9264-8.
[2012]: ‘Conservativeness, stability, and abstraction’, The British Journal
for the Philosophy of Science 63, 673–696.
DOI: https://doi.org/10.1093/bjps/axr047.
Cook, Roy T., and Philip A. Ebert [2005]: ‘Abstraction and identity’, Dialectica 59,
121–139. DOI: https://doi.org/10.1111/j.1746-8361.2005.1023.x.
Dummett, Michael [1978]: ‘Frege’s philosophy’, in Michael Dummett, Truth and Other
Enigmas, pp. 87–115. London: Duckworth.
Ebels-Duggan, Sean C. [2021]: ‘Identifying finite cardinal abstracts’, Philosophical
Studies 178, 1603–1630. DOI: https://doi.org/10.1007/s11098-020-01503-1.
Ebert, Philip A., and Marcus Rossberg [2016]: Abstractionism: Essays in Philosophy
of Mathematics. Oxford University Press.
Fine, Kit [2002]: The Limits of Abstraction. Oxford University Press.
Frege, Gottlob [1980]: The Foundations of Arithmetic. J.L. Austin, trans. 2nd rev.
ed. Evanston, Ill.: Northwestern University Press.
[2013]: Basic Laws of Arithmetic. Philip A. Ebert and Marcus Rossberg,
trans. Oxford University Press.
Hale, Bob [1987]: Abstract Objects. Oxford: Blackwell.
[1997]: ‘Grundlagen §64’, Proceedings of the Aristotelian Society 97, 243–
261. Reprinted in [Hale and Wright, 2001a], pp. 91–116. Page references to reprint.
DOI: https://doi.org/10.1111/1467-9264.00015.
[2000]: ‘Reals by abstraction’, Philosophia Mathematica (3) 8, 100–123.
DOI: https://doi.org/10.1093/philmat/8.2.100.
Hale, Bob, and Crispin Wright [2001a]: The Reason’s Proper Study: Essays towards
a Neo-Fregean Philosophy of Mathematics. Oxford University Press.
[2001b]: ‘To bury Caesar’, in [Hale and Wright, 2001a], pp. 335–396.
Heck, Richard Kimberly [1997a]: ‘The Julius Caesar objection’, in [Heck, 1997b],
pp. 273–308. Originally published under the name Richard G. Heck, Jr.
, ed. [1997b]: Language, Thought, and Logic: Essays in Honour of Michael
Dummett. Oxford University Press. Originally published under the name Richard
G. Heck, Jr.
[1997c]: ‘Finitude and Hume’s principle’, Journal of Philosophical Logic
26, 589–617. Originally published under the name Richard G. Heck, Jr. DOI:
https://doi.org/10.1023/A:1004299720847.
[2005]: ‘Julius Caesar and Basic Law V’, Dialectica 59, 161–178. Originally
published under the name Richard G. Heck, Jr.
DOI: https://doi.org/10.1111/j.1746-8361.2005.01025.x.
266 • Studd

[2011a]: ‘The existence (and non-existence) of abstract objects’, in [Heck,

2011c], pp. 200–226. Originally published under the name Richard G. Heck, Jr.
[2011b]: ‘Frege’s Theorem: An overview’, in [Heck, 2011c], pp. 1–39.
Originally published under the name Richard G. Heck, Jr.
[2011c]: Frege’s Theorem. Oxford University Press. Originally published
under the name Richard G. Heck, Jr.
Linnebo, Øystein [2005]: ‘To be is to be an F’, Dialectica 59, 201–222.
DOI: https://doi.org/10.1111/j.1746-8361.2005.01027.x.

Downloaded from https://academic.oup.com/philmat/article/31/2/236/7116333 by guest on 06 September 2024

[2018]: Thin Objects: An Abstractionist Account. Oxford University Press.
Linnebo, Øystein, and Agustı́n Rayo, [2012]: ‘Hierarchies ontological and ideological’,
Mind 121, 269–308. DOI: https://doi.org/10.1093/mind/fzs050.
Linnebo, Øystein, and Gabriel Uzquiano [2009]: ‘Which abstraction principles are
acceptable? Some limitative results’, The British Journal for the Philosophy of
Science 60, 239–252. DOI: https://doi.org/10.1093/bjps/axp001.
MacBride, Fraser [2003]: ‘Speaking with shadows: A study of neo-logicism’, The
British Journal for the Philosophy of Science 54, 103–163.
DOI: https://doi.org/10.1093/bjps/54.1.103.
[2006a]: ‘The Julius Caesar objection: More problematic than ever’, in
[MacBride, 2006b], pp. 174–202.
, ed. [2006b]: Identity and Modality. Oxford University Press.
Miller, Alexander, ed. [2020]: Logic, Language, and Mathematics: Themes from the
Philosophy of Crispin Wright. Oxford University Press.
Parsons, Charles [1964]: ‘Frege’s theory of number’, in Max Black, ed., Philosophy in
America, pp. 180–203. London: Allen and Unwin.
Rosen, Gideon, and Stephen Yablo [2020]: ‘Solving the Caesar Problem — with
metaphysics’, in [Miller, 2020], pp. 116–131.
Shapiro, Stewart [1997]: Philosophy of Mathematics: Structure and Ontology. Oxford
University Press.
[2000]: ‘Frege meets Dedekind: A neologicist treatment of real analysis’,
Notre Dame Journal of Formal Logic 41, 335–364.
DOI: https://doi.org/10.1305/ndjfl/1038336880.
[2006]: ‘Structure and identity’, in [MacBride, 2006b], pp. 109–145.
Studd, J.P. [2016]: ‘Abstraction reconceived’, The British Journal for the Philosophy
of Science 67, 579–615. DOI: https://doi.org/10.1093/bjps/axu035.
Warren, Jared [2017]: ‘Quantifier variance and indefinite extensibility’, Philosophical
Review 126, 81–122. DOI: https://doi.org/10.1215/00318108-3683622.
Wehmeier, Kai Frederick, and Peter Schroeder-Heister [2005]: ‘Frege’s permutation
argument revisited’, Synthese 147, 43–61.
DOI: https://doi.org/10.1007/s11229-004-6206-6.
Weiner, Joan [1984]: ‘The philosopher behind the last logicist’, The Philosophical
Quarterly 34, 242–264. DOI: https://doi.org/10.2307/2218759.
Williams, J. Robert G. [2007]: ‘Eligibility and inscrutability’, The Philosophical
Review 116, 361–399. DOI: https://doi.org/10.1215/00318108-2007-002.
Wright, Crispin [1983]: Frege’s Conception of Numbers As Objects. Aberdeen Univer-
sity Press.
[1997]: ‘On the philosophical significance of Frege’s theorem’, in [Heck,
1997b], pp. 201–244. Reprinted in [Hale and Wright, 2001a]. Page references to
reprint.
 The Caesar Problem • 267

[1998]: ‘On the (harmless) impredicativity of Hume’s principle’, in M. Schirn,

ed., Philosophy of Mathematics Today, pp. 339–368. Oxford: Clarendon Press.
Reprinted in [Hale and Wright, 2001a]. Page references to reprint.
[1999]: ‘Is Hume’s principle analytic?’, Notre Dame Journal of Formal Logic
40, 6–30. Reprinted in [Hale and Wright, 2001a]. Page references to reprint.
DOI: https://doi.org/10.1305/ndjfl/1039096303.
[2020]: ‘Replies to Part I: Frege and Logicism’, in [Miller, 2020], pp. 279–353.
DOI: https://doi.org/10.1093/oso/9780199278343.001.0001.

Downloaded from https://academic.oup.com/philmat/article/31/2/236/7116333 by guest on 06 September 2024