Synthese (2012) 187:693–713

DOI 10.1007/s11229-011-0009-3

Theory structuralism in a rigid framework

Christian Damböck

Received: 3 June 2011 / Accepted: 29 August 2011 / Published online: 10 September 2011
© Springer Science+Business Media B.V. 2011

Abstract This paper develops the first parts of a logical framework for the empirical
sciences, by means of a redefinition of theory structuralism as originally developed
by Joseph Sneed, Wolfgang Stegmüller, and others, in the context of a ‘rigid’ logic
as based on a fixed (therefore rigid) ontology. The paper defends a formal conception
of the empirical sciences that has an irreducible ontological basis and is unable, in
general, to provide purely structural characterizations of the domain of a theory. The
extreme rationalist utopia of a characterization of the real world ‘up to isomorphism’,
therefore, is rejected.

Keywords Formal frameworks for the empirical sciences · Theory structuralism ·

Semi-interpreted languages · Rigid logic · Joseph Sneed · Wolfgang Stegmüller

1 Introduction

The following account of theory structuralism deviates from the classical presentation1
in two respects. There are (1) some inessential formal simplifications in my account
that may have some didactic advantages, but there is (2) also one essential restriction,
namely, ‘rigidity’: we define structures and classes of structures only relative to some
fixed sets of objects—the ‘sorts’ of a so-called ‘structure frame’.

1 The main source for this paper is Balzer et al. (1987). Henceforth I quote this book always in the form
(BMS, page), (BMS, section), (BMS, page, definition), etc. Other works on theory structuralism that
I used here are especially Sneed (1971), Stegmüller (1973, 1979, 1986), Balzer and Moulines (1996),
Balzer et al. (2000), Moulines (2002).

C. Damböck (B)
Institute Vienna Circle, c/o University of Vienna, University Campus, Spitalgasse 2-4, Court 1,
Entrance 1.13, 1090 Vienna, Austria
e-mail: christian.damboeck@univie.ac.at

123
694 Synthese (2012) 187:693–713

This rigid framework, firstly, has some technical merits. It allows us to develop a
considerably simpler presentation of theory structuralism; it especially allows us to
define a logical framework that is comprehensive, insofar as it enables us to develop
every formal aspect of theory structuralism in a logical way, whereas in the classical
framework a logical representation may hardly cover more than the basic non-modal
aspects of the framework.2
Secondly, beside that (rather technical) aspect of a (complete) ‘logification’, there
is another, more philosophical argument for the use of the rigid framework in the
context of theory structuralism. Theory structuralism is intended as a framework for
the empirical sciences, and, in its original form, it is based on model theory. However,
some crucial notions of theory structuralism, e.g., the notion of the ‘empirical claim’
of a theory or the notion of ‘reduction’ of one theory to another, are defined in such a
way that the whole framework becomes rigid in fact, because in all these cases a theory
is not merely structurally compared with another theory (that represents the empirical
world or the theory it is reduced to), but some concrete models of a theory are directly
linked to other concrete models. The crucial concept is not that of a structural isomor-
phism between models (as it is commonplace in model theory), but that of a (partial)
identity between models or that of being linked to another model by means of a fixed
reduction relation. In other words, theory structuralism in fact makes use only of such
a fragment of the model theoretic framework that is structurally more or less identical
with a rigid framework, in the sense specified in the present paper. Thus, the claim is
that the rigid framework is the more genuine framework for theory structuralism than
the framework of classical model-theory. In particular, one of the main oddities of
theory structuralism, namely, the task of ruling out ‘unintended models’, immediately
vanishes if we switch to the rigid framework, because in the latter we can provide the
whole framework with such a restrictive ontological layout that unintended models
do not appear at all.
There are three different general aspects under which the present paper may be
read. First, it may be considered as a demonstration of the use of the rigid framework
as a framework for the empirical sciences. Theory structuralism rather functions as
an example here, as an object of demonstration. It was chosen because it is one of the
best developed formal accounts of scientific theories, and possibly the only one that
deals exclusively with the empirical sciences. The purpose of the present paper, seen
from the perspective of the rigid framework, is to illustrate the expressive power of
this formalism as a framework for the empirical sciences and to illustrate the family
resemblances with the framework of theory structuralism. Second, the paper may be
seen as a formal and philosophical critique of theory structuralism, as presented in
(BMS). Along that line of thought, the paper points out a number of failures of the
original account and shows how we may improve them, if we transform this account
into the rigid context. Third, this paper can be seen as the preliminary part of a research
project that intends to develop a comprehensive logical framework for the empirical

2 For a discussion of the problem of a logification of theory structuralism, in the context of classical logic
and set theory, see Rantala (1980). Cf. also Balzer and Moulines (1996, Chaps. 12 and 13), where rigorous
set theoretical and category theoretical foundations of theory structuralism are discussed. In contrast to
these accounts, the present proposal may be seen as a rigorous logical foundation for theory structuralism.

123
Synthese (2012) 187:693–713 695

sciences. In this sense, the present work is part of a philosophical research tradition
that begins with Rudolf Carnap’s seminal book The Logical Structure of the World. In
Carnap’s terminology, the task of a ‘rational reconstruction’ of the empirical sciences
is divided into two parts: ‘constitutional theory’, which is only concerned with the
purely formal aspects of the rational reconstruction, and the development of concrete
‘constitutional systems’ that illustrate the meta-theory with some concrete examples
and philosophical explanations.3 The present paper focuses on the former task. Its
strategy is to provide a framework that has the greatest possible expressive power, so
that research does not cover unnecessary formal hindrances if it goes on to develop
the more concrete parts of the program.
What we need is an abstract framework that makes as few formal presuppositions
as possible that may force us later on to give up a particular idea because it does not
fit into the framework. It is not easy to understand, however, what exactly is meant
by this claim, if we adopt it for the idea of a framework for the empirical sciences.
It is not meant, of course, that we must be able to express the most complex for-
mal structures (in the mathematical sense), but rather that we can express the most
complex ontologies. A formal framework that allows us to express complex formal
structures is classical first- or second-order logic. Complex ontologies for empirical
cases, however, in general have quite simple formal structures: they are often finite or
countable, and they are virtually always based on the simple idea of having a number of
concrete objects and ascribing a number of concrete properties to them. The complex-
ity, in that case, comes from a completely different side. For example, we may have
quite complex modal notions to express here that force us to quantify not only over
basic objects, but also over structures, as built from basic objects (‘possible worlds’),
over formulas that express properties of basic objects, over the whole range of basic
objects of any type, and over some power-sets and other set-theoretical constructs that
we obtain from objects of that kind. Operations of that kind can be established in the
context of a structurally powerful language like first- or second-order logic only in a
quite limited way. However, rigid logic, i.e., a logic that is based on a fixed collection
of basic objects (and thus can be seen as a fragment of first-order logic), does allow
operations of the just-mentioned kind, without any limitation, simply because the
object level in such a logic is simpler and more concretely construed than in the usual
structurally powerful logics. Thus, we finally arrive at a situation where we have to
choose definitely between two different types of limitation. Our formal framework can
be either structurally comprehensive and ontologically limited or structurally limited
and ontologically comprehensive. The former is the case in first-order and (to an even
higher extent) in second-order logic. The latter situation we obtain in the context of a
framework of rigid logic (that, as it turns out, is structurally weaker but ontologically
stronger than first-order logic). And this is why rigid logics may turn out to provide
better frameworks for the empirical sciences than the tools of classical (first- or sec-
ond-order) logic. In particular, the framework of theory structuralism may appear to
be much more powerful and much more manageable if we reconsider and redefine it
in a rigid setting.

3 See Carnap (1928, §§ 100 and 179). See also Leitgeb (2011), where the ‘Aufbau’ is reconsidered, on the
basis of an up-to-date formal framework.

123
696 Synthese (2012) 187:693–713

In Sect. 2 the principal formal aspects of the rigid framework are developed. Sec-
tion 3 contains a general discussion of theory structuralism and defines some fun-
damental layout principles for the present framework. On this basis, in Sect. 4, some
portions of theory structuralism are reconsidered, in the context of the rigid framework,
in a more formal way.

2 The rigid framework

In this section we specify a simplified version of the rigid framework that, in particular,
does not contain a detailed specification of a formal language. A more comprehensive
specification and discussion of this framework is developed elsewhere.4 The some-
what unusual technicalities of Sect. 2.1 are followed by a number of informal remarks
and clarifications in Sect. 2.2.

2.1 Structure frames and structures

The basic objects of theory structuralism in the classical setting (BMS, I) are the
structures

D1 , . . . , Dk , A 1 , . . . , A m , R 1 , . . . , R n

of a particular structure type. Here Di are the principal sets, Ai are the auxiliary sets
(i.e., the sets of mathematical objects), and Ri are relations of some specified types.
By contrast, the rigid framework introduces Di , Ai , and Ri by means of some fixed
sets that contain either ‘basic objects’ or ‘labels’ of ‘relations’. These sets we will call
‘sorts’ here. The elements of the sorts are considered as rigid designators, i.e., names
that have a fixed reference. Thus, the elements of the sorts provide both the names and
the primitive objects of our framework.
Note also that rigid designators are of a rather technical nature in our framework.
In particular, we do not claim here that the names of a particular language have to
be rigid, i.e., to have a fixed reference. Rather, the rigid layout of our framework is
merely a technical device for the introduction of objects into our language, to be able
to specify every kind of ontology directly, on the level of the object language. How-
ever, an extremely important special case of rigid ontologies as specified in this way
will be ontologies that contain non-rigid names. This will be realized by an operator
↓ that interprets any object/name c of our language as a ‘zero place function’ such
that ↓c points to the reference of c (in case there is such a reference). Thus, if c,
for example, represents the notion of space, the reference ↓c may be a completely
different manifold in the case of a theory that characterizes Newtonian physics and
in a theory that describes general relativity. In other words, non-rigid designators are
quite fundamental cases of our rigid framework, and they are not at all ruled out here.
Whether a particular name c is rigid or not will be one of the technical decisions that

4 See Damböck (2011).

123
Synthese (2012) 187:693–713 697

we will have to make in the context of the development of the formal specification of
a particular theory.
The sorts of our framework are typified by a set of ‘types’ of relations between
sorts. Each type is defined as a finite sequence of sorts. If (s, s1 , . . . , sn ) is a type,
then the first sort s always represents the names or ‘labels’ of the relations of this
type. Therefore, the ‘individuals’ of a structure species, i.e., Di and Ai , are introduced
here by means of sorts (thus, are restricted to fixed collections of basic objects); the
‘relational objects’ of a structure species, i.e., Ri , are introduced here by means of
types.
A structure in our framework is simply a function that specifies (1) the domain sets
as subsets of their respective sorts and (2) a suitable relation over the domain sets, for
every specified type. Thus, the whole framework is based on a ‘structure frame’ that
consists of a set of ‘sorts’ and a set of ‘types’. More formally:
A structure frame F = (S, T ) consists of a non-empty set S of non-empty and
disjoint sets—the sorts of the structure frame—and a set T of types that is defined as
a subset of the set of all finite sequences of elements of S such that (s) ∈ T , for every
S (i.e., every sort specifies a unary type).
The elements of the sorts we call the atomic objects of the structure frame. Thus,
we obtain the set Sa of all atomic objects of F as

Sa := s.
s∈S

We designate as τ the function that assigns to every atomic object o ∈ Sa its (well-
defined) sort τ (o) ∈ S.
If t = (s, s1 , . . . , sn ) is a type with n > 0, then we call s the labeling sort of t.
(Unary types do not have labeling sorts.) To obtain a useful type structure we require
the following typification rules to hold:
(1) There are some sorts that do not function as labeling sorts of any type. These are
the basic sorts. (These sorts correspond to first-order sorts in the classical sense.)
(2) Every sort is the labeling sort of, at most, one type. (This ensures that labels
identify types in a unique way.)
(3) S contains at least one basic sort and one labeling sort.
The set of all basic sorts we call B, and the set of all labels, i.e., objects that are
contained in labeling sorts, we call L.
A structure S over F = (S, T ) is defined as a function that firstly assigns to every
sort s ∈ S a set S(s) ⊆ s. Secondly, the structure S assigns to every sequence
(s1 , . . . , sn ) ∈ T with n > 1 a set

(S) S(s1 , . . . , sn ) ⊆ S(s1 ) × · · · × S(sn ).

2.2 Remarks on the above definitions

(1) Ultimately, the only completely unusual feature of this notion of a structure is its
rigidity: for every atomic sort s the domain S(s) of a structure S is specified as

123
698 Synthese (2012) 187:693–713

a subset of the set s, which builds the same fixed basis, for every structure. This
feature of rigidity is shared by van Fraassen’s conception of semi-interpreted
languages (see Sect. 3.1, below), and it can also be found in the context of quan-
tified modal logic with variable domains, where also the domain of a particular
possible world must be a subset of a fixed basic domain D (i.e., the union set of
the domains of all possible worlds).5
(2) The main advantage of rigid structure frames, in the context of empirical theo-
ries, is that the ontology of a theory can be specified here completely by means
of non-logical terms, i.e., objects of the structure frame. This is impossible, in
the classical framework of model theory, because there we do not have rigid
designators. Thus, in the classical framework we have to specify the ontology
indirectly, on the basis of purely structural specifications (by means of Ramsey
sentences and the like). This leads to the awkward situation in which every formal
specification of an empirical theory is burdened with a vast amount of unintended
models, i.e., models of a theory that have nothing to do with its intended applica-
tions. These unintended models can be ruled out only on an informal level, then,
but by no means in a formal way. In sharp contrast to this unfortunate situation
the rigid framework allows us to rule out unintended models from the beginning,
because we only introduce the objects we really want to talk about. Thus, the
actual models of a theory, in the rigid case, are always assumed to be also its
intended models (cf. Sect. 3.2, below).6
(3) If (s, s1 , . . . , sn ) is defined as a type, then there is exactly one relation over
s, s1 , . . . , sn specified in a structure (i.e., the structure specifies a subset of the
Cartesian product s × s1 × · · · × sn ). In non-rigid logic, by contrast, a struc-
ture ranges either over all relations or (in the case of Henkin-semantics) over
a subset of all relations of a particular type t = (t1 , . . . , tn ). In rigid logic, on
the other hand, each type contains a first sort s of labels that represents the
relations of a particular type t. Thus, the rigid type that corresponds to the just-
mentioned non-rigid type t has the form (s, t). Take, for example, the rigid type
(colors, objects), where ‘colors’ is a set of colors: ‘red’, ‘blue’, ‘green’, etc., and
‘objects’ is a set of objects: ‘table’, ‘chair’, ‘house’, etc. A rigid structure S, then,
first specifies subsets of ‘colors’ and ‘objects’: the colors and objects that exist
in that structure. Then, the structure specifies a subset of the Cartesian product
S(colors) × S(objects): the pairs of existing objects for which the relation of
type (colors, objects) holds. If (red, chair), for example, is an element of that
relation, then we may interpret this in such a way that the object ‘chair’ has
the property ‘red’. In other words, n + 1-place types of the rigid framework are
always interpreted in the sense of n-place types of a non-rigid framework; in the
rigid framework, the first place simply contains the names of the relations of the

5 The classical presentation of this type of quantified modal logic is Kripke (1963).
6 Note also that in the rigid framework purely structural descriptions are possible as well. Because a typical
rigid logic is higher order the formulation of Ramsey sentences is not at all a problem here. Thus, one may
decide to characterize the whole amount of theoretical and empirical objects exclusively by means of rigid
designators (my proposal), but one also may decide to have a more traditional layout in which either the
theoretical or the empirical objects or both are characterized in a purely structural way, by means of Ramsey
sentences.

123
Synthese (2012) 187:693–713 699

type. (If we want to be able to quantify over the whole range of all relations of
a particular type (s, s1 , . . . , sn ), then we have to introduce a sufficiently large
set s, i.e., the cardinality of s has to be greater or equal to the cardinality of
s1 × · · · × sn .)
Unary types in the classical sense are introduced in a structure frame as binary
types where the first place represents the labels of the unary type and the sec-
ond place its range. On the other hand, every atomic sort of a structure frame of
the rigid framework is defined as a rigid unary type. Rigid unary types of that
kind function as ‘zero-place’ or ‘empty’ types in the classical sense, and we may
interpret them in two different ways. First we may interpret them as ‘existence
predicates’ that pick out all these objects that may have positive properties in
the structure; rule (S) stipulates that if an object o of sort s does not exist in a
structure S, i.e., o ∈ / S(s), no sequence of objects of the form (. . . , o, . . .) can
be contained in the structure S. Second, we may interpret rigid unary types as
‘propositions’ such that any o ∈ s is ‘true’, iff o ∈ S(s).
(4) It is essential that the sorts of a structure frame are disjoint, because we take the
atomic objects of a structure frame as rigid designators, i.e., as names that have a
fixed reference. The notion ‘rigid designator’ was introduced by Saul Kripke.7 In
our framework, however, rigid designators have a rather technical nature. Unlike
Kripke, we do not claim that any name (of a particular language) is rigid in fact
(i.e., it refers causally to a particular object). Rather, rigid designators are merely
the basic technical element of our framework. An important special case for these
objects are the above-mentioned non-rigid designators, i.e., objects/names c that
have a varying reference ↓c in different theories or even in different models of
one and the same theory. Formally, the operator ↓ is introduced in the context of
sorts s that correspond to a type (s, s  ), which is specified as a partial function
from s to s  . For each object c ∈ s, then, we define ↓c as the definite object of
s  that is picked out by the partial function (s, s  ), or by a dummy-object NULL,
respectively, if there is no such object.
(5) In general, we assume that S contains a suitably large number of mathematical
standard sets like R, R3 , and the like. We do not always mention such sets explic-
itly, if we describe a particular structure frame. Additionally, we assume that, in
the context of a structure frame, there is defined a suitably large number of math-
ematical standard functions and relations over the mathematical standard sets.
All these elements we call auxiliary elements of a structure frame. The remaining
elements we call its principal elements (BMS, 10). The set of all principal atomic
objects of a structure frame we call S p .
(6) We cannot provide a detailed specification of a rigid logic in the context of the
present paper. We just sketch the principal layout of the simplest form of such a
logic. A logic over a rigid structure frame F, in its simplest form, can be viewed
as an instance of propositional logic with possibly uncountable propositional
constants. Consider the set

7 See Kripke (1980).

123
700 Synthese (2012) 187:693–713


Tx := s1 × · · · × si ,
(s j )ij=1 ∈T

which is well-defined for every structure frame F = (S, T ). Structures, then, can
be redefined as subsets of Tx . (Note that not every subset of Tx is defined as a
structure, because of the unary relations that form ‘existence predicates’.) Thus,
a rigid logic can be defined as propositional logic over the set Tx of proposi-
tional constants. We only have to include a device for building conjunctions
of arbitrary large (and possibly uncountable) sets of formulas.8
This basic rigid logic is bound to a particular structure frame F. Thus, we call it
L r (F). On this basis we can introduce a first level of modal quantification over
structures in the context of a first order modal structure frame (F, Sm , Tm ). Here,
the elements of Sm are ‘modal sorts’, i.e., sets that contain either accessibility
relations and the like or such things as the structures, formulas, and other logical
elements of the basic rigid logic; the elements of Tm are ‘modal types’, specified
by analogy with non-modal types. We obtain a set of ‘modal structures’ and a
set of atomic formulas that allow us to quantify over structures and formulas of
the basic logic, while the layout remains that of propositional logic, in principle.
The result of this specification is a ‘first-order modal rigid logic’ L r (F, Sm , Tm ).
By analogy we also may introduce a second-order and a third-order modal rigid
logic, etc.
In the following considerations we always assume that a suitable (modal) struc-
ture frame is given, together with a suitable formal language and a suitable defi-
nition of satisfaction of a formula in a (modal) structure.

3 A general discussion of theory structuralism

3.1 More structuralism or less structuralism?

Theory structuralism originates from the tradition of formal semantics and model the-
ory. One of its main forerunners is Patrick Suppes, who propagated in Suppes (1957)
a reconstruction of the empirical sciences by means of ‘set theoretic predicates’, with
the slogan “that philosophy of science should use mathematics, and not meta-mathe-
matics”.9 Another obvious source of theory structuralism is the structuralist approach
of the Bourbaki group.10 Sneed and Stegmüller adopted this approach and enriched
it with some vocabulary of more recent model theory. However, the mere usage of
model theoretic vocabulary does not imply that a theory is model theoretic in a full-
fledged way. And indeed, theory structuralism is not. A full-fledged model theoretic
account describes some portions of reality, exclusively by means of their pure structural
properties. Thus, for example, the only possibility of ascribing a particular property to
some portions of reality, is to identify the latter as isomorphic or at least approximately

8 The technique of reducing a logic to propositional logic is developed in Damböck (2009).

9 I found this slogan only in its indirect quotation in van Fraassen (1980, p. 65).
10 See Bourbaki (1968, Chap. IV), Stegmüller (1979).

123
Synthese (2012) 187:693–713 701

isomorphic with a particular structure or structure-characterization. This principle pro-

hibits, in particular, our identifying two portions of reality as structurally (dis-)similar
or (in-)commensurable on the mere basis of a description of their ontology. In other
words, a purely structuralist account, in this sense, has to express everything that it
wants to express, exclusively with appeal to structural properties (by means of Ram-
sey sentences), and not with appeal to ontological properties (by means of sentences
that ascribe concrete properties to concrete objects). In this respect, theory structur-
alism clearly is not a purely structuralist account, because most of its specifications
for the comparison of structures do not refer to isomorphy and other purely structural
properties, but rather compare structures directly by means of their ontology. Theory
structuralism is ultimately sort of a mixture between a purely structuralist account, as
based on formal semantics and model theory, and an ontological account, as based
on the idea of providing some concrete ontological descriptions of objects and their
(actual or possible) relations.
For example, the ‘constraint’ of a theory (BMS, II.2) generally specifies a particular
class or set of models that (1) share their ontology (i.e., their domains) and (2) share
some properties. The former is an ontological property, and only the latter a structural
one. On the other hand, the ‘reduction’ of a theory to another one (BMS, 277, DVI-5)
is defined (at least at a first glance) in structuralist terms, because there has to be a
reduction-relation ρ that maps the models of one theory to the models of the other in
a particular way. The problem that appears here is that the notion of reduction of an
empirical theory to another, in purely structural terms, tends to trivialize the whole
idea of reduction.11 The question that we have to ask is the following: Is what we want
to say when we describe constraints or reduction relations in fact something purely
structural, is it in fact a purely ontological thing or a mixture of both? In a way, the latter
must be evidently true, because every empirical theory clearly has both ontological and
structural aspects: it ascribes some properties to some objects, therefore identifies an
ontology with a structure. However, this sort of combination of structural and ontolog-
ical aspects clearly must be a feature of every kind of a ‘pure’ ontological framework,
and it is a feature of the rigid framework, in particular. Whereas it is true, in this sense,
that there is no purely ontological framework at all (unless we may be willing to call
a mere collection of objects, without any structural commitment, a ‘framework’), it
is also quite clear that theory structuralism mixes the ontological and the structural
aspects in a different way. The problem here seems to be that theory structuralism (at
least in some sense) wants to provide a purely structuralist framework, but fails to
do so.
Let us take this for granted and ask how we may improve this failure of theory struc-
turalism. There seem to be two possible strategies. First, one may try to improve theory
structuralism by making it even more structuralist. Second, one may try to improve it
by simply giving up the structuralist intention and reconsidering the whole account in
a different framework that has a more ontological nature. The first strategy obviously
would lead to a more or less complete counteraction of the original task of theory

11 Cf. Niebergall (2000) where it is demonstrated how to ‘reduce’ a particular empirical theory to a purely
structural and empirically meaningless thing like ‘propositional logic’. Another critical discussion of the
structuralist account of reduction is provided in Hoering (1984).

123
702 Synthese (2012) 187:693–713

structuralism, namely, to provide a formal framework for the empirical sciences, sim-
ply because it is the very foundation of this task to take the empirical sciences as
different from pure mathematics in having quite essential ontological aspects. In other
words, it cannot be a vital aim of theory structuralism to provide a purely structural
account of the empirical sciences, insofar as this would be just question-begging: the
empirical sciences are not purely structural—that’s it. In the empirical sciences, we
(per definition) do not talk about pure structures, but rather about some objects in
the real world having these and those properties. All this ontological talk, of course,
also has a structural aspect, which may have led theory structuralists to adopt the
model theoretical framework for their purposes. The very aim of theory structuralism,
however, never was the complete structuralization of empirical knowledge in such a
way that the world is characterized up to isomorphism, but a much more modest one,
namely, to “develop a ‘representation scheme’ for scientific knowledge” (BMS, xvii).
This representational attitude implies, of course, that the whole structuralist frame-
work must be designed in such a way that it always enables us to ‘encapsulate’ some
aspects of the empirical world by simply referring to it (because there is no complete
structuralization that allows us to replace direct reference with a structure).
Given these considerations there can be no doubt that the second strategy is best
suited for an improvement of the structuralist framework. Following this strategy, the
present paper aims to reconceptualize some portions of the original framework of the-
ory structuralism in a rigid framework. In this framework, constraints, to come back
to the first of our examples, simply describe the same structural properties of concrete
ontologies, as in the original conception. We may say that the notion of a constraint,
as presented in (BMS), like many other notions, is a rigid notion, in fact. On the other
hand, such things like reduction of a theory to another one clearly must obtain in the
present framework a much more ontological nature. There must be a concrete relation
ρ that reduces the ontology of one theory to the ontology of the other. Only on the
basis of this concrete relation ρ (that must be seen as the direct expression of some
empirical facts) are we able to define the notion of being a reduction of a theory (by
means of having corresponding models, in a rather obvious way).12
A historical remark. Theory structuralism is by far not the only example that allows
us to support the claim that a framework for the empirical sciences has to be less struc-
tural than a framework for pure mathematics, and therefore is rigid in fact. The logics
that were considered by Frege, Russell, Carnap, and other classical philosophers of
logic, until the end of the 1920s, where rigid.13 Rudolf Carnap, who developed in his
seminal Carnap (1928) an account of theories of the empirical sciences, which was
firstly introduced as a purely structural one (§ 16), came to the conclusion, at the end
of his book, that a purely structural characterization fails and has to be replaced by an
account that is based on so-called founded relations (§ 154), and therefore is rigid in

12 The discussion of reduction is one task that has to be postponed in the present paper, even though it
seems to be one of the crucial aspects of the framework of theory structuralism. In the present context we
only shall discuss the specification of single theory-elements, in a rigid framework. See also note 17, below.
13 For an account of the development of the modern semantic picture of logic, from that rigid starting
point, see Goldfarb (1979). The present account, indeed, is a plea to go back to the rigid roots of pure
logic, wherever we are concerned with theories of the empirical sciences and not with theories of pure
mathematics.

123
Synthese (2012) 187:693–713 703

fact. Moreover, the whole philosophical discussion in the context of the so-called ‘new
theory of reference’ (Kripke, Putnam) is nothing else but a broad establishment of the
claim that a purely structural identification of empirical objects is generally impos-
sible.14 Probably the most explicit example of a logical framework for the empirical
sciences that is rigidly construed in fact is Bas van Fraassen’s use of semi-interpreted
languages (as inspired by Evert Willem Beth). Semi-interpreted languages are formal
languages that are semantically interpreted only in the context of a fixed set F of
objects. Thus, semi-interpreted languages and rigid logics have the same conceptual
basis.15

3.2 The problem of unintended models

Theory structuralism, like the rigid framework, is intended as a framework for the
empirical sciences, but not for theories of pure mathematics.16 Nevertheless, theory
structuralism is based on the classical framework of model theory, so that, given a
particular structure type of a theory, every structure

D1 , . . . , Dk , A 1 , . . . , A m , R 1 , . . . , R n

of that type must be considered. Theory structuralism, then, provides some restrictions
to structures in an obvious way. Firstly, it defines some characterizations C, i.e., some
rules that stipulate such things, like that a particular relation Ri is defined as a bijec-
tive function or that a particular set D j has to be finite. Secondly, and in particular, it
describes the respective laws L of a theory T and defines the class M of all models of
T, essentially, as all models of the sentences C ∪ L.
A main problem of theory structuralism is that the class M inevitably contains a
vast number of unintended models, i.e., models that characterize ‘realities’ that have
nothing but their formal structure in common with the reality that the theory intends
to describe. In other words, what clearly would be a virtue in the context of a theory
of pure mathematics, namely, to provide a purely structural and not an ontological
characterization of a theory, turns out to be a serious shortcoming in the context of a
theory of the empirical sciences.
Note also that this problem of unintended models becomes more serious the simpler
a particular theory is. On the one hand, we may construe the utopian and extremely
rationalist picture of a holistic, all-encompassing theory of the empirical world that
‘converges’ with a mathematical theory, because it has just one model: the real world
as it actually is. In that case, a purely structuralist account would be the right choice,
of course. On the extreme opposite side, if we take an extremely simple example of
an empirical theory, e.g., a theory that only ascribes one property to one object, in the
context of the claim

14 See Kripke (1980). One should also note here that quantified modal logic is a version of predicate logic
that shows a number of parallels with rigid logic. Cf. the first remark in Sect. 2.2, below.
15 See van Fraassen (1967, 1969, 1970) and Beth (1960).
16 See (BMS, XVf).

123
704 Synthese (2012) 187:693–713

‘Snow is white’
then we may represent this theory structurally, by means of the structures of type

D, R,

where D is a set with one element and R is a unary function over D. The law of the
theory, of course, is the claim

∀x ∈ D : R(x).

The ‘intended model’ I of the theory is the set D that contains the object ‘snow’ and
the relation R that represents the property ‘white’. However, besides this intended
model I the models of the theory obviously contain a vast amount of models, namely,
all structures of the just-specified type, where R is true for the only element of D.
This is not to say that it is impossible to describe an empirical theory in such a
framework. Clearly, however, as long as we do not have a holistic and all-encompass-
ing theory in the above-mentioned sense, the structuralist setting always provides us
with this enormous bunch of unintended models that are completely irrelevant for the
empirical claim of a theory, and are therefore useless.
The rigid framework provides a solution to that problem of unintended models,
which is obviously valid for all possible examples of empirical theories below the
threshold of the above-mentioned extreme rationalist case. In our example we would
take a structure frame F = (S, T ) where S consists of only two sorts s and s  : s con-
tains the single object ‘snow’ and s  the single object ‘white’. T contains the single
type (s  , s), and the law of the theory is simply that the claim (white, snow) is true.
Thus, we obtain exactly one model here that, of course, is also the only intended model
of the theory.
We then may enrich our framework for more and more complex cases in the context
of more and more complex structure frames and laws. Nonetheless, however complex
the setting for an empirical theory may be, it seems to be always possible here to find
a characterization of a theory that completely rules out unintended models, because it
divides the structures of a structure frame into two subsets: (1) the structures in which
the laws of a theory are false, and (2) the intended models of a theory. We assume this
remark with the following first layout principle for a redefinition of theory structural-
ism in the rigid framework:
In the rigid framework, in general, the models and the intended models of
a theory are identical.
This affects the whole layout of our presentation of theory structuralism in Sect. 4
below, because the distinction between theory-cores and intended applications falls
away here: the ‘actual models’ and the ‘(theoretical) content’ of a theory, in the rigid
setting, are exactly everything the theory tells us about the world.
In the rigid framework the ontology of a theory can be specified in such a restrictive
way that the models of a theory are exactly its intended models. The (intended) models
of a theory are the picture that a theory draws from the world. This picture, then, may
be true or false, empirically adequate or not.

123
Synthese (2012) 187:693–713 705

Note also that this convergence between models and intended models clearly does
not imply that it is impossible here to distinguish between different intended appli-
cations of a theory. The theory of classical particle mechanics, for example, may be
refined in such a way that we can distinguish formally between the applications to the
solar system (and its subsystems), the pendulum, and the harmonic oscillator. We must
be able, of course, to distinguish between all these things even in the rigid framework
(by means of some constraints or some restrictions of the ontology of a theory). The
only thing that we claim here is that the set of all models of a theory, in the rigid
context, ultimately shall contain all and exactly all its intended applications (see also
the remarks at the end of Sect. 4.2, below).

3.3 The problem of internal model restriction

We now turn to the problem of model restriction, not on the external level of compar-
ison of the models of a theory with the intended (empirical) world, but on the internal
level of the formal establishment of the models of a theory. Internal model restriction,
in this sense, is realized, in the structuralist framework, on (at least) five different
levels:
(1) The level of specification of the ‘ontology’ of a theory
(2) The level of ‘characterization’ of ‘potential models’ of a theory
(3) The level of the specification of the ‘laws’ of a theory
(4) The level of the specification of the ‘constraints’ of a theory
(5) The level of the specification of the ‘intertheoretical links’ of a theory
The main philosophical and formal problem that we may identify here is situated
already on the first level. The structuralist framework, as specified in a classical model
theoretic setting, implies that we can specify only parts of the ontology of a theory on
the elementary level of formal theory specification. If we take the structures

D1 , . . . , Dk , A 1 , . . . , A m , R 1 , . . . , R n

of a particular type, then we stipulate as rigid elements of a theory (a) the auxiliary
sets (i.e., sets like R, R3 , N, etc., that are present in every structure of a theory) and
(b) the relations R1 , . . . , Rn that have to be specified in the context of every structure.
These objects are fixed ontological parts of a theory; every structure and every model
of a theory must contain them and (in the case of relations) must interpret them. So
far the structuralist framework is completely rigid, from scratch. In sharp contrast to
this, we have the principal sets D1 , . . . , Dk of the theory, which are completely unre-
stricted sets that may contain anything whatsoever. (We at most may establish some
formal restrictions of the cardinality of a particular Di and the like.) Thus, as already
explicated in the previous section, the ontology of a theory is simply ill-founded in the
context of a specification of theory structuralism, as established in a classical model
theoretic setting.
Thus, we henceforth take for granted that, to compare the ontologies of differ-
ent theories and restrict ourselves to the intended models of a theory, it is inevitable

123
706 Synthese (2012) 187:693–713

to establish a completely restricted ontology for our theory. In the rigid setting we
realize this by introducing even the sets D1 , . . . , Dk on the basis of a predefined set of
sorts so that every Di is defined as a subset of a particular sort s. Ultimately, one may
well say that this is the only real difference between our presentation and the classical
one (although this difference, of course, has a number of serious consequences).
Levels (2) to (5) of internal model restriction, are realized in the context of four
different levels of specification of formal rules and axioms:
(2) The ‘characterizations’ of the ‘potential models’ of a theory are nothing more
than further restrictions and specifications that we provide for some particular ele-
ments of the ontology of a theory. A particular relation of the theory, for example,
may be specified as a bijective function, another as a reflexive relation; a particular
set of a theory may be specified as a finite or countable set, and so on. The only
restriction that takes place here is that the respective rules and axioms must not
contain references to more than one principal element (either a set or a relation) of
the structure species of a theory (BMS, 14, DI-4).
(3) The ‘laws’ of a theory, then, are also rules and axioms that specify properties
of the elements of the ontology of a theory. The only difference between a law and
characterizations is that a law always has to contain references to at least two dif-
ferent principal elements of the ontology (BMS, 16). For example, E = mc2 may
be seen as a law, c ≈ 3 × 105 m/s as a characterization.
(4) According to theory structuralism, however, the models of a theory generally
are not sufficiently specified, on the basis of its characterizations and laws. In most
cases of empirical theories we need (at least) two further specifications that restrict
the models of a theory. The first one consists of ‘constraints’: rules that describe the
formal relations between the models of a theory. A typical example of a constraint is
the equality of mass constraint in classical particle mechanics (BMS, II.2.1), which
stipulates that if two models of that theory both contain a particle p, then p must
have the same mass in both models. Thus, the constraints specify ‘consistent sub-
classes of models’, and they restrict the class M of a theory (that we obtained from
the characterizations and laws) to a class of consistent subclasses of M. (Note also
that, formally, the constraints lift the specification of models of a theory, from the
level of classes of structures to the level of classes of classes of structures).
(5) The second important group of further specifications that restrict the models of a
theory are ‘intertheoretical links’ that define connections between a theory and other
theories. These connections, for example, may reduce the non-theoretical parts of a
theory to another theory that allows us to calculate or to visualize these non-theoret-
ical elements. Formally, intertheoretical links are nothing more than relations that
are specified between the (potential) models of different theories. Thus, the models
of a theory are restricted here to only these structures that are actually linked to
other theories.
In contrast to our criticism of level (1), none of these four aspects of the axiomatic
specification of the models of a theory will be questioned here, in principle. We take
for granted that the identification of these four aspects of model restriction is one of the
great merits of theory structuralism as a framework for the empirical sciences, because
they provide these specifications with refinements that we miss in other frameworks.

123
Synthese (2012) 187:693–713 707

What I want to point out here is only the rather technical question of the establishment
of these four aspects, in the context of a particular formal framework. If this frame-
work provides a logic that is expressive enough, then all these axioms and rules may
be established simply as formulas (axioms), namely:
(2) Formulas that refer to only one principal element of the theory
(3) Formulas that refer to at least two principal elements of the theory
(4) Formulas that refer to sets of models of a theory
(5) Formulas that refer to the models of this and other theories
Thus, the following second layout principle for the below-specified framework can
be formulated:
The framework should be at least that powerful so that axioms of type (2)
to (5) can be introduced as formulas of the framework.
This layout principle implies, in particular, that our framework must allow us to quan-
tify (without any serious restriction) over the structures of a basic language, in the
context of the object language of a suitable (first order) modal rigid logic, as specified
over it.

4 Theory structuralism in a rigid framework

In this final section we develop a presentation of theory structuralism, which is based

on rigid logic and the above specified layout principles and provides counterparts for
some of the definitions in (BMS, Chaps. I and II). Because of questions of space we
do not discuss the most important notions of theory-nets, reduction, and evolution of
theories. These aspects are/will be discussed elsewhere.17

4.1 Theory-elements

A crucial aspect of theory structuralism is that it is not restricted to the specification

of one single ‘theory-element’, but considers arbitrarily large collections of theories
and theory-elements, and relations between them. As a consequence of this we decide
to construe our logical framework in such a way that it is based on a structure frame
which is sufficiently large so that it is possible to specify every theory-element that is
relevant, in the context of our enquiry. This implies, in particular, that the ontology
of a theory-element must be specified independently of a structure frame (because
its structure frame, in general, will have a much larger ontology than this one single
theory-element). Formally, we specify the ontology of a theory-element on the basis
of a pair

(B p , L p ),

17 Some aspects of the dynamic of scientific theories are discussed in Damböck (submitted). At any rate, a
more detailed account of ‘theory nets’ and reduction requires a detailed formalization of rigid logic, which is
provided in Damböck (2011). An adoption of rigid logic to more complex problems of theory structuralism
is intended for future work of the author. This shall include, in particular, the problem of reduction, theory
nets, and invariance principles as considered in (BMS, Chaps. IV–VI) and Sneed (1979).

123
708 Synthese (2012) 187:693–713

where B p is a set of subsets of basic sorts of the structure frame and L p is a set of
principal labels of the structure frame. These two objects clearly form the direct coun-
terparts to the sets D1 , . . . , Dk and the relations R1 , . . . , Rn of a structure (species),
in the classical presentation of theory structuralism. (The auxiliary sets and relations
are the same, for every theory-element that we specify in the context of a structure
frame, thus we do not mention them explicitly here.)
B p and L p specify the whole ontology of a theory (with the exception of its auxiliary
objects). However, because we are concerned with empirical theories, the ontolog-
ical specification has to be supplemented by a third fundamental set of objects: the
theoretical terms of a theory. This set T shall be specified simply as a set of atomic
objects of the basic structure frame, i.e., it may contain any label or any element of
a basic sort. Thus, theoretical objects may be either ‘properties’ or ‘things’, without
any general restriction.
In addition to these three sets that specify the ontology of a theory-element we will
have only one more set A of axioms that contains all the necessary specifications of
characterizations, laws, constraints and links, in the sense of the types of rules and
axioms (2) to (5) that we mentioned in Sect. 3.3, above. More precisely, a theory-ele-
ment T (as based on a modal structure frame Fm = (F, Sm , Tm ) and a modal structure
M) is specified in the following form:

T = (B p , L p , T, A),

and we have:
(1) for every s ∈ B p there is a basic sort s  with s ⊆ s  . This sort shall be picked out
by the function τ (s).
(2) L p ⊆ L is a set of labels of principal types.
(3) T ⊆ Sa is a set of atomic objects of the basic structure frame.
(4) A is a set of L r (F, Sm , Tm )-formulas.
Given these specifications we define the ontology O of a theory as
⎛ ⎞

O := ⎝ s⎠ ∪ L p.
s∈B p

We sometimes may also consider the abbreviated description of a theory-element, in

the form

T = (O, T, A).

For obvious reasons, we usually may also require that T ⊆ O.

As explicated above, the set A of axioms of the theory-element contains charac-
terizations, laws, constraints, and links, in the sense of the classical framework. We
distinguish between these four types of axioms, in the following way:
(1) φ ∈ A is a characterization iff there is either an object x ∈ L p or a set of objects
x ∈ B p such that every non-logical and non-auxiliary term in φ refers to x.

123
Synthese (2012) 187:693–713 709

(2) φ ∈ A is a law iff it is not a characterization and every non-logical and non-aux-
iliary term in φ refers only to objects in O.
(3) φ ∈ A specifies a constraint iff it specifies a unary property of sets of F-structures.
(4) φ ∈ A specifies a link iff it specifies a binary relation between F-structures.
We define the set C of characterizations, the set L of laws, the set E of constraint
specifications (‘E’ for ‘Einschränkungen’) and the set V of link specifications (‘V’ for
‘Verbindungen’) as the respective subsets of A. (Note that if there are formulas in A
that are not specified as elements of one of these sets, then they have no effect for the
specification of a theory.)—Given these specifications we sometimes may expand the
specification of a theory-element to:

T = (B p , L p , T, C, L, E, V).

Here, the different sets of axioms C, L, E, V are introduced explicitly. (If a theory-
element does not specify axioms of one of these four types we may leave out the
respective set.)
The definition of a theory-element in (BMS, 89, DII-17) does not mention any of
the elements that are contained in our definition. However, beside of the ‘intended
applications’ that fall away in our account, we are able to introduce all the remaining
elements of a ‘theory-core’

(Mp , M, Mpp , GC, GL)

by means of some more or less obvious explicit definitions.

We start with the potential models Mp (T) (BMS, 15, DI-7) of a theory. Given
the function μ that assigns to every L r (F, Sm , Tm )-formula its models, the set of all
potential models must be based on the set

μ(C)

of all models of the characterization-axioms. However, we need some more restric-

tions here, because the potential models also have to restrict the models of a theory
to exactly all these structures that specify positive properties, only for the principal
objects of O(T). In other words: Mp (T) must not specify any positive property of
any atomic object of a structure frame that is not contained in O(T). Thus, we define
Mp (T) as the set of all models of C such that it additionally holds, for every S ∈ Mp :

∀o ∈ S p : o ∈
/ O(T) → o ∈
/ S(τ (o)),

i.e., every principal atomic object that does not belong to the ontology of T also does
not exist in a structure of Mp (and this implies that it cannot have any positive property
in that structure). Alternatively, the definition of Mp (T) can also be given by means
of the following one-liner:

Mp (T) := {S | S ∈ μ(C) ∧ [∀o ∈ S p : o ∈

/ O(T) → o ∈
/ S(τ (o))]}.

123
710 Synthese (2012) 187:693–713

The actual models M(T) (BMS, 20) of a theory T, then, are defined as:

M(T) := Mp (T) ∩ μ(L).

For the definition of the ‘partial potential models’ of a theory we need the following
auxiliary device: If O ⊆ Sa is a set of objects (i.e., of elements of any atomic sort) of
a structure frame then S \ O is the (well-defined) structure S such that:

(1) S (s) = S(s) \ O, for every s ∈ S

(2) for every sequence of objects c1 , . . . , cn with ci ∈ S (τ (ci )) it holds

(c1 , . . . , cn ) ∈ S (τ (c1 ), . . . , τ (cn )) ↔

(c1 , . . . , cn ) ∈ S(τ (c1 ), . . . , τ (cn ))

Intuitively, S \ O is the structure that results, if we ‘subtract’ the vocabulary out of O

from it. (Note that O may contain both labels of relations and elements of basic sorts.)
On this basis, the partial potential models Mpp (T) (BMS, 57, DII-3) of a theory T are
defined as:

Mpp (T) := {S | ∃S : S ∈ Mp (T) ∧ S = S \ T}

The partial potential models are all these models of a theory-element that does not
ascribe positive properties to any theoretical object of the theory-element.
The constraints and links are integrated into this picture in two steps. First, we
define, for every φ ∈ E and every φ  ∈ V, the constraint C(T, φ) (BMS, 47, DII-2)
and the abstract link L(T, φ  ) (BMS, 61, DII-4):

C(T, φ) := {X ⊆ S | φ(X )}
L(T, φ  ) := {S ∈ S | φ  (S)}

Here, S is the set of all F-structures. On this basis we define the global constraints
GC(T) (BMS, 78, DII-10) and the global links GL(T) (BMS, 78f, DII-11) of the
theory:


GC(T) := C(T, φ)
φ∈E (T)

GL(T) := L(T, φ)
φ∈V (T)

And we obtain obvious definitions for the theoretical content Cnth (T) (BMS, 82,
DII-13), and the content Cn(T) (BMS, 85, DII-15) of a theory:

123
Synthese (2012) 187:693–713 711

Cnth (T) := ℘ (M(T)) ∩ GC(T) ∩ ℘ (GL(T))

Cn(T) := {X | ∃X  ∈ Cnth (T) ∃θ :
θ is a bijective function from X to X 
∧∀ S, S : S = θ (S ) → S = S \ T(T)}

The theoretical content simply combines all these restrictions for the models of a the-
ory that are contained in the axioms: characterizations, laws, links, and constraints.
The content, then, restricts the theoretical content to these sub-structures that we obtain
if we subtract the theoretical vocabulary T. Thus, the really significant aspects of a
theory-element, concerning questions of truth and empirical adequacy, are Cnth (T)
and Cn(T).

4.2 An example: classical particle mechanics

We finally describe one typical example of a theory-element, in order to show how

to transform examples from the classical presentation (BMS) to the rigid framework.
The example is classical particle mechanics, as presented in (BMS, III.3). The theory-
element of classical particle mechanics CPM is defined as:

CPM := (B p , L p , T, C, L, E).

Here are the specifications of the respective components:

B p := {P, T, S}.
L p := {c1 , c2 , s, m, f }.
T := {m, f }
C := (1) P is a finite, non-empty set.
(2) c1 : T → R and c2 : S → R3 are bijective.
(3) s : P × T → S and c2 ◦ s p ◦ č1 is smooth for all p ∈ P.
(4) m : P → R+ .
(5) f : P × T × N → R3
L := ∀ p, α : m( p)D 2 c2 ◦ s p ◦ č1 (α) = f ( p, č1 (α), i)
i∈N
E := (1) ∀S, S ∈ X ∀ p : p ∈ S(P) ∩
S (P) → S(m( p)) = S (m( p)).
(2) ∃◦ : π × π → π where π := {S(P) | S ∈ Mp (CPM)} and:
∀S ∈ X ∀ p, p  ∈ S(P) : ( p ◦ p  ∈ S(P)) →
S(m( p ◦ p  )) = S(m( p)) + S(m( p  ))
(3) ∀S, S ∈ X ∀ p ∈ S(P) ∩ S (P) ∀t ∈ S(T ) ∩ S (T ) ∀i :
S( f ( p, t, i)) = S ( f ( p, t, i))
These specifications are more or less literal reproductions from (BMS, III.3). For our
present purpose, the only important aspect of this specification is that it is in full
accordance with the above specified formal rules. Note in particular that the three
constraint-axioms all contain the free variable X . This variable is of type ℘ (S), i.e. it
ranges over the power set of the set of all structures of the structure frame. As pointed
out above (p. 706, remark (4)) a constraint is a property of sets of models of a theory,

123
712 Synthese (2012) 187:693–713

thus it formally has to be specified as a ℘ (S)-property. (Following (BMS), we do not

introduce links, in the context of this example.)
Characterization rule (1) refers to the set S(P) that is picked out by a particular
structure S. Thus, in spite of the limited size of S(P), the respective sort P may be
an arbitrarily large set. In general, the sort T may contain every possible temporal
state of any system relevant to CPM, the sort P every possible spatial object of any
system relevant to CPM. The result should be a theory-element whose content Cn(T)
converges with the intended applications of CPM, as characterized in (BMS, 107).
Note also that the choice of the respective sets P, T , and S is the main pragmatic
aspect in our specification. The stipulation of P, T and S restricts the amount of pos-
sible intended applications, but still leaves open enough space for specifying different
instances of a theory with different sets of intended applications. In other words, those
unintended models in the classical framework that are completely unintended (because
their ontology does not match the ontology of the theory in question) are ruled out
here, whereas we do obtain a particular amount of structures as possible candidates
for different sets of intended applications. The ‘larger’ our structure frame is the larger
the scope for the specification of different sets of intended applications will be (and
vice versa).

Acknowledgments This paper is partly based on my PhD-thesis Damböck (2005). An earlier version
of this paper was presented at a workshop on theory structuralism on 6 May 2011 at the Institute Vienna
Circle, University of Vienna. For comments I am grateful to Hans-Joachim Dahms, Richard Dawid, Thomas
Mormann, Carlos Ulises Moulines, Paul Weingartner, and two anonymous referees of Synthese.

References

Balzer, W. & Moulines, C. U. (Eds.). (1996). Structuralist theory of science. Focal issues, new
results. Berlin: Walter de Gruyter.
Balzer, W., Moulines, C. U., & Sneed, J. D. (1987). An architectonic for science. The structuralist
program. Dordrecht: D. Reidel Publishing Company.
Balzer, W., Moulines, C. U., & Sneed, J. D. (Eds.). (2000). Structuralist knowledge representation.
Paradigmatic examples. Amsterdam: Rodopi.
Beth, E. W. (1960). Semantics of physical theories. Synthese, 12, 172–175.
Bourbaki, N. (1968). Theory of sets. Berlin: Springer-Verlag.
Carnap, R. (1998 [1928]). Der logische Aufbau der Welt. Hamburg: Felix Meiner Verlag.
Damböck, C. (submitted). Progress without truth: A reflection of formal epistemology. (submitted).
Damböck, C. (2005). Semantische Strategien. Philosophische Logik vor dem Hintergrund von Endlichkeit
und Starrheit. PhD-thesis. University of Vienna.
Damböck, C. (2009). Philosophical logic in a framework of propositional logic. Logique Et Ana-
lyse, 205, 21–37.
Damböck, C. (2011). Rigid logic as a framework for the empirical sciences. (submitted).
Goldfarb, W. (1979). Logic in the twentieth: The nature of the quantifier. Journal of Symbolic
Logic, 44, 351–368.
Hoering, W. (1984). Anomalies of reduction. Wolfgang Balzer et al: Reduction in science. Structure,
examples, philosophical problems (pp. 33–50). Dordrecht: D. Reidel Publishing Company.
Kripke, S. A. (1963). Semantical considerations on modal logic. Acta Philosophica Fennica, 26, 83–94.
Kripke, S. A. (1980). Naming and necessity. Cambridge, MA: Harvard University Press.
Leitgeb, H. (2011). New life for Carnap’s Aufbau?. Synthese, 180, 265–299.
Moulines, C. U. (2002). Structuralism (=Synthese 130, 1–172). Dordrecht: Kluwer Academic Publishers.
Niebergall, K.-G. (2000). Structuralism, model theory and reduction. Synthese, 130, 135–162.
Rantala, V. (1980). On the logical basis of the structuralist philosophy of science. Erkenntnis, 15, 269–286.

123
Synthese (2012) 187:693–713 713

Sneed, J. D. (1979). Theoretization and invariance principles. In I. Niiniluoto & R. Tuomela (Eds.),
The logic and epistemology of scientific change. Acta Philosophica Fennica, 30, 130–178.
Sneed, J. D. (1979 [1971]). The logical structure of mathematical physics (2nd ed.). Dordrecht: D.
Reidel Publishing Company.
Stegmüller, W. (1979). The structuralist view of theories. A possible analogue of the Bourbaki programme
in physical science. Berlin: Springer-Verlag.
Stegmüller, W. (1985 [1973]). Theorienstrukturen und Theoriendynamik. Probleme und Resultate der
Wissenschaftstheorie und Analytischen Philosophie, Band II: Theorie und Erfahrung. Zweiter
Teilband (2nd ed.). Berlin: Springer-Verlag.
Stegmüller, W. (1986). Die Entwicklung des neuen Strukturalismus seit 1973. Probleme und Resultate
der Wissenschaftstheorie und Analytischen Philosophie Band II: Theorie und Erfahrung. Dritter
Teilband. Berlin: Springer-Verlag.
Suppes, P. (1957). Introduction to logic. New York, NY: Dover Publications.
van Fraassen, B. C. (1967). Meaning relations and predicates. Noûs, 1, 161–179.
van Fraassen, B. C. (1969). Meaning relations and modalities. Noûs, 3, 155–167.
van Fraassen, B. C. (1970). On the extension of Beth’s semantics of physical theories. Philosophy of
Science, 37, 325–339.
van Fraassen, B. C. (1980). The scientific image. New York, NY: Oxford University Press.

123