Australasian Journal of Philosophy

ISSN: 0004-8402 (Print) 1471-6828 (Online) Journal homepage: http://www.tandfonline.com/loi/rajp20

Inferential Transitions

Jake Quilty-Dunn & Eric Mandelbaum

To cite this article: Jake Quilty-Dunn & Eric Mandelbaum (2017): Inferential Transitions,
Australasian Journal of Philosophy, DOI: 10.1080/00048402.2017.1358754

To link to this article: http://dx.doi.org/10.1080/00048402.2017.1358754

Published online: 27 Aug 2017.

Submit your article to this journal

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

http://www.tandfonline.com/action/journalInformation?journalCode=rajp20

Download by: [the Bodleian Libraries of the University of Oxford] Date: 28 August 2017, At: 07:01
 AUSTRALASIAN JOURNAL OF PHILOSOPHY, 2017
https://doi.org/10.1080/00048402.2017.1358754

Inferential Transitions
Jake Quilty-Dunna and Eric Mandelbaumb
Downloaded by [the Bodleian Libraries of the University of Oxford] at 07:01 28 August 2017

a
University of Oxford; bThe Graduate Center and Baruch College, CUNY

ABSTRACT
This paper provides a naturalistic account of inference. We posit that the core of
inference is constituted by bare inferential transitions (BITs), transitions between
discursive mental representations guided by rules built into the architecture of
cognitive systems. In further developing the concept of BITs, we provide an account of
what Boghossian [2014] calls ‘taking’—that is, the appreciation of the rule that guides
an inferential transition. We argue that BITs are sufficient for implicit taking, and then,
to analyse explicit taking, we posit rich inferential transitions (RITs), which are
transitions that the subject is disposed to endorse.

ARTICLE HISTORY Received 21 August 2016; Revised 11 July 2017

KEYWORDS inference; thought; rule-following; association

And what is thinking? Well, don’t you ever think? Can’t you observe yourself and see what is
going on? It should be quite simple.
Ludwig Wittgenstein, Philosophical Investigations, x327

1. Methodology and the Analysis of ‘Inference’

Thinking is easy to do, but difficult to understand. Often, our minds are consumed with
cascading thoughts, at least some of which are fully conscious. Yet we’re hard pressed to
say exactly what it is that we’re doing when we’re thinking. As far as philosophy of
mind and cognitive science are concerned, the question ‘What is thinking?’ decomposes
into at least two questions. What kinds of mental states are thoughts? In what kinds of
mental transitions do those states figure? This paper is concerned with the latter
question.
At least one kind of transition between thoughts—perhaps the fundamental kind—is
inference. There are at least two main ways of approaching inquiry into the nature of
inference. According to one approach, we should ask how inferential transitions suc-
cessfully transmit epistemic warrant between thoughts. This epistemological approach
fixes the referent of ‘inference’ by means of a description such as this: whatever type(s)
of mental transition(s) are apt to transmit epistemic warrant from premises to conclu-
sions. This approach fosters interest in active, reflective, and conscious aspects of
human reasoning, particularly given internalist epistemological assumptions on which

© 2017 Australasian Association of Philosophy

 2 JAKE QUILTY-DUNN AND ERIC MANDELBAUM

consciousness, reflection, and/or personal activity are necessary for the transmission of
warrant (for instance, Boghossian [2014] and Valaris [2014]).1
A second approach is to ask what the psychological parameters of inferential transi-
tions are, irrespective of how (or whether) they successfully or unsuccessfully transmit
warrant. Those adopting this naturalistic approach seek to specify the purely descrip-
tive features of inference as a type of logical and reason-responsive transition between
mental states. Part of this project is to demonstrate how inference differs experimen-
tally from other types of transitions, such as associative transitions or noninferential
Downloaded by [the Bodleian Libraries of the University of Oxford] at 07:01 28 August 2017

computations. Cognitive scientists have been interested in such a notion of inference to

account for the reason-responsiveness of propositional attitudes, which cannot be
understood in purely associationist terms [De Houwer 2009; Mandelbaum 2016]. This
psychological approach focuses attention on the involuntary, unconscious, and perhaps
normatively degenerate aspects of inferential transitions as they’re empirically shown
to occur in cognition (for example, Kornblith [2012]; for a similar but more a priori
approach, see Richard [forthcoming]).
Nonetheless, there are few philosophers today who use empirical results to charac-
terize inference as a natural kind. This is no accident: some philosophers in the first
camp think that inference is inherently non-naturalizable [Boghossian 2014], in which
case the project of situating inference in a scientific taxonomy of mental transitions
looks hopeless. The data, however, may call for inferential transitions to be enumerated
alongside other sorts of empirically tractable mental operations.
Our project in this paper is to sketch a naturalistic theory of inference. We aim to
isolate a distinctive type of transition between thoughts by appeal to (i) intuitive para-
digm cases, (ii) other cases uncovered or suggested by recent empirical data previously
unmentioned in the literature, and (iii) contrasts with noninferential transitions, espe-
cially associative ones. We’ll use Boghossian [ibid.] as a foil for our naturalistic
approach. Like him, we take the basic phenomenon of inference to be the kind of men-
tal transition typically involved in paradigm cases such as a person’s movement from
thinking (1) that it’s raining, and (2) that if it’s raining then the streets are wet, to think-
ing (3) that the streets are wet. Unlike Boghossian, however, we think that unconscious
inferences are every bit as central to an account of inference as the conscious cases.
There’s a difference between merely going from the thought that p to the thought
that q—via conditioned association, say, or a kind of arbitrary mental ‘jogging’
[Broome 2013: 226]—and undergoing an inferential transition from p to q, even if the
latter doesn’t involve conscious deliberate thought. We’ll argue that the data call for a
characterization of inference as a basic kind of transition between thoughts that’s nei-
ther conscious and deliberate reasoning nor mere mental jogging. Such an account
might not turn out to satisfy certain internalist assumptions about transmission of war-
rant. Indeed, trying to satisfy the internalist desiderata leads Boghossian to conclude
that there’s a hard problem for inference analogous to the hard problem for conscious-
ness. Perhaps this is right, or perhaps those desiderata should be rejected. Our descrip-
tive project is silent on warrant and other normative notions.
To telegraph where we’re going, what we’ll call bare inferential transitions are a spe-
cies of non-associative, rule-governed transitions between thoughts, and what we’ll call

1
Siegel [2017] is an exception, in that she offers an epistemological account that allows for unconscious
inference.
 AUSTRALASIAN JOURNAL OF PHILOSOPHY 3

rich inferential transitions also involve a richer form of ‘taking’ [Boghossian 2014: 3ff]
in the form of explicit endorsement. Throughout, we assume that the constituents of
inferential transitions are structured mental representations with both syntactic and
semantic properties. We’ll argue in section 2 that the right way to understand inference
is in contrast to association, leading to the notion of bare inferential transitions. In sec-
tion 3, we’ll develop the account further, discussing how our view handles different
sorts of inference, including misinference. Then, in section 4, we’ll provide an account
of taking that leads to the notion of rich inferential transitions.
Downloaded by [the Bodleian Libraries of the University of Oxford] at 07:01 28 August 2017

2. Inference and Association

2.1 Dual Systems
Boghossian [ibid.] begins by discussing the dual-systems approach to cognition. ‘System
1’ is a cognitive system responsible for quick, automatic, unconscious, and associative
processes, whereas ‘System 2’ is slow, voluntary, conscious, and rule-based [Evans and
Stanovich 2013]. Boghossian says that he’s interested in ‘System 1.5 and up’ [2014: 2],
meaning a system whose operations are conscious and voluntary but possibly also
quick and automatic without necessarily being effortful and attention-demanding.
One putative difference between the two systems is that transitions in System 1 are
associative while transitions in System 2 are rule-based. The dual-systems claim we care
about here is that fast, automatic, unconscious transitions are associative, while genu-
inely inferential rule-based transitions are slow, reflective, and conscious. The tendency
among philosophers to focus on reflective inference as ‘the Platonic Form’ of inference
[Boghossian 2016: 48] can lead to the idea that unconscious or automatic transitions
are merely associative, an idea that’s bolstered by the dual-systems approach. We ques-
tion the usefulness of both this tendency and the dual-systems approach. Our immedi-
ate question, then, is that of whether inferential transitions must be slow, reflective, or
conscious.
From a common sense perspective, the answer seems to be ‘no.’ Suppose that you’re
so engrossed in Descartes’ Meditations that the door to your apartment opens and
shuts without your consciously registering this. A moment later, when you emerge
from contemplation, you find yourself with the belief, MY ROOMMATE IS HOME.2 It seems
reasonable to suppose that you inferred that your roommate is home from the fact that
someone opened the door. Perhaps this is merely a case of association between the
door opening and your roommate being home. But suppose that your roommate typi-
cally spent her time at her partner’s apartment, and that in fact you more often heard
your door opened by your nosey landlord than by your roommate; nonetheless, on this
particular occasion, you knew that your landlord was out of the country. It seems plau-
sible that you unconsciously access the information that, since your landlord is gone,
anyone who opens the door must be your roommate, and then used that information
to infer that your roommate is home. This interpretation is plausible even though nei-
ther the perception that triggered the inference nor the inferential process itself was
conscious, slow, or reflective.
Stepping outside of common sense, consider the phenomenon of effort justification
[Aronson and Mills 1959]. Marcus endures a harsh regimen of hazing and degradation

2
We follow the convention of using small caps to denote structural descriptions of concepts.
 4 JAKE QUILTY-DUNN AND ERIC MANDELBAUM

to join a fraternity. Afterwards, he feels very positive about the fraternity—indeed, even
more positive than Trevor, who joined a similar fraternity but did not suffer any hazing,
feels about his. The cognitive-dissonance explanation of this case is that Marcus
believes that he isn’t stupid; he also believes that if he underwent hazing to join a group
that was less than excellent, he’d be stupid; so he concludes that the group is excellent.
This unconscious, unreflective process is genuinely inferential. If the processes in
Marcus’s mind were purely associative, then—given that hazing triggers negative feel-
ings and hazing becomes associated with the fraternity—you would expect negative
Downloaded by [the Bodleian Libraries of the University of Oxford] at 07:01 28 August 2017

feelings to be associated with the fraternity. It’s only because Marcus is drawing infer-
ences from his beliefs about himself and the effort he has expended that he ends up
increasing his positive feelings toward the fraternity. Our reasoning isn’t simply that
association and inference are exhaustive, so that anything non-associative is inferential.
This transition isn’t merely non-associative. It involves rational connections between
Marcus’s beliefs that lead him to acquire a new belief and that instantiate a paradig-
matic form of inference (namely, modus tollens).3
Some final examples from the psychological literature on reasoning might help to
drive home the point. There’s evidence that people make deductions without conscious
awareness when those deductions map onto certain inferential forms, particularly
modus ponens [Reverberi et al. 2012]. Subjects who are given a major premise ‘If p then
q’ supraliminally (for 2.5s), and then given the minor premise ‘p’ subliminally (for
50ms, flanked by masks on both sides), have the conclusion q facilitated. However, sub-
jects who also saw ‘If p then q’ supraliminally but saw q subliminally, fail to have the
conclusion ‘p’ facilitated. The latter subjects, those who encounter the affirming-the-
consequent form of the argument, don’t have the conclusion facilitated even though
the relevant concepts (in ‘p’ and q) have been primed from the major premise.
Similarly, in another study subjects read modus ponens arguments and were
instructed to say either whether the conclusion logically followed from the premises, or
whether it was ‘believable’ in light of background knowledge [Handley et al. 2011].
Logic-based judgments were quicker and more accurate than belief-based judgments.
In cases where logical validity and believability diverged (say, the conclusion that a
feather is heavy, following from a valid argument), the conflict hampered speed and
accuracy for belief-based judgments while logic-based judgments were hardly affected
at all. Thus, logic-based judgments occurred automatically, without interference from
other cognitive processes. Contra some versions of dual-systems theory, humans auto-
matically and immediately run inferences from separate statements when those infer-
ences satisfy certain mental logical rules (see also Lea [1995]). Again, these processes
aren’t merely non-associative—they instantiate paradigmatic inferential rules in taking
thinkers from one propositional thought to the next.
It would appear, then, that transitions between thoughts can be rule-based and non-
associative—and hence genuinely inferential—without being slow, reflective, or con-
scious. Boghossian’s ‘System 1.5’ is thus too high a place to start. We need a notion of
inference that’s rule-based and non-associative but without necessarily being conscious
or voluntary. To put this another way: What is the difference between fast, unconscious,
automatic associative transitions and fast, unconscious, automatic inferential transitions?

3
And in fact, when subjects think that they’re stupid (or have lowered self-esteem), effort justification ceases to
change attitudes [Glass 1964], just as one would predict would occur if a premise in a deductive chain of reason-
ing was deleted.
 AUSTRALASIAN JOURNAL OF PHILOSOPHY 5

2.2 Logic and Inference

Using basic principles of associationist psychology as a starting point, we rely on two
key differences between inferential and associative transitions. The first difference,
already mentioned, is that the former are constitutively rule-based and logic-obeying,
while the latter are not. The second is that inference is reason-responsive and thus can
be modified by evidence, while associations don’t respond to reason and instead can
only be modified in certain arational ways. We’ll examine each of these differences in
turn.4
Downloaded by [the Bodleian Libraries of the University of Oxford] at 07:01 28 August 2017

The first key difference is that inferential transitions are rule-based, and obey some
kind of logic (whether that logic is normatively respectable, however, is an open ques-
tion).5 The inference from IT IS RAINING and IF IT IS RAINING THEN THE STREETS ARE WET
to THE STREETS ARE WET operates in accordance with some logical rule—namely, modus
ponens. When two representations are associated, however, the transition from one to
the other isn’t dependent on any logic. You might, perhaps through participation in
some bizarre psychology experiment, come to associate the thought DONALD TRUMP IS
THE AMERICAN PRESIDENT with the thought THE SUN WILL ONE DAY EXPLODE, even though
there’s no logical connection between these two thoughts. The transition is simply an
artefact of an associative process and isn’t due to the semantic or syntactic properties of
the thoughts.
One way that this difference can be cashed out empirically is in how these kinds of
transitions can be changed, which leads to the second difference: inference is responsive
to evidence while association is responsive only to forms of conditioning. If you learn
that bananas are in fact not yellow but red, and you have been the victim of an elaborate
prank in your past experience with them, then you will, ceteris paribus, cease to infer
that something is yellow from its being a banana. Associative transitions, by contrast,
are not amenable to reason. Donald Trump’s presidency bears no interesting relation
to the inevitable explosion of our sun. But being rationally convinced of that fact won’t
suffice to break a pre-existing associative link between those two thoughts. Associative
links are modulated through counterconditioning and extinction. Roughly, if S associ-
ates A and B, one breaks that association not by giving good reason not to associate A
and B, but rather by introducing A without B and B without A. A link between two
concepts that cannot be affected by any amount of extinction or counterconditioning is
ipso facto not an associative link.6
Note that it’s irrelevant whether there actually happen to be rich metaphysical rela-
tions between the states of affairs targeted by the associated representations; what

4
A further difference is that associative links are ideally symmetric: ceteris paribus, activating one associate will
activate the other and vice versa. We lack space to develop this point here (although see Mandelbaum [2017]).
5
Whether the logic is good old-fashioned classical logic, some non-classical logic, or rather a proprietary mental
logic [Braine and O’Brien 1998] is immaterial—although we suspect that mental logic is indeed proprietary and
inconsistent with wide swathes of classical logic (e.g. people don’t seem to reason by the principle of explosion).
6
‘Association’ is a technical term whose meaning is understood against the background of associationist psy-
chology going back to Pavlov [Mandelbaum 2017]. It also has a very loose meaning in ordinary language, which
is best avoided in a careful discussion of the difference between inference, association, and other sorts of transi-
tions. For example, we might wonder whether there’s an innate ‘association’ between spiders and fear. But if this
innate link between SPIDER and the activation of fear cannot be modulated by counterconditioning or extinction, it
simply isn’t an association in the sense that figures in psychology. In fact, these innate associations (such as taste
aversions) were the first empirical counterexamples to associationism (e.g. Garcia and Koelling [1966]). If it turned
out that no structures were modulable by counterconditioning and extinction, then we’d have to conclude that
there were no associative structures.
 6 JAKE QUILTY-DUNN AND ERIC MANDELBAUM

matters is whether the structural or semantic relations between the representations are
causally operative in the transitions. Suppose, for instance, that you eat a banana every
morning, and spend evenings contemplating tomorrow’s breakfast. After a while, you
come to associate I WILL EAT A BANANA TOMORROW with I WILL EAT SOMETHING YELLOW
TOMORROW, so that thinking the former causes a non-inferential associative transition
to the latter. Given that bananas are yellow, there’s a non-arbitrary relationship
between those contents, and it’s originally responsible for the association. Once the
associative tie is established, however, it operates independently of logic.
Downloaded by [the Bodleian Libraries of the University of Oxford] at 07:01 28 August 2017

Finally, we say ‘logic-obeying’ in addition to ‘rule-based’ because not every rule is a

logical one. One might construct a representational system that moves from IT IS RAIN-
ING to THE STREETS ARE WET as a built-in routine. This transition may be rule-based and,
if it’s not modifiable through extinction or counterconditioning, it’s not associative. But
the rule, If IT IS RAINING then THE STREETS ARE WET, isn’t a logical rule. What distinguishes
logical rules from other sorts of rule is that they abstract away from specific non-logical
semantic contents and instead describe formal properties. It’s famously difficult to say
precisely what formality consists in [Beall and Restall 2006: 18–26], but it seems to be a
matter of the structure of representations irrespective of what things, properties, or
relations the representations are about. Sticking to modus ponens as our paradigm case
of an inferential rule, subjects infer according to it, regardless of content. Thus, one of
our footholds in characterizing inferential transitions is not only that they’re rule-
based, but also that the relevant rules are logical. There are rule-based non-associative
processes in early perceptual systems, for example, which aren’t genuine inferences.
Accounts like Kornblith’s, which take ‘transitions involving the interaction among rep-
resentational states on the basis of their content’ [2012: 55] to be inferential, thus miss
out on a joint in nature between mere rule-based transitions between representations
and logical-rule-based inferential transitions between thoughts.
So, in short, we need a notion of inferential transitions that captures the fact that
they constitutively obey some logic and respond to reasons. These conditions can be
met if we take inferential transitions to be transitions that are sensitive to the constitu-
ent structure of representations.
Constituent structure is an essential property of representations with a discursive
representational format, as opposed to an iconic format [Fodor 2007; Quilty-Dunn
2016]. Consider the contrast between the sentence ‘Bananas are often yellow’ and a pic-
ture of a yellow banana. The sentence is composed of four words, which are the atomic
semantic units of the whole representation, since they aren’t themselves composed of
meaningful representations. ‘Bananas are often yellow’ has at least one canonical
decomposition, or right way of carving it into parts: ‘Bananas’, ‘are’, ‘often yellow’
(which in turn decomposes into ‘often’ and ‘yellow’). There are many wrong ways to
carve the sentence, too—say, ‘Bana’, ‘nas a’, ‘re oft’, ‘en ye’, ‘llow’. A picture of a yellow
banana, by contrast, can be carved up in any way that you like, and the parts that are
separated will still be meaningful iconic representations. Every part of the picture repre-
sents some part of the scene, and, as a result, the representation lacks a canonical
decomposition. The constituents of the sentence are the parts that are individuated in
its canonical decomposition; the picture, on the other hand, has parts but no constitu-
ents. The constituent structure of the sentence is the structure that it has in virtue of
the structural relations in which the constituents stand to each other.
Some argue that mental representations do not have constituent structures. We
don’t intend to get embroiled in this debate here (although see Fodor [1975]; Goodman
 AUSTRALASIAN JOURNAL OF PHILOSOPHY 7

et al. [2015]; Mandelbaum [2016]; and Quilty-Dunn and Mandelbaum [ms.]). For pres-
ent purposes, however, we simply appeal to the explanatory virtues of our own account.
There is, we argue, a logical rule-based character to inferential transitions, and an
account in terms of the constituent structure of mental representations can explain this
feature of inference. Even if one is not swayed by independent evidence in favour of the
structured mental representation hypothesis, we aim to motivate the hypothesis by pro-
viding a successful account of inferential transitions that presupposes it.
If transitions between thoughts are sensitive to constituent structure, those transi-
Downloaded by [the Bodleian Libraries of the University of Oxford] at 07:01 28 August 2017

tions must obey some logic. This is true because a logical rule just is a kind of rule that
is sensitive to constituent structures. For instance, suppose that a rule of mental logic is
the following: If X is an AN, then X is an N. Suppose, further, that you token the
thought BERTHA IS A BROWN COW. You will then, ceteris paribus, token the thought BER-
THA IS A COW. This transition is logical because it occurs in virtue of the fact that the
constituent structure of the input representation satisfies the antecedent of the rule,
and the output is generated because its constituent structure satisfies the consequent of
the rule. Transitions between discursive representations that are triggered because their
constituent structures instantiate some rule of mental logic thus suffice to make those
transitions rule-based and logic-obeying.
Furthermore, such transitions enable the reason-responsiveness of inference. Sup-
pose that a thinker has a belief that G(x). If she acquires evidence that leads to the belief
that F(x) and that if F(x) then not-G(x), then the constituent structures of the acquired
beliefs are such that they logically entail not-G(x). Assuming that her mind is con-
structed in such a way that she infers according to modus ponens, these facts explain
how she’ll come to revise her initial belief that G(x).

2.3 Bare Inferential Transitions

The above discussion presupposes the psychological reality of logical rules that pertain
to the constituent structure of discursive mental representations. There are, however,
familiar issues about how such rules are psychologically realized. It must not be the
case that these rules are always explicitly represented (meaning that they’re the contents
of some mental representation in the system).7 As Lewis Carroll [1895] showed, that
condition would generate a vicious regress. Instead, we propose that the basic rules of
mental logic are built into the architecture. A rule is built into the architecture of a rep-
resentational system iff whenever a mental representation is tokened that satisfies the
antecedent of the rule, then, ceteris paribus, the system will token a representation that
satisfies the consequent of the rule.
The ceteris paribus clause is important because there will inevitably be cases in which
the transition doesn’t successfully occur. Given that the rules are built into the architec-
ture, the ceteris paribus clause can only be violated in certain ways. Unlike explicit rules,
for example, those built into the architecture cannot have an intervening intentional
state be the reason why ceteris isn’t paribus. Take the practical syllogism: if you desire
that Q and you believe that doing P will bring about Q, then ceteris paribus you’ll do P.
A ceteris paribus clause here may be invoked merely because one also believes that
doing P will bring about R and one desires not-R. In contrast, however, no such inten-
tional factors could shortcut processes built into the architecture. A rule built into the

7
‘Explicit’, as we use it, doesn’t mean ‘conscious’. An explicit representation is just a concrete mental token,
which may be conscious or may be unconscious.
 8 JAKE QUILTY-DUNN AND ERIC MANDELBAUM

architecture specifying If P then Q won’t be shorted because of a belief that Q will lead
to R and a desire that not-R. Roughly speaking, the ceteris paribus clause of rules built
into the architecture will be invoked only by variables at a level or more ‘below’ inten-
tional psychology (such as processing constraints on memory, architectural boundaries,
or neurological snafus).
Rules that are built into the architecture, therefore, are propositions that accurately
describe all transitions within the relevant system in possible worlds where the relevant
ceteris paribus clause is not violated. Our notion of rule-following is thus a functionalist
Downloaded by [the Bodleian Libraries of the University of Oxford] at 07:01 28 August 2017

one: a system’s following a rule is a matter of the truth of counterfactuals that specify
transitions between mental representations. We do not presuppose any more robust
notion of rule-following. Any system that is constructed such that the proposition that
p accurately describes transitions in worlds where ceteris paribus clauses are not vio-
lated is a system that has the rule p built into its architecture. Below, we characterize
performance errors in a way that allows a substantive characterization of relevant cete-
ris paribus clauses. For now, what matters is that the psychological reality of rules built
into the architecture of a system is a matter of how the system would move from one
representation to another in cases that do not involve performance errors or otherwise
violate ceteris paribus clauses.
Not every rule built into the architecture of some mental system is necessarily a rule
of inference, nor need every rule instantiate some logical principle. We argued above
that inferential transitions are essentially transitions between discursive representations
in virtue of their constituent structure. Thus, an inferential rule is built into the archi-
tecture iff whenever a mental representation is tokened whose constituent structure sat-
isfies the antecedent of the rule, then, ceteris paribus, the system will token a
representation whose constituent structure satisfies the consequent. As mentioned, any
cognitive system will exhibit ideal regularities such that token representations of one
type lead, barring intervening factors, to token representations of another type. Our
claim is that, when those representations have a discursive format and the regularities
pertain solely to the constituent structure of those representations, the transitions
between them are inferential.
Finally, one might assume that the relevant representations must be not only discursive
but fully propositional, given the reasonable assumption that inferences must operate over
truth-apt (and hence propositional) representations. One might quarrel with this and
argue that the transition from thinking BROWN COW to thinking COW counts as inferential
despite lacking any propositional structure. We will sidestep this debate and simply use the
term ‘discursive’, while assuming that at least the paradigm cases are fully propositional.
The foregoing furnishes us with a simple account of inferential transitions:

(1) The transition from state A to state B is inferential iff (i) A and B are discursive,
(ii) some rule is built into the architecture such that A satisfies its antecedent in
virtue of A’s constituent structure and B satisfies its consequent in virtue of B’s
constituent structure (modulo logical constants), and (iii) there is no intervening
factor responsible for the transition from A to B.8

8
We add ‘modulo logical constants’ because BITs will be sensitive to elements of thoughts that aren’t purely
syntactic, such as IF and THEN in a conditional, negation, etc. Since logical constants can be given narrow identity
conditions, and since the only semantics involved is that of logical constants, this condition doesn’t undermine
the formal computational character of BITs.
 AUSTRALASIAN JOURNAL OF PHILOSOPHY 9

We’ll call transitions as described in (1) bare inferential transitions, or BITs. We’ll make
three remarks on rules and BITs before moving on.
First, the notion of BITs doesn’t rely on a primitivist notion of rule-following. We
don’t (contra Boghossian) take there to be some special, fundamental relation that
thinkers stand in to a certain rule when inferring in line with it, nor do we think the
rule must be the content of some intentional state, such as a belief. Because we reject
the taking condition as necessary for BITs, it’s enough for our purposes that the rule
accurately describes ceteris paribus regularities of transitions in the cognitive sys-
Downloaded by [the Bodleian Libraries of the University of Oxford] at 07:01 28 August 2017

tem—in other words, that it’s built into the architecture. BITs need only conform to
the rule in a particular way. It’s not sufficient for a transition to count as inferential
that it can be loosely described as following some rule, since one could even
perversely describe associative spreading between two logically related thoughts as
following a rule. Of course, this associative transition would not be inferential. It
must be a feature of the cognitive system that putting in representations with one
type of constituent structure will, all else being equal, result in representations with
another type of constituent structure. Neither the system, nor the thinker, nor any
of the intentional states figuring in the transition need represent or otherwise follow
the rule in some more robust sense. The rule needs only to be built into the architec-
ture, such that representations in the system will ceteris paribus act in accordance
with it.
Second, we’ll put aside sceptical worries about rules, as canonically articulated by
Kripke’s [1982] Wittgenstein. We don’t see much cause for despair in the face of these
worries. It’s important for cognitive science to have some account of inferential transi-
tions that are unconscious, fast, and automatic, for we know that people do have
unconscious, fast, automatic transitions that aren’t associative transitions but are sensi-
tive to logical form. An account that analyses these transitions needs to be developed
even if, ultimately, there will be some indeterminacy in the rules that are built into the
architecture.9
Third, the notion of a rule being built into the architecture is not a mere appeal to
dispositions. It is, crucially, a counterfactual notion: in a world where there are no
performance errors, the rule will accurately describe every transition within its
scope. A mind can have such a rule built into it, even if the rule accurately describes
only a small percentage of the transitions that mind is disposed to make in the
actual world, due to systematic performance errors. A system can be disposed to
make transitions in line with a rule without having that rule built into its architec-
ture, and a system can have the rule built into its architecture without making tran-
sitions in line with it with any statistical regularity. The mere possession of the
disposition to accord with a rule is neither necessary nor sufficient for its being built
into the architecture.10

9
Note that there being some indeterminacy is very different from the more radical Kripkensteinian claim that
there’s no naturalistic way of delineating performance errors as opposed to instances of following some bizarre
rule.
10
There’s a very loose reading of ‘disposition’, on which our account might be dispositional simply in virtue
of our appeal to counterfactual support. But, on that loose reading, even an arch-representationalist and
anti-dispositionalist like Fodor [1975] provides a dispositional account. We have in mind the more robust
notion of disposition employed in, for example, Schwitzgebel’s [2002] account of belief (cf. Quilty-Dunn and
Mandelbaum [ms.]).
 10 JAKE QUILTY-DUNN AND ERIC MANDELBAUM

3. Problem Cases: Semantic Entailment, Misinference, and Induction

3.1 Semantic Entailment
BITs are governed by structural features of the representations that figure in them and
by the rules that are built into the architecture pertaining to those structural features.
One might reasonably object that this story can only apply to syntactic entailment, or
entailment in virtue of the syntactic (constituent) structure of premises and conclu-
sions, not to semantic entailment, or entailment in virtue of the contents of premises
Downloaded by [the Bodleian Libraries of the University of Oxford] at 07:01 28 August 2017

and conclusions.
An example of semantic entailment is someone’s inferring from the fact that an
apple is red to the fact that it’s coloured. Thus described, the transition is not due to
structural features of the premises and conclusions. The structures involved are simply,
X IS Y and X IS Z. The rule If X is Y then X is Z is clearly not built into the architecture—
if it were, then predicating any property of something would cause you to predicate, of
that thing, every other property that you can represent. Instead, the contents of RED
and COLOURED are semantically related such that the inference is valid even though the
general schema is not. This sort of fact suggests a distinction between semantic entail-
ment and syntactic entailment.
We think that cases of semantic entailment will branch into cases of syntactic entail-
ment (which are genuine inferential transitions) and noninferential associative transi-
tions; there’s no category of transition called ‘semantic entailment’ that’s both
genuinely inferential and not due to constituent structure. There can clearly be transi-
tions from APPLES ARE RED to APPLES ARE COLOURED that are noninferential. Just as
semantically unrelated thoughts like DONALD TRUMP IS THE AMERICAN PRESIDENT and THE
SUN WILL ONE DAY EXPLODE can become associated, thoughts that happen to be semanti-
cally related, like APPLES ARE RED and APPLES ARE COLOURED, could become associated.
For example, one might have a stored association between RED and COLOURED; so,
thinking APPLES ARE RED will trigger an associative transition to APPLES ARE COLOURED
without being mediated by inference [Mahon and Caramazza 2003].
The question at hand, then, is this. What makes it the case that a transition from
APPLES ARE RED to APPLES ARE COLOURED is genuinely inferential? Intuitively, you cannot
infer that apples are coloured from the fact that apples are red, unless you know that
red things are coloured and you employ that knowledge in the transition. This intuitive
requirement on inference fits comfortably with (1). What it means for you to know
that red things are coloured and to employ that knowledge, we assume, is in part for
you to have the thought IF X IS RED THEN X IS COLOURED. So, for this semantic entailment
to count as an inference, you have to think both APPLES ARE RED and IF X IS RED, THEN X
IS COLOURED. These premises provide an instance of the antecedent of the rule, If F(x),
and if F(x) then G(x), then G(x). So, at the point when both thoughts are tokened simul-
taneously, the structure-sensitive architecture takes over, and the system delivers the
thought APPLES ARE COLOURED. This transition is simply a BIT. We hypothesize that
cases of semantic entailment that are genuinely inferential transitions will involve an
explicit representation of the entailment, thereby satisfying the antecedent of some rule
that’s built into the architecture, resulting in a BIT. One might reply that we have sim-
ply stipulated that only the BIT cases of semantic entailment, and none of the associa-
tive cases, count as inferential. But we arrived at this position by seeing what
independent way there is to distinguish inferential semantic entailments from associa-
tive ones: if our proposal is incorrect, there must be some alternate account of this
 AUSTRALASIAN JOURNAL OF PHILOSOPHY 11

distinction. In the absence of such an account, we assume that a semantic entailment is

inferential only if the thinker employs her knowledge of the entailment, and thus only
if it’s a BIT.11
Two competing intuitions conspire against our proposal for how to understand
inferential transitions. The first is that some cases of semantic entailment are genuinely
inferential—surely one can infer that apples are coloured from the fact that they’re red.
The second is that going from the thought that apples are red to the thought that
they’re coloured doesn’t always involve forming the thought that if something is red
Downloaded by [the Bodleian Libraries of the University of Oxford] at 07:01 28 August 2017

then it’s coloured. We can accommodate these intuitions by denying that the cases
overlap. Some cases of semantic entailment are inferential, and some cases don’t
involve explicitly representing the entailment, but no cases are both. Furthermore,
common sense is on our side in so far as it’s common sense to think that moving from
APPLES ARE RED to APPLES ARE COLOURED only counts as inferential if the thinker employs
her knowledge that red things are coloured. The intuition that we can infer without
representing the entailment may arise from associative transitions that develop out of
repeated inferences. The transition continues to feel like an inference even though,
strictly speaking, it has become an association. One might have the intuition that some
semantic entailments are neither BITs nor associative transitions. However, the matter
need not rest on intuition. Our theory has clear empirical commitments: mental transi-
tions that instantiate semantic entailments will either involve explicit representation of
the intervening premise (and thus be BITs), or else they will be modulable through
extinction (and thus be associations).

3.2 Misinference
A theory of inference needs to distinguish inferences from misinferences and other
types of transitions. We take misinference to be a performance error. A person with
aphasia might have competence with English but be unable to produce a sentence,
and so lack the ability to perform linguistically. Competence is a standing state of a
given cognitive system, the state of being disposed to operate in accordance with
certain rules (such as rules that are built into its architecture). In some types of
aphasia (say, Broca’s), the language faculty’s competence is unharmed while neuro-
logical damage prevents the competence from being manifested in linguistic perfor-
mance. We understand performance errors as follows: a performance error, relative
to a given system S that exhibits a particular competence, is a behaviour or mental
event caused either by an intervention by another system that interrupts the normal
functioning of S, or by some factor one or more psychological levels down (perhaps
down to a neural level).
Not every apparent mistake will count as a genuine misinference. For instance, if
one thinks IF P THEN Q and NOT-Q, but one has an association between NOT-Q and P, one
might associatively activate P, even though the modus tollens rule mandates an inferen-
tial transition to NOT-P. This case would be a logically problematic transition, but would
simply be a case of association, not genuine misinference as we understand it.

11
One option between the propositional and the associative poles holds that the concept APPLE functions as a
pointer that enables access to various predicates, such that activating the propositional structure APPLES ARE RED
facilitates access to the predicate COLOURED via the pointer APPLE. We lack space to develop this intriguing possibility
here (but see Green and Quilty-Dunn [forthcoming]).
 12 JAKE QUILTY-DUNN AND ERIC MANDELBAUM

This notion of misinference rules out certain exotic candidate rules from being built
into the architecture (cf. Kripke [1982]). For example, suppose that you typically infer
in line with modus ponens, but being struck on the head at the right time causes you to
move from thoughts of the form F(X) and IF F(X) THEN G(X) to BOB DYLAN IS JESUS. A
sceptic might ask whether the rule If F(X) and IF F(X) THEN G(X) then either G(X) or, if hit
on the head in way W, then BOB DYLAN IS JESUS is built into the architecture. In this case,
the intervening factor is not explicable in psychological terms. Instead, something from
a level down—the neural state that is induced by being hit on the head a certain way—
Downloaded by [the Bodleian Libraries of the University of Oxford] at 07:01 28 August 2017

is responsible for the transition. This case thus constitutes a performance error and
thereby violates the ceteris paribus clause of psychological-level rules built into the
architecture of the relevant cognitive system.

3.3 Induction
Inductive inferences aren’t simply performance errors. So, if they aren’t BITs and they
aren’t performance errors, how should we model them? With respect to induction and
other forms of probabilistic reasoning, one possibility is that they’re really BITs after
all, just probabilistic ones. For what it’s worth, this take appears to be advocated by one
of the few descriptive models of Bayesianism out there. Take Goodman et al. [2015],
who attempt to provide a descriptive account of Bayesian mental processing. Their
account explicitly characterizes the role of mental representations in probabilistic rea-
soning. According to them, probabilistic reasoning relies on a discursive language of
thought in which probability operators play a crucial role. Their account thus seems
wholly compatible with our theory of inferential transitions.
We’re committed to the feasibility of some account such as Goodman et al.’s on
which probabilistic inference such as induction is at bottom a matter of sensitivity to
constituent structure and involves some additional psychological apparatus for evaluat-
ing and computing probability in cases of probabilistic reasoning and induction. On
such accounts, computations of probability operators run alongside inferential transi-
tions between the constituent structures of propositional thoughts to which those oper-
ators attach.

4. The Taking Condition

Our account thus far has focused on BITs, which are structure-sensitive transitions
between discursive representations. BITs may not suffice for inference in the richer
sense required by Boghossian [2014] and his interlocutors (e.g. Broome [2014]). We
think BITs are the kinds of inferential transitions that matter for cognitive-scientific
theorizing, but perhaps don’t satisfy a richer philosophical notion of reasoning
[Harman 1986] as, among other things, a quasi-epistemological and quasi-psychologi-
cal act consciously performed by a rational agent. Boghossian says that genuine infer-
ence satisfies the Taking Condition, according to which a thinker must take the
premises of his inference to support the conclusion [2014: 5]. This matters in part
because, according to Boghossian, understanding inference in this richer sense requires
positing a metaphysically fundamental relation of rule-following, ‘an unanalyzable
primitive’ which cannot be naturalistically accounted for [ibid.: 17]. If one already
accepts primitive rule-following, one might argue that it’s all that we need to account
for inferential transitions in general, and so the notion of BITs won’t do any important
 AUSTRALASIAN JOURNAL OF PHILOSOPHY 13

theoretical work. For our account of inferential transitions to integrate with inference in
the richer sense, we have to sketch an account of the richer kind of inference and how
BITs interact with it.
At the heart of the richer sense of inference is ‘taking’, and BITs seem to do a sub-
stantial chunk of the descriptive work that taking is invoked to do. If a rule is built into
someone’s mental architecture, they needn’t explicitly represent the rule or stand in
any attitudinal relation to it. That’s just as well, for even according to Boghossian taking
doesn’t involve explicit representation [ibid.: 14ff]. Perhaps taking involves some
Downloaded by [the Bodleian Libraries of the University of Oxford] at 07:01 28 August 2017

implicit appreciation of the rule that’s nonetheless more robust than mere accordance
with the rule, and is thus more robust than mere mental ‘jogging’ [Broome 2013: 225]
from premises to conclusions. A rule’s being built into the architecture of a thinker’s
central cognitive system seems sufficient for that thinker to implicitly appreciate that
rule. A BIT that instantiates modus ponens is not a case of mere accordance. One might
associatively move from two thoughts to a third thought, and the thoughts might hap-
pen to be of the form P, IF P THEN Q, and Q, respectively. But in that case the modus
ponens rule was not really involved; the associative transition at most merely accorded
with it. In the case of the BIT between the same thoughts, it’s precisely because modus
ponens is built into the architecture of the thinker’s cognitive system that she draws the
conclusion. The rule thus plays a direct role in producing the conclusion. We aren’t
sure whether this suffices for full-blooded ‘rule following’. BITs nevertheless explain
‘how such a rule could guide a person’s inferences’ [Boghossian 2014: 13], while
explaining why inferential rules don’t guide associations. The core descriptive function
of taking—allowing rules to guide inferences directly and thus to distinguish them
from noninferential transitions like associations—is performed by BITs.12
Other psychological factors considered in the recent inference literature seem more
obviously to go beyond mere BITs. In particular, there does seem to be a difference
between transitions that merely happen and those that one endorses. Imagine a logician
who propounds a radical logic on which modus ponens is invalid. Suppose, nonetheless,
that the modus ponens rule is still built into the architecture of her central cognitive sys-
tem. Thus, when she tokens the thought that apples are red, and the thought that if
something is red then it’s coloured, she’ll trigger a BIT to the thought that apples are
coloured. While she might endorse the premises and the conclusion, she would reject
the transition itself because of her views about logic. There seems to be a real psycho-
logical difference between her case and the case of someone who undergoes the same
transition while accepting the validity of modus ponens. The difference isn’t in their
premises or conclusions (and we can stipulate that both transitions are unconscious).
The difference seems to be that the second person is disposed to endorse the inferential
transition itself, while the radical logician is not. This seems like a meaningful and
important sense in which the one person takes the premises to support the conclusion
while the other doesn’t. Satisfying the Taking Condition in a richer sense, then, is being
disposed to endorse the inferential transition.

12
In addition to the question of whether the rules involved in inferential transitions are followed, there’s the
question of whether the rules are followed by the thinker, or are followed instead by the cognitive system. We
intend talk of rules being built into the architecture to be neutral on whether the rule is thereby followed by the
thinker or just a subsystem of the thinker (or even, as noted, whether it’s ‘followed’ in some more robust sense at
all). Perhaps, since thinking is something that the agent does even when it’s involuntary (e.g. you are involuntarily
thinking about polar bears now that we’ve mentioned them), BITs are things that agents do despite the BITs
being architectural. We lack the space to pursue this question here.
 14 JAKE QUILTY-DUNN AND ERIC MANDELBAUM

We think that the best way to understand the disposition to endorse an inferential
transition is as follows:

(2) A person is disposed to endorse an inferential transition, from F(x) and If F(x)
then G(x), to G(x) iff she is disposed to form the thought F(X) THEREFORE G(X).

What’s doing the heavy lifting in (2) is the THEREFORE concept (see also Neta [2013]).
We think that there must be some concept that represents a relation of support between
Downloaded by [the Bodleian Libraries of the University of Oxford] at 07:01 28 August 2017

facts without being merely conditional. The difference between thinking F(X) THEREFORE
G(X) and thinking IF F(X) THEN G(X) is that the former isn’t neutral on the truth of the
individual propositions. It’s represented as being the case that F(x), and that its being
the case that F(x) provides support (of some kind) for its being the case that G(x). One
might think that if this apple is red then it’s coloured; but one might also think that the
apple is red, so it’s coloured. Presumably people have a THEREFORE concept, which yokes
together facts by representing one as implying the other. The ‘therefore’ thought entails
the conditional thought, but differs from it in not being merely hypothetical. Call infer-
ential transitions that the person is disposed to endorse—in line with (2)—rich inferen-
tial transitions (RITs). While the taking afforded by BITs is implicit taking, the richer
form of taking afforded by RITs is explicit taking.
Note that neither BITs nor RITs need be conscious. We see this as a virtue of the
present account. For one thing, if RITs do special epistemic work, then the fact that
they need not be conscious makes them compatible with epistemological theories that
allow unconscious mental states and processes to do epistemic work (for example,
Siegel [2017]). And while some aspects of inference are surely intentional—deciding
what to think about, how much attention to devote, whether to allow your mind to
wander—our account entails that the actual inferential transitions from premise
thoughts to conclusion thoughts are determined by the cognitive architecture, and as
such aren’t intentional acts. This seems to us to be intuitively the right answer. It’s hard
to imagine a mind like ours that could activate thoughts of the form p and if p then q,
attend to them without distraction (and lack relevant neuropathology, resource deple-
tion, emotional manipulation, etc.), and yet decide not to draw the conclusion that q.

5. Conclusion
The notion of inference is foundational in cognitive science and philosophy, so it’s
refreshing to see a revival of philosophical analyses of inference. However, we think
that recent analyses of inference have been overly intellectualized, and thus obscure the
vast role that inference plays in our cognitive economy (particularly in unconscious
cognition). We have tried to do justice to the notion of inference by highlighting the
dual roles that it plays. On the one hand, it’s a central and distinctive way in which we
move between thoughts, which is often completely unconscious, involuntary, and
understood in contrast to associative transitions. On the other hand, it can be under-
stood as a process that involves explicitly taking a set of premises as reasons for believ-
ing some conclusion.
We haven’t accounted for inferential transmission of warrant. We believe, however,
that any such story should proceed on the basis of a naturalistic descriptive account of
psychological reality rather than by allowing a priori epistemology to dictate philoso-
phy of mind. There’s a deep methodological divide in philosophical theorizing between
 AUSTRALASIAN JOURNAL OF PHILOSOPHY 15

those who take normative factors to be primary and those who think that descriptive
questions must be answered before we can see which norms actually constrain us and
which turn out to be beyond our limited capacity. Fodor asks, in a similar vein, how
one could be ‘bound by norms that one is, in point of nomological necessity, unable to
satisfy?’ [2007: 115]. We don’t make the bold statement here that internalist theories of
inferential warrant transmission are false or incompatible with our theory. Perhaps
there’s a way of squaring BITs with internalism, or perhaps there are other mental
capacities that fit better with internalist construals of epistemic norms. But we don’t
Downloaded by [the Bodleian Libraries of the University of Oxford] at 07:01 28 August 2017

think any descriptive account, particularly one committed to naturalistic intelligibility

and explanatory continuity with cognitive science, should be thrown out simply on the
ground of a lack of fit with a priori epistemological theories.
Descriptively speaking, there are many aspects of conscious reasoning that our
account doesn’t explain. Thinking through a problem, even in mundane circumstances,
involves many complex operations, such as searching through vast stores of thoughts
and concepts, evaluating probabilities, drawing analogies, imagining various states of
affairs in both propositional and sensory imagination, activating some thoughts and
concepts and not others, integrating desires and goals with beliefs, activating episodic
memories, and doubtless other kinds of mental processes including some yet to be
named. Some of these processes are partly understood, and others aren’t understood at
all. Some might be conscious intentional acts, and others might be unconscious auto-
mated processes. Conscious human reasoning is a variegated ‘mental chaos’ [Siegel
2017: 99] that integrates many distinct processes in ways that are opaque to introspec-
tion. Providing a complete account of conscious human reasoning is an extremely
ambitious project, of which an analysis of inference is only one part. Nonetheless, infer-
ential transitions between thoughts are a central facet of human thinking and, we sus-
pect, a central part of what makes for rational thought. Despite our failing to fully
capture the mental chaos of conscious reasoning, we have given an account of the core
of inference, one that separates inference from other sorts of transitions and respects its
essentially logical nature.
Most of all, we think that we have accurately described the phenomenon while also
responding to a deep worry that Boghossian poses. He writes [2014: 18]
If the present account of reasoning is along the right lines, it opens up the possibility that rea-
soning poses as much of a challenge to a naturalistic worldview as does consciousness. It makes
it difficult to see what naturalistic process inference could consist in.

By introducing the BIT notion of inferential transitions, we have aimed to secure the
groundwork for a naturalistic theory of inference, one that helps to illuminate how rea-
soning works, without adding to our hard problems.13

Funding
This paper was conceived and executed while JQD and EM were on EM’s PSC-CUNY Award 67331-00
45. PSC CUNY is hereby thanked for their largesse.

13
Helpful comments and criticisms were offered by Tim Bayne, Joseph Benda~na, Jake Berger, Paul Boghossian,
Ryan DeChant, Zoe Jenkin, David Papineau, Jesse Rappaport, David Ripley, Susanna Siegel, and the NYU Con-
sciousness discussion group. Special thanks to the editor and referees at AJP for their useful and open-minded
comments.
 16 JAKE QUILTY-DUNN AND ERIC MANDELBAUM

References
Aronson, E. and J. Mills 1959. The Effect of Severity of Initiation on Liking for a Group, The Journal of
Abnormal and Social Psychology 59/2: 177–81.
Beall, J.C. and G. Restall 2006. Logical Pluralism, Oxford: Clarendon Press.
Boghossian, P. 2014. What Is Inference? Philosophical Studies 169/1: 1–18.
Boghossian, P. 2016. Reasoning and Reflection: A Reply to Kornblith, Analysis 76/1: 41–54.
Braine, M.D.S. and D.P. O’Brien, eds, 1998. Mental Logic, Mahwah, NJ: Lawrence Erlbaum Associates.
Broome, J. 2013. Rationality through Reasoning, Malden, MA: Wiley Blackwell.
Broome, J. 2014. Comments on Boghossian, Philosophical Studies 169/1: 19–25.
Downloaded by [the Bodleian Libraries of the University of Oxford] at 07:01 28 August 2017

Carroll, L. 1895. What the Tortoise Said to Achilles, Mind 4/14: 278–80.
De Houwer, J. 2009. The Propositional Approach to Associative Learning as an Alternative for Associ-
ation Formation Models, Learning & Behavior 37/1: 1–20.
Evans, J.St.B.T. and K.E. Stanovich 2013. Dual-Process Theories of Higher Cognition: Advancing the
Debate, Perspectives on Psychological Science 8/3: 223–41.
Fodor, J. 1975. The Language of Thought, Cambridge, MA: The MIT Press.
Fodor, J. 2007. The Revenge of the Given, in Contemporary Debates in Philosophy of Mind, ed. B.P.
McLaughlin and J. Cohen, Oxford: Blackwell: 105–16.
Garcia, J. and R.A. Koelling 1966. Relation of Cue to Consequence in Avoidance Learning, Psycho-
nomic Science 4/1: 123–4.
Glass, D.C. 1964. Changes in Liking as a Means of Reducing Cognitive Discrepancies between Self-
Esteem and Aggression, Journal of Personality 32/4: 531–49.
Goodman, N.D., J.B. Tenenbaum, and T. Gerstenberg 2015. Concepts in a Probabilistic Language of
Thought, in The Conceptual Mind: New Directions in the Study of Concepts, ed. E. Margolis and S.
Laurence, Cambridge, MA: The MIT Press: 623–54.
Green, E.J. and J. Quilty-Dunn forthcoming. What Is an Object File? British Journal for the Philosophy
of Science.
Handley, S.J., S.E. Newstead, and D. Trippas 2011. Logic, Beliefs, and Instruction: A Test of the Default
Interventionist Account of Belief Bias, Journal of Experimental Psychology: Learning, Memory, and
Cognition 37/1: 28–43.
Harman, G. 1986. Change in View: Principles of Reasoning, Cambridge, MA: The MIT Press.
Kripke, S.A. 1982. Wittgenstein on Rules and Private Language: An Elementary Exposition, Cambridge,
MA: Harvard University Press.
Kornblith, H. 2012. On Reflection, Oxford: Oxford University Press.
Lea, R.B. 1995. On-line Evidence for Elaborative Logical Inferences in Text, Journal of Experimental
Psychology: Learning, Memory, and Cognition 21/6: 1469–82.
Mandelbaum, E. 2017. Associationist Theories of Thought, The Stanford Encyclopedia of Philosophy
(Summer 2017 edition), ed. E.N. Zalta, URL = https://plato.stanford.edu/archives/sum2017/entries/
associationist-thought
Mandelbaum, E. 2016. Attitude, Inference, and Association: On the Propositional Structure of Implicit
Bias, No^us 50/3: 629–58.
Mahon, B.Z. and A. Caramazza 2003. Constraining Questions about the Organisation and Representa-
tion of Conceptual Knowledge, Cognitive Neuropsychology 20/3–6: 433–50.
Neta, R. 2013. What Is an Inference? Philosophical Issues 23/1: 388–407.
Quilty-Dunn, J. 2016. Iconicity and the Format of Perception, Journal of Consciousness Studies 23/3–4:
255–63.
Quilty-Dunn, J. and E. Mandelbaum ms. Against Dispositionalism: Belief in Cognitive Science.
Reverberi, C., D. Pischedda, M. Burigo, and P. Cherubini 2012. Deduction without Awareness, Acta
Psychologica 139/1: 244–53.
Richard, M. forthcoming. Is Reasoning a Form of Agency? in Reasoning: Essays on Theoretical and
Practical Thinking, ed. M. Balcerak-Jackson and B. Balcerak-Jackson, Oxford: Oxford University
Press.
Schwitzgebel, E. 2002. A Phenomenal, Dispositional Account of Belief, No^ us 36/2: 249–75.
Siegel, S. 2017. The Rationality of Perception, Oxford: Oxford University Press.
Valaris, M. 2014. Reasoning and Regress, Mind 123/489: 101–27.
Wittgenstein, L. 1953 (1958). Philosophical Investigations, trans. G.E.M. Anscombe, 2nd edn, Oxford:
Basil Blackwell.