Argumentation (2013) 27:167–181

DOI 10.1007/s10503-012-9269-4

A Priori Abduction

David Botting

Published online: 15 April 2012

 Springer Science+Business Media B.V. 2012

Abstract While ‘‘All events have a cause’’ is a synthetic statement making a

factual claim about the world, ‘‘All effects have a cause’’ is analytic. When we take
an event as an effect, no inference is required to deduce that it has a cause since this
is what it means to be an effect. Some examples often given in the literature as
examples of abduction work in the same way through semantic facts that follow
from the way our beliefs represent those effects; from this we may deduce not only
that it has a cause, but what that cause is.

Keywords Abduction  Inference to the best explanation  Causal reasoning 

Coherence

1 Two Kinds of Abduction

In the literature on abduction and inference to the best explanation, two kinds of
example are often given but treated as if they were the same. One is the kind of
which the Semmelweiss case can be taken as the exemplar: there is some surprising
fact to explain which a number of hypotheses are generated and then a preference
ordering imposed on the grounds of ‘explanatory virtue’, the most virtuous of which
is taken (in some guarded sense of ‘taken’) to be true. This is, I think, better
described as inference to the best explanation than abduction, but I don’t wish to
argue this terminological point. I intend to deal with this kind of example elsewhere.
The other kind does not seem to require the fact to be especially surprising, but
only that it be a sign of something else, e.g., the track of an animal, the symptoms of
a disease. Here there is very often no generation of a number of hypotheses then to

D. Botting (&)
Institut de Filosofie da Linguagem, FCSH, Universidade Nova de Lisboa, 1069-061 Lisbon,
Portugal
e-mail: davidbotting33@yahoo.co.uk

123
168 D. Botting

be whittled down because the alternative explanations are independently unlikely;

for example, it does not seem in general that we generate, and then discard, the
hypothesis that we could be the victims of an elaborate hoax. Nevertheless, the
reasoning is often reconstructed as if inference to the best explanation has taken
place—the best explanation of why patient P is exhibiting the symptoms of disease
D is because P is suffering from disease D rather than because they are simulating
their symptoms in order to attract attention (although this obviously happens
sometimes). It is this kind of example—that might be called reasoning with causal
concepts—with which this paper is concerned. To reconstruct such reasoning as
inference to the best explanation, or for that matter to inductive inference, is, I will
argue, to misdescribe it, or at least—there being no guarantee that real reasoners
will reach their conclusions in the simplest way possible—it seems to make the
inference appear more complex than necessary and more worthy of special attention
than it probably deserves. In fact, it is only in a very emaciated sense of inference
(the sense in which there is sometimes said to be a ‘materially valid inference’ from
‘‘Tom is a bachelor’’ to ‘‘Tom is unmarried’’) that it should be reconstructed as
performing any kind of inference at all. I call such reasoning an a priori abduction.
I want to avoid verbal issues whether this kind of reasoning is really an ‘‘abduction’’
and I certainly do not claim that everything that has been called an abductive
inference is of this kind; I claim only that such reasoning has been called
abductive.1 By reducing the inference to triviality it might be argued that I remove
the philosophical interest from it, and indeed, I expect the vast majority of my
argumentation to strike the reader as so obvious as to be not worth saying, but the
claim or demonstration that something is trivial is not itself a trivial claim or
demonstration. Besides, examples of this kind, of which the following is typical,
remain prevalent in the literature.

2 The Example

Here is the example. Henry likes to hike. He likes to get back to nature and is
familiar with the sounds, smells, and sights of the forest, and can identify their
sources: the bird by its song, the flower by its fragrance, and the animal by its tracks.
Out hiking one day Henry sees a track and thinks ‘‘Aha! That is a bear track! A
bear has passed this way recently.’’ What is the relationship between these two
statements?
Many accounts seem to say that there is an inference from effect to cause and/or
from sign to thing signified that the inference from ‘‘That is a bear track’’ to ‘‘A bear
has passed this way recently’’ exemplifies. If there is such an inference, then the
question is whether this inference is deductive, inductive, or of some distinct third
kind that may be labelled abductive.
1
Nor do I wish to take any side on whether ‘‘abduction’’ and ‘‘inference to the best explanation’’ should
be taken to name the same thing. Since reasoning with causal concepts does not involve the explicit
elimination of hypotheses or appeal to explanatory virtues, I prefer to call it a kind of abduction, but it
would be equally valid to consider a priori abduction distinct from both or even to consider it a kind of
deduction, since being a priori it is deductively (materially rather than formally) valid.

123
A Priori Abduction 169

2.1 The First Analysis

Walton et al. (2008, 170) say that it is an abductive inference: ‘‘The given data are
the bear tracks, and the best explanation of the data is that a bear passed this way
producing the tracks…. Reasoning backward from the given data of the tracks, we
infer a causal explanation for how the tracks came about.’’2 For them, abductive
inference is neither deductive nor inductive but a sui generis inference.

2.2 The Second Analysis

Fumerton argues that abduction or inference to the best explanation is not some
third, distinct kind of inference but can always be reduced to an enumerative
inductive argument such as: (1) In all (or most) cases in which we have observed
bear tracks there were bears present just prior to the existence of the tracks; (2) Here
is another case of bear tracks; (3) Therefore, a bear was present just prior to the
existence of these tracks. According to Fumerton (1980), the inductive scheme
explains why we do not infer to some more specific causal hypothesis. For instance,
the passage of some particular bear—call it Halle—would causally explain the
presence of the tracks just as well as some unspecified bear, yet Henry does not, on
perceiving the bear track, say ‘‘Halle has passed this way recently’’ unless he has
some additional reason to rule out alternative bears.

2.3 Comparison and Critique

Caution must be shown about what is being claimed here. It might be claimed that
IBE and enumerative induction are (i) the same form of inference differently
expressed; (ii) different forms of inference but with a common form of justification;
or (iii) different forms of inference with different justifications. Walton et al. (2008,
170–172) seem to be arguing for (iii) since they take abduction to be a species of
presumptive argument, itself a genus of argument distinct from deductive or
inductive arguments. Fumerton’s position is a little more difficult to pin down. If his
analysis is meant, as he seems to say at first, as a reduction of the abductive
argument then he is arguing for (i), but if he is only showing that there is a logically
or epistemically prior inductive argument that authorizes this use of abduction then
2
It is possible that in this example the authors are misspeaking and would not deny what I will shortly be
claiming, namely that there is an a priori connection between the two statements in question. I am
prepared to be charitable to Walton et al. on this score. However, this is not the only place this kind of
example has been put forward. Are the authors of all of these simply misspeaking?
The inference with which we are concerned and that Walton et al. take themselves to be illustrating is
from ‘‘That is a bear track’’ to ‘‘A bear has passed this way.’’ I concede that the addition ‘recently’ may
require an inference. I also remain agnostic about how one comes to believe ‘‘That is a bear track’’ to be
true in the first place. A fashionable view is that there is an inference of some kind between the perceptual
experience and the corresponding belief, and that this inference is abductive. Equally fashionable is the
view that there is an (often non-demonstrative) inference from what people say to what we understand
them as meaning. These are issues in the philosophy of perception and language respectively about which
I have nothing to say. Neither is relevant to my purposes here, which is to make logical points about the
nature of the illustrated inference and not about the status of the premises.

123
170 D. Botting

he is arguing for (ii).3 I think that his position is that since they have the same form
of justification then there is good but not conclusive reason to think that IBE is
enumerative induction differently expressed, i.e., (i), but that (ii) is true if (i) is not.
Hence, he remarks that although proving that IBE and induction lead to the same
inferential consequences does not necessarily imply that IBE cannot be an
independent kind of reasoning ‘‘we have surely cast even more doubt on what must
be an initially implausible view, namely that we can have a priori knowledge that
theories exemplifying some set of properties, C, are more likely to be true than not’’
(Fumerton 1980, 598). Fumerton seems to be saying that if an inference can be
grounded on induction, and we need to ground at least some inferences on
induction, then we should ground the inference in question on induction as well.
This is curious in a couple of ways. Firstly, it begs the question against someone
like Harman (1965) who would claim that induction collapses into IBE—if IBE and
induction ground exactly the same set of inferences then you could just as equally
argue that IBE is the only form of non-demonstrative inference and comprises the
common justification. However, analogously to Fumerton, Harman himself can be
interpreted in two ways: either as arguing that induction reduces to inference to the
best explanation or that the inductive inference depends on a prior inference to the
best explanation. When Harman says that we qualify the inductive inference
depending on whether we have decided that we have a fair sample it seems to be the
second of these that is meant. However, it is the first view that is usually (though I
am inclined to think wrongly) attributed to Harman. Secondly, it almost seems as if
Fumerton is saying that induction is a better explanation than IBE of why particular
inferences are justified, even if IBE cannot be ruled out. But this seems to have the
form of IBE! But perhaps he would respond that this is IBE only in the
uncontroversial sense of Occam’s razor.
In this paper, I will show that both of these ways of looking at it are false
descriptions of how we actually reason in this kind of situation, and that it is only in
a particularly trivial sense of the word ‘inference’ that an inference is performed at
all. We reason in an a priori manner by elaborating the meanings of our concepts,
that is to say, when we grasp the concept of bear track, eo ipso we grasp whatever is
contained in that concept and are no longer free to withhold attribution of any part
of the content of the concept when we attribute the concept itself; in assenting that
something is a bear track, assent to its being caused by a bear is forced upon us. As
far as thought goes (the case is different with regard with what we are prepared to
say, as will be explained later), this is something that we do not generally need to do
consciously. We do not say to ourselves ‘‘If this is true, then that is or probably is
true’’. We do not need an argument, as I will now show.
The point is simply that ‘‘bear track’’ is a causal concept—a description of an
effect—and just means ‘‘a track caused by a bear by its passage.’’ So, at least part of
what Henry means when he thinks ‘‘That is a bear track’’ is ‘‘A bear has passed this
way.’’ This is an entirely a priori matter. It is simply a mistake to think that we
3
For instance, an inductive argument may authorize my use of a crystal ball for predicting the future by
looking at the incidence of true predictions in the class of all predictions made using this method, but this
would not mean that gazing at a crystal ball is an inductive method. It is not clear whether Fumerton fully
appreciates this point.

123
A Priori Abduction 171

reason ‘‘That is a bear track. Therefore, a bear has passed this way recently.’’ We do
not apply a rule of inference but, at best, substitute an identical.
Now, it may be objected that I am applying the rule {Bp, p ? q} entails Bq that
is not, in general, true. In the case of bear tracks and footprints whose causal history
is borne on their faces, as it were, as part of their semantic value, it seems to me that
this part of their semantic value should be substitutable salve veritate because
implied intensionally and not just extensionally or materially. The truth of ‘‘A bear
passed this way’’ is implied by, but does not imply, the truth of ‘‘That is a track
caused by a bear by its passage’’—which is equivalent to ‘‘That is a bear track’’—
because it is part of its truth-conditions and not because it has to be inferred by some
logical rule of inference. This copes in a very straightforward way with Fumerton’s
point that we do not infer that Halle passed that way: Halle is not included in the
concept we associate with ‘‘bear track.’’ Yet isn’t it possible that we could believe
that that is a bear track and yet not believe that a bear has passed this way, or even
believe that it is false that a bear has passed this way? Intuitively it might seem like
what I have said above is false and there could be failure to preserve truth under
substitution. And if so—if the realization that a bear passed this way is something
that we have to come to, e.g., by drawing it as a conclusion—don’t we have an
inference after all?
This objection is based on a confusion between thought and speech. Henry does
actually believe that a bear has passed this way, and the illusion that he is not is
brought about because he is not necessarily as likely to produce the sentence
‘‘A bear passed this way’’ as he is ‘‘That is a bear track,’’ and because Henry may
have to think before he can say ‘‘A bear has passed this way.’’ Also, because it may
even appear to him that the sentence ‘‘A bear has passed this way’’ is false, it may
appear to him that he has to infer his belief that a bear has passed this way and that a
mere substitution will not do the job. The reason substitution does not work for
sentences is simply that what we are substituting are not identical—the sentential
functions ‘‘___ is a bear track’’ and ‘‘___ is a track made by the passage of a bear’’
being distinct even if the propositions they express after identicals have been
substituted into their argument places are not—so here there must be an inference.
As O’Shaughnessy (1980, 153–154) puts it, just because the propositions are
identical does not mean that all the sentences we would take to express them can
indifferently be intentionally used; we should not confuse an episode of thinking
with an episode of sentence-production. Confusion between these things leads to
confusion between genuine cases of failure to preserve truth under substitution and
its mere appearance. Henry can believe that a bear passed that way even if he is not
yet ready to say ‘‘A bear has passed this way’’ and even, conceivably, when he
would deny ‘‘A bear has passed this way.’’ Similarly, we may say that Mary is
beautiful with more alacrity than we may say that Mary is pulchritudinous because
of unfamiliarity with the latter term, but this is not the same as coming to believe
that Mary is pulchritudinous—we believed this all along but were just not ready to
express it in these words. Indeed, we may have a mistaken idea of the meaning
of ‘‘pulchritudinous’’ and would deny the statement ‘‘Mary is pulchritudinous.’’
Realising that we can apply a word is not the same as reasoning to an indirect belief.

123
172 D. Botting

Strictly speaking, I have shown all that I need to show, it being irrelevant in what
way, whether justified or unjustified, Henry comes to have the thought ‘‘That is a
bear track’’; the nature of an inference is independent of the status of the premises
(whether it is a necessary truth, an inductive generalization, or a mere guess) and of
how one has come to believe the premises. However, an objector might like to
change the problem by arguing something like this: ‘‘That is a bear track’’ really
only means ‘‘That looks like a bear track’’ or even less committally, ‘‘That is
something that looks to me like a bear track.’’ In other words, the objector would say
that our representational states are more modest in their ontological commitments
and it is only for pragmatic reasons, e.g., because we want to avoid being stared at,
that we do not walk around saying things like ‘‘That looks like my hand’’ or ‘‘I
believe that I am now sitting at my desk, but I could be dreaming.’’ We say more
than we are strictly entitled to, more than we can take ourselves to know, more than
our representations entail; a wedge can be driven between the ‘thick’ contents of
what we say and the ‘thin’ contents of what we think.
What motivates the drive for ‘thin’ content is, I think, the desire for a content that
is infallible and is thus able to satisfy the requirements of a strong foundationalist
theory of knowledge, as if epistemology should dictate to the theory of mind what
representational states ought to be without actually considering what they are. This
is to confuse a biological question with an epistemological question. I do not see any
reason to believe that the contents of our representational states to be anything other
than what we say they are or what we would think they are before philosophers tried
to persuade us otherwise; knowing ‘thick’ contents and the ontological claims they
make to be fallible should not deter us from the view that these contents are
nonetheless how our experience is represented. The best way, I think, to argue for
what representational states are or are not in the absence of empirical data is to
consider it from the point of view of evolutionary advantage. Is there any
evolutionary advantage to be gained from believing ‘‘That looks like a bear track’’
in contrast to ‘‘That is a bear track’’? Perhaps yes if our aim to maximize the ratio of
true beliefs to false ones, no if our aim is to avoid being eaten by bears. In other
words, the latter belief allows us to navigate the world just as well if not better than
the former, and besides, ‘‘That looks like a bear track’’ reports an objective, mind-
independent resemblance and is false if what is referred to does not look like a bear
track. To achieve infallibility we have to retreat further to something like ‘‘That is
something that looks to me like a bear track.’’ But this expresses an uninteresting
fact about your mental life whereas perceptual beliefs and statements would more
naturally be taken as beliefs and statements about the world, the self-same world
that your beliefs are supposed to help you navigate. This is just the truism that in our
everyday dealings with macroscopic, observable objects we are under no
uncertainty that those objects have a mind-independent existence. We can be made
uncertain by consideration of illusions, for example, and sometimes such
considerations rightly defeat our perceptual beliefs, but our starting point is always
acceptance and it requires an argument to infer its falsity. A belief doesn’t have to
be infallible in order to be foundational, as more moderate theories of foundation-
alism point out.

123
A Priori Abduction 173

So I have no objection to even basic perceptual beliefs having thick contents, and
moreover it seems the natural way to interpret a case such as Henry’s, who we
should remember is an expert on identifying animals by their tracks. However, let us
look at ‘thinner’ contents to see where they get us. If, for example, the spoor is
disturbed Henry may think to himself ‘‘Well, I’m not sure, but it looks like/could be/
might be a bear track.’’4 From such content the fact that a bear passed that way is not
guaranteed, but it is, I claim, warranted in much the same way as it was from ‘‘That
is a bear track.’’ This can be seen from the trivial syntactical fact that ‘‘looks like
they were made by a bear’’ contains ‘‘made by a bear’’, which leads to the equally
trivial semantic fact that the concept denoted by the former phrase is parasitic on the
concept denoted by the latter phrase. If there were no conceptual connection
between looking like a bear track and being a bear track it would be possible that
nothing that was a bear track looked like a bear track, but this is not just an
empirical improbability but a conceptual impossibility. If the question is raised of
what caused whatever it is that looks like a bear track then this conceptual
connection warrants the reply, with appropriate admissions of fallibility, ‘‘Given my
currently available evidence, probably a bear’’; no additional factual information, no
statistical information about the actual reliability of the inference in this case, no
information about contra-indications, is required. If one has that information then
one may withdraw or adjust one’s inference accordingly, but lacking that
information is no hindrance to drawing the inference. The warrant is a priori in
this case just as much as in the last.5 Given the evidence that we have and don’t
have, there is every reason to think that the inference will move from a truth to
another truth.
Consider a kind of case where the content does not contain ‘‘bear’’ at all. Suppose
that Henry takes his son Ian on one of his hiking trips and tries to teach him about
the forest. Henry points at the track and asks his son what animal he thinks caused it.
Unskilled, Ian cannot recognize it as being a bear track, but can tell that it is a track
of a large animal. Just as Henry can make the inference from unspecified bear to
Halle if and only if he has reason to believe that Halle is the only bear that could
have made those tracks, equally Ian can make the inference from an unspecified
large animal to a bear if and only if he has reason to believe that there are bears in
the forest and only they could have made those tracks, or at least are far more likely
to have made those tracks than any other large animal that could have done so. For
Fumerton, this would be an additional premise that is supported by evidence from
enumerative induction. This seems fine as far as it goes and I think we can accept
this as an account of how such an inference is justified, i.e., as demonstrating (ii)
4
Although they slip from this in his ensuing explanation as quoted earlier, when Walton et al. (2008,
170) refer to ‘‘tracks that look like they were made by a bear’’ as the data from which the conclusion is
inferred they seem to be following this interpretation, insisting on a thinner perceptual content that is
nevertheless transparently about a mind-independent object. The sequel will show that this makes no
difference to the nature of the inference.
5
The conditional that says ‘‘If it walks like a duck and quacks like a duck, then it is a duck’’ is easily
recognized as a methodological principle of reasoning rather than as an empirical generalization. It is
available to any reasoner irrespective of whether they have observed ducks or even know what the word
‘‘duck’’ means, provided that they know what the word ‘‘like’’ means. This does not preclude the
possibility of being radically mistaken about what a duck really is.

123
174 D. Botting

above, but it leaves out the question of what that evidence is. When we have filled
this in, I think that Fumerton’s account is once again descriptively inadequate; I do
not think that Fumerton succeeds in establishing (i). We do not reason via the
enumerative induction, or indeed via any general statement, but rather our belief in
the existence of such evidence follows a priori from other beliefs.
This has an epistemological and a psychological aspect. The epistemological
point is that we do not need to know what the justifying evidence is in order to draw
conclusions from it but only that it exists and that it supports the conclusion; for
example, if I know that there is a valid proof where C is deduced from P, and I know
P, then I can deduce C, even if I do not know what the proof is. The proof itself
justifies the inference, but my knowledge that it exists may have a different
justification and in turn justifies my use of the inferential rule to conclude C. The
psychological point is that when we take the evidence to exist then we are
predisposed to look out for it and interpret what we find as such evidence. This is
especially the case when we ‘‘prefer to be safe than sorry.’’
There are two possible sources of evidence: perception and testimony. With
regards to perception, suppose that, in the course of the hike, Ian has observed bears
in the forest such that all or most of his perceptions of the activities or signs of large
animals in the forest have been perceptions of the activities or signs of bears. On
perceiving the track of a large animal, Ian will naturally infer that this is also the
perception of the track of a bear. In this case, ‘‘That is a bear track’’ is inferred but
not, it seems to me, in the sense that Ian forms a general statement from which he
then reasons, but in the sense that in Ian’s epistemic context the concepts of ‘‘large
animal’’ and ‘‘bear’’ have become closely associated. In some cases we might even
say that they are contextually co-extensive, but the extensions need not be exactly
the same in order for there to be an a priori warrant; it is not necessary that Ian
contextually take ‘‘large animal’’ to mean, or to be identical with, ‘‘bear’’, any more
than ‘‘looks like a track made by a bear’’ means ‘‘is a track made by a bear.’’ A
warrant may be a priori even without guaranteeing the truth of what is warranted. It
should be noted that if this is true it greatly extends the range and power of this kind
of abduction, since it amounts to saying that when we are reasoning in contexts (and
when are we not?) we can sometimes treat one causal concept (e.g., track of large
animal), and perhaps even a non-causal concept (e.g., shape on ground), as a causal
concept (e.g., bear track) and then reason with it in the way I have described.
Sometimes this substitution is conscious—as is probably the case here—or there
may be cases, perhaps because we have been primed by expectation (see the
psychological point above) or conditioned by stimulus–response conditioning,
where it becomes unconscious and, once more, it is the perceptual belief itself that is
‘‘There is a bear track’’. Undoubtedly, there are also cases where we are faced with a
choice of substituents and then we have to decide which is most likely, or perhaps
which is the most explanatory. There is clearly a limit to what can be achieved by
a priori abduction on its own.
Suppose instead that Ian’s evidence for the presence of bears is testimony, e.g.,
he observed, on entering the forest, a sign saying ‘‘Warning! Bears!’’ We do not
perform an inference as such from a warning of the presence of bears to the actual
presence of bears for reasons similar to those stated earlier. Although ultimately our

123
A Priori Abduction 175

faith in the warning sign is justified by some kind of statistical data regarding the
reliability of such signs and the frequency of bears, we do not epistemically have
this statistical data, but this does not matter as long as we are justified in believing
that such statistical data exists, and we are justified in this because we are justified in
thinking that whoever is responsible for the sign is observing the Gricean Maxim of
Quality that states that you should not say that for which you lack adequate
evidence. It is precisely in having this evidence that the sign qualifies as a warning
rather than a joke or mere squiggles stuck together. Although they may not be
observing this maxim, and we do not know the statistical data regarding the
frequency with which this maxim is actually observed, it is not an a posteriori but an
a priori matter that we are justified in believing the warning sign once we have
perceived the warning sign as a warning sign and as an attempt to communicate.
Just as it follows as an a priori matter that when we perceive a shape in the ground
as a bear track we perceive something made by a bear, it follows that when we
perceive a collection of symbols as a warning sign then we perceive something
intended to warn,6 and we are justified in believing, for instance, that there are bears
in the woods.
Let me try to summarize where we have got to and where we need to go: whereas
before the substitution was warranted by semantic connections, now it is warranted,
if it is warranted, by a strong inductive argument, and this becomes the source of
justification mentioned in (ii). But we do not entertain this inductive argument but
only reason that there is an inductive argument, or at least empirical evidence that
justifies our beliefs, and at least in some cases this will be shown to be an a priori
abduction. So, our reasoning will not in general be induction differently expressed,
as claimed in (i). Yet arguably all that Ian can justifiably believe on the basis of this
abduction is that bears are present and not that any tracks of large animals that he
finds are likely to have been made by bears. Although objectively justified by

6
This abduction can be compared with the form of what Walton (2010, 168–174) calls the parascheme
for argument from expert opinion. A parascheme is a quick heuristic that is meant to approximate in its
judgments the complete argumentation scheme. For instance, the parascheme for argument from expert
opinion is given as:
E is an expert
E asserts A
A is true
This abbreviated version is less reliable than the full argumentation scheme because it does not consider
critical questions or implicit premises. But, I would add, the parascheme can be abbreviated even further
as:
A is an expert opinion
A is probably true
To take A to be an expert opinion just is to take it as probably (given our current state of knowledge)
true; no inference is required and we do not need to go through the list of critical questions in order to
labour the obvious point that the expert must be an expert. If the expert is not an expert then A is not an
expert opinion, simply because of what it means to be an expert opinion, and what this entails is that we
have made a mistake about A which is not the same as committing a fallacy, as Walton (2010, 174)
himself says: ‘‘[T]he notion of fallacy is generally taken in logic to represent a fallacious inference of
some sort, an argument from premises to a conclusion, and not merely a false or insufficiently
substantiated explicit premise in the argument.’’ A fallacy would be committed if, for example, we make
a hasty generalization from some statistical data to the reliability of a particular person such that we
mistakenly call him an expert.

123
176 D. Botting

induction, Ian is subjectively justified, I will argue, by making the substitution ex

suppositione and then showing that the beliefs he gets by doing this are more
coherent than if he had not; the perceptual beliefs and a priori abductions all
converge, so to speak, on a particular interpretation of the evidence.
Ultimately it is this convergence that retroactively subjectively justifies taking
the perception of the sign to be a perception of a warning and not a joke and the
perception of bears to be a perception of bears and not cunningly disguised zebras.
For Ian now has two facts: (1) his perceptual belief that ‘‘That is a track made by a
large animal’’; (2) his perceptual belief or a priori abduction from testimony that
‘‘There are bears in the forest.’’ He also has the background knowledge that bears
are large animals. Hence, a specification of (1) is (1*) ‘‘That is a track made by a
bear.’’ There is a relation of mutual support that (2) has to (1*) that it does not have
to either (1) or the thin (10 ) ‘‘That is a track that looks like it was made by a large
animal (or, for that matter, by a bear).’’ It is because Ian automatically takes these
facts to be connected that he can infer that a bear has passed this way recently, and it
should be noted that neither (1*) nor (2) are general statements. Once he has
accepted this, he is further reassured that he has not been guilty of misperception. So
again, Ian does not reason through any generalization.
What we have here, though, is a description: because he automatically takes these
facts to be connected he will ‘closely associate’ one concept with another. But is he
justified from his own point of view in taking these facts to be connected? And is he
justified in taking the sign to be a warning in the first place? My view is that an
answer to the first question is also an answer to the second; what subjectively justifies
our interpretation of speech acts is that they lead to beliefs that are more ‘connected’,
that is to say, more coherent. He is objectively justified by the inductive argument, as
already said, but he is subjectively justified by coherence. The next section will aim
to answer why we are in fact subjectively justified in preferring more coherent beliefs
to less. This is basically because it leads to an economy of thought that is naturally
selected. Perhaps it will be objected that this is not really justification because
economy of thought is not truth. If so, let me make a more modest claim: Ian is
rational when he concludes that a bear has passed this way.
Now, it is of course possible that Ian is in neither of these situations or epistemic
contexts, that is to say, he has not observed bears or any testimony that there are
bears. It is also possible that he reaches the conclusion that a bear has passed this
way recently. The question is: has he reached this conclusion by making a correct
inference, or is he just guessing, bear being perhaps the first large animal that he
thinks of? If the latter, it is obviously no counterexample to a priori abduction,
which is an account of making correct inferences; no analysis of inference is meant
to capture lucky guesses. But as usual I leave it open whether other modes of
reasoning such as IBE may nevertheless be used to give a correct inference.

3 Testimony and Coherence

The a priori abduction described above relies on the sign being correctly categorized
as a particular kind of speech act. I have said above that greater coherence can

123
A Priori Abduction 177

justifiably be taken as evidence that there has been no misperception or

miscategorization. But does coherence have the epistemically normative power to
do this? I will argue that we are rational and, from our own point of view at least,
justified when we take signs at their face value and draw inferences from them.
First, I want to compare the a priori abduction from testimony described above to
another analysis of testimony.
First, we assume that in most instances one’s conversational partner to be
cooperative unless there is a reason to believe otherwise. Gelfert (2010, 8) describes
something similar to what I have said above as the evidential theory of testimony:
[T]he speaker S aims to get her hearer H to believe something (e.g., that p) by
getting him to recognise that this is what she is trying to do. The way in which
H is intended to arrive at his belief that p is by taking S’s act of telling (e.g.,
her perceptible utterance of certain sounds) as evidence of her intention to get
H to believe that p, and subsequently to take her having that intention as
evidence that S herself has the belief that p (or, on a somewhat stronger
interpretation, takes herself to know that p), which in turn is then to be taken
as good evidence that p is indeed true. What S is relying on when she intends
to communicate her belief that p, is the assumption that H will take the fact
that S wants him to believe that p as something which, given proper
background beliefs and competencies on the part of H, is a good reason for
H to believe that p. Note that, on this interpretation of Grice’s account, the
speaker need not offer any direct evidence that would independently support
what she claims, nor need she offer any natural evidence as to her own
sincerity or competence.
From the last sentence it can be seen that the existence of evidence, either for what
has been testified, or for the reliability of the testimony, is inferable and justifiably
believed although the evidence itself is never and nowhere given. We do not
entertain the evidence but infer its existence. This account seems to me basically
correct, although I think there is still a question of justifying why a particular
utterance should be taken as a particular speech act, e.g., as a warning or an
assertion. In these cases I will argue that what we take to be testimony is that which
results in the believer’s beliefs as a whole being more coherent.
Gelfert (rejecting the evidential theory) here uses an inference to the best
explanation—the best explanation of why the speaker said that p is because p is true.
My own explanation, as will be seen, appeals to coherence. These alternatives may
well boil down to the same thing, since Gelfert (2010, 14) goes on to say:
[C]oherence with our background beliefs, or comparing new instances of
testimony with past (remembered) instances of testimony: if the reports of
people (on the same topic, or in general) cohere with each other and with our
own observations and background beliefs, this provides considerable abduc-
tive grounds for accepting what we are told.
It is not clear why these are abductive grounds, rather than warrant-conferring in
their own right. However, it has been argued, e.g., by Psillos (2002) that abduction
is warrant-conferring precisely because it improves coherence. This account of

123
178 D. Botting

testimony relies on coherence being an epistemic norm. This is the fundamental

issue we keep coming back to.
According to Psillos (2002) coherence is an epistemic norm in its own right
distinct from likeliness of truth. On the other hand, it has been shown by Glass
(2007, 2009) that there is a coherence measure that strongly approximates to the
results of Bayesian conditionalization, where this is taken normatively as likeliness
of truth.7 If so, likeliness of truth can be retained as the only epistemic norm.
However, this sketchy treatment is only suggestive of ways in which the very
problematic question of objectively justifying coherence considerations might be
tackled. My intention in this paper is much more modest, namely how we can
subjectively justify coherence considerations. For even supposing that they are
objectively justified it is not very plausible that it is because of these objective
justifications that we take ourselves to be justified or that it is in these ways that we
justify considering coherence as epistemically valuable. We could perhaps base our
subjective justification on an enumerative induction showing that those theories that
have been most successful and have a high level of likeliness also are coherent in
best satisfying the multiple constraints placed upon theories by explanatory
considerations, as has been strongly suggested by the work of Thagard. Yet it seems
nearly as unlikely as before that the reasoner himself reasons that way because of
this justification or is aware of this justification or even its mere existence; the
reasoner seems lucky that the reasoning process that is actually used has this kind of
justification. What, from the reasoner’s point of view, allows him to think that his
reasoning processes are getting at the truth? Ultimately, as before, the only option
seems to be the naturalistic one that our reasoning processes have been selected.
However, I would not make too strong a connection between survival value and
truth, but I think that there is a strong connection between survival value and
coherence, as I hope to show, and on the assumption that it is rational to survive, it is
rational to prefer more coherent beliefs over less.
Let us return to Ian. I have left it open whether Ian is objectively justified in
aiming at greater coherence, although I have given good reasons to think that he is. I
want to consider instead whether Ian is subjectively justified. Although Fumerton
may be right in that there is not an a priori connection between explanatory virtue
and inductive success, there is an a priori connection between explanatory virtue
and the necessary conditions for inductive success, and vice versa. The world’s
being structured (e.g., having laws and natural kinds) is a necessary condition of
both, so both are sufficient conditions for the world’s being structured. This does not
mean that explanatory virtue and inductive success are logically inter-dependent. If
A is a sufficient condition for B and B is a necessary condition of C, A is neither a
sufficient condition of C, nor a necessary condition of C, nor even an insufficient but

7
The measure is called the overlap coherence measure. The coherence here is coherence between the
hypothesis and the evidence rather than between the hypothesis and background information and does not
consider explanatory virtues but only the same kind of probabilities that Bayes’ rule uses. Therefore, it
does not confer any advantage with respect to usability and is probably no better as a description of our
actual inferences than was the Bayesian. But if this measure may itself be approximated by judgments of
explanatory virtue, then explanationism and Bayesianism can be friends—Bayesianism being the correct
normative account and explanationism the best descriptive account.

123
A Priori Abduction 179

necessary condition of an unnecessary but sufficient condition of C. However, by

establishing B we have succeeded in making C more likely: C is only true in worlds
where B is true, so eliminating worlds where B is false (e.g., unstructured worlds) is
tantamount to raising the chance that C is true. This shows only a symmetry
between them and does not show that induction is successful or that there are any
virtuous explanations; it shows that taking something to be a genuine explanation
(whatever we may take to constitute a genuine explanation) presupposes that
induction is at least likely to be successful and taking something to be a valid
induction is to take it that there is likely to be a genuine explanation. This follows
deductively from the probability calculus and is a priori.
How is this relevant? Because taking facts to be connected is virtually the same
thing as taking those facts and their connection to be something that requires a
unifying explanation and is prerequisite to applying the concept of explanation. Out
of the potential explanations, the one that we are most disposed to take as the
genuine explanation is the one that gives us the most coherent set of beliefs overall.
This disposition is, I think, just an empirical fact about reasoners in the actual world
and follows from the assumption that the purpose of our set of beliefs is to help us
navigate in the world, it being more efficient and economical to deal with a few
general rules that cover many cases (even if the rule gives the wrong result in some
cases) than to try and evaluate each new situation ab novo.8 The basic idea is that
coherence governs the choice of potential explanations, and that once chosen
everything else follows from conceptual connection. Taking certain facts to be
coherent is a necessary precursor to taking something to be an explanation or taking
there to be an explanation, and rejecting coherence virtually amounts to rejecting
explanation and induction as concepts applicable to our world. Coherence
subjectively justifies the belief that a common explanation exists, and the existence
of an explanation makes it more likely that the world is structured. Just as when we
take a collection of symbols as a warning sign then we take it as something intended
to warn, equally when we take something as an explanation then we take it as
something intended to explain, viz., a description of a world that is mostly true and
can be investigated and corrected using inductive methods.
It is true that I still have not shown that coherence is epistemically normative
(although the work of Thagard, Psillos, and Glass suggests that it is). I do not intend
to answer this question, but I will suggest what an answer might look like and what
its consequences would be. What would I have to show? What I would not have to
show is that the more coherent belief systems are epistemically better than
incoherent belief systems in every possible world but only that they are better in the

8
Perhaps this is not a very strong argument, but one way to strengthen it is to draw an analogy to
deduction. What deduction tells you is that if the premises are true then the conclusion must be true. But
given a deductive argument, if the reasoner will not accept the conclusion then it is always open to him to
reject one of the premises, or even to reject a rule of inference; our doubts about a particular pattern of
reasoning may be inhibited or reinforced by our beliefs about the conclusions of other arguments
instantiating the same pattern, and our doubts about a particular premise may be inhibited or reinforced by
our beliefs about the conclusions of other arguments utilizing the same premise. As Quine remarked, we
tend to choose the option that requires the smallest adjustment to our web of beliefs as a whole.

123
180 D. Botting

actual world9 and that if such an argument is forthcoming it follows from the fact
that our beliefs in the actual world tend towards coherence (and that each is
warranted) that we are epistemically reliable reasoners.

4 Conclusion

It is from semantics, from a priori reasonings about the meanings of our concepts,
rather than from applying a formal system of deductive, inductive, or indeed
abductive logic that we make at least some of our abductive inferences. I do not
make any sweeping statements that all abductive inferences can be given this trivial-
seeming analysis. It is not inconsistent with my view should there turn out to be an
abductive inference (such as some say is involved in generating perceptual beliefs)
that cannot be analysed in the way I have here. However, I think that my analysis
works even in more sophisticated cases than those I have up to now considered. The
following case was once put to me:
[A]rriving home from work I notice that my door is unusually ajar and that
when I enter the house I notice that things are out of place and that a few items
are missing. It does seem like I infer as the best explanation that someone
burgled my house. However, it does not seem even remotely plausible (as it
does in the bear tracks case) to claim that the data I appeal to somehow means
that my house was burgled.
This is very plausible, but we do not have to change the description by much to
make it into an a priori abduction. Instead of noticing that the door is ajar I notice
(or substitute into the content of what I notice) that it has been opened, and since
this strikes me as unusual it follows a priori from this surprise that it was not I who
opened it, or anyone else to whom I had given permission, and these exhaust the
possible people who could have opened it legally. Similarly, instead of noticing that
items are missing I notice that they have been taken, again illegally. So, I notice that
my home has been entered illegally and items removed from it illegally. But this is
only a long-winded way of saying that I notice that I have been burgled; the
inference is once more a priori. Of course, I could have misinterpreted the data and
after my initial surprise I may consider alternative ways of interpreting the data and
either confirm or reject my initial conclusion—I may wonder whether I had locked
the door when I left, or remember that I had lent my key to my drug-addicted
sister—but if I do reject the conclusion then I also reject the perceptual belief, just
the same as if I reject the conclusion that a bear passed this way I reject the
proposition that what I was observing was really a bear track and not just something
that looked like one. That the perceptual belief from which the a priori abduction

9
This is a point made by Psillos (2002, 608). According to him a method can still be ‘epistemically
probative’—despite the fact that it is not more likely to be true—because it can identify, search for, and
cope with, potential defeaters. This seems to constitute for Psillos another dimension of epistemic
normativity, rather than, e.g., a pragmatic norm. This opens the question of what should occur when these
norms come into conflict. If Psillos is right then I have the argument I require.

123
A Priori Abduction 181

was made turned out later to be false is beside the point and does not alter the fact
that it was made and was made correctly.

Acknowledgment The author would like to acknowledge financial support received from the FCT
under the grant ‘‘Argumentation, Communication and Context’’ PTDC/FIL-FIL/110117/2009.

References

Fumerton, R.A. 1980. Induction and reasoning to the best explanation. Philosophy of Science 47(4):
589–600.
Gelfert, Axel. 2010. Reconsidering the role of inference to the best explanation in the epistemology of
testimony. Studies in History and Philosophy of Science 41(4): 386–396.
Glass, David H. 2007. Coherence measures and inference to the best explanation. Synthese 157: 275–296.
Glass, David H. 2009. Inference to the best explanation: A comparison of approaches. Proceedings of the
2nd symposium on computing and philosophy.
Harman, Gilbert. 1965. The inference to the best explanation. The Philosophical Review 74(1): 88–95.
O’Shaughnessy, Brian. 1980. The will: A dual aspect theory, vol. 2. Cambridge: Cambridge University
Press.
Psillos, S. 2002. Simply the best: A case for abduction. In Computational logic: Logic programming and
beyond (Essays in Honour of Robert A. Kowalski, Part I), ed. A.C. Kakas, and F. Sadri. Berlin:
Springer.
Walton, Douglas. 2010. Why fallacies appear to be better arguments than they are. Informal Logic 30(2):
159–184.
Walton, Douglas, Chris Reed, and Fabrizio Macagno. 2008. Argumentation schemes. Cambridge:
Cambridge University Press.

123