Professional Documents
Culture Documents
Introduction
In his Sophistical Refutations Aristotle introduces a list of thirteen fallacies. Six of them he classifies as depending on the formulation, fallacies which rely on purely linguistic features of the statements involved
in the reasoning. Two of these are the sophisms of combination and
division. Aristotle does not characterize them abstractly in any way, but
his introductory examples seem clear enough:
Depending on combination [is] ... for example [i] being able to
walk while sitting and [ii] [being able] to write while not writing
( ). (SE 4,
166a23-5)
And:
Depending on division [are the arguments] [iii] that five is two and
three, and odd and even, and [iv] [that] the larger [is] equal, as it is
Dedicated to Erik Krabbe on the occasion of his retirement as Professor of Argumentation Theory at the University of Groningen (February 2008) with many
thanks for comments and the most enjoyable cooperation in trying to understand
Aristotles dialectic and theory of fallacy.
106
The two examples of combination (i and ii) exhibit the same pattern, as
the phrase being able to while not-ing can be interpreted in two
ways:
(1a) while not-ing, being able to
(1b) being able to [ while not-ing]
The two examples of division (iii and iv) likewise are structurally similar, as the phrase a is b and c is interpretable in two ways:
(2a) a is [b and c]
(2b) a is b, and c
For if five is two and three is understood along (2b), then we may
claim that five is even, because it is two, and odd, because it is three.
And if what is larger than x is x and something in addition is similarly understood, then it follows that what is larger than x is something
equal to x.
In terms of how they work these introductory examples may be sufficiently perspicuous, but that is not the case with the two fallacies of
combination and division in general. For one, quite a few of the further
examples Aristotle gives are difficult to understand and have caused
commentators trouble as we shall see, some cases of combination or
division have not even been identified as such. More serious, however,
is that what little Aristotle sets forth on a theoretical level about combination and division seems incomplete and inconsistent. One point on
which Aristotles account is incomplete is that he, on the one hand, tells
us that the fallacies of combination and division arise because of thinking that a combined or divided does not differ at all (SE 7, 169a2526), but, on the other hand, does not give any criterion for counting one
reading as divided and another as combined. Of course, in the case of
(i) and (ii) it is natural to suppose that (1b) constitutes the combined
reading, as it is somehow linguistically intuitive, which is confirmed by
the fact that the fallaciously drawn conclusion, which is said to depend
on combination, derives from reading (1b). (Similarly with (iii) and (iv)
reading (2b) would be the divided reading.) But a clear criterion would
have been useful with examples as the following:
These two questions, which are each others converse, can be analysed
in two ways, depending on where with which and with that respectively are to be placed, but both these ways can be phrased in terms of
combination as well as of division: combined with you saw/did you
see or with him being hit, or, alternatively, divided from him being
hit or from you saw/did you see.1
Aristotle is, secondly, also incomplete in so far as he fails to provide
an explanation why there is in the one case a fallacy of combination
and in another one of division. It is true that Aristotle does say of most
examples of fallacies he gives whether they are dependent on combination or on division, and sometimes, as with examples (i)-(iv), a natural
explanation for their classification suggests itself. However, with some
of the examples where such a qualification is lacking an explicit explanation would have been most helpful. In addition, it remains to be
shown that whatever explanation we may want to ascribe to Aristotle
on the basis of the examples explicitly called the one or the other, can be
extended to all examples unproblematically.
Now incompleteness may perhaps be compensated for, but inconsistency is quite another matter. The first point on which Aristotle seems to
contradict himself derives from the fact that even though every fallacy
of combination or division involves two readings of one sentence and
thus underlying both fallacies is the same kind of what one may term
doubleness, Aristotle counts them as two separate fallacies. For when
Aristotle wants to establish that there are merely these six linguistic fallacies he has just listed, he seems to argue from types of doubleness:
Of this [i.e. that there are these six fallacies] there is a proof both [] by
way of induction and [] as a deduction, [that is,] both [] when some
other [case] is taken up () and [] that in so many
ways we may indicate with the same words and sentences what is not
Cf. S.G. Schreiber, Aristotle on False Reasoning. Language and the World in the Sophistical Refutations (New York 2003), 91-2.
108
the same (
). 2 (SE 4, 165b27-30)
How then are we going to end up with six rather than five linguistic
fallacies if those of combination and division involve the same way of
indicating what is not the same?
A similar inconsistency seems to occur because Aristotle often likens
the doubleness involved in the fallacies of combination and division to
that in the fallacy of accent. Both, he states, do not concern one statement or one word with several meanings and are thus not based on ambiguity. Rather they involve two statements or words, corresponding
to two readings of the same sentence (as a string of words) or written
word, in the former case a combined and a divided reading, in the latter
a reading with higher pitch and one with lower pitch. This doctrine he
states most explicitly in the following passage:
Of the [apparent deductions and refutations] which reside in the
formulation, some are dependent on something double, such as
homonymy, amphiboly ( ) and similarity of formation (
) ..., whereas combination and division and accent
[come about] because of the statement not being the same or [because
of] the word [being] what is different (
). (SE 6, 168a23-8)3
For alternative translations of this sentence, see e.g. L.-A.Dorion, Aristote: Les
rfutations sophistiques (Paris and Laval 1995), 124; J. Barnes, ed., The Complete
Works of Aristotle: The Revised Oxford Translation [CWA] (Princeton, NJ: 1984), 280;
Schreiber, False Reasoning, 20; E. Poste, Aristotle on Fallacies or the Sophistici Elenchi
(London 1866), 7; E.S. Forster, Aristotle: On Sophistical Refutations, On Comingto-be and Passing-away [together with: D.J. Furley, Aristotle: On the Cosmos]
(Cambridge, MA 1955), 17; and P. Fait transl. and comm, Aristotele: Le confutazioni
sofistiche. Organon VI (Rome and Bari 2007), 7, who at 107-9 offers an overview
of all attempted interpretations. All these translations ignore the clearly parallel
double ... ... structure. One should confer, moreover, Topica I 8, 103b3-8. It
might be objected against my translation that I cannot supply any masculine word
with from the context. However, in a classification of fallacious arguments
the word easily springs to mind (cf. a few lines further, at 165b31).
But why then does Aristotle not distinguish likewise two fallacies of
accent, one depending on a sharp accent and one depending on a low
accent?4
To add to the confusion, sometimes Aristotle seems to forget about
this claim that combination and division involve two readings or statements rather than one. For example, he states that a fallacy may occur when a divided and combined signifies something else (SE
20, 177a34-5).5 Moreover, this inconsistency points to a further issue on
which Aristotles account is not complete. For why do we have, rather
than one ambiguous statement, two separate statements in the case of
combination and division? We could ask the same question by comparing combination and division with the fallacy of amphiboly, which occurs when the same statement can be construed grammatically in two
different ways. Aristotle himself seems to acknowledge that they are
quite similar, for after giving the two examples of hitting someone with
the same as that with which one saw him being hit, he continues:
It has, then, also something belonging to amphibolous questions (
), though it is dependent
on combination.6 (SE 20, 177a38-b1)
At a38 I hesitatingly read , with manuscripts D, u and V, and Boethius translation, against the reading of A and B, which is adopted by W.D. Ross, ed.,
Aristotelis Topica et Sophistici Elenchi (Oxford 1958) (for the letters standing for
the manuscripts, see J. Brunschwig, ed and trans, Aristote: Topiques I Livres IIV [Paris 1967] the reports about V, which has not yet been used in any edition of the Sophistici Elenchi, are based on my own inspection of the manuscript).
110
Dorion, Les rfutations, 340, argues that should here mean ambiguous rather than amphibolous, on the grounds that amphiboly only concerns
statements in which terms can be assigned two different grammatical functions
and that and retain their grammatical function, whether they belong to
or . However, not only does this seem too strict an application
of the concept of grammatical function, but it also leaves Aristotle with a case of
ambiguity which he cannot accommodate in his scheme. For the only two cases
of ambiguity Aristotle distinguishes in the Sophistical Refutations are homonymy
(lexical ambiguity) and amphiboly (syntactical ambiguity), and these two examples obviously do not involve any kind of lexical ambiguity.
7
111
See A. Schiaparelli, Aristotle on the Fallacies of Combination and Division in Sophistici Elenchi 4, History and Philosophy of Logic 24 (2003) 111-129 though she
was perhaps preceded as far as the main idea is concerned by J.D.G. Evans, The
Classification of False Refutations in Aristotles De Sophisticis Elenchis, Proceedings
of the Cambridge Philological Society NS 21 (1975) 42-52, at 50.
112
are not identical to operators and cannot even be made identical to each
other, so it seems that on Schiaparellis account Aristotle is committed
to there being three kinds of division and combination fallacies, without a uniform nature.
What is more, her two proposed analogues are problematic in themselves. To take the first analogue first, in five is two and three Aristotle does not consider five as the one- or two-place predicate which is
predicated either of two and three separately or of two and three
together, but as the subject which either has two and three as two
separate predicates or two and three as one single predicate. For otherwise he cannot deduce, as we saw he does, the further point that five
is odd and even from the divided reading of five is two and three
without committing a further fallacy.
The analogue of an elliptical expression with empty slots is even
more deficient, as it lacks a crucial feature of the analysis in terms of
operators with wide and narrow scope. For while that analysis allows
us to say that there is a combined reading if some string of words as a
whole is governed by an operator, and a divided reading if only part of
that same string is governed by the same operator, this feature is lost
with the analogue. Thus in the example of I made you free while being
a slave, it does not matter whether we divide the string of words free
while being a slave over one or two slots in the elliptical expression
I made you (...) (...): there is nothing in the so-called divided reading
which necessitates us to take while being a slave with you rather
than with free, because a specification to that effect is not part of the
elliptical expression.
This is not to say, however, that Schiaparelli is not on to something
with her proposal to analyse these fallacies as involving an operator
with wide and narrow scope. For with the clear criterion for what
counts as a combined reading and what as a divided reading there is
definitely something in her proposal worth preserving.9 This element
is that whereas on a divided reading two strings of words make independent contributions to the sentence as a whole, on a combined
reading the same two strings of words contribute something together.
That is, what needs to be brought out if we want to come up with an
informative as well as uniform account of the fallacies of division and
combination is that there is more than one way the same sentence (in
And because of which I was woken from my, perhaps not dogmatic, but certainly
ignorant, slumber.
113
This is exactly the distinction Schiaparelli wants to capture by employing the terminology of operators with a scope. But unlike with her
proposal, the analysis in terms of grammatical trees works equally well
in order to bring out the difference between the two readings of (iii)
. For that example can be represented by the
following trees:
114
and:
2.2
10
Notably the examples which I have numbered (v), (xii) and (xiv) see below. For
the limited heuristic value of the approach of analysing these examples in terms
of logical operators with scope, see Schiaparellis rather strained attempts (in her
Combination and Division, 117-19) at finding anything like a logical operator in
examples (v) and (xii).
115
And [v] learning now letters, since he was learning what he knows
(). 11 (SE 4,
166a30-1)
imperfectum
or as:
imperfectum
In the former case it is about someone who was learning things he now
knows, while in the latter case it says that this person was in the process
of learning things he already knows. If someone makes the concession
that someone was learning things he knows as interpreted in the latter way, he could just as well concede that one may ascribe instances of
the predicate learning now things one knows to someone, for example
11
12
116
or as:
This is Aristotles analysis of what we would call a quantifier-shift. Interpreted in the first way, there is no inconsistency with the earlier concession that this person can only carry one thing (unless we would wish
to propose a similar distinction between two readings for that statement
as well), but if interpreted in the second way, it is in straight contradiction with it. Thus someone could be refuted in a sophistical way.
That this is the correct analysis is confirmed by another example of
the same quantifier-shift, but which has not been identified yet as a case
of combination and division: example (vii). It appears in De Generatione et Corruptione I 2 and concerns the property everywhere divisible
():
13
117
(b)
14 For a full discussion of the issues involved with this passage and of the mysterious
phrases in it, see my Aristotles Diagnosis of Atomism, Apeiron 39 (2006) 121-156.
118
He does so by introducing, with the notion of points, possible divisions and by correlating them to the positions (which we would call
points) on a magnitude. The basic correlation is that between a possible
division and a position, which Aristotle describes in terms of there being a point anywhere, which is common to both senses we could
represent this correlation with the property of there being a possible division at x, that is, being divisible at x. Now in order to describe sense (a) of
everywhere divisible Aristotle quantifies over these possible divisions
or over these correlations between possible divisions and positions: all
[points or possible divisions] are like each in that they belong to a position. So we get:
all
possibility of division
position
On the other hand, in order to describe sense (b) of everywhere divisible Aristotle does not quantify over these possible divisions or over
these correlations, but only over the positions: not only is there a point
or possible division anywhere, but there is also a point everywhere. So
we get:
possibility of division
all
position
or:
119
As this seems a general remark, it is not clear under which fallacy each
of these two examples (viii and ix) should be classified probably they
are not even meant as examples of fallacies, but merely of sentences
which can be read in two ways.17 But again they may be analysed with
the help of trees. The first example (viii) can be interpreted either as:
15
16
This translation is not literally correct, but is meant to capture the ambiguous
structure of the Greek example.
17
120
or18 as:
On the first analysis this statement turns out to be the absurd claim that
I have made you free while at the same time being a slave, while on
the second analysis it merely says that I have made you, while being
a slave, free. (That there are here in fact two trees is meant to signify
that the state indicated by has nothing to do with, or is at
least relatively independent from, the act referred to with , but is
merely tagged on to .)
The second example of this passage (ix) allows both for the following tree:
and for:
<>
18
or as:
19
As Schiaparelli, Combination and Division, 126, note 32, remarks correctly, because of its hexametric form one need not expect the example to adhere to the normal rule that in composed numerals the smaller unit (here:
) may only precede the larger (here: ) if linked by way of
It is also possible to have a tree of the second type with and
<> switching rles: { } [ <>]
: Of fifty men the divine Achilles left a hundred behind (see e.g.
Schiaparelli, Combination and Division, 126), but as this tree does not differ in
structure from the one given, and yields a completely absurd and therefore uninteresting reading, it does not seem likely that Aristotle had this one in mind.
122
On the first analysis, this part of the sentence refers to that with which
he was being hit, as you saw it, while on the second analysis it is about
that with which you saw it happening. Thus on the first analysis one
would be asking the trivial question whether that with which he was
being hit, as you saw it, was indeed the same thing as that with which
he was being hit. On the second analysis, on the other hand, the question would be whether that with which you saw that happening was
the same thing as that with which he was being hit a question which
can only be answered with an emphatic no.
The next example (xii) Aristotle gives of this kind of fallacy is notoriously short:
And there is Euthydemus argument: [xii] Do you know now being in Sicily the triremes being in Peiraeus? (
.) (SE 20, 177b12-13)
In principle there are three elements one may move around in a possible
tree which for the rest consists of and its object :
and , for all three could in principle
both belong and not belong to what is known ( may take both a
noun-phrase or a participle-phrase for its object and could latch on
to either kind of phrase). However, that alone would be the element
to have different places in a tree seems unlikely, as Aristotle gives such
an example only a little later, at 177b20-22. That could
belong to different branches in a tree seems again unlikely, as it would
involve positing a grammatically strained participle-phrase
(with or without it seems equally strange)
in at least one tree. Therefore it is preferable to interpret the example in
such a way that may appear in two different roles.
Aristotles example seems most effective if it is analysed20 either as:
or as:
20
In the following trees I have, for the sake of brevity and simplicity, left out the
phrase , which should be understood as tagged on to .
(in which latter case we have again two trees, one for the main statement and another for a statement tagged on). Thus according to the first
analysis you, rather problematically, know that some triremes are now
in Peiraeus, even though you are in Sicily, while on the second analysis
you know, while you are in Sicily, of some triremes, which, as a matter
of fact, without you knowing so, are now in Peiraeus.
That this is the right interpretation of the point of the example is confirmed by Aristotles other reference to the argument of Euthydemus,
which we find in the Rhetoric:
Another [way of giving an apparent enthymeme] is talking about
something divided while combining it or about something combined
while dividing it. For since often it seems to be the same without being the same, one should do that, whichever of the two ways is more
useful. This is the argument of Euthydemus, such as knowing that
there is a trireme in Peiraeus for one knows each (
<>
). (II 24, 1401a24-8)
21
Cf. Dorion, Les rfutations, 344, who despairs of getting anything even relatively
secure from the passage in the Rhetoric.
124
Peiraeus.22 However, that is not the only possible way of reading this
remark; it could also be a theoretical remark about grammar, to the effect that verbs of knowing may also be complemented with a nounphrase: For one knows of x.23 And one may know of something under
different descriptions: one may know a trireme merely as that trireme
(because one has seen it once, for example, or even because one has
heard about it), but also as it is now, namely as a trireme which is in
Peiraeus. It is this construction which is abused by Euthydemus in his
argument, by phrasing the question in such a way that it is ambiguous
under what description you know some triremes, whether just like that
or as triremes which are now in Peiraeus.
The next example (xiii) is very simple:
And again: [xiii] Is it possible to be a miserable cobbler who is good?
( ;) But [then] there
could be some miserable cobbler who is good (
). Hence there will be a good miserable cobbler.
(SE 20, 177b13-15)
The initial question (and also the first affirmative statement) could be
read either as:
<>
or as:
<>
22
See e.g. C. Rapp, Aristoteles: Rhetorik II (Darmstadt 2002), 782; cf. Schreiber, False
Reasoning, 71-2, Fait, Le confutazione, 186, and the translation in CWA, 2233.
23
This use of for, as we would call it, a free variable without any generalising import can be seen as an extension of the use of with widest possible
scope.
The second reading then licences the final conclusion through a fallacy
dependent on accident.
The next example (xiv), however, is far more obscure, and even involves textual issues. The text as read by all the authoritative manuscripts is as follows:
[xiv]
24
25 . (SE 20, 177b16-20)
Commentators, not being able to find anything remotely similar to combination or division in the rest, have only paid attention to the inference
from what is bad is both bad and a thing learned to what is bad is a
thing learned which is bad,26 even though they are quite aware that it
is in fact a fallacy dependent on accident. There may be a problem here,
but even if there is, it seems quite unlikely that that inference constitutes the core of the example. For if it were, Aristotles presentation of
examples in this chapter would be inconsistent, for they are all adduced
in the form of questions27 in reply to which the answerer must draw a
distinction ( at 177b10).28 Moreover,
24
With the exception of c1 and u, which merely have , all manuscripts (A, B,
D, V and also c2) as well as Boethius translation and Michael of Ephesus commentary have . Ross, however, following Poste, reads
. The thought behind this emendation is that otherwise the initial question does not seem to play any real part in the argument, but that is only correct
if one assumes that the question is syntactically unambiguous which is not the
case, as we shall see below.
25
Ross reads, with Michael of Ephesus, after , but none of the manuscripts have it, and it is not necessary at all, since Aristotle could just as well be
talking about a science of bad things. I suspect that Ross insertion of derives
from his adoption of Postes emendation at 177b17 (see the previous note).
26
Dorion, Les rfutations, 345-6; Schreiber, False Reasoning, 66-7; Fait, Le confutazioni, 187.
27
28
This phrase should not be translated as: The answerer must bring about a division in the sense of giving a divided reading, as it seems to be almost universally
rendered (e.g. Dorion, Les rfutations 172, CWA, 302, and Forster, Sophistical Refu-
126
<>
<>
Conceding this syntactically ambiguous question thus allows the sophistical interlocutor to use both constructions, the one in which
goes with an genitive of object, and the other in which a
tations, 105; see, however, Fait, Le confutazione, 63). In addition to the fact that
then Aristotle is advising something else than what he himself does in the subsequent examples, it is then not possible either to make sense of Aristotles reason for
this advice:
(177b10-12). By adding quotation marks, Ross tries to get
two versions out of the identical phrases , in order that one may translate: for ... is not the same as ... (cf. Dorion, Les rfutations,
172-3). However, this interpretation would leave one wondering what the function
of were indeed, Ross suggests that it should perhaps be omitted. This
problem is avoided by understanding Aristotle as saying that
, that is, saying is not [something which
is one and] the same [thing]. (Thus I take to be used in an only implicitly
comparative way one should supply understood with combination or division
to indicate the domain of comparison.) Thus interpreted, the reason can only be
a reason for the advice that the answerer must draw a distinction between two
different readings a pertinent meaning of the verb in the Sophistical
Refutations (e.g., 18, 176b36 and 177a4-5).
29
For the construction, see W.W. Goodwin, A Greek Grammar (London 1879), 919.
30
One might be tempted, as I initially was, to posit a switch in meaning of the term
, from what is learned (or study) to the object of what is learned (or
object of study), but then it would be hard to understand what Aristotle thought
possibly problematic about the second statement and about the second analysis
of the initial concession. Moreover, in Greek is what is learned, and that
comprises, just as in the case of what is known and what is said, both the things
or facts one acquires knowledge of and the content of the knowledge one acquires
about these things or facts. I suppose that if Aristotle did feel a difference in meaning, he would locate it precisely where he seems to locate it here, in the difference
between as identifiable with something, and with a genitive of
object: where with genitive can only be interpreted as the content learned
about things, as identified with things can be ambiguous.
31
It is in order to play down the emphasis on the fallacious inference that I have
translated the final conclusion as: Hence what is bad is a thing learned which
is bad, rather than more straightforwardly as: Hence what is bad is a bad thing
learned. One may even be tempted to deny that this inference is fallacious, by insisting that what is bad is a bad thing to learn as an art (the art of evil is obviously
evil itself), but this interpretation would involve a shift in meaning in from
128
Finally Aristotle concludes his discussion with a generalised version (xvi) of the two fallacies depending on combination (i and ii) with
which he introduced them in chapter 4 of the Sophistical Refutations:
[xvi] May you do the things which you can [do] in the way in which
you can [do them]? (
) But while not playing the guitar you have the ability of
playing the guitar. Therefore you may play the guitar while not playing the guitar. Or: he does not have the ability of that: of playing the
guitar while not playing the guitar, but when he is not doing it [the
ability] of playing the guitar.(SE 20, 177b22-6)
The crucial relative clause in the initial question one may either analyse
as:
<> <>
<> <>33
or as:
33
130
Now that all examples Aristotle gives have been discussed and shown
to be susceptible to an analysis by way of trees of grammatical composition, we are in a position to take up the theoretical issues concerning
the fallacies of division and combination. First I shall formulate the criteria Aristotle apparently presupposes for calling a reading combined
or divided and for classifying a fallacy as dependent on combination
or division; thus the first two counts of incompleteness will be taken
care of. Then I shall address Aristotles inconsistency in saying that
combination and division involve one or two separate statements, as
well as the related charge of incompleteness, that he has not given us
a reason to consider the divided and combined readings as two separate statements rather than as one ambiguous statement, as is the case
with amphiboly. Finally I shall try to absolve Aristotle from the two remaining accusations of inconsistency, by showing that despite the fact
that combination and division, just as the fallacy of accent, involve the
same kind of doubleness, there are nevertheless grounds for counting
two fallacies of combination and division, without these considerations
likewise leading to two fallacies of accent.
3.1 Combined and divided readings
As already stated at the introduction of the analysis in terms of grammatical trees as the framework in which to understand these fallacies,
the first benefit to be reaped from this way of analysing them is that it
should yield a criterion as to what will count as a combined and what
as a divided reading. For the idea suggested by Schiaparellis analysis
in terms of operators and their scope which I strove to retain by applying grammatical trees was that:
34
132
35
36
It is interesting to note that it is possible to construct examples with multiple grammatical determinators, so that a certain reading may be both combined (relative
to one grammatical determinator) and divided (relative to another): On what
ground do you think he assumes that the decision is taken?
134
dependent on combination,37 even though the questions can only be answered affirmatively because of their combined reading and can only
be used to derive unacceptable conclusions if they are understood in a
divided sense. The difficult case is (xiv), for there the sentence which
lends itself to two readings may be accepted because of only one of
its readings, but it is not the case that it is subsequently not used on
only one of them to derive a further, unwished for conclusion; rather
it serves as a kind of derivational principle which allows for jumping
from the one grammatical construction ( with a genitive of object) to the other ( identified with an object).
There are three remarks to be made about the two apparent counter-examples. The first is that if (x) is dependent on combination, also
(xvi) should be dependent on combination, for it has exactly the same
structure: both consist of a relative clause (starting with and respectively) which has a divided as well as a combined reading and a
main clause which, by leaving out the grammatical determinator, unambiguously combines the two parts which can be either divided or
combined in the relative clause. Now in the case of (xvi) we have the
following argument:
(a)
You may do the things which you can do in the way in which
you can do them.
(b)
While not playing the guitar you have the ability of playing the
guitar.
(c)
Therefore, while not playing the guitar you can play the
guitar.
(d)
By (a) and (c), you may, while not playing the guitar, play the
guitar.
The easiest way to describe the use of (a) is to take (c) to be a reformulation of (b) to make (a), including its possibility of being read in two
ways, applicable for use in modus ponens. In a quasi-formal way (with
gd standing for the grammatical determinator) the argument can be
represented as follows:
(a*)
37
(c*)
Gd[X1Y1] / X1[gdY1]
(d*)
Therefore X1Y1
And the initial concession in (x) is put to the same kind of use as (a) in
(xvi), in order to arrive at a conclusion featuring a problematic combination (X1Y1). Aristotle hints at the use (x) is put only a few lines further:
, that is, saying
is not [something which is one and] the same [thing]. (SE
20, 177b10-12)38
That you saw with eyes him being hit is only acceptable on a divided
reading, but leads through (x) immediately to the conclusion that he
was being hit with eyes.
The second remark is that, though Aristotle does not say when discussing (xvi) whether it constitutes a fallacy dependent on division or
combination, we know from (i) and (ii), which tacitly presuppose a general scheme as formulated in (xvi), that he considers this use a fallacy
of combination. The reason is that (c) is crucially readible in two ways
and, as Aristotle assumes when discussing (i) and (ii), is acceptable
on the divided reading alone, but is subsequently abused in its combined reading. Therefore it is not surprising that he also states that (x)
is dependent on combination, for it is used to argue from a proposition
which is accepted because of its divided reading, to a conclusion which
is based on its combined reading.
The third remark is, however, that we shall have to admit that example (xi) cannot be called dependent on combination, precisely because it
is the converse of (x) and can only be put to use for converse purposes,
like arguing from the proposition that he was being hit with a stone to
the conclusion that you saw it with a stone. Such an argument depends
on the divided reading of , and should
therefore be called dependent on division. Having said that, though, it
seems to me perfectly possible that Aristotle only had (x) in mind when
saying that we have here an argument dependent on combination.
The difficult case (xiv) can also be made sense of by describing similarly in a quasi-formal way the use to which its initial concession, readable in two ways, is put. As said in section 2.2, the concession
38
136
39
138
with combination and division there are two statements rather than one
ambiguous one, is that the difference between the two readings always
shows up in pronunciation, others have seen this variation as an indication that the syntactical fallacies of combination and division are in fact
merely superficially different from the syntactical fallacy which does
depend on there being one statement with two different meanings, amphiboly.40 In reply I want to argue in this sub-section that there are other
indications that at least for Aristotle the point that combination and division involve two separate statements expressed with one sentence is
one of importance, that the assumption that the difference between the
combined and divided readings always shows up in pronunciation is
mistaken, and that we may ascribe to Aristotle a more relevant account
as to what counts as a single in the sense of a single statement.
Aristotle may be loose in using both for the sentence allowing two
readings and for the statement constituting one of these two readings,
but that does not mean that the underlying account is inconsistent or
unclear.
One sentence, two statements. A first small indication that in the case
of combination and division Aristotle distinguishes between the two
statements and the one sentence expressing them appears, if my translation is correct, in the following remark, quoted before:
, that is, saying
, is not [something which is one and] the same [thing]. (SE
20, 177b10-12)41
40
41
The additions of as in most cases and not always imply that there are
cases in which the combined reading and the divided reading do not
differ at all in meaning. The same conclusion can be drawn from the
conditional remark:
If a divided and combined signifies something different, ... (SE
20, 177a34-5)
The one Aristotle is referring to in these passages, therefore, cannot be one statement with two different meanings, as with amphiboly,
but must be the sentence which can be construed in two different ways
so as to produce two different statements which may or may not have
the same meaning.
A third and final consideration can be found in an argument Aristotle gives us as to why some argument does not constitute a fallacy of
amphiboly but one of combination:
It [i.e., example (x)] has, then, also something belonging to amphibolous questions, though it is dependent on combination. For what
is dependent on division is not double (
). For not the same statement comes into being when it
is divided ( ), as it is not
the case that as well as , as pronounced with the accent,
signify something else (
).42 Rather, in the case of writing there is the
42
There are textual issues here. I omit the accents and breathings on , as Aristotle himself did not read them. The rest of the text I have adopted it is as read by
manuscript V, with the exception of and interchanged. It is also the text translated, of course without articles and with other examples, by Boethius: si quidem
non et malum et malum secundum accentum prolata significant aliud. With the
exception of the plural it is also read by manuscripts D and c. On the other hand, all the other manuscripts have the singular and, moreover, A and
B as well as Michael of Ephesus leave out . If we were to follow those readings,
we would get: ,
which would make the very same point as the text adopted: since it is not the case
that also , that is, pronounced with the accent, signifies something else.
It thus does not matter a great deal which text we adopt, though in defence of my
140
reading I want to point out that Boethius translation is actually our oldest authority and represents a tradition which apparently survives in the old Byzanthine
manuscript tradition (V is from the tenth century). I cannot exclude the possibility,
however, of Boethius translation and V featuring convergent modifications of the
alternative text. What is anyway clear is that we do not have reason to follow Ross
in emending to <> [] .
43
In order to understand Aristotles argument we need to take into account the context in which Aristotle presents it. This concerns the solution of fallacious arguments depending on combination and division:
It is clear as well how one should solve [fallacies] depending on division and combination. For if a sentence divided and combined signifies something different, one should, once the conclusion is being
drawn, state the contrary (). (SE
20, 177a33-5)
142
With this picture in mind even the third puzzling point ceases to be
troublesome: it is precisely because it is possible to solve a fallacy depending on combination or division by giving its contrary, while it is
not with amphiboly, that one cannot run the same argument for fallacies dependent on amphiboly. Thus the difference between amphiboly
44
The translation is meant to be ambiguous between speaking by silence and speaking of silence as a subject. Unfortunately the element of amphiboly is thus lost
(the example is Aristotles standard case of amphiboly, as may both be
construed as the subject of and as its object).
45
46
Dorion, Les rfutations, 82, 228; Schreiber, False Reasoning, 64; R.B. Edlow, Galen
on Language and Ambiguity. An English Translation of Galens De Captionibus (On
Fallacies) with Introduction, Text and Commentary (Leiden 1977), 26; C.L. Hamblin,
Fallacies (London 1970), 83
47
The reference is to Aeneis I 476-7: fertur equis curruque haeret resupinus inani /
lora tenens tamen;
144
Quintilianus treats the examples as cases of real ambiguity, but for the
rest they are identical in structure to Aristotles examples, one of which
even reappears in an only slightly changed form. And indeed Quintilianus states that the ambiguity can be removed by a division, that
is, a clear way of relating a word or a string of words to the one side
rather than the other, through a break in the flow of the sentence as
pronounced. However, it would go too far to use this as even an indication that also Aristotle might be thinking along these lines. First of all,
Quintilianus does not claim that division by way of an audible break in
the sentence is the only solution, for he mentions the transposition of
words as well. Apparently division might not work everywhere. Secondly, as Quintilianus presents it, a division is brought about by a conscious effort, with the clear goal to disambiguate, rather than that every
confusion between two readings disappears in spoken language.
If we then turn to Aristotle himself, we do have evidence that he
thought that it was possible to disambiguate sentences with ambiguous syntax by way of pronunciation. The clearest evidence comes from
a passage in the Rhetoric:
In general what is written should be easy to read and easy to deliver
( ... ) that is the same. Precisely that ...
[sentences] which are not easy to punctuate () do not [have],
48
Here flexus cannot refer to the inflection of the voice, as Atherton, Stoics, 479 and
H.E. Butler in: Quintilian: Institutio Oratoria Books VII-IX (Cambridge, MA 1921)
have it, for that could not be the source of the ambiguity. For my translation, see
the lemma in the Oxford Latin Dictionary.
49
The translation differs slightly from the example in Latin, in order to make it fit the
translation of the ambiguous sentence as a whole. In Latin the division is: statuam,
deinde auream hastam; vel statuam auream, deinde hastam.
50
I follow the text as edited by R. Kassel, Aristotelis Ars Rhetorica (Berlin 1976).
51
Cf. SE 21, 177b35-7: Dependent on accent there are no arguments, not among
those in written form nor among those in spoken form, save that some few might
come about.
52
146
In other words, the tree only shows how the grammatical functions are
distributed over the words of the statement, not what these grammatical functions are.
According to Aristotle this is also true for those cases where we
would feel the difference between the two interpretations depends on
whether or not a word ought to be supplied, as with the following example:
And: For one can signify both the
person knowing and the thing known as knowing with this statement.
(SE 4, 166a7-9)
53
148
[C/D]
This occurs in examples (viii), (ix) and (xii), though the last example is
probably the result of an argument having the divided sentence for its
conclusion which is then understood in a combined sense. The second
context is that from (1) an inferential principle is derived to the effect
that we may go from a statement with the same structure as the divided
reading to a statement with the same structure as the combined reading
and vice versa:
(2)
As explained above, we encounter this use in example (xiv). And thirdly, there is the context in which a premiss which lends itself to both
This structure we see in all the other examples, even if we are not given
arguments by Aristotle, as with example (xi), and if the actual structure
of examples (x), (xi) and (xvi) is slightly more complicated because of
quantification.
In the case of accent, on the other hand, we do not have so many
examples to go on, but a little consideration is enough to conclude that
there are likewise three contexts in which accent may play a part. First
again is the simple confusion of two ways of having an accent:
(1)
[A1/A2]
54
55
150
Now bearing in mind that Aristotle is after a classification of fallacies, that is, deceptive arguments, rather than a classification of types
of confusion, we see that there is a striking difference between the two
lists of contexts. In the case of combination and division we have with
argumentation pattern (3) two separate types of argument, one called
dependent on division as it illicitly uses the inferential possibilities of
the divided reading, the other called dependent on combination as it
illicitly uses the inferential possibilities of the combined reading. Also
with (2) we have reason to call, as explained already before, to call the
use of the one derived implication [C*] [D*] division and the
use of the other derived implication [D*] [C*] combination. In
the case of accent, on the other hand, we do not have any reason to call
argumentation patterns (2) and (3) after the one accent rather than the
other. They are perfectly symmetrical, so there is no reason to privilege
one over the other as the ground of the fallaciously derived conclusion.
It may be objected that I have left out in the case of accent a fourth
type of argument which does show asymmetry. For why could there be
not an argument of the following pattern:
[A1] and [A2] , therefore
However, every such an argument can be reduced to a two-step argument, the first consisting of the substitution of a synonymous word or
phrase for the confusing word (on one of its readings):
[A1] and A2 = B, therefore [B]
[B] and [B] , therefore
And this first step is of type (3), and can thus not be called after one of
the two readings of the confusing word.
On the other hand, such a reduction is not possible for (3) in case
of combination and division, for, unlike with accent, there are no substitutions with a synonym possible. We could of course paraphrase a
sentence, in the way Aristotle paraphrases the divided reading of being able to write while not writing ( )
as: [having] the ability, when one does not write, of writing ([]
) (SE 4, 166a29-30). But this goes
much further than the mere substitution for one word in a larger context we have a whole new sentence.
Thus in the case of combination and division there are two fallacies
based on single kind of doubleness because there are logically two distinct ways of making use of the confusion caused by this doubleness.
We should therefore understand Aristotles claim that there are six linguistic fallacies because there are six ways in which we may indicate
with the same words and sentences what is not the same in a similar
vein: from a logical perspective there are six ways that is, two ways
based on combination and division in which we may delude ourselves into thinking that while using the same word or statement we
are employing something with the same meaning in our arguments.
These delusions may occur on three levels: on the level of meaning, on
the level of words or statements and on the level of argumentative use.
We may be using the same word or statement, but are not aware that it
has several meanings (homonymy, amphiboly and form of expression);
we may be confusing the one word or statement for another (accent and
combination/division); and we may employ the confusion between
combined and divided statements in opposite directions (combination
and division).
4
Conclusion
152