The Complete Aristotle (eng.)

and constructive purposes: for if,
when we have suggested a division, the predicate appears to hold in
all or in a large number of cases, we may then claim that the other
should actually assert it universally, or else bring a negative
instance to show in what case it is not so: for if he does neither
of these things, a refusal to assert it will make him look
absurd.
    Another rule is to make definitions both of an accident and of
its subject, either of both separately or else of one of them, and
then look and see if anything untrue has been assumed as true in
the definitions. Thus (e.g.) to see if it is possible to wrong a
god, ask what is ‘to wrong’? For if it be ‘to injure deliberately’,
clearly it is not possible for a god to be wronged: for it is
impossible that God should be injured. Again, to see if the good
man is jealous, ask who is the ‘jealous’ man and what is
‘jealousy’. For if ‘jealousy’ is pain at the apparent success of
some well-behaved person, clearly the good man is not jealous: for
then he would be bad. Again, to see if the indignant man is
jealous, ask who each of them is: for then it will be obvious
whether the statement is true or false; e.g. if he is ‘jealous’ who
grieves at the successes of the good, and he is ‘indignant’ who
grieves at the successes of the evil, then clearly the indignant
man would not be jealous. A man should substitute definitions also
for the terms contained in his definitions, and not stop until he
comes to a familiar term: for often if the definition be rendered
whole, the point at issue is not cleared up, whereas if for one of
the terms used in the definition a definition be stated, it becomes
obvious.
    Moreover, a man should make the problem into a proposition for
himself, and then bring a negative instance against it: for the
negative instance will be a ground of attack upon the assertion.
This rule is very nearly the same as the rule to look into cases
where a predicate has been attributed or denied universally: but it
differs in the turn of the argument.
    Moreover, you should define what kind of things should be called
as most men call them, and what should not. For this is useful both
for establishing and for overthrowing a view: e.g. you should say
that we ought to use our terms to mean the same things as most
people mean by them, but when we ask what kind of things are or are
not of such and such a kind, we should not here go with the
multitude: e.g. it is right to call ‘healthy’ whatever tends to
produce health, as do most men: but in saying whether the object
before us tends to produce health or not, we should adopt the
language no longer of the multitude but of the doctor.
<
    div id="section21" class="section" title="3">
3
    Moreover, if a term be used in several senses, and it has been
laid down that it is or that it is not an attribute of S, you
should show your case of one of its several senses, if you cannot
show it of both. This rule is to be observed in cases where the
difference of meaning is undetected; for supposing this to be
obvious, then the other man will object that the point which he
himself questioned has not been discussed, but only the other
point. This commonplace rule is convertible for purposes both of
establishing and of overthrowing a view. For if we want to
establish a statement, we shall show that in one sense the
attribute belongs, if we cannot show it of both senses: whereas if
we are overthrowing a statement, we shall show that in one sense
the attribute does not belong, if we cannot show it of both senses.
Of course, in overthrowing a statement there is no need to start
the discussion by securing any admission, either when the statement
asserts or when it denies the attribute universally: for if we show
that in any case whatever the attribute does not belong, we shall
have demolished the universal assertion of it, and likewise also if
we show that it belongs in a single case, we shall demolish the
universal denial of it. Whereas in establishing a statement we
ought to secure a preliminary admission that if it belongs in any
case whatever, it belongs universally, supposing this claim to be a
plausible one. For it is not enough to discuss a single instance in
order to show that an attribute belongs universally; e.g. to argue
that if the soul of man be immortal, then every soul is immortal,
so that a previous admission must be secured that if any soul
whatever be immortal, then every soul is immortal. This