them as strong as the correlations between expression and content.
A typology of the various coding levels can be found in Eco (1976, ch. 3), with a theory of the modes of sign production. This review of the possibilities of sign production shows that there is a semiosic continuum which goes from the strongest kind of coding to the most open and inde-terminate. The task of a general semiotics is that of tracing a single formal structure which underlies all these phenomena, this structure being that of the inference which generates interpretation.
The task of specific semiotics, on the other hand, will be that of establishing—according to the sign system in question —the rules of greater or lesser semiotic necessity for inferences (institutionalization rules).
When — still in semiotic terms — the class of the consequents is very imprecise, we have a sign which is either not yet coded, vaguely coded (the ‘symbol’), or in the process of being coded (see Eco 1979, ch. 3 for the process of code invention; and Chapter 4 of this book). Usually, code invention takes the form of the most daring of inferences, that of abduction or hypothesis.
1.11. Abduction and inferential nature of signs
Compared to deduction and induction, abduction gives rise to three dif-ferent inferential schemes (Figure 1.3), where the solid-line boxes indi-cate propositions which are already verified and where the broken-line boxes indicate tentative propositions produced in the process of reasoning.
If signs were rooted in mere equivalence, their understanding would represent a simple case of modus ponens: every time one utters /man/ one means «rational mortal animal». But one uttered /man/; therefore, one meant «rational mortal animal». This is in fact the absolutely deductive process we implement when dealing with substitutional tables, as it happens with the dots and dashes of the Morse alphabet. But it does not seem that we do the same with all the other signs, that is, when we are not invited to recognize the conventional equivalence between two ex-pressions belonging to two different semiotic systems, but when we have to decide what content should be correlated to a given expression.
If we did not know the meaning of a sign and had to reconstruct it through repeated experiences, the correct process to develop would ap-pear to be of an inductive type.
Apparently, this is how ostensive defini-tions work. Each time a native speaking a language unknown to us utters the expression /x/, he or she either points to the object ‘у’ or there is a recurrence of experience ‘y’. Therefore, that word can reasonably be in-terpreted as meaning that class of objects or of actions. The induction by ostensive interpretation is shown to be very precarious by Augustine in De Magistro. When Augustine asks Adeodatus how he would explain the leaning of the verb /to walk/, Adeodatus answers that he would start walking. And when Augustine asks him what he would do if the ques-tion were posed to him while walking, Adeodatus answers that he would start walking faster. Augustine then replies that the answer could be understood to mean that the sense of /to walk/ is «to hurry up». Obvi-ously, the accumulation of ostensive signs does not clarify the meaning of a term by simple induction. A frame of reference is necessary, a
Deduction Induction Abduction
Rule Rule Rule
Case Case Case
Result Result Result
metalinguistic (or, rather, metasemiotic) rule expressed in some way, prescribing what rule should be used in order to understand ostension. But at this point we have already arrived at the mechanism of abduction. Only if I hypothesize that Adeodatus’ behavior (in which his hurry acts as a metasemiotic mechanism which should make evident the act of walking) constitutes the interpretation of the linguistic term am I able to suppose that what he calls to my attention (Result) is a case of the hypothesized Rule. This procedure occurs in the decoding of known linguistic terms as well, when one is uncertain about what language they belong to. When someone tells me /cane!/ in an excited voice, in order to understand whether it is a Latin imperative («sing!») or an Italian holophrastic indexical proposition («dog!»), I must hypothesize a lan-guage as a frame of reference. The fact that there are circumstantial and contextual clues to direct me toward the determination of the rule does not change in principle the structure of the interpretive process.
Abduction is, therefore, the tentative and hazardous tracing of a sys-tem of signification rules which will allow the sign to acquire its mean-ing.
Abduction occurs with those natural signs which the Stoics called indi-cative and which are thought to be signs, yet without knowing what they signify. Kepler noticed that the orbit of Mars passes through points x and у (this example is given by Peirce, С. Р. 2.96): this was the Result, but the Rule of which this was a Case was not yet known (the consequents of this antecedent being, therefore, equally unknown). Points x and у could have been points of, among other possible geometrical figures, an ellipse. Kepler hypothesized the Rule (and this was an act of imaginative courage): they are the points of an ellipse. Therefore, if the orbit of Mars were in point of fact elliptical, then its passing through x and у (Result) would have been a case of that Rule. The abduction, of course, had to be verified. In the light of the hypothesized rule, x and у were ‘signs’ of the further passage of Mars through the points z and k. It was obviously necessary to wait for Mars at the spot where the first ‘sign’ had led one to expect its appearance. Once the hypothesis was verified, the abduction had to be widened (and verified): the behavior of Mars was hypothetically thought to be shared by all the other planets. The be-havior of a planet thus became a sign for the general behavior of planets.
Abduction is a very complex mode of inference, and there are prob-ably many types of abduction, ranking from the simplest and easiest to the more complex and tentative. Thagard (1978) has distinguished be-tween hypothesis and abduction in connection with the distinction be-tween overcoding and undercoding (Eco 1976, 2.14); Bonfantini and Proni (1983) have outlined three types of abduction, and I have elabo-rated on this proposal (Eco 1983). The three types of abduction I shall outline here represent a typological abstraction, a sort of rough segmen-tation of a more finely segmentable continuum. In other words, one could find concrete examples of abduction which cover intermediate positions between the first and the second or between the second and the third types, as well as complex processes that combine these different types.
(a) There is a hypothesisor an overcoded abduction when the law is given automatically or quasi-automatically. This ‘quasi’ is to be taken very seriously. Let us suppose that a verbal language represents a system of absolute equivalences and that in this language /man/ stands for «rational mortal animal». When someone utters /man/-, I must first as-sume that this utterance is the token of a type of English word. It seems that usually we do this kind of interpretive labor automatically, but it is enough that we are living in an international milieu in which people are supposed to speak different languages, and we realize that our choice is not completely an automatic one. To recognize a given phenomenon as the token of a given type presupposes some hypothesis about the cir-cumstances of utterance, the nature of the speaker, and the discursive co-text. Thagard suggests that this kind of hypothesis corresponds to my notion of overcoding and says that, since the rule is already given, the inference concerns only the decision to recognize the Result as the Case of that Rule. I agree, but I insist on the fact that, since one has to decide to connect that Rule with that Result through the mediation of the Case, then the process is never fully automatic.
Abductions of this type are also implemented in co-textual interpreta-tion. The example provided by Augustine apropos of ex (see 1.9) repre-sents an interesting instance of such an inferential process. Augustine knew that /ex/ meant the separation of something from something else; he still had to decide how to identify the two terms of this relationship within the co-text he was interpreting. His decision was quasi-automatic; however, he had to figure out a hypothesis — even though a hardly chal-lengeable one.
(b) There is an undercoded abduction when the rule must be selected among a series of equiprobable alternatives. In Chapter 2 of this book it is maintained that our semantic representations do not follow the model of a dictionary but of an encyclopedia. Therefore, we have no guarantees that the meaning of /man/ is necessarily, and in every context, «rational mortal animal». According to different contextual and circumstantial selections (see Eco 1976, 2. n), a man also can be a very virile person, a brave male, a two-footed creature, and so on. Therefore, when one ut-ters /this is a man/, we have to decide whether one says that this is a rational animal, a mortal creature, or a good example of virility, and so on. Likewise, /this is not a man/ can mean either «this is not rational» or «this is not mortal», depending on whether the sentence is about a monster produced by Doctor Frankenstein or about an angel. The deci-sion as to whether certain properties (belonging to the
