Naturally it is legitimate to inquire whether we are entitled to deduce this idea of an open-ended encyclopedia from a few allusions in Leibniz and an elegant metaphor in the Encyclopédie, or whether instead we are attributing to our ancestors ideas that were only developed considerably later. But the fact that, starting from the medieval dogmatics of the Arbor Porphyriana and by way of the last attempts at classification of the Renaissance, we slowly evolved toward an open-ended conception of knowledge, has its roots in the Copernican revolution. The model of the tree, in the sense of a supposedly closed catalogue, reflected the notion of an ordered and self-contained cosmos with a finite and unalterable number of concentric spheres. With the Copernican revolution the Earth was first moved to the periphery, encouraging changing perspectives on the universe, then the circular orbits of the planets became elliptical, putting yet another criterion of perfect symmetry in crisis, and finally—first at the dawn of the modern world, with Nicholas of Cusa’s idea of a universe with its center everywhere and its circumference nowhere, and then with Giordano Bruno’s vision of an infinity of worlds, the universe of knowledge too strives little by little to imitate the model of the planetary universe.
1.6. The New Encyclopedic Models
Whether or not this was the unconscious model for a new ideal of encyclopedic knowledge, it must be said that the first real efforts at creating semantic representations in encyclopedic form did not get underway until the second half of the twentieth century and only after a fierce debate regarding the shortcomings of any dictionary representation.33
Clearly, although the idea of the encyclopedia as postulate and ideal model is infinite, all that could be attempted were limited and local representations, which however did not exclude the possibility of their progressive and potentially limitless enrichment.
The new encyclopedic models assumed a number of formats, among them:
(i) Matrices representing the presence or absence of traits chosen ad hoc to account for the differences among items belonging to the same semantic subset, such as chair, armchair, sofa, etc. (cf. Pottier 1965).
(ii) Contextual selection models (specifying the various meanings a given lexeme may take on in different contexts) (cf. Eco 1975, 1984a).
(iii) Models by Cases that include Agents, Objects, Instruments, Purposes (the verb “to accuse,” for example, is defined as an action in which a human Agent communicates to a human Object by means of a verbal Instrument with the Purpose of revealing to him that the action of another human Object is evil; whereas “to criticize” is explained as the action of a human Agent who by means of a verbal Instrument speaks to a human Object with the Purpose of demonstrating that the action of another human Object is open to censure; or else the verb “to kill” is analyzed as the action of a human Agent which causes a change of state, from living to dead, of an animated X—further specifying, by the use of the English verb “to assassinate,” that the X in question must be a political figure) (cf. Fillmore 1968, 1969, 1977).
(iv) Representations that take into account, in the case, for instance, of a term like “water,” the properties that determine its extension or its referent (its being H2O); labels of a quasi-dictionary variety, such as being Natural and Liquid; as well as stereotypical notions like Colorless, Transparent, Tasteless, Odorless, Thirst-Quenching (cf. Putnam 1975, 12).
(v) Representations that take into account all possible properties of a term and specify, for a chemical element for example, odor, color, natural state, atomic number, effects, history, etc. (cf. Neubauer and Petöfi 1981).
None of these proposals, however, had had recourse to network structures. It is in the field of artificial intelligence that frame-, script- or scenario-type representations appear, registering each stage of a sequence of typical events (for instance, what does “going out to a restaurant” mean: entering, sitting down at a table, ordering from the menu, eating, requesting the bill, etc.)—all models that have proved successful in the field of artificial intelligence, where, in order for a computer to understand a text and draw conclusions from it, it must first be provided with all of the competences with which (even without their being aware of it) the average human being is endowed (cf. Schank and Abelson 1977; Schank and Childers 1984).
But it is with Quillian (1968) that the notion of a semantic network, structured as a labyrinth of interconnected nodes, first appears. To simplify things, all we have to do is take another look at Figure 1.17. Any node can be taken as the point of departure or type of a series of other nodes (tokens) that define it (let’s say the point of departure is dog and that this node is defined by its links with animal, quadruped, able to bark, faithful, etc.). Each of the defining terms may in its turn become the type of another series of tokens. For instance, animal could be exemplified by dog, but also by cat, and would include quadruped but also biped; or, if a node cat were to be identified, it would be defined by a number of nodes it shared with the definition of dog, such as animal and quadruped, but it would also refer to nodes like feline, which it shares with tiger, and so on.
A network model implies the definition of every concept (represented by a term) through its interconnection with the universe of all the concepts that interpret it, each of them ready to become the concept interpreted by all the others.
If we were to expand the network of linked nodes ad infinitum, from a concept assumed as type it would be possible to retrace, from the center to the outermost periphery, the entire universe of the other concepts, each of which may in its turn become the center, thereby generating infinite peripheries.
Such a model is also susceptible of a two-dimensional graphic configuration when we examine a local portion of it (and in a computer simulation, in which the number of tokens chosen is limited, it is possible to give it a describable structure). But it is not in fact possible to represent it in all its complexity. It would have to be shown as a kind of polydimensional network, endowed with topological properties, in which the paths become longer or shorter, and every term gains in proximity with the others, by way of shortcuts and immediate contacts, while remaining at the same time linked to all the others according to historically mutable relationships.
It has been said that, if we assume a maximal notion of competence about the world, the meaning of a term would then consist of all the true propositions in which it has appeared or could appear. In fact, this would presuppose the ideal model of the encyclopedia. But in scientific practice and the way in which, in our daily lives, we try to make sense of sentences, we do not make a global appeal to the encyclopedia for every sentence, and it is the content that selects the local zones of competence that must be activated. Two flexible criteria may be assumed: (i) information is potentially part of the average encyclopedic competence if it can be supposed to be sufficiently shared by a collectivity (which may also be a “regional” collectivity—in this sense the definition of neutrino would form part only of the regional competence of a community of nuclear physicists—see the concept of Specialized Encyclopedia discussed below in section 1.9); (ii) the format of the network to be activated is prescribed by the contexts and the circumstances of the proposition (accordingly, if someone uses the word torus in speaking of topology a network is constituted which is concerned with mathematical objects, and all concepts regarding the fields of architecture, anatomy, and botany are excluded).
While in an ideal encyclopedia there are no differences between necessary and contingent properties, it must be admitted that, within a specific culture, certain properties appear to be more resistant to negation than others, on account of the fact that they are more salient: it could feasibly be denied, for instance, in the light of a new system of classification, that a sheep is ovine, or again this particular trait might not be deemed necessary to the understanding of the term sheep in the sentence: “the sheep was bleating in the field.” There can be no doubt, however, that it is hard to deny that a sheep is an animal—and the characteristic also remains implicit for the comprehension of the example we just cited. It has also been observed (Violi 1997: sect. 2.2.2.3) that some traits seem to be more resistant than others, and that these uncancelable traits are not only
