To think with accuracy about collective cognition it is necessary to define with precision the subject of this cognition. This latter is not a totality of separate individual subjects constituted by the unique common reference to an object or several objects of knowledge. Individuals communicate, knowledge spreads in human communities to the extent that D.Sperber (1987) has been able to advance the metaphor - perhaps a bit risky - of a "epidemiology of representations" . It concerns of course the capture of cognitive functions that are accomplished by collective entities as groups, enterprises, organizations or institutions (P. Levy 1990); for a lot of reasons these functions have to be distinguished from purely individual cognitive phenomena. However, reflections in this area (cognitive ecology) lead to a collective entity provided with a mind and the possibility of expression in thought like any individual. This allows us to underline the fact that in this passage from the individual to the collective, which is most often done in a spontaneous manner and without examination, the collective is provided with the same attributes and the same abilities as the individual. In other words, in this agglutination of individual thought in collective thought, and whatever the mode of agglutination, essential characteristics of the individual thought are preserved. There is a sort of a priori of mereologic nature that deserves to be examined in order to justify it. For that purpose we will put forward the notion of network that will allow us to conceive rationally the human totalities as collective totalities.

The notion of network is not required only to constitute collective individual entities; it is also necessary to form the basis of the individual cognition on which collective cognition is based in its turn. Indeed, all knowledge with regard to any object (material or non) for a given individual is the result of a direct experience of this object, and of the information on the object that he acquires as a member of the network. This latter (mediate experience) absolutely determines (1) the meaning when it organizes the original experience. It is knowledge by signs : messages transported by the network are the amount of information about the object which they stand for under a certain (cognitive) aspect. The effect of a sign on an individual is the production of an interpretant (a determination of its mind of a particular kind that Peirce qualifies "a cognition of a mind" in C.P. 2 - 242). In the semiosis (interpretation of a sign) the individual is informed that the qualities or configurations of qualities that the sign that he perceives posseses are qualities or configurations of qualities possessed by another object which is the object of knowledge absent from the field of its current experience. He builds its conception of the object and organizes its knowledge by accumulation and successive restructurings that "stick" constantly to its conception any information that he gets .

For a rigorous approach to collective cognition, we believe that it is therefore necessary to take social networks into consideration on the one hand and the semiotic fact that is coextensive to each acquisition of knowledge on the other hand. This is why by analogy with the mathematical flow theory on a graph we suggest the formalization of collective cognition in terms of flows of signs on a network. The analogy is purely formal and excludes for the moment any quantitative approach like that of the theory of information. Notice that this approach organizes formally the conception of Peirce according to which, the individual lives under a "perfusion of signs" . Note also, from these common sense verifications, to what extent semiotics is the most notable absentee among the cognitive sciences, probably because linguistics badly fulfills the role that, to our way of thinking, should belong to semiotics.

In the perspective that we have just considered we therefore suppose that nodes of the networks are occupied by agents rather than by individuals and that its arcs are channels of communication effectively covered by signs. We thus underline the active character of the nodes of the network which can be interpreters as well as sign transmitters. In addition we will suppose that the network is related, that is to say that no agent or subnetwork of agents is isolated. By organizing a priori agents in networks we have ensured that there will be incorporated in the modelization of collective cognition not only interindividual communications that bring a part of the cognitive information on objects of the world but also communications between individuals, groups, organizations and institutions.

It will suffice to notice that to every group, organization or social institution one can associate a related subnetwork and to consider that all agents of this subnetwork are equivalent, (under a certain aspect that corresponds to specific communication channels which are theirs due to their membership of the group, the organization or the institution). That does not prevent individuals from preserving their interpersonal communication channels. The phenomenon of multiple memberships allows all agents to belong to several subnetworks of this type. That will authorize us to replace some subnetworks linked to social groups by an unique collective agent able to emit and to interpret signs in the same way as an individual agent. That same agent, for each of its memberships, will be able therefore to play the role of representative of each collective agent (it will be, in each of these roles, like P.Bourdieu's "Sir as" : as consumer, as viewer, as professor, as taxpayer, etc, etc,...).

We will not seek to specify unduly the formal structure of the network beyond coherence or consistency demands because that would lead us to pilot the construction of the model by considerations that do not stem from the observation of facts. We can admit a certain vagueness as to representations in the network of the different collective agents and individual agents that constitute them. It is clear however that, in many areas, it is possible to implement what are clearly more advanced abstract formalisms (cf. D. Parrochia, 1993).

In our introduction, we have made implicit reference to a notion of sign that situates us in the triadic semiotics of Peirce and this, roughly speaking as above-mentioned: a sign is something that stands for another thing for a given person. Note immediately that there is no question of introducing any pairing: signifiant/signified or any pairing: expression/content. Indeed, the saussurean signified as well as the hjelmslevean content have a universal value for a culture: they are normative in the sense that, in the semiological analysis, they are considered as the "already-there is" to which a given subject accesses or does not access according to its anterior cognitive trip. It is difficult, from there, to conceive collective cognition in its relationships with individual cognition because the first emerges from a universalism that ignores the singularity (therefore also the negativity) of the second. On the other hand, the peircean triadic sign as we have formalized and completed it (Marty, 1990) is perfectly adapted to our purpose since it allows us to individualize the relations of every agent to any object of knowledge according to its personal implication in the institutions of the meaning that rule the relationship between the signs and their objects (that we could call cultural codes if the notion of code was less rigid, less reified). We will analyze further the role and the cognitive function of these relationships.

It is now advisable to expose briefly and in a more precise manner the triadic conception of the sign that we will implement. To say that a sign is triadic is to say that it consists of three elements: the object, the sign itself and the interpretant. The object is the reality or fragment of the reality that determines the sign (a "knowable"); it can belong or not to the physical world. The sign is the concrete thing that represents and the interpretant is, in the first moment of the analysis, a connection "already-there is" between object and sign, a connection instituted by the culture (a micro-social institution in the meaning of the Institutional Analysis, cf. R.Lourau, 1970) and internalized in varying degrees by members of a community (2). The agent is therefore the place in which micro-institutions express themselves; it is the place where their particularity is to be found.

It is at this level possibly that the negation of the institution as a universal norm appears. Then one sees that this conception of the sign brings social dialectic to the heart of the model because it is easy to conceive that a denied norm, transformed by a significant majority of agents of the network, opens up the possibility, in the short term, of a transformation of the norm in question and this allows it to take charge of the temporal dynamic of meaning. In short, the interpretant is a social norm established under its universal aspect; and under its psychological aspect, internalized hic et nunc by an interpreter; it is the particular determination of a mind conditioned both by this norm that it has internalized (in the "Teacher Society" in the meaning of Lourau) and by its own experience of the field to which the norm applies.

In conclusion the signs that circulate in the network refer, in the way we have just shown, to the objects of knowledge with which they maintain an instituted relationship (but always by way of an institution) because they carry some characteristics of these objects. They allow agents of the network that receive them to enter into a mediate cognitive relationship, that is to say a relationship formed by the culture, with these objects.

How signs bring knowledge of real objects to an agent of the network?

A prerequisite is that this agent has internalized and/or builds a connection signs/object; in others terms it is necessary that what we have called a phenomenology of second intention (Marty, 1990) happens, according to whether a perceived thing - the sign - is present to the senses or a thing generally absent from the field of the perception - the object - that is present to the mind.

This prerequisite being supposed one can begin to give the general elements of a reply by calling on a universally adopted taxinomy of signs, borrowed from Peirce (and used sometimes without proper judgement) namely icon - index - symbols. An icon is a sign that possesses qualities or a configuration of qualities that the object also possesses (it is a sign by "resemblance" , a term which must be used with a lot of precaution, just to give some idea because otherwise it is too vague). An index directs the attention to the object with which it is really connected. A symbol is an interpreted sign, an instance of a social convention, law or collective habitus (one sees here the necessity of the prerequisite); more generally a symbol is a sign to which a community attributes a regular value for an indefinite period.

Iconic signs carry information about their objects since they possess qualities of this object. These qualities, configured or not, are selected de facto at the moment of the actualization of the connection with the object of which they constitute the essence. The index allows us to identify objects of which information is carried and the symbols mobilize for their own purposes, concepts, laws or habitus that are applicable to them (for example the membership of the object of a class of objects labeled by a name in the culture; a symbol states therefore that its object "falls within a concept"; in other words, is an instance of the law).

It is important to note that these categories do not constitute a tripartition of signs. More precisely and by definition, an index necessarily contains an icon and a symbol necessarily contains an index and therefore also an icon.

However for a given, somewhat complex sign, it can happen that parts of this sign function as icons, others as index, others again as symbols: these icons do not necessarily have any relationship with those that are logically incorporated in the index. The same holds good for symbols in relation to the index. The perfect sign is that which functions simultaneously at the three levels (3); it does not exist necessarily for each object to which one wants to transmit the knowledge. It will be necessary then to resort to a combination of the different categories of signs to give a complete cognitive representation. It is considerations of this order that are going to allow us to complexify the model.

Signs that circulate in the network represent therefore, for agents that are in place in the nodes of the network, objects to which they are connected, under a certain aspect. It follows that the representation of an object and, consequently, the cognition of this object, depends a priori on the performance of the sign in this connection. The latter is conditioned, in a way, by the capacities of the sign to connect, and it is clear that, in the case of connections by social convention at least, it is what motivates the choice of signs. Modes of being cover categorizations of possibilities a priori for a thing to be connected with an other. It is advisable therefore to grant them the greatest importance since they entirely determine the mediate cognition.

As far as we are concerned, since a sign is a concrete thing, we are going to categorize relational capacities of concrete things. Previously, it was necessary to notice and keep constantly in mind that a sign is a fragment of the physical universe that "lends" a part of its being to another fragment of the universe. If Peirce could write: " it is one thing to be and it is another thing to be represented ", it is precisely because the connection between a sign and an object, even if is it highly faithful , can practically never represent the whole of being of the object. Such a thing would be possible only in a "singleton" universe with an unique element that would be both sign, object and interpretant.

Modes of being are therefore categories of connective possibilities. One can define them rigorously from purely formal considerations by situating them in relational algebra from a formalization of the perception of the physical world in terms of relational structures (Marty, 1990, 1992 and still to appear). We give here only the results by underlining that it is in perfect concordance with the phaneroscopic or coenopythagorean categories of Peirce, that is to say categories of elements of phenomena. One distinguishes unary or monadic modes of being corresponding to the Peircean Firstness that connect a sign to its object by reference to common qualitative possibilities (connection by the quality: two things producing, for example, a "feeling of red" belong to the class of red things that bases the "redness" on a general quality and this common membership connects them ); binary or dyadic modes of being that are connections de facto in which two things merge in an event whose protagonists they are and that they constitute (as in the case of every action-reaction in the physical universe that illustrates the Peircean Secondness ) and ternary or triadic modes of being that institute the mediations between two other things (as in the case of a mental act that connects a sign and its object in accordance with a law or regularity of which it is an instance : this case illustrates perfectly the Peircean Thirdness ).

The laws or regularities in question preexist in the explicit form (laws of the physical world, concepts, institutionalized social conventions ), or the implicit form (habitus, instituted social , "practical" ideologies",...). There are no other "relational" modes of being possible. Indeed, a relational algebra theorem of which Peirce had the intuition and established in different but convergent theoretical contexts (Herzberger 1981, Marty 1990, Burch 1992) allows us to establish that any other mode of being can be described as a combination of the three fundamental modes of being (like that which , in chemistry, puts together atoms in molecules).

In addition these three modes of being are organized into a hierarchy by non reciprocal relationships of presupposition: the Thirdness presupposes the Secondness which presupposes the Firstness. It is obvious since all regularity necessarily concerns the Existent and Facts which presuppose the incarnate general qualities of matter.

To take into account the modes of being in constitutive relationships of the sign leads logically to a categorization of signs that complexifies the initial purpose and presents a great interest for the study of collective cognition.

A finer analysis of the triadic sign shows that the triad is established by means a twofold determination : first, from the sign by the object, second, from the interpretant by the sign. These dyadic relationships are implied, incorporated in the triad. They express modes of being in the sign of the elements of the sign. Now, the hierarchy of modes of being impose that these determinations respect the order 3,2,1 according to which a Third mode of being is able to determine a Third, a Second or a First, a Second mode of being is able to determine a Second or a First and a First can determine only a First. Consequently only 10 combinations are phenomenologically possible. There are therefore only 10 classes of signs possible. In addition these 10 classes of signs are ordered in an algebraic structure called a lattice (these results are established and widened in Marty, 1990 using algebraic category theory).

The lattice of 10 triadic classes of signs constitutes therefore a real grammar that governs the relationships of signs . It follows that every set of signs given to the perception as a collective totality is structured a priori by these relationships. It seems to us that a result of this nature that concerns notably the representation of knowledge could be usefully taken into account in cognitive science.

The knowledge relative to an object that reaches the agents of the network by means of signs is therefore determined by modes of being of the objects that they represent and by modes of being of the signs that support the representation. An object "third" (a regularity) will be able therefore to be represented by an other object third, an object "second" or an object "first", an object second by another second or by a first, a first only by a first. The same happens in the relationship of the sign to the interpretant and the two modes of being are concatenated. There can exist a sort of phenomenological entropy (for example, a third that is represented by a second can be interpreted as a second or a first that is incorporated in this second). In other words each agent builds, according to its particular relation to institutions of the meaning, a set of objects determined by the modes of being that it identifies. These objects are structured by these modes of being; they belong to classes of signs that are regulated by the lattice. These objects are, in a way, "above" signs to different levels determined by their phenomenology. They constitute the very "foliation" of knowledge transported by the representation (for an application to semantic networks, see Marty, 1992). In addition, the different levels of this foliation are linked by relationships of the lattice a bit in the manner in which pages of calculation can be linked in a spreadsheet It follows that flows of signs that reach an agent of the network are in fact "multiflows" in a different and more precise meaning that which D. Parrochia expresses(1993, p. 213). That allows a clearly more structured approach to the "communication in movement"(Ibid., p. 71)

We are now in position to formalize in new terms the real situation in which agents of the network receive and emit signs by means of channels that are at their disposition. What stands to reason, it is that one will not be able to speak of collective cognition if all the agents of the network have not obtained the same information, directly or indirectly, and if they have not all interpreted it likewise (that is to say if they have not all attributed to the same objects, the qualities, -configured or not - of the same signs, and/or if they have not all taken into account its objective relationships with other objects, and/or if they have not interiorized the same law or regularities and they have not considered that these incorporate the object and its relations in their application area).

However, we have seen furthermore that, according to the relations that agents maintain with micro-institutions of the meaning (notably corresponding to their social status) agents can spontaneously "degenerate" signs (in the sense of the phenomenological entropy, that is to say that they construct signs by taking into account only the underlining modes of being that have determined the sign to its emission). It follows that collective cognition is situated necessarily at the lowest phenomenological level: it is the common sign, inferior to the others; the highest in the lattice that characterizes collective cognition in relation to this object.

Conversely the lowest sign relative to the others characterizes the highest knowledge level reached by at least one agent of the network.

Every experience is cognitive because it consists of a relation with the object that brings necessarily some information on this object. With Peirce we define the conscience as a "bundle of feelings" and therefore the conscience of an any object is a certain bundle of qualities of feelings of which this object is the cause. Now, in addition, the signs of the lower level are precisely the qualisigns or signs of quality. It suffices therefore that all agents of the network have had a unique relation with an object in order that one could speak of the collective cognition of this object. There is, in a way, a guaranteed cognitive level. It is the product of an "emotional" immediate interpretant that builds the qualisigns materialized in the iconic sinsigns (singular things that possess qualities of the object). These latter are, approximately, images of common sense. It follows that the "civilization of the image" is that which produces surely the collective cognition of a lower cognitive level. This conclusion which is not simply the result of intuition, matches up precisely the observations of many observers of social communication (to see also Marty, to appear).

It is clear that the introduction of the new cultural mass technologies (P.Levy, 1990) by modifying existing channels and by creating new channels of communication increases the possibilities of agents of obtaining a mediatized experience of most objects. However this modification cannot generate a cognition of the superior level (in the meaning of the lattice of the classes of signs) only if it is accompanied by elements that would make it possible, namely the possibility of disposing of legisigns of all types (there are six) which would allow the elaboration of a rational knowledge of objects by means of "logical" interpretants. Indeed agents can create their own legisigns but, current conditions of the reception that isolate agents, have for consequence that laws that they elaborate spontaneously are rarely confronted with others, hence a certain epistemological solipsism.

In conclusion, we note that the interest of the semiotic formalization of dynamic cognitive phenomena that concerns the sociology of cognition is not limited simply to the reformulation in terms of more or less sophisticated common sense observations. Because one can, for example, delve into the complexity of the network and its subnetworks linked to groups, organizations and institutions so as to release properly semiotic characteristics of the different relations to knowledge of these social categories. Similarly one could imagine that some questions of Distributed Artificial Intelligence could be somewhat clarified. It is necessary also to underline strongly that the model can complicate matters - by using, for example, the hexadic sign (cf. Marty 1990)- according to the requirements of the research.

(1) That is highlighted, for example, in this well known experience of psychologists in which a person mixed with ten accomplices finishes by admitting, with all the other participants, that a round object is square.

(2) One can consider that a cultural community and the bundle of its micro-institutions of the meaning are interdefined.

(3) Reflections of Peirce on signs were notably motivated by the research of an ideal notation system for needs of logic.

Burch R. W. A Peircean reduction thesis and the Foundations of the Topological logic, Texas Tech University Press, Lubbock, Texas (1992).

Herzberger H.G. Peirce Remarkable Theorem, Pragmatism and Purpose, pp. 41-58, University of Toronto Press, Canada (1981).

Lévy P. - Les technologies de l'intelligence, l'avenir de la pensée à l'ère informatique, La découverte, Paris (1990).

- Les systèmes à base de connaissances comme modes de transmission de l'expertise, Intellectica, 1991/92, pp. 181-214, Association pour la Recherche Cognitive, Paris(1992).

Lourau R., L'Analyse Institutionnelle, Editions de Minuit, Paris (1990).

Marty R. - L'Algèbre des signes, John Benjamins B.V., Amsterdam - Philadelphie (1990).

- Foliated Semantic Networks : concepts, facts and qualities, Semantic Networks in Artificial Intelligence, Fritz Lehmann (éd.) pp. 679-696, Pergamon Press, London (1992).

- Imagen y ecología cognitiva, Associación Vasca de Semiótica, Bilbao (to appear)

Parrochia D., Philosophie des réseaux, Presses Universitaires de France, Paris (1993).

Peirce C.S. Collected Papers, Burcks W., Hartshorne H., Weiss P. (éd.) Harvard University Press. Cambridge, MA. (1931-58).

Sperber D., Anthropology and psychology : Towards an epidemiology of representations, Man (M.S.), n° 20 pp. 73-89 (1987).