Two problems are already apparent after the first two chapters. The first concerns Foucault's equivocal use of the terms 'statement, event, and discourse.' The second problem is to set out the kinds of 'relations that may be legitimately described between statements.' A key issue here concerns the criteria by which we might say that two statements are 'in continuity,' that they are part of a group (keeping in mind that we've rejected all received notions of continuity in the last chapter). Foucault examines four 'hypotheses' on this matter. The first says that two statements belong to the same group if they refer to the same object. Two statements both belong to psychopathology, for example, if they refer to madness. This hypothesis falls apart, however, because there is no consistent thing called 'madness' throughout history. 'Madness' is not a single object, but a whole slew of different objects, emerging successively or simultaneously at different points in history. Thus, the unity of discourses on madness must actually be based upon 'the interplay of the rules that make possible the appearance of objects during a given period of time.' Far from there being a consistent 'madness' that defines all statements about it as statements belonging to psychopathology, there is only a broad range of statements (in 'daily practice, in law, in religious casuistry, in medical diagnosis') whose relations define the development of successive versions of madness. To describe a discursive unity, we then describe those relations. But this leaves us only with a paradox: such a description must address the gaps and differences that define the dispersion of the statements under consideration, thus attempting to define the unity of a set of statements by 'formulating their law of division.'
A second hypothesis would be to define a group of statements by certain relatively constant relations of similarity between them (rather than by what they refer to). Thus, a discursive unity might be defined by a certain style, a certain 'way of looking,' or a similarity in vocabulary or metaphor. But this method, too, breaks down into the sheer multiplicity of statements. At one point, for example, Foucault thought that medical discourse might be defined by a certain kind of descriptive mode. But this theory was countered by a recognition that medical discourse is not just a series of descriptive statements, but also included 'ethical choices, therapeutic decisions, institutional regulations, teaching models,' and so on. Furthermore, the very notion of description itself changed throughout history as new models and standards emerged. Any organizing system that sought to define which statements were medical 'disintegrated as soon as it appeared.' Again, rather than individualizing a group of statements based on a simplistic idea of their similarity to each other, we must individualize the specific 'coexistence of these dispersed and heterogeneous statements; the system that governs their division the way in which they interlock or exclude one another the play of their location, arrangement, and replacement.'
A third hypothesis would unify a group of statements via the constant, abiding concepts that govern their method; grammar, with concepts like the noun, the verb, or even the word (as the sign of representation), is the clearest example here. But again, no sooner do we choose our stable concepts than we can note transformations of them and emergences of antithetical concepts further on. Again, any discursive unity must operate on the level of these very transformations and incompatibilities, on the variable differences that separate statements. Finally, there is a fourth hypothesis: statements or discourses could be grouped by their 'theme,' the theory that 'direct[s] research from afar.' Thus, we could put all discourses on 'evolution' or 'economics' in a unitary group. If we did, however, we would be ignoring or eliding the fact that a theme like that of 'evolution' can actually cover multiple, even opposed, discourses. Thus, evolution in the eighteenth century marked a discourse about the continuity of species, whereas in the nineteenth century it marked a discourse on the interaction of species with the environment. Discourses on economics, too, even when utilizing the same set of concepts, can operate on two entirely different theories of value. Again, we must direct our attention to these shifts and differentials, which in the case of thematics shows us not a constant theme but a 'field of strategic possibilities that permit the activation of incompatible themes.'
These four hypotheses, then, have each failed, and each generated a new hypothesis. Rather than pursuing any of these four kinds of discursive unity, Foucault's method will be to describe 'system[s] of dispersion' between statements, and seek regularities only there. Wherever such regularities of dispersion can be found, we will say there is a 'discursive formation.' The rules that govern this dispersion will be the 'rules of formation.' These rules are 'rules of existence' for a given formation, but also rules 'of coexistence, maintenance, modification, and disappearance.' Such an approach, Foucault notes, is dangerous in that it may not lead us back to the discursive unities we thought we knew. In fact, it may threaten to leave the historian with only 'a blank, indifferent space, lacking in both interiority and promise.'
Having dismissed most major, traditional forms of historical continuity in the previous chapter, Foucault continues by rejecting a series of much more subtle possibilities for unity (possibilities that he himself tried and rejected). The chapter shows that Foucault's aim is not simply to throw out tired old ideas about what unifies various historical statements, but also to reconceive of what it means to look for historical unity or continuity in any form.
The most immediate problem to emerge from Foucault's initial suspension of 'discursive unities' in the last chapter is the difficulty of saying anything about the statements made within disciplines relate to one another. Foucault now begins to address what might seem the most obvious place to start: the divisions of discourse into categories like economics, medicine, or grammar. Although Foucault admits that these groupings of statements seem 'quite obvious' (after all, many historical statements classify themselves under these divisions), this chapter proceeds to show that they are in fact so tricky to define that they cannot be taken for granted at all. Thus, this chapter is partly framed in terms of false starts: Foucault tries out four possible ways to begin specifying coherent groups of statements (by a common object of study, a common style or viewpoint, a constant set of operative concepts, or a common theme), and four times he finds that the relations between the statements are too multiplicitous, shifting, and even dissonant to submit to such organizing principles. A given discourse, even if it can be identified as such, develops as much through sudden irruptions, transformations, contradictions, and differences as it does through constancy or regularity.