The heterogeneity of discourse: expanding the field of discourse analysis

This contribution aims to outline some limits of discourse analysis today. Discourse analysis may not privilege any specific kind of data, but, like modern linguistics since the beginning of the 20th century, most discourse analysts focus on a restricted area within the manifold manifestations of discourse. They take for granted that discourse must be modeled after conversational practices, and/or that the basic relevant unit of discourse analysis is the pair text/genre. This restriction can be explained by the historical context in which discursive approaches have emerged. However, it can be argued that such a standard is not relevant for a wide range of data, that discourse practices can be divided into various regimes, which correspond to various models of communication and must accordingly be analyzed with specific concepts and toolkits. This article discusses two main phenomena that call this standard into question: (1) the notion of genre on the Internet; (2) “textless sentences” and authorship. Indeed, these all imply another conception of discourse activity that deserves to be taken into account if we are to consider discourse activity in the full diversity of its manifestations.


Introduction
A t this point of the twenty-first century, the field of discourse analysis is faced with a paradox. While discourse analysts are supposed to study any manifestation of discourse, they often limit themselves to either "ordinary" conversations or certain genres. The discourse genres studied by discourse researchers often belong to media and politics and other social areas such as education, business, justice, the health system. 1 I think that this problem does not strike most discourse analysts because they presuppose that discourse is basically homogeneous. Such a presupposition is based on two postulates. One is that discourse must be modelled after conversational practices, and/or that the basic relevant unit of discourse analysis is the pair text/genre. Each of these postulates may be linked to a specific tradition within discourse studies. North-American research, for instance, tends to give priority to conversation while 2 many European researchers focus on text and genre. 3 In our current globalized world, they both combine and contaminate each other: indeed, most discourse analysts studying conversation consider conversation as a genre or as covering a wide range of genres while, conversely, those who study genres from institutional settings tend to focus on talk. 4 However, the many researchers belonging to both traditions (and I include here myself too) share a certain conception of discourse as a human practice. Specifically, they see it as an interaction between flesh-and-blood people who act in well defined settings, and develop strategies in keeping with their own aims and interests, to influence each other or modify the situation. However, this implicit model does not fit for a wide range of other substantial data.
In this contribution I will start by showing that many researchers who focus on immediate interaction bypass many other types of discourse, while others, notably register theorists, do tackle the question of heterogeneity head on, but in a rather superficial way. Contrary to these approaches, I claim that discourse is basically heterogeneous, that it can be divided into various regimes, which correspond to various models of communication. I will mention several phenomena in support of this proposition. I will first make a distinction between conversation and "instituted genres", and emphasize the heterogeneity of the very notion of genre. I will then demonstrate that the functioning of the Internet challenges some basic presuppositions of discourse analysis, which has developed since the 1960s in a world where recorded orality and printed texts prevailed. In the last section I will propose a distinction between "attached" and "detached" utterances, for which there are neither "speakers" nor "addressees" in the ordinary sense of each word.
How heterogeneity is overlooked I will begin here with a quick survey of past issues of Discourse Studies, the most important journal in the field, which proves very revealing. 5 This journal focuses on talk, although it claims not to be affiliated with any specific school, defines itself as "an international journal for the general study of text and talk", and aims "to publish outstanding research in any domain of the study of spoken and written discourse". Six issues are published every year. I examined the years 2010, 2014 and 2015 (all in all 98 articles); to reach 100, I added the first two articles of the year 2016.
Out of these 100 articles, 55 deal with oral interactions, which have been transcribed and analysed with the help of methods from the toolkit of Conversation Analysis. To these can be added 10 articles dealing with monologic oral utterances (actually most of these "monologues" have been detached from interviews). Overall, 65% of the articles therefore deal with oral data. Now, let us have a look at the remaining 35%: -Five articles involve the study of data from the Internet, but in the form of forum discussions or email; these are tackled with concepts and methods from Conversation Analysis; and not data that cannot be studied with the help of the usual toolkit of discourse analysts. -11 articles ignore the distinction between spoken and written utterances: they study discourse markers, from the perspective of linguistic pragmatics or text linguistics; -Out of the remaining 19 articles, two articles are devoted to the study of images, and one makes a taxonomy of political speakers.
As a result, only 16 articles study written data, which are based on various sectors: academia, newspapers, the health sector and politics. There are no studies on the discourses of Art, literature, Law, religion, philosophy or science. Scientific discourse is not dealt with specifically, but simply as a type of academic discourse (conferences or handbooks), or as corpora for the study of discourse markers. The only article about religious discourse (Loeb, 2014), using Conversation Analysis methods, analyses "call and response" sequences in "Bible study meetings" of an Afro-American community. Such a restriction in terms of the data being studied can be explained by the convergence of the founding principles of modern linguistics that assumed that only oral data were relevant for the study of language, and those of the micro-sociological trends, which played a key role in the development of discourse studies. The pragmatic conception of language also favoured the emphasis on oral interaction. Significantly, Grice's (1975) theory of implicatures, which draws on the "cooperative principle", is organized around "conversational maxims".
But other reasons can be invoked to explain why discourse analysis focuses on oral interaction: -Discourse analysis appeared within a given historical context, in the 1960s in Western countries, when TV had become preeminent and for the first time in history researchers had at their disposal technical resources to easily record long spontaneous interactions. -Discourse analysis has mainly developed in the area of Social sciences, and has remained relatively absent from the departments that up until then had been studying prestigious texts (Literature, Law, Philosophy…). In contrast, doing discourse analysis implied paying high attention to areas of discourse that before were considered as peripheral, and were now considered as "authentic" discourse activity: ordinary conversation, media, politics, education… As is often the case, this has become a routine and a norm for discourse analysts: in this way, they set themselves apart from tradition, and conveyed a specific ethos. -Studying conversations, political discourse or TV shows makes it easier to link linguistic phenomena with social contexts, which is the main purpose of most discourse analysts.
Obviously, this link is much more difficult when one studies literary, scientific or religious corpora.
One could object that my diagnosis is excessive, that many discourse analysts take into account the "heterogeneity of language" (Stubbs, 1993: 11). Indeed, discourse analysts are always distinguishing spoken and written utterances, formal and informal settings, monologic and dialogic discourse and so on. ARTICLE PALGRAVE COMMUNICATIONS | DOI: 10.1057DOI: 10. /palcomms.2017 This is particularly the case of corpus linguistics (Biber, 1988, Sinclair, 1991, which itself has been influenced by the theory of "registers", within the systemic-functional linguistics of M.A.K. Halliday. Registers are hybrid realities: from an extra-linguistic viewpoint, they are identifiable upon the basis of the situation type they are associated with; from a linguistic viewpoint, they are characterized by a set of distinctive linguistic features. In fact, in this kind of approach, saying that language is "heterogeneous" is only a way of acknowledging that the system varies according to situations. If for instance literature or religion is defined as a register or a "style" (Crystal and Davy, 1969), it is because literature and religion are considered as specific uses of English language, not because there exists a "literary discourse" or a "religious discourse" forming a part of institutions that play a specific role in society. Furthermore, despite efforts to consider any area of language use as a register, the status of conversation remains a problem: "there is broad agreement among linguists that conversation is not just any register but a prominent one for various reasons (…) Further, conversation is regarded by many linguists as the fundamental basis for other registers" (Rühlemann, 2007: 8).
On the need to recognize the heterogeneity of discourse "Conversational" versus "instituted" practices. I consider that the kind of heterogeneity discourse analysts are usually referring to-whether they use the notion of register or not-is not sufficient. My own conception of heterogeneity is stronger: discourse practices can be divided into various regimes, which correspond to various models of communication and must accordingly be analysed with specific concepts and toolkits (Maingueneau, 2014). A fortiori, such a conception of heterogeneity has nothing to do with the philosophical assumption of the Althusserian trend of discourse analysis (Pêcheux, 1975) which claims, by referring to psychoanalysis, that the discourse and the subject are submitted to a "radical heterogeneity" (Authier, 1984, in Angermuller et al. (eds) 2014.
Nobody can deny the importance of conversation in the construction of subjectivity and social order, but this does not mean that it is the centre and the model of discourse activity, or that the concepts and the methods of discourse analysis must be based on this kind of data.
Instead of emphasizing the unity of discourse "practices", two discourse regimes, subject to different constraints, can be distinguished: conversational practices, on the one hand, and instituted practices on the other. Conversations are not closely related to institutions, roles or stable scripts; their textual organization and their contents are usually rather fuzzy; their frame constantly evolves during interaction, as the participants constantly negotiate their roles. Conversations are subject to predominantly local and horizontal constraints, whereas the constraints of instituted genres are predominantly global and vertical: they integrate speakers into more or less ritualized communication devices where speakers are given specific roles. Obviously, the distinction between these two regimes is not clear-cut, and verbal practices that have the properties of both regimes can easily be found. Moreover, both regimes can be used in the same speech event. But generally, while in instituted practices the very notion of genre of discourse is fully relevant, in the conversational regime it is highly problematic. Conversation cannot be easily divided into distinct categories. It is important to note, however, that we should abstain from establishing a hierarchy between these two regimes, which are intertwined and complementary. People's lives are made of the interaction between these two ways of shaping subjectivity.
"Scenes of enunciation". The very notion of genre in "instituted practices" must also be refined to take into account the heterogeneity of discourse. Various kinds of discourse genres must be distinguished, according to the way speech is staged. I propose (Maingueneau, 1998) to analyse a genre as a "scene of enunciation". The scene of enunciation can be broken down into three components: -An "enclosing scene" 6 : roughly speaking, this corresponds to main areas of discourse within a society (religious, political, administrative, medical and so on); -A "generic scene", which assigns roles to actors, prescribes the right place and the right moment, the medium, the text superstructure and so on; -A "scenography": texts belonging to the same generic scene may stage different scenographies. For example, preaching in a church, can be staged through a prophetic scenography or a meditative scenography. In the former case, the speaker will speak in the way prophets do in the Bible and will give the corresponding role to his addressees; in the latter case, he will feign to be speaking to himself.
On this basis a distinction can be made between four types of genres depending on the degree of variation and/or originality that is permitted within the genre: -Type 1 genres: these are not subject to variation, or only to very little variation; their speakers follow strictly pre-established formulas and schemes: telephone directories, birth certificates and so on. In fact, we cannot really speak of "authors" for such texts. -Type 2 genres: speakers produce singular utterances while obeying a script, a routine: television news, business correspondence and so on. But some of these genres may tolerate distortions and give speakers the possibility of using an original scenography: a travel guide, for example, may be presented as a friendly conversation, a novel and so on. -Type 3 genres: they require the invention of original scenographies: advertising, folk songs, entertainment programs on television and so on. If you make an advertisement for a face cream or a car, you must invent the scenography through which it will be presented. Of course, many scenographies are stereotypical, but the logic of such genres urges people to innovate perpetually. However, these innovations are not supposed to modify the frames imposed by the generic scene or to question them. -Type 4 genres: instead of following a strict pattern, an author with individual experience self-categorizes his or her own verbal production, as "essay", "fantasy", "thoughts", "story" and so on. Generic labels such as "newspaper", "talk show" or "lecture" are given to activities that exist independently of these labels (actually, many discursive practices have no name at all). But when a religious author, a politician or a moralist calls his or her text a "meditation", an "utopia" or a "report", these names cannot be replaced by another one because they are the consequence of a personal decision, the manifestation of an act of positioning inside a certain field. Such labels contribute significantly to the way the text is to be interpreted, but they do not refer to the actual communication activity: if I name "meditations" an academic book about philosophy, this does not mean that it is not an academic book.
Genres of the third and the fourth types are similar in many aspects: both must set up stimulating scenographies to convince their audience, and give sense to their own discursive activity by proposing a frame in harmony with the very content of the utterance. But, while advertising texts (third type) have a specific purpose (chiefly making people buy something) and are always searching for the best way to achieve this objective, religious writers or novelists cannot really define what they are aiming at when publishing their texts: "there remain some genres for which purpose is unsuited as a primary criterion" and which "defy ascription of communicative purposes" (Swales, 1990: 47).
Significantly, most discourse analysts focus on the first three types of genres which prescribe roles, settings, registers…-in particular the second one-and ignore that texts can also be created by the invention of original scenographies.
In addition, some categories that discourse analysts use do not correspond to genres. This hierarchical distinction between the components of the enunciation scene-Enclosing scene4Generic scene4Scenography-appears to be insufficient to take into account categories such as "interview", "letter", "diary", "report"…, which cannot be considered to be genres, but rather as "hypergenres" (Maingueneau, 1998) (or "disembedded genres" (Fairclough, 2003: 68-70)). They can be used during long periods, in different discourse areas, to frame texts belonging to a wide range of genres. The constraints that they impose are very poor; to categorize a written text as a dialogue, for example, one only needs to stage at least two speakers.
We can go further by claiming that the universe of discourse is not a monotonous space where any discourse activity (from conversation to highly elaborated texts) can occur. On the contrary, one can claim that some areas have a specific status. For example, it can be argued that there exist atopic discourses (pornography, for instance) (Maingueneau, 2008), which do not occupy a legitimate place in society. As for "selfconstituting discourses" (aesthetic, religious, scientific and so on) (Maingueneau Cossutta, 1995;Maingueneau, 1999) they are paratopic, in that they must show they both "belong" and at the same time do not "belong" to ordinary society. Bordering on what is unspeakable, they must negotiate the paradoxes that such a position implies. To found other discourses without being founded by them, they must set themselves up as intimately bound with a legitimizing transcendent source and show that they are in accordance with it, owing to the operations by which they structure their texts and legitimate their own context.
In the following pages I will question another facet of the presupposition that the universe of discourse is homogeneous: the idea that any utterance (spoken or written) is a text that belongs to a genre. I think that this postulate can be challenged by taking into account various kinds of data. I shall consider here the Internet and what I call "detached enunciation".

The challenges of the Internet
Paradoxically, now that the field of discourse studies has high visibility, the universe from which discourse studies emerged is vanishing. New communication devices subvert the very distinction between orality and writing, and so we have to rethink many categories: textuality, speaker, addressee, utterance, memory, storage, circulation, etc. We can no longer consider technology as just an element of the "context": it now needs to be considered as a true actor in the communication process. Such a transformation relates to the data-since the Internet offers new kinds of semiotic productions-but also to the very conditions of research, which depend increasingly on sophisticated programs and data bases. The problem is that most discourse analysts seem to live in a world where traditional faceto-face talk is still the norm of communication. If we consider the handbooks and the articles published in the field, a peripheral role is given to corpora produced by new technologies, except if they can be tackled by using the toolkit of Conversation Analysis. This is particularly the case with chats, forums, emails, phone textmessages and so on. As the focus of discourse analysis is not on the most important aspects of the Web, its study belongs mainly to specialists from other fields. The Discourse reader (Jaworski and Coupland, 1999) does not mention the Internet. But, surprisingly, this is also the case 13 years later in The Routledge Handbook of Discourse Analysis: the introduction does not mention the existence of new communication technologies and none of the 46 chapters deals with this topic.
In my view, most discourse analysts do not pay much attention to the Internet, probably because their concepts and their methods are poorly adapted to this kind of data. This is particularly the case for the notion of genre.
As said earlier, genre is structured by a double hierarchy: between Enclosing scene4Generic scene4Scenography, on the one hand, and between hypergenre and genre, on the other hand. The problem is that this double distinction is hardly compatible with the Web. To the detriment of generic and enclosing scenes, on the Web, scenography plays the key role: the main problem for the producers of Websites is the way they stage their communication. This can be explained by the fact that all units are of the same kind (they are "websites") and are subject to strict technical constraints. The need to circulate from site to site reinforces this homogenization. Of course, Websites can be divided into main categories (blogs, social networks, news…), but these categories are not genres in the true sense of the term: they are hypergenres.
A blog, for example, is generally considered as a "bridging genre" (Herring et al. 2005), which traverses thematic categories (personal, institutional, commercial, educational and so on (Myers, 2010)). Its communicative properties are minimal: some being (with a proper name) speaks about him/herself to somebody who is visiting his/her website. What matters for the people who manage blogs is the invention of suitable scenographies inside the frame imposed by this hypergenre. For example, in her study of 80 blogs of French politicians which were on line in September 2007, Lehti (2011) has distinguished five kinds of scenographies: "diary", "scrapbook", "notice-board", "essay" and "polemic". By imposing their own scenography, the designers of websites give meaning to their own activity, to the images of themselves and of their addressees, to the relationship between them and so forth.
Such a transformation also impacts on textuality.Websites can no longer be reduced to "speech": they are screen pages, images that integrate texts, pictures, sounds and videos. Whereas the classical conception of genres focuses on verbal scenography, on the Web, verbal scenography is part of a "digital" scenography, which can be analysed as two inseparable components: iconotextual: the site includes images, and it is itself an image, limited by the screen; reticular: a site is a network of pages and is linked with other sites.
The very notion of "page" is questioned. A "page" on the Web cannot be taken in at a glance: the screen displays only a part of a whole that cannot be considered as a unity. On most sites, pages are mosaics of heterogeneous modules: ads, slogans, videos, quotations, beginnings of articles, diagrams, pictures… Many modules are not autonomous: they are only fragments which, when we click on them, give access to other pages of the same site or other sites. The digital world is increasingly a world of decontextualised chunks which can be combined in countless ways, independently of a reference to a text totality they would be a part of.
Even the identity of a page is a problem. What can be seen on the screen exists only at a specific point in time. According to the ARTICLE PALGRAVE COMMUNICATIONS | DOI: 10.1057DOI: 10. /palcomms.2017 kind of sites we consider, modules can change at any moment: some remain stable for a certain period, whereas others are always moving (the score of a tennis match, the Stock Exchange…). This calls into question what was considered as one of the main properties of a text: stability.
Researchers often analyse what I call designed textuality (the planned production of oral or written texts), as opposed to immersed textuality of oral interaction. But the Internet is characterized by the browsing textuality of hypertexts.
This implies new ways of "reading", the possibility of passing instantly from one "page" to another in an open space. "Hypertext" has been intensively studied (see for example Landow, 1994Landow, , 2006Barnet, 2013): it is up to every single user to make the hypertext he/she is "reading". Such a practice questions an assumption which is at the heart of traditional Humanism: the constitutive relationship between one Subject (the author or the reader) and one well-defined text that is already given, waiting to be deciphered.
The problem is that the conception of textuality that commonly prevails in discourse analysis is implicitly based on "immersed" textuality in American approaches or "designed" textuality, whose core is the generic scene 7 , in European trends. Whereas the usual conception of genre implies a map of verbal activities (the universe of discourse can be divided into various main areas, which in turn can be divided into well differentiated speech institutions and genres), the Web, by giving a key role to scenography and hypergenre, implies a "de-differentiation" of speech areas. On the screen one can see more or less transient images, moving mosaics of modules, nodes in networks, rather than texts that could be anchored in specific places within welldefined institutional territories.
I have just evoked a very limited aspect of the Web: the relevance of the notion of genre. But technical innovations are increasingly challenging our traditional categories. For example, our world swarms with messages produced by non-human beings that do not communicate in the usual sense of the term, and that, strictly speaking, are not speakers. We could instead call them "angelic speakers" (Maingueneau, 2014: 195), which cannot exist independently of the production of their utterances. Actually, this term covers a wide range of distinct phenomena: messages that appear on computer screens or displays in airports, stations, shops and so on, messages that we hear on automatic voice messengers answering phone calls, messages voiced by G.P.S. devices or automatic checkouts for payment and so on. In these examples, only pre-formatted utterances are produced, but in other cases complex messages can be generated, belonging to a given genre (a written abstract, a weather forecast, a technical report and so on.). Some programs write poems or novels. When we think about it, the invention of writing many centuries ago had already offered the opportunity to create a strange kind of "speaker": authors, who could for example be fictive, collective, or abstract.

Detached utterances
Aphorising enunciation. From a very different point of view, the question of "textless sentences" also allows us to challenge the way notions such as text, genre and speaker are commonly used.
This question draws on the perspective of French "enunciative pragmatics" (Angermuller, 2014), which gives prominence to the distinction between "speaker" and "enunciator". While the speaker is the individual, considered as someone belonging to the world outside language, the enunciator exists only through enunciation: this term refers to the role of appropriating and setting language in motion during the process of enunciation (Benveniste, 1966), or to a being that is responsible for a viewpoint which is staged in the utterance (Ducrot, 1984): the enunciator of proverbs, for example, is not the speaker but "popular wisdom" or "common sense".
In any society, many "detached sentences" circulate. By "detached sentence", I mean a sentence that does not belong to a text, that is, to a cohesive sequence of sentences. Mottos or slogans immediately spring to mind: they are autonomous, that is, detached by nature. But many detached utterances have been extracted from texts, such as maxims, soundbites, titles in newspapers, photo captions and so on.
Most of the time the people who detach a fragment modify it, even when the original text is right next to it. Let us have a look at this title of an interview in a magazine: J. Foster: Once fear has touched you, you realize it's been hiding beneath the surface of daily life Clearly, the content of the detached utterance is not exactly the same as the corresponding fragment of the interview: And once that fear touched you, you realize that it's been there all along, hiding beneath the surface of your everyday life. (Newsweek, 10 September 2007, p. 64) The ethos of the enunciator is slightly different too: in the title, Jodie Foster seems to utter a sort of maxim.
These modifications are quite normal if we assume that detached utterances do not imply the same kind of enunciation as usual utterances do, and if we therefore make a distinction between what I call "aphorising enunciation"-or "aphorisation"-and "textualizing enunciation". While in "texualizing enunciation" speakers produce texts belonging to genres, aphorising enunciations are not texts or fragments of texts, but utterances for which the very notion of text is irrelevant.
To this claim that they are foreign to genres and texts, the objection could be raised that aphorisations are part of texts: that is, those texts into which they are inserted. Indeed, it cannot be denied that aphorisations are always parts of texts, but their authority is based on their claiming to be foreign to texualizing enunciation. What matters is the tension between their insertion into a text and their constitutive exclusion of textuality.
Unlike texts, aphorisations are not produced by speakers. Primary aphorisations (for example, sayings, slogans) are basically polyphonic: their responsibility is attributed to another, anonymous instance (for example the so-called "Wisdom of nations"). Secondary aphorisations, those which have been detached from a text, are not attributed to true speakers, but to aphorisers, who have been created by the very detachment of the sentence. Aphorisers are supposed to speak independently of any genre and to speak in an absolute way. They are presented as Subjects in full right: in Latin, sub-iectum refers to something that does not vary. 8 In aphorisations, the linguistic Subject coincides with the ethical and legal Subject: by his/her aphorisation, aphorisers express their values in the face of the world, without leaving room for an answer. They do not target a specific addressee, defined by a genre, but a kind of "universal audience" (Perelman and Olbrechts-Tyteca, 1958), the whole community for which their utterances are supposed to be fully meaningful.
When courts condemn somebody for what he/she has said, as a rule they condemn aphorisations, not texts. For example, some years ago, the French far-right leader Jean-Marie Le Pen was condemned for having said that the extermination of the Jews was "a detail" of the Second World War. His lawyers objected on the grounds that this sentence was unduly detached from its context. The problem is that Law needs to judge Subjects, independently of any context, and not roles in genres. If you consider a sentence within a text, you have no access to a full Subject but to a role played inside a genre.
The distinction between speaker and aphoriser allows us to better understand why in posters, newspapers or websites, there are often photos close to aphorised utterances of the face of their authors: the journalists are prompted to show the faces of the Subjects who are responsible for what has been said. Typically, the face is the noble part of the body, that which shows the identity of the speaker, and not the individual in a particular setting: aphorisers are staged as being far from circumstances and immediate interactions, in contact with transcendent values.
Texts that are detached into "Works". Unlike that of speaker, the category of "author" does not really have a status in discourse analysis. Traditionally, authorship is considered as a topic for literary theorists. Indeed, the notion of "author" questions usual distinctions: it refers to a being that is not the "enunciator", that is, a linguistic category, nor a flesh-and-blood human being, outside language, but a function that tightly combines agency (X is the cause of an utterance) and a legal dimension (X must answer for it). To any monologic utterance one must be able to attribute an author, be it an individual, a group or an abstraction.
Not all kinds of authors are associated with Works. "Works" is a category that discourse analysts generally ignore, probably because in their mind it is tightly connected for example with literature, philosophy or religion. This is surprising if we take into account the fact that any culture depends on a restricted set of major personalities-including the ones that discourse analysis draws on. "Works" is a unit which brings together a multiplicity of texts that are supposed to express a singular view of the world. It is the kind of authorship that Foucault reflects upon in a famous article, "What is an author?" (1969). For the sake of clarity, we shall refer to it by using the Latin word Auctor, in order to distinguish it from ordinary authors, who are associated with definite descriptions: "the author of this letter, this report…" Auctors are separated from the multitude of speakers whose utterances will not enter into collective memory. And among these auctors very few become "great authors" or "authorities". "Great authors" are characterised by the fact that all the types of their texts can be published: school homework, letters, diaries, and so on.
Like the notion of aphorisation, that of "Works" pushes discourse genres into the background. The "Works" of an auctor detach the original texts from their original context to give them a new pragmatic status. As a fragment of the "Works" of an auctor, a text is no longer read as the product of a situated speech activity but as the expression of an extra-ordinary Subject. If, in the Complete Works of Jane Austen or Thomas Hardy, we read some letters that they wrote to their friends or some pages of their diary, we do not read them as typical of a specific genre but as fragments of their "Works". When Jane Austen wrote her letters, she was involved in a routine of her social environment; when we read them in her Complete Works or in handbooks of English literature, they are separated from this discourse practice and referred to in the world's view because they represent a great personality and are part of the cultural heritage. As for the addressees of texts belonging to "Works", they are not specified by the corresponding genres, but are an undeterminable audience.
In addition, like aphorisers, auctors cannot institute themselves as such: this operation requires the intervention of third parties, who decide that these or these utterances are worth being detached from their genres and presented in a new configuration, referred to as an auctor.
As we can see, each in their own manner, aphorisers and auctors are allowed to escape the constraint imposed by genres.
Auctors escape it upwards, by converting texts into fragments of a higher unit, "Works". Aphorisers escape it from below: texts are decomposed into aphorisations. In both cases, the usual notion of "speaker" is not relevant. Aphorisation and "Works" are highly valued productions of their Subjects, and are foreign to the logic of ordinary exchange. Through them, speech is taken to a new stage: aphorisations are quotations; "Works" re-present texts inside a new frame. They do not call for the response of other speakers: they have to be commented on. As detached utterances, they differ from the countless attached utterances, those which are submitted to the logic of text and genre.

Conclusion
Discourse analysis, like modern linguistics, emerged in a context where ignoring the corpora which had been associated with traditional hermeneutic practices was a way to position and legitimate oneself. By organizing the field around conversations and genres from institutional settings, it gave a relative homogeneity to its data and its toolkit. But I think that we should take a more realistic view of the possibilities offered by discourse analysis, by paying attention to phenomena that challenge its prevailing routines. In the preceding sections I have evoked various manifestations of discourse heterogeneity: -The distinction between conversation and (instituted) genres; -The difference between various kinds of (instituted) genres; -The existence of self-constituting discourses, which implies that the universe of discourse is not "monotonous", that is, qualitatively undifferentiated; -The specificity of the Web, whose functioning is not compatible with the traditional conception of genre and of textuality; -Aphorisation, which is not submitted to the logic of text and genre; -The existence of producers of utterances who are not "speakers" in the usual sense of the term: "enunciators", "aphorisers", "auctors", and "angelic speakers". -Utterances that have no addressee in the usual sense of the term.
Indeed, these phenomena do not belong to the same category or to the same level, but they force us to open up the field of discourse analysis. The current mainstream of discourse analysis gives an illusory centre to discourse and casts to the fringes the data which do not fit its usual toolkit. But the universe of discourse is too complex to be represented in this way: it has no centre, no fringes.
Discourse analysis must take this reality into account if it does not want to be reduced to a set of "qualitative methods" for a restricted area of social sciences. This implies the acknowledgement that all manifestations of discourse are relevant, and that one of the key challenges of discourse theory and discourse analysis is to reflect on the conditions of the heterogeneity of discourse.
4 This tendency is favoured by the fact that Conversation Analysis, which provides a toolkit to most discourse research, is divided into two main branches: one studies ordinary conversation, the other one, known as "institutional conversation analysis", is specialized in interactions in institutions such as law or medicine (Heritage, 2005;Heritage and Clayman, 2010). 5 It should be noted that the editor of the journal, T. Van Dijk, one of the most prominent researchers in Critical Discourse Analysis, has done research mainly on written corpora. 6 In French, scène englobante. 7 In the case of immersed textuality, texts are the product of the transcription. 8 It is only one of the meanings of subiectum, which, like substantia, in philosophy corresponds to the greek hupokeimenon: the underlying thing, what persists through the change. See, for example, the article subicio in the Oxford Latin Dictionary (P. G.W. Glare (ed), 2nd edition, 2012).