Ann Banfield
University of Berkeley

Les chifres en parenthèses rouges renvoient aux notes en bas de page

 

A Grammatical Definition of the Genre "Novel"

In Le Degré Zéro de L'Écriture, Roland Barthes refers to the "signes formels de la Littérature", listing "passé simple, style indirect, rythme écrit" (1953 et 1964, p. 58). Earlier in the same text, he discusses, as a variant of "style indirect," the "troisième personne du Roman", there clearly attaching these formal signs to "L'Écriture du Roman" (33). The "signs" Barthes identifies are in fact grammatical phenomena. We might posit that each literary genre exploits (as opposed to uses) a different set of such grammatical "signs" out of the repertoire of possible forms provided by language.
           Kristin Hanson and Paul Kiparsky (1997, p. 17) posit that verse is a form of language with "Regular Recurrence of Linguistic Equivalences." The interesting question becomes what things can qualify as equivalent. Hanson and Kiparsky's answer is: units language treats as such. For instance, for meter, "isochrony is often invoked as an important principle of certain metrical systems, but time is not a linguistic entity" (p. 22). Rather, the equivalent units will, as in language, be somewhat abstract entities like segments treated as clusters of features, morae, syllables defined by various properties. I.e., "for rhyme or alliteration to become a principle of verse structure, the interval of recurrence must be measured by some other linguistic element; and it is meter which normally serves this purpose." (p. 33) Poetic meter is, in particular, "a stylization of the rhythmic structure which language has naturally", Hanson and Kiparsky write (1997, p. 56).[1] I.e., poetry heightens features of natural language rather than being either "contra natura" - the hypothesis of the violation of rules - or merely conventional patterns arbitrarily imposed - the conventionalist or constructivist thesis. The elegance of the "naturalness" argument for verse lies in the demonstration that the relevant notions required for metrical theory like "syllable" or "foot" are those already proposed by phonological theory and thus independently justified.
           Jean-Claude Milner, in a similar vein, furnishes an internal linguistic criterion to differentiate prose and verse: the non-coincidence of syntactic and phonological limits. "En bref," Milner summarizes his claim, "il y a vers dès qu'il y a possibilité d'enjambement" (1982, p. 301) Milner contrasts the "insignes conventionnels de la versification" with the possibility of enjambement, which is not a "conventional" sign (1982, p. 300). "On comprend l'importance du blanc typographique: il n'est pas une marque conventionnelle, mais le signal de l'enjambement possible", Milner writes (1982, p. 301). In other words, these grammatical signs of verse are "natural", in the sense that they arise as possibilities of natural language. There might also exist conventional signs of verse - Milner seems to suggest that there do - but these in no way imply that the category "verse" is purely conventional. For one could imagine that a theory of conventional forms could be maintained, but that the repertoire of possible conventions would have to be constrained so as to permit only those which conform to the features of natural forms defined by the language. There might in addition exist natural aspects of poetic form determined by other features of human cognition than language. The same patterns of regular recurrences that constitute what Halle and Keyser call the "abstract metrical formula" or various rhyme schemes or stanzaic forms might be abstract forms realized in different arts by other than linguistic material - by tiles of differing colors or shapes or lumps of wool of different configurations, e.g. knit/purl/knit/purl, and yet still be considered to each represent one of a limited set of possible patterns producible and recognizable by the human mind. Language, per se, however, decides which sounds are treatable as recurrences in the language arts.[2]
           The idea of natural language imposing constraints on possible literary forms arises in the opening of Le Degré Zéro de l'Écriture when Barthes isolates his key term "écriture" from two other notions, "language" and "style". "Écriture" represents the arena of the writer's freedom, a space carved out of both language and style - language, which each writer finds already there, setting "une limite", tracing "un horizon" (1953 et 1964, p. 13) to his or her innovations, and style, which is personal to each writer, a kind of unconscious signature, formed both, Barthes suggests, by inherited tendencies and learned features of vocabulary and construction - "un lexique naissent du corps et du passé, non d'une Histoire."[3] One might see the history of writing which Barthes traces in this Ur-structuralist text as providing those conventions which each generation comes up against and attempts to change, for it is precisely conventions which the writer is "free" to change. But Barthes is not a radical conventionalist, for language, which is "comme une nature" (1953 et 1964, p. 13) and style, "au niveau d'une biologie ou d'un passé" (1953 et 1964, p. 14), are posited as outside the arena of freedom, that is, of what can be, by an act of will, changed. They define the possible forms that writers can freely develop.
           Barthes' list of the signs of Literature, which he attributes specifically to the novel, might be taken as conventional signs, but I want to suggest that, like the notions appropriate to poetry Hanson and Kiparksy treat, they are natural signs or, if conventional, have a basis in natural features of language. They are syntactic, as opposed to phonological "signs", which determine the repertoire of different literary forms. Nor was Barthes the first to find in grammar the contours of a literary form. As early as Plato and Aristotle drama and epic were distinguished by the different exploitation of the very grammatical phenomena (omitting tense) Barthes sees as exploited by the novel - pronouns and reported speech. Plato claims that "poetry and fiction fall into three classes. First, that which employs representation only, tragedy and comedy, as you say. Secondly, that in which the poet speaks in his own person; the best example is lyric poetry. Thirdly, that which employs both methods, epic and various other kinds of poetry" (Republic, Book III, 394c, Desmond Lee, trans.). The epic uses the method of drama by using direct speech, where the first person is not the narrating I but a character. Aristotle echoes Plato, although he makes no reference to lyric but only epic and drama, claiming that one "difference in these arts is in the manner in which each kind of object is represented. Given both the same means and the same kind of object for imitation,[4] one may either (1) speak at one moment in narrative and at another in an assumed character, as Homer does; or (2) one may remain the same throughout, without any such change; or (3) the imitators may represent the whole story dramatically, as though they were actually doing the things described" (1448a20-24, Ingram Bywater, trans.). The first of Aristotle's distinctions might be understood to refer to the epic; the second to pure narration and the third to the drama.[5]
           We assume as a working hypothesis that these divisions of genre correspond to real divisions among the forms of the language arts, much as linguistics adopts the traditional "parts of speech". But we are concerned with finding in the differing exploitation of the syntactic phenomena of person and reported speech a basis for distinguishing the related genres "novel" and "short story", unknown to Plato and Aristotle, from the epic. To Plato's and Aristotle's implicit focus on the alternation of the personal pronouns and reported speech, Barthes, we saw, adds tense. We might ask why the categories of person and tense are the particular "signs" that come into play, if this is indeed the case, in this division of genres. I will suggest a hypothesis that brings us back to Hanson and Kiparsky's claim that verse crucially manipulates the units language treats as such and Milner's hypothesis about verse as the possibility of enjambement. Milner's definition of poetry also invokes linguistic units. For enjambement, the pertinent syntactic limits call into play the major categories, NP, VP, AP, PP and S, or, in the framework of the Bar-Notation, the maximal projections of X. Epic, on the one hand, and novel and short story, on the other, likewise bring into play the question of limits and thus of grammatical units. Banfield (1982), analyzing the conjunction of reported speech, the personal pronouns and tense, formulates what the relevant limits or units which serve to distinguish the novel are: the node E and the unit there called "Text". The category E is a non-embeddable sentence, one that permits constructions such as exclamations which cannot occur in subordinate clauses such as those of indirect speech and one that dominates the sentence itself. The Text is a larger unit of related Es that falls outside the tree-based, hierarchically-ordered syntax of the sentence. By contrast with verse, no unit smaller than the E need be mentioned as setting the limits crucial to the form "novel". If a smaller category such as the pronoun [N"] seems crucially involved, it is, we will see, only insofar as it enters into the statement of a principle for the interpretation of E or Text.
           Hence, the theory of the novel dependent on E and Text does not exclusively center "on sentences taken individually", without taking into account "narrative context", as has been frequently claimed.[6] The role of Text in the theory of unspeakable sentences has been largely ignored. (There is no reference to the term as used in Banfield, 1982, in Fludernik, 1993). What is of significance for any proposed theoretical construct is whether it is confirmed by the evidence. That confirmation may take time - i.e., a notion may be proposed at one point to explain a certain phenomenon in a fashion that might seem ad hoc. Only later might other phenomena be discovered which are accounted for by this same notion, thus giving it a generality it seemed initially to lack. I hope in the course of this paper to so provide further confirmation for the unit "Text", which proposed some minimal formal notions of "context" in a quite precise sense. The proposals suggested will at the same time point to directions for further research. It is at the interface between E and Text that, I will argue, the novel and short story distinguish themselves from the related narrative genre that is the epic. E and Text are both categories independently required by the grammar. What will turn out to be decisive is that E and Text serve to distinguish two distinct levels of pronouns and deictics. This is, to my knowledge, a hitherto unobserved fact. If one translates Plato's formulation in terms of whether the poet speaks or not in propria persona to turn on the first person, then his divisions focus on a pronoun which we will characterize as a Text-level deictic.
           The specific innovation of novelistic style is to have suppressed the first person and, in the process, to have discovered in the linguistic repertoire a third person pronoun which is not an anaphor but what I have called an "E-level deictic".[7] This discovery permits the language of the novel to overturn the monopoly of the first person and to orchestrate within the confines of a single Text the "shift in point of view", a traditional notion a linguistic-based theory of the novel can give formal content to. It is the possibility of shifts in point of view within a single Text which sets the novel and the short story apart from other genres, although not every novel or short story may exploit this possibility. The fact that they don't becomes itself significant - hence the marked case of the novel of "limited point of view".

The Domain of the Third Person Deictic Pronoun
The notion "Text" was initially introduced in Banfield (1973 and 1982) to account for direct speech: a direct quotation is not just a new E, separate from its introductory clause, but also a new Text. This is because direct speech may introduce a new referent for the first person pronoun. Thus, one of the defining features of the unit called "Text" is the obligatory coreference of every instance of the first person pronoun within it. This is captured by the principle of "Concordance of Person", which stipulates "1 Text/1 Speaker". (1982, p. 59) Hence, it might appear that Text is the level at which point of view is determined, and the Shift in Point of View an intertextual and not an intratextual phenomenon. But this conclusion fails to take in the evidence of the language of the novel. The phenomenon of represented speech and thought with its possible "third person point of view" means that a new point of view does not require a new referent of the first person and a new Text.
           "1 Text/1 Speaker", when its full implications are seized, captures something of the peculiar status of the first person pronoun. But these implications were not sufficiently foregrounded to register with the commentators, perhaps because all its consequences were not taken cognizance of in Banfield (1982). More recently, in Banfield (1998 and 2001), I have returned to the question of the first person, first via a philosophical criticism of the Cartesian cogito and then via a linguistic analysis of "I" vs. the other deictics. It emerges that the "s/he" of represented speech and thought, while it resembles "I" in being the reference point for the interpretation of subjective elements, differs from "I" in a way significant for the development of the novel form with its multiple viewpoints. Taking the term "shifter" Jakobson borrowed from Jespersen for the deictics to emphasize their ability to shift referent (and ignoring certain complications), we note that the deictics shift referent with each new E (treating, as in Banfield, 1982, each conjoined sentence as a separate E), i.e., at "E-level", as indicated below:

Unique among the shifters, "I" may not shift with each "E" but changes referent only at Text-level, e.g., in Direct Speech. Otherwise, in a sequence of related Es understood to form a unit, i. e., a Text, as in (2), all instances of "I" must be coreferential:

(2) provides evidence for the Principle of Concordance of Person. Evidence of this sort was treated in Banfield (1982) as largely crucial for defining the notion "Text." What was sufficiently recognized was the different behavior of I with regard to the other shifters. For the fact that the other deictics also seemed to shift reference from introductory clause to quoted clause of direct speech masked their difference, which the examples in (1) reveal. For a new Text is necessarily a new E. Putting the fact accounted for by 1 Text/1 Speaker in terms of the distinction between Text-level and E-level shifters reveals certain hitherto unnoticed logical consequences of the former generalization and gives a formal status to the distinction between two concepts made in Milner (1978): a) Lacan's concept of the "sujet de l'énonciation", which is "le point de subjectivité auquel on rapporte un énoncé en acte et dont on ne suppose rien - ni permanence ni conscience ni individualité" and b) the concept, "tout descriptif, de "locuteur"'".[9] I.e., the role of I as Self or subject is distinct from its role as speaker. This also explains why only "I" among the traditional deictics can cooccur with the passé simple in what Benveniste calls "histoire". This is a revision of Benveniste's claim that the passé simple cooccurs with no deictic, but the exceptional cooccurance of I with the passé simple follows from the fact that it is the only Text-level shifter. (See Banfield, 1982, pp. 146ff.).

Further Defining Features of the Text
Not all sequences of Es constitute a Text. A collection of sentences, e.g., of proverbs, is not so constrained, nor a sequence of linguistic examples. If they are ordered, it is by principles external to language - in being, for instance, all examples of grammatical constructions, as in (3):

By contrast, (3a and b) can be understood to form a Text if the NPs it contains are interpreted as related by coreference, as in (4) below:

The question of the coreference of "John" and "him" or of "me" and "I" and of the coreferentiality of "We" with "John" and the first person singular does not arise in (3), because no relation of the sort in (4) is implied by a list of sentences. In (4), the coreference of the proper name and the third person pronoun is optional in principle, but in practice it is obligatory if no other relevant antecedent for "him" is supplied by a larger context, as it would be if the sequence in (4) were preceded by the first sentence in (5):

A similar ambiguity of referent would become possible for the third person plural pronoun of the sequence in (4) if it were proceeded by the first sentence in (6):

("Him" could equally be interpreted as coreferential with "John," and "we" with "John" and "me".)
           All instances of the first person singular, on the other hand, must be interpreted as coreferential if the three sentences in 3 are treated as a linguistically related sequence. It is for this reason that direct speech constitutes a shift to a new Text, because it permits a shift to a new referent for "I".
           Concordance of Person is one formal principle defining which sequences of Es constitute a Text. There are other such principles which link E to E. One is that which governs the interpretation of sequences of tenses. This is presented in Banfield (1982) as Concordance of Tense. The verbs in the examples in (5) and (6) above are interpreted as referring to sequential events. In the passages below, the verbs in the passé simple in (7a), in the narrative present tense in (7b), and in the English simple past in (7c) are similarly understood to refer to events which occur in sequential order. The verbs so interpreted in (7a) occur in a single E, but (7b and c) make clear that the principle applies as well to sequences of Es.

The linear order established is non-deictic, the order of the integers. It would thus use something like Bertrand Russell's formulation for events in time: it happened at time T, T1, T2, Tn. . .
           Milner (1990, p. 61, n. 26) notes as another feature of such sequences definitization, giving as an example the following direct speech construction: "(aa) A man saw a unicorn in his garden and said to his wife: 'a/*the unicorn is in our garden'." This seems to indicate, Milner comments, "that a change of speaker (which, in Banfield's terms, implies a change of E) prevents definitization." But, he adds, "there is a distinction to be made between two cases: (i) two Es that are structurally distinct, define, however, the same reference for the 1st person pronoun (and for other shifters); (ii) two structurally distinct Es define different references for the 1st person pronoun." "Definitization would only be possible in the first case.", he concludes. (p. 61, n. 26)
           The distinction between (i) and (ii) is precisely that between a sequence of Es constituting a single Text and the particular sequence of Texts consisting of introductory and quoted clauses of direct speech.[10] Thus, the indefinite noun phrases in the sequences below are interpretable as coreferential with the appropriate definite noun phrases in the sentences [Es] following them.

The indefinite and definite noun phrases in (8) can be understood as coreferential because the sequences in (8a-e) form each a single Text. Milner states correctly that a change of speaker "in Banfield's terms, implies a change of E". But technically it is a new Text, which, a fortiori, means a new E, rather than a new E, that permits a new speaker. Such examples of definitization are not possible, by contrast, across the frontiers of two Texts, including between introductory clause and quoted clause of direct speech:[11]

A quite different set of phenomena the prediction of whose behavior would require reference to the unit Text is the class of sentence or coordinating conjunctions that includes not only and, or, but but also words like for, thus, however, French donc as discussed in Nølke and Olsen (2000) and mais in Jorgensen (in this issue). There are differences in the various members of this class. But they plausibly play a role in the linking of E to E. Furthermore, many, if not all, are excluded, it seems to me, from the initial position in the Text. This is a subject deserving of further research.

The Epic: the Genre of Direct Speech
We have now arrived at a point where the differences between the units E and Text can be invoked to distinguish the genres "novel" and "short story" from the epic. We can hypothesize that the notion of "character" within narrative in general crucially involves the linguistic principles for which the constructs Text (as well as E) are required, principles that involve the behavior and interpretation of pronouns and other referring categories, e.g., the coreference of indefinite and definite noun phrases of the sort Russell calls "descriptions" with the special case of definite NPs, proper nouns, and pronouns, (along with adjective phrases and other constructions predicated of the NPs). It can be demonstrated that they are not simply pragmatic principles. Tense also plays a role in the development of character, in adding the idea of persistence as well as change over time, an idea Proust foregrounded and problematized. (A narratively realized character contrasts with the static notion of character in texts like those of La Bruyère.) The units E and Text are not, of course, restricted to the genre novel. But the specificity of novelistic character results from their being exploited in idiosyncratic ways. In particular, the conception of character peculiar to the novel is dependent the properties of the +human E-level shifter, Barthes' "third person of the novel", which does not appear in the epic. This third person pronoun is not a simple anaphor, dependent on an NP antecedent. For this reason, it, like the first person, may occur in initial position in a Text, as in these opening sentences of stories by D. H. Lawrence.

A speaker cannot, by contrast, begin a conversation by saying "She was too good for him." Indeed, incipit is a notion that is definable with respect to the unit Text.[13]
           This specificity of the E-level pronoun shifter, what I have called elsewhere "the name of the subject", Lacan's "sujet de l'énonciation", might be considered a core of the first person itself; what the first person, Lacan's "locuteur", adds is obligatory coreference from E to E. One can see this split in the I in those cases of represented speech and thought in the first person. The sentence "How my heart was beating now, I realized" is an example of represented thought; the parenthetical presents a past response to a simultaneous, past event. The first person is speaker or narrator as well as subject or Self. 1Text/1 Speaker prevents its core of subjectivity from detaching itself from its role as speaker and behaving like an E-level shifter.[14].
           Aristotle's characterization of the epic in effect singles out the alternation between sentences of narration with a first person narrator and sentences of direct speech, i.e., of the shift from Text to Text, as essential to the form. This is indeed an observably striking feature of the oral epic, one readily apparent if we examine some frequent formulas and formulaic lines which consist of introductory clauses of direct speech, from the Homeric formula "epea pteroenta proseuda" ("spoke winged words") repeated several times verbatim to introduce quoted speech (Iliad 1.201, 4.69, 5.242, 12.365; Odyssey 1.122, 4.77, 10.265, 16.7), to Beowulf's formulaic "Beowulf maÞelode" ["Beowulf spoke"] (405, 529), "Hrothgar maÞelode" ["Hrothgar spoke"] (372, 925, 1840), "Weard maÞelode" ["The guard spoke"] (286), or "Ða se wisa spraec" ["Then the wise one spoke"] (1698), "werodes wisa, wordhord onleac" ["The company's leader unlocked his wordhoard"] (259), to the many formulas introducing direct speech in La Chanson de Roland.

Subjectivity or point of view is not "represented" in the oral epic; instead there are shifts to the character's spoken viewpoint. These frequent passages of direct speech mark the "public" aspect of the epic hero or heroine, who is constituted by his or her words and deeds. Novelistic character is, by contrast, "private".[15] (We might also see a further connection between the epic's public performance and the novel's private "consumption".) This difference between epic hero and heroine and novelistic character is captured in the difference between the first person used for the former in direct speech and the E-level shifter used for the second.

The Shift in Point of View
The epic's manipulation of direct speech allows the shift to a new Text and hence to a new speaker and that speaker's spoken point of view. There is another possibility for the novel. Once it has eliminated the first person elsewhere than its use of direct speech, it neutralizes or renders null 1 Text/1 Speaker. Novelistic character's subjectivity is separated from its role as speaker and subjectivity disconnected from speech. Point of view or subjectivity can then be located in the special third person pronoun, the E-level shifter. In the process, the link between E and E forged by 1 Text/1 Speaker is dissolved. Each E in theory can now, in the absence of a first person, represent an independent point of view. I say "in theory", because in fact if point of view shifted radically from every E to every E, it would frequently be impossible either to detect the shift or to decide whose point of view was represented. Yet there seems to be no formal principle that constrains the shift in point of view within a single Text. Here is plausibly a case where pragmatic principles intervene. Consider the examples of shifts in point of view in (12). At some point in the Text we pass from an E representing one (or no) Self's point of view to one representing another Self's point of view. (I indicate where the shifts occur by placing in bold face the name of the character whose point of view is represented; I further indicate whether the point of view has the form of represented speech (R.S.) or represented thought (R. T.) or whether a sentence of Pure Narration intervenes between those representing point of view. In some further cases, I indicate where the shift is marked by a distinction between reflective or non-reflective consciousness, as discussed in Banfield (1982, pp. 183ff.)

       12.        Shifts in Point of View

                     [Linda] Linda looked up at the fat swelling plant with its cruel leaves. High above them, as though becalmed in the air, and yet holding so fast to the earth it grew from, it might have had claws instead of roots. The curving leaves seemed to be hiding something . . ." Katherine Mansfield, "Prelude" pp.74-5

There are various ways of choreographing a shift in point of view, some of which are illustrated above. Sometimes we pass from an E of represented thought with one point of view or Self to another with another Self via a sentence of narration, as in (12a), via a sentence of direct speech, as in (12c) or via the use of parentheticals, as in many of the examples of (12). Sometimes we pass from Narration to the representation of non-reflective consciousness, as in (12a), sometimes from a point of view to pure Narration, as in (12b) and (12c). (12c) is particularly revealing, because the last sentence invokes "the branch of a tree which couldn't be seen." If the narration had been in the first person, the knowledge of this branch would be registered by at least one character, the first person narrator. In the novel as it appears, there is no supposition that any character is aware of this branch.
           In this way, the novelistic universe contains many points of view, some occupied by different individuals at different moments - "Nows" - and some unoccupied. The total knowledge of that universe is therefore possessed by no one character in it, none is "omniscient"; it is possessed by the novel or story as a whole. This implicit epistemology of the novel follows from the linguistic properties of the units E and Text, once the elimination of the first person renders null 1 Text/1 Speaker. The whole - the linguistic artifact through which this universe is represented, the Text - is held together by other principles than 1 Text/1 Speaker, those determining Sequence of Tense, the Coreference of NPs with lexical head nouns and their pronoun anaphoras, definitization, etc. That whole nonetheless permits the internal shifting of point of view, because point of view is attached no longer rigidly to the Text via its I, but to an E-level Shifter, grammar's name for the subject.

Bibliography
Aristotle (1954), The Rhetoric & The Poetics of Aristotle, Rhys Roberts and Ingram Bywater, trans., Random House, New York.
Banfield, Ann (1998), "The Name of the Subject: the 'il'", The Place of Maurice Blanchot, Tom Pepper (ed.), Yale French Studies, No. 93, fall, 133-174.
Banfield, Ann (2001), "Le Nom Propre du Réel", Cahier Jean-Claude Milner, Jean-Marie Marandin (ed.) Paris: Éditions Verdier, pp. 229-266.
Banfield, Ann (1973), "Narrative Style and the Grammar of Direct and Indirect Speech," Foundations of Language, 10, pp. 1-39.
Banfield, Ann (1982), Unspeakable Sentences: Narration and Representation in the Language of Fiction. Routledge & Kegan Paul, London.
Barthes, Roland (153 et 1964), Le Degré Zéro de L'Écriture suivi de Éléments de Sémiologie. Éditions Gonthier, Paris.
Duggan, Joseph (1969), Concordance to the Chanson de Roland, Ohio State Univ. Press, Columbus, Ohio.
Emonds, Joseph. (1987). "Parts of Speech in Generative Grammar", Linguistic Analysis 17: 3-42.
Fludernik, Monica (1993), The Fictions of Language and the Languages of Fiction Routledge, London.\
Fludernik, Monica (1996), Towards a 'Natural' Narratology, New York.
Hamburger, Käte (1973), The Logic of Literature, Marilyn Rose, trans., Indiana University Press, Bloomington.
Hanson, Kristin (1996), "Une théorie phonologique du mètre et ses conséquences pour le pentamètre iambique anglais", Bulletin de la société stylistique anglaise, pp. 57-72.
Hanson, Kristin (1997), "From Dante to Pinsky: A theoretical perspective on the history of the modern English iambic pentameter", Rivista di Linguistica, 9.1, pp. 53-97.
Hanson, Kristin and Kiparsky, Paul (1997), "The Nature of Verse and its Consequences for the Mixed Form", in Reichl, Karl and Joseph Harris (eds), Prosimetrum: cross-cultural perspectives on narrative in verse and prose. D. S. Brewer, Cambridge, pp. 17-44.
Jahn, Manfred (1992), "Contextualizing Represented Speech and Thought", Journal of Pragmatics 17, pp. 347-67.
Jørgensen, Katherine Ravn (in this issue), "Le connecteur mais et le discours indirect libre"
Marnette, Sophie (1996), "Réflexions sur le discours indirect libre en français médiévale", Romania, Vol. 114, pp. 1-149.
Marnette, Sophie (1998), Narrateur et points de vue dans la littérature française médiévale, Peter Lang, Berne.
Milner, Jean-Claude. (1978), De la syntax à l'interprétation, Seuil, Paris.
Milner, Jean-Claude (1982), "Réflexions sur le fonctionnement du vers français," in Ordres et Raisons de Langue, Seuil, Paris.
Milner, Jean-Claude (1990), "Some Remarks on Principle C", in Binding in Romance: Essays in Honour of Judith McA'Nulty, Di Sciullo, Anne-Marie and Anne Rochette, Canadian Linguistics Association, Ottawa, pp. 41-66.
Nølke, Henning and Olsen, Michel (2000), "Donc pour conclure. Polyphonie et style indirect libre : analyses littéraire et linguistique ", Actes du XIVe Congrès des Romanistes scandinaves, Stockholm 10-15 août 1999. Acta Universitatis Stockholmiensis. Romanica Stockholmiensia 19, éd. Jane Nystedt. Cd-rom.
Nølke, Henning and Olsen, Michel (to appear), "Puisque: indice de polyphonie?", Faits de Langue.
Plato (1974), The Republic, Desmond Lee, trans., 2nd revised edition, Penguin, New York.
Ruppenhofer, Josef (2001), Unpublished handout, Berkeley-Stanford Language Change Workshop I, April 28, 2001.
Violi, Patrizia (1986), "Review article: Unspeakable Sentences and Unspeakable Texts", Semiotica 60. 3-4, pp. 361-78.

Corpus
Beckett, Samuel, Company, Grove Weidenfeld, New York, 1980.
Beowulf and the Fight at Finnsburg, Fr. Klaeber (ed.), D. C. Heath and Company, Boston, 1922.
de Goncourt, Edmond et Jules, Manette Salomon, Préface de Michel Crouzet, Édition établie et annotée par Stéphanie Champeua avec le concours d'Adrien Goetz. Gallimard, Paris, 1996.
Lawrence, D.H., The Complete Stories of D. H. Lawrence, vol. 1, Harmondsworth, Middlesex: Penguin, 1922 (1961).
La Chanson de Roland, Cesare Segre (ed.), Riccardo Ricciardi Editore, Milano, Napoli, 1971. or Cesare Segre, La Chanson de Roland (2 vols.; Textes Littéraires Français, 368; Geneva: Droz, 1989).
Mansfield, Katherine, Stories, edited by Elizabeth Bowen, Random House, New York, 1956.
Scott, Walter, The Antiquary, Penguin Books.
Woolf, Virginia, Jacob's Room, New York: Harcourt Brace Jovanovich, 1922 (1950).
Woolf, Virginia, Mrs. Dalloway, New York: Harcourt Brace Jovanovich, 1925 (1985).
Woolf, Virginia, To the Lighthouse, New York: Harcourt Brace Jovanovich, 1927 (1955).
Woolf, Virginia, The Years, New York: Harcourt Brace Jovanovich, 1937.
Woolf, Virginia, The Waves, New York: Harcourt Brace Jovanovich, 1931.
Zola, Émile, L'Oeuvre, Préface de Brunou Foucart; Édition établie et annotée par Henri Mitterand, Éditions Gallimard, Paris, 1983.

Notes:

1. See also Hanson (1996, p. 57).

2. Against the naturalist hypothesis is ranged the current conventionalist orthodoxy.  One among many examples of it is that presented in Fludernik (1993).  She "foreground[s] the invented quality of all representational processes and insist[s] on their consequent fictionalization . . . . reported discourse constitutes a fiction that language fabricates in accordance with discourse strategic requirements etc."  Fludernik (1996), despite the title - Towards a 'Natural' Narrotology-, continues this "by firmly reiterating my insights about the artificial nature, inventedness and fictionality of the mimetic", now extending it "to cover, quite generally, all mimetic representation." (p. 10)  In "propos[ing] to redefine narrativity in terms of cognitive ('natural') parameters, moving beyond formal narratology into the realm of pragmatics, reception theory and constructivism" and in admitting that "the term 'natural' in the title feeds from three separate disciplines and areas of knowledge", Fludernik puts the language of anti-conventionalism-"cognitive parameters", "natural", "a theory of naturalness" - contradictorily in the service of constructivism.

3. These are the features which can be imitated or parodied, as Proust knew, and which certain computer studies seek to identify.

4. The formal distinctions that are being made here can be considered independently of any theory of representation or mimesis.

5. The fact that Aristotle seems to exclude lyric poetry from "mimesis" leads Hamburger (1973) to treat first person narration as logically distinct from third person narration.  See pp. 10ff.

6. Fludernik (1993) claims that when a formal analysis of the E "eventually has to have recourse to contextual criteria, these are alien to the framework [of Banfield (1982)], whereas the conventional account of free indirect discourse can much more easily handle interference from the 'previous sentence' or, indeed, paragraph". (p. 362) Here Fludernik does not understand the constraints under which a theory which aims not merely to repeat the intuitions of traditional criticism but to be minimally "formal" and empirical operates. Yes, as Fludernik writes, "[a]s a linguistic theory about free indirect discourse, Banfield's model therefore exposes itself to empirical verification and falsification." (1993, p. 362)  It only must be added that any theory of the novel, whatever its points of reference, if it is to be non-trivial, must "expose itself" to falsification.  Fludernik mentions Jahn (1992) and Violi (1986) as "regret[ing] the sentence-by-sentence approach in Banfield's model, which - as already noted - is a direct consequence of her modified Chomskyan framework." (Fludernik, 1993, p. 364)  But Violi's reference to the notion of text is quite different from either Jahn's or Fludernik's.

7. See Banfield (1998 and 2001).

8. There are cases where the temporal deictic "now" does not shift referent.  Cf. "All is solid now. Instinctively my palate now requires and anticipates sweetness and lightness . . . . Now I can look steadily into the mill-race that foams beneath." The Waves, p. 138  It seems to me strained to read the first two nows as other than simultaneous.

9. I ignore a third concept Milner distinguishes, that of the "sujet parlant" defined as "un point de subjectivité par rapport à la capacité de parler, indépendamment de toute mise en acte; c'est donc le concept d'une permanence qui, par-delà les énoncés singuliers, les unifie." (1978, p. 229, n. 1)

10. Introductory clause and quoted clause of direct speech are not any two random Texts; they are two related but separate Texts.  In Banfield (1973 and 1982), there is a relation of coreference between a demonstrative in the introductory clause and the entire quoted E.

11. Certain cases of coreference can also operate between introductory clause and quoted clause of direct speech, i.e., between two related Texts.  For example, the subject of a communication verb in the introductory clause is interpreted as having the same referent as the first person in the quoted clause:
The directori said to mej, "Ii will guarantee that youj will receive the order by the eighth."  But the coreference of the sort found in (4-6) where a pronoun corefers with a proper name, for instance, is in contrast to that holding between introductory clause and quoted clause in direct speech, where the pronoun seems less acceptable:  "Mary asked Bill's sister, 'Is Bill (?he) your younger brother?'"
   By contrast, dialogue has somewhat different constraints.  An indefinite noun phrase used by a first speaker may be coreferential with a noun phrase containing a demonstrative but not with one containing a noun phrase with a definite article, as in the relation of E to E in the same Text in the examples of (8).  Thus, if speaker A says "I saw an old woman", speaker B may felicitously respond "That old woman is my mother" but not "The old woman is my mother."

12. It seems also to be possible to begin a novel or short story with a pronoun without antecedent if it occurs within the point of view of a character.  Thus, in (11a), either pronoun, his or she, is eligible to be interpreted as the E-level shifter, i.e.; as the Self of the E; the other would refer to the object of the Self's thought.  A third possibility is that the sentence represents the plural point of view of them.

13. Phenomena like definitization also seem to show special behavior in this position.  My intuition is that Texts beginning with a definite as opposed to an indefinite NP, even including the proper noun, mark themselves as specifically novelistic, like those in (11).  "Once upon a time there lived a little girl.  The girl..." is not a novelistic opening, whereas "The girl left the cool of the house and walked into the hot streets of the city" is.

14. Marcel Vuillaume has pointed out to me the following example:

                       "Superbe, ai-je-dit. Magnifique. Par accident, je prends en filature deux encaisseurs d'un gros gang, je n'en sais rien. Le même gros gang vient de faire disparaître une nénette, et vous me branchez directement sur la disparition de la nénette, et je ne suis toujours au courant de rien. Vous savez ce que j'aurais fait à leur place ?- Ben - Je me serais abattu, ai-je bredouillé. J'aurais abattu Eugène Tarpon, tout de suite, très vite? (Jean-Patrick MANCHETTE, Que d'Os!, pp 164-165)

           The sentence "Je me serais abattu!, Vuillaume maintains, is to be understood unambiguously with the two instances of the first person having disjoint reference, with the first occurence of "je" referring to the gang and the second to the narrator-here, Eugène Tarpon, who is a private detective.  "J'(i) aurais abattu Eugène Tarpon(j)", where "les hommes du gang" = (i) and "Eugène Tarpon" = (j)  In English, there are two possible translations of the sentence, "I would have killed myself" and "I would have killed me", both more or less equivalent in meaning, although the reading with disjoint reference is clearer with either the non-reflexive pronoun or when the reflexive pronoun is contrastively stressed.  The example makes clear that, despite the splitting of reference, 1 Text/1 I nonetheless holds.  For the contrast with the third person counterpart "He would have killed him", which only permits a reading with disjoint reference for the two third person pronouns, by contrast with "He would have killed himself", shows that, insofar as the first person refers to the speaker or "locuteur", there is still coreference in "Je me serais abattu."

15. Of course, direct speech occurs in the novel as well.  It is the absence of represented thought in the epic that is crucial.  But we can observe differences between the use of direct speech in these two forms.  The novel places a value on variety-in vocabulary, in wording, in the order of words.  An interesting case is furnished by the difference between the formulas introducing direct speech in the epic style and the "parentheticals" which often accompany both direct speech and  represented speech and thought in the novel, as in "Oh how happy she was, she realized."  Ruppenhofer (2001) writes that the way in which quoted speech is presented "differs quite significantly between the spoken and the written mode.  In conversation, only a limited number of different speech reporting verbs occur."  He lists the verbs "ask, bawl, be like, decide, figure, go, it be, it be like, say, tell, think, wonder" (p. 1). "The reporting verbs used are also more varied in number and type" in writing. Ruppenhofer's sample includes "add, admit, agree, answer, ask, assure, bark, comment, confess, confirm, continue, counter, croak, demand, echo, explain, hiss, lie, query, remark, remind, reply, retort, say, shout, shriek, sob, tap out, tell, think, whisper".  The examples below of such parentheticals illustrate the novel's tendency to vary them, even though nothing in their nature requires that they not be the same formula each time.  It is also apparent that this tendency was well-established long before contemporary fiction, even if, as Ruppenhofer claims, certain kinds of communication verbs have "gained in frequency particularly . . . Manner and Noise verbs". (p. 7)  The following list from Sir Walter Scott's The Antiquary indicates the great variety of these parentheticals even in a short space: "answered the Naiad", "retorted the Antiquary", "rejoined the virago", (p. 88); "blubbered the boy", (p. 116).  As suggested in Banfield (1982, p. 86), by the time of Joyce's Ulysses, the practice was established enough to be parodied.