Film Restoration and Textual Criticism
In 1980 the London Film Festival screened a new print of Napoléon vu par Abel Gance (Abel Gance, 1927), a film originally shot during 1925-1926. The near five-hour version of the film projected that evening had been “restored”, and reconstructed, by Kevin Brownlow, who had been working on the project since 1969. In his research Brownlow had scoured archives and private collections to find material to be included in the new print; he had consulted extensively with Gance and others who had worked on the original production, and he had followed up hundreds of leads to written and printed documents related to the film. The showing, which featured an orchestral accompaniment especially composed for the occasion, was a deserved success for the film, for Gance – and for Brownlow, who published a book about Napoléon and its restoration three years later. 
The London Film Festival screening of Napoléon brought film restoration a public recognition that it had never had before. There has been at least one special event in film restoration virtually every year since that time. An annual festival at Pordenone currently provides a showcase for restored and reconstructed prints from the world film archives and a forum for the discussion of film preservation and film history. The publicity and special television and film events surrounding the “centenary of cinema” in 1995 tended to make even more people aware of the restoration and preservation work of national film archives.
Nor should we be surprised that there is something of a vogue for restored versions of movies at this time. Those who write and read articles like this are producers and products of an international film culture one aim of which is to represent film as the peer of other arts. History is being made by film reconstructions: the past is being recognised and a cinematic culture is defining its domain.
The widespread distribution of video tapes and laser discs has promoted recent popular interest in the different versions a film can take. A magazine called Video Watchdog introduced itself in 1990 as “a consumer’s guide to horror, science fiction and fantasy films on video tapes and discs” and went on to justify its existence with a litany of film editorial malpractice.
The process begins even before the films hit the theatres, with distributors changing their titles to make the subject matter more emphatic (i.e., Dario Argento’s Tenebrae becomes Unsane 1982). Then the MPAA steps in; suddenly the film is ten minutes shorter – removing not only bloody violence, but directorial flourishes and plot.
The video release is free to reincorporate the missing footage – does it? How can you possibly know when most companies print a generic “90 minutes” running time on their box? In America, widescreen films are usually cropped to make use of the full TV screen – has any important visual information been cropped out? How can you tell? The same movie is released to video in other countries, yes – but which version is most complete? 
At the same time, scant attention is usually paid in print to the versions of films that academic film historians and analysts use in their work. One of the earliest such discussions is still one of the lengthiest to have appeared in a book of serious film analysis: Lotte Eisner’s chapter on different prints of F. W. Murnau’s Nosferatu (1922) in her book on the director and his work.  Before 1995, the journal Film History had published just two substantial pieces on the topic; since that date two more have appeared, evidence of a new understanding of scholarship within the academic film community. 
The bulk of academic writing on the cinema is textual analysis – that is to say, work that one would presume must be dependent upon the prior existence of a fixed text – yet textual analysis in film rarely acknowledges that the physical film text may be problematic. Much of the most careful and interesting writing in the field does not indicate the specific sources of the specific prints used as objects of study, nor does it take time to describe the features or the antecedents of those prints.
It does not compare existing prints. It is usually not concerned to establish the authority of the print or prints to which its analysis refers. When variants are mentioned, as they are, for example, by Patrice Petro in her discussion of Joyless Street (Georg Wilhelm Pabst, 1925),  it is most often as evidence of one or another form of censorship.
This attitude is, I believe, largely unthought – a matter of convenience rather than of principle. But at the same time, tacit authority is granted to a written analysis of “a film” which must be, at least in part, derived from the authority (or lack of it) that can be established for the actual print(s) upon which that analysis is grounded.  Of course, it is possible, as a matter of principle, to deny authority to one’s analysis, to offer it playfully rather than absolutely. It is also possible, as a matter of principle, to deny authority to any print of the film under discussion. Lawrence Alloway has made a case for recognising the tattered condition of the prints of older films as one of the necessary, significant and positive circumstances of the discipline,  but no one ever cites Alloway on this point, and I suspect that even in these postmodern days most film academics would be uncomfortable with the idea that films are textually unstable as a implacable consequence of the circumstances of their production, distribution and exhibition. 
The textual analyst’s lack of interest in the specifics of specific prints has the effect of transforming a mundane and imperfect physical thing into an imaginary and perfect “text” (“the film”) which is/could be accessible to all readers/viewers. The analyst ignores the print as a physical object entirely and regards it, rather, as the end of the accomplishment of a desire – a fetish – seeing through the print to the imaginary work with which the analyst seeks to effect an impossible conjunction. It may be argued that this is an inescapable condition of the analysis of the products of modern communications media: one only works on a token of a type, a specific instance standing in the place of a general instance. Yet textual analysis as it is currently practised in the writing on film does not even acknowledge the type/token problem, much less attempt to understand a work through it, although it is undeniable that the relation of token to type, of specific to general, fibrillates in the heart of the dialectic of film writing.
If textual analysts have ignored the issues evoked by film reconstruction, those engaged in the reconstruction of films have not been overmuch concerned about such matters either. They elaborate no theories of textual editing in film. They do not feel called upon to defend what they have done, even when they explain how and why it has been done. No one seems to have published detailed notes on the reconstruction of any film comparable to the scholarly apparatus of literary textual editing; I do not know whether notes of this kind are even made. The attitude of film “restorers” (the term is, significantly, borrowed from painting) is pragmatic. In view of the critical film community’s attitude – or lack of it – towards what they do, they cannot be held to blame for that: it permits them to get on with what is a very necessary job. Yet what results from film reconstruction is often publicised as the “definitive version” of a film – through no fault of those involved with the film’s restoration – as though what had been done was unproblematic and required no documentation or validation of its own.
In literature certain procedures for establishing “responsible”, “critical” or even “definitive” versions of texts have evolved and have even been somewhat theorised. Such practices, which we may call “textual criticism” when the aim is merely to describe a copy of a text and “textual editing” when the goal is the production of a new edition, arise in the first place because of an undeniable practical need to know what it is that is being discussed in discussions of literary texts. Such a need is particularly strongly felt at those points when the authority of texts is an issue (either strongly asserted or strongly challenged), for example in early Renaissance Europe or in Islamic hadith.
At first glance there would seem to be no significant theoretical problem associated with literary textual criticism. It deals, after all, with a concrete physical entity – a book or manuscript. It endeavors to establish the authoritative form of this entity: what it would look like if it were in pristine condition. This seems to be a question more suited to collectors, those archetypal fetishists, than to theorists or critics. If there are pages missing, for instance, the copy is worth less to both academics and collectors.
And the notion of a specific, concrete and physical entity has been until fairly recently very much a guiding assumption in the practice of textual criticism and editing. The goal of textual editing is an “edition”, a single thing upon which we can come as close as is possible to complete agreement. Textual editors are not naive. As a rule they are much more sophisticated, even more cynical, than the rest of us. They know that they will not be able to provide unassailable answers to all the physical problems manifested in the different copies of a literary text. But they do want to identify all the questions at least. They want to collect and classify all the variations. Still, it would be a mistake to think that because of their passion for completeness all textual editors are advocates of closure. Doubtless some are. But some others display a love amounting to obsession for variation, a gleeful acknowledgement of variety and bewilderment, a deep satisfaction in the bafflement of their projects.
These critics and editors have perhaps come to accept a certain tension inherent in their work arising from the circumstance that the real physical thing they seek is in its origin and purpose imaginary.
For a “critical edition” presupposes that all previous – that is, existing – editions are spurious, flawed. There is no one thing to be found here. Or, rather, there are many things to be found which are all to be taken as no things, as negations. By this I mean that the errors that each existing text obediently reveals to the textual critic’s Oedipal gaze are undeniably, concretely present, but are read as their own cancellation, as evidence of absent, correct instances. (So Oedipus reads his sins as signs of right actions that should have been taken and resolves not to be mistaken again).
The literary critical edition then, is composed of two distinct types of signs: the first are unproblematic signs that represent what they at first represented, and the second are problematic signs that do not now represent what they once represented (corrected mistakes: typographical errors, for example). In practice existing texts are often all subordinated by textual criticism to a non-existent text: the intended text. A traditional textual critic is thus an augur of intention and, ultimately, its manifestor as well. Such a textual editor is no person other than the true author of the text (not the actual author’s double, but her apotheosis).
Reading someone’s or something’s intention, although we engage in it constantly and although it is one of the necessary conditions of communication, is fraught with difficulty. For intention is prototypically causal, and there is no escaping getting involved in arguments about effects and causes once the work of establishing intent is undertaken. (Curiously, in the case of errors the intended text is established through the inefficiency of the operation of cause and effect, for an error is precisely what was not intended; the precision in this case allowing a causal inference to be drawn that postulates both a different effect – the correction – which is taken as identical to the original intention, and a different cause – the deflection of the original intention.) At its best textual criticism demonstrates a refined appreciation of the difficulties of attributing intention (as, for instance, when dealing with corrections or interpolations in the author’s own hand or in the notorious case of Shakespeare’s compositors). 
But it cannot be denied that attributing intention is what traditional textual criticism does most of the time. It is not, then, simply a practical business of comparing existing versions and taking the one that seems to be the best, as you or I might do when selecting something to read. Rather, textual critics and editors seek to enforce a moral imperative: comparing existing versions to arrive at what should have been on the basis of what was intended to be.
So the text, or pre-text, of a traditional critical edition in literature is an imaginary (and good) object, the traditional theory of textual criticism and editing in literature is a theory of intention, and the practice of such textual criticism and editing is the establishment of intention through a cause and effect chain of valid inferences.
I do not believe that textual criticism and editing in film needs to be based on the same presuppositions. Indeed, I believe that the physical conditions of the production, distribution and exhibition of film prints provide the grounds for adopting quite a different theory and practice. However, the practice of textual criticism in the writing on film, such as it is, often seems to have been informed by certain of the basic presuppositions of a traditional literary approach to the topic. Two of these – the idea of the author’s intended text and the idea of the original, or copy text are illustrated in the two cases that follow.
The case of Red River (Howard Hawks, 1948)
For many film analysts Red River stands as a sterling example of what is best and most typical in the work of its director, Howard Hawks. It now appears, however, that two versions of Red River have been in circulation for some years. One, which was the version actually released in 1948, is narrated by Walter Brennan and climaxes with a gunfight in which nothing is said. The other, a version shown at previews before the official release and at one time available fairly readily in pirated 16mm prints, contains written narration (the viewer reads from a handwritten manuscript) and ends with a gunfight in which one character is repeatedly challenged by another to draw and fight back.
In Howard Hawks, Storyteller, the late Gerald Mast discusses the two versions and makes a case for preferring the latter – which he, like others, calls “the Book Version”.  Earlier, in a letter of which Mast was unaware when he wrote his discussion, John Belton had argued for the primacy of “the Voice Version”.  The arguments are of interest here, especially Mast’s, which appears to have been informed by certain precepts of traditional literary textual criticism.
Mast begins by pointing out that the Book Version is longer and “contains everything that the shorter version does and more”. But he qualifies this by saying that “just because the version came first does not mean that it seemed preferable to Hawks or anyone else” (p. 340). Having asserted that the Voice Version was the result of “extra-artistic” pressures (audience restiveness at previews), he notes that this does not mean that it will necessarily be “artistically inferior to the original”. There follows a carefully worded passage: “Hawks had no objections to the voice-over version, may even have preferred it . . . and never expressed any feelings at all about the preferability of either of the versions”. (p. 341) 
He then cites “minor inconsistencies” in the Voice Version.
These are matters of taste and style rather than textual or interpretative problems, and Mast’s authority for deeming something inconsistent is whether Hawks would have done it or not. This line of argument may be undercut slightly when he makes the claim a little later that “Walter Brennan’s narration . . . pushes Red River closer to the smaller, more domestic, more chamberlike, conversational western that Rio Bravo [Howard Hawks, 1959] is”, and that he prefers the Book Version “precisely because it is more epic”. (p. 342) (He also calls it “lyrical” in contrast to the “conversational” Voice Version.)
Finally Mast argues that the climactic gunfight in the Book Version is better because it includes “essential narrative details” (p. 343) that the ending of the Voice Version does not. For Mast this latter judgement is explicitly not a question of “taste,” although surely it is just that – even if most of us would agree with Mast’s taste for narrative completeness in this instance.  He concludes with a certain epic daring of his own: “The Book Version of Red River seems to me not only the version of the film Hawks initially intended but the version he really intended… If you haven’t seen the Book Version of Red River, you have probably not seen Red River“. (pp. 345-346)
Red River and the argument for authorial intention
Mast overtly claims authorial intention as the principle grounding of his argument. In this, as well as in his admirably scrupulous manner of establishing authorial intent, he is faithful to a scholarly practice common in the editing of literary texts.
The theoretical issue posed here is apparently whether in film study we should prefer something like an author’s manuscript to something like a first edition. In American textual editing of literature this position has been prominently argued by Fredson Bowers, whose thesis it is that the final authority in literary textual criticism must be the author’s intentions, preferably as embodied in a manuscript, rather than any printed edition. 
Concern for establishing authorial intention has been in this way linked to an older textual tradition interested in establishing the authority of classical manuscripts, in which chronological priority is the final arbiter and the reconstruction of ideal texts an accepted and necessary practice. 
Belton’s argument, by contrast, never stands on authorial intention – and this is particularly remarkable because the potential case for such intention would seem at least as strong as the one Mast must make. Hawks did authorise the release of the Voice Version and, given an opportunity to rescind that authorization, confirmed it instead. But Belton does not say this. Instead he concedes that “the structure of the film” is essentially unchanged by the change in the mode of narration – “it is its tone which changes slightly” – and he expresses a preference for the tone of the Voice Version. His argument, in other words, is entirely based on taste.
Now, it seems to me that Mast’s is too, even if not overtly. The actual grounding of Mast’s claim is not Hawks’ “real” intention, which is precisely what is not known, but the aesthetic merit of the Book Version. It may well be that the Book Version is better – although I am not sure what that qualification may mean beyond an assertion that it has pleased someone more – but quality is not supposed to be an issue in the “scientific” tradition of textual criticism that Mast appears to be following in this instance. 
Moreover, the Voice Version was the one originally released. In the absence of strong testimony about Hawks’ preference for the Book Version a heavy burden of proof descends upon Mast, which his argument fails to bear. His preference for the Book Version seems to me to strain the authority of Bowers’ preference for the author’s manuscript beyond the limits of what it was devised to take.
In a discussion of the parallel situation in literature, Thorpe calls for a distinction between “potential” – that is, unissued – versions and “actual”, released, ones, and concludes that the work of a responsible textual editor must be based on actual versions whenever such versions exist.  It seems to me that this principle has particular force when applied to film.
It is at the point of release that the parallel between film texts and printed texts breaks down. Printing merely captures a moment in the life of a written text as a snapshot freezes the movements of a dancer. Any printed edition, including a critical one, may be presumed to contain mistakes. Any printed edition is subject to correction. This is not the case for commercially released film texts. What is released is – and somehow must be – in this instance, the “final version”, mistakes and all.
This sounds as though I am articulating a prescriptive theoretical rule, but I am actually attempting to describe the conditions of film distribution and their effects upon scholarship. I can see no very compelling reason why conditions should be so very different in film and print publication, except that the notion that an individual may own a film text as an author owns what she writes is not common within the cinematic institution. Instead, ownership is usually vested in corporate bodies in the way ownership of some written work is vested in publishing companies. But this notion of ownership is quite different from the notion of authorial ownership (which is why we can entertain the idea that a book, for example, might be subject to legitimate control simultaneously by both an author and a publisher – not to mention the owner of a copy of the book). The point is that publishing practice, enforced by legal and cultural sanctions, supports the rights of an author over the text she has written: the text never entirely escapes her control. In the commercial cinema that situation does not obtain. Once films have been publicly released, individual human authors (writers, directors, producers, whoever) have few, or no, rights over the texts upon which they have worked. The releasing agent holds most of the rights involved – and thereby hangs an anomaly.
For those rights are total in one sense and non-existent in another. The corporate owner of a commercial film text may do whatever it likes with that text. It may make change after change before, after and during release. But each time the film text is shown is an instance of publication, and each variant thus constitutes a separate edition. Each instance of publication in effect freezes that version of the text “for all time”. Because there is no authorial right to the film text, the intended, imaginary, text cannot take precedence over any existing version, and the changes made by the corporate owner during the life of the text – including gross abuses and errors – are all made with the weight of the same authority (that is, none may be said to be more acceptable than another). For the purposes of film scholarship, this has the effect of nullifying the corporate owner’s rights at the same time that it asserts them – for each “published” version of a film is as definitive as any other.
I am referring to inescapable and undeniable conditions of film distribution and to the implications of those conditions for film scholarship. What I have said does not rule out the possibility that someone – Mast or Belton, for instance – may evince a preference for one version over another and may argue for the superiority of one version over another, even on the basis that the version in question is what the “author” of the film intended. What is inadmissable is the attempt to prove that one or another version – in this case the Book Version or the Voice Version of Red River – is the authoritative one. Both versions were publicly shown, both were published, both are definitive, “final”.
The case of Life of an American Fireman (Edwin S. Porter, George S. Fleming, 1902)
Life of an American Fireman, directed by Edwin S. Porter for the Edison Company in the early part of this century, raises a different set of issues. For some at least, nothing less than the history of the language of film is at stake. It was widely accepted for several decades that Porter had invented “cross-cutting”, probably without much cogitation, in 1902, and that the evidence was to be found in Life of an American Fireman. The first and best known argument for this claim was advanced by Lewis Jacobs in 1939. Writing in the absence of any prints of the film, Jacobs went to the Edison Catalogue for 1903 and in it discovered an account of “Scene 7: ARRIVAL AT THE FIRE”, which seems to describe a sequence of three shots edited together so as to show alternating spaces (outside and inside a burning house) while preserving temporal continuity. 
Not long after, a print of the film was discovered with an even more elaborate montage than that described in the Edison Catalogue. However, Kemp Niver’s restoration of the paper print collection of the United States Library of Congress brought to light a version deposited for copyright purposes in 1903 that also does not correspond to the Catalogue description. The final sequence of this print too consists of three shots, but here the third (outside the burning building) recapitulates the actions of the second (inside), giving two views of the same action and destroying chronological narration in the process – an “instant replay” effect.
The discovery of this material (and, later, of a nitrate print from around the same period that is identical for Scene 7) fueled renewed discussion about Porter’s role in film history. In 1979 André Gaudreault, referring to previous articles by Roman Gubern and Barthélémy Amengual, published an exhaustive and considered study of the arguments for and against the two existing prints which is a model of traditional textual criticism applied to film. At the same time (indeed, side by side in English and in French in the same publications), Charles Musser produced an impressively researched and reasoned “revisionist” account of Porter’s career in film to 1903 that included a detailed case for accepting the Library of Congress version (which Gaudreault calls “the Copyright Version”) as authoritative. 
Gaudreault and Musser both think that the Copyright Version is “the original version”.  Gaudreault can find no evidence that Porter was using cross-cutting at that time in the way that the Edison Catalogue description of Scene 7 seems to suggest (at least to us today). On the other hand, he sees a definite structural parallel between Life of an American Fireman in the Copyright Version and The Great Train Robbery (Edwin S. Porter, 1903). His arguments are, then, somewhat like Mast’s in that both consist of attributions of authorial intention. Porter is seen as the cause of Life of an American Fireman and his actions are the authority upon which can be determined what is “original”.
Musser, in a complex and compelling argument, takes a somewhat different line. He argues that early cinema does not develop according to a linear pattern, with each “innovation” building on the last, but that specific solutions arise in response to specific problems: put simply, “No inventor of film editing existed”. (p. 3) In so arguing he is rejecting the basic presuppositions of film historians like Jacobs who are concerned to establish priorities in a quest for a straight line, cause and effect history. His case for the Copyright Version first stresses the authenticity of the paper print deposited for copyright in 1903 by asserting that such prints were made by Edison at that time directly from the negatives used to make release prints. This is an important point, for such was not always the practice of other firms: unedited footage was sometimes used to make the paper prints, and sometimes only the tops and tails of shots were provided. He suggests that since the Copyright Version is longer it is more likely to be the original. This is a principle that is often at least tacitly followed in establishing authoritative editions of films and which deserves closer attention than I can pay to it here. He says that whoever wrote the description of Scene 7 for the Edison Catalogue would have had a difficult task in describing the scene as it appears in the Copyright Version and that the resulting written description more nearly matches that version than it does the other, called by Gaudreault “the Cross-Cut Version”.  But his major grounds are those of consistency: “Today it should be clear that the LoC paperprint [Copyright Version] is internally consistent, is consistent with Porter’s own development as a filmmaker, and consistent with the development of international cinema during the 1901-1903 period”. (p. 31)
This is overtly an argument about the priority of a certain physical text, not an intention; but it is also, I think, an argument based on a perception of style. Musser points out that actions are repeated earlier in the film when the firemen first jump down the fire pole in shot 3 and are then shown sliding down the pole in shot 4, after a lapse of time. He relates the repetition of action in Scene 7 of the Copyright Version to similar repetitions in How They Go Things on the Bowery (1902), also made by Porter. Then he says that continuity, established by sequential shots showing repeated actions, “signaled a decisive shift in editorial control from exhibitor to cameraman . . . In the process, creative control became centralized primarily in the production companies”. (p. 34) Life of an American Fireman displays “a mode of representation that was unstable, transitory, a direction in narrative cinema that was briefly explored, discarded and gradually forgotten”. (p. 34) 
In this argument Musser does use authorial authority: what Porter had already done, who and what he was (his piece is about Porter, after all), but he mixes these warrants in with others about industrial practices, both marketing and textual. Clearly it would have benefited the claim that editorial control shifts from exhibitors to producers in this period if Life of an American Fireman could have been shown to have utilised such a complex device for establishing continuity as cross-cutting, but Musser resists the temptation to fudge the data as perhaps others have not.  The thrust of his contentions about consistency are, I think, ultimately authorial, however, even when they do not posit Porter as an individual, humanistic author in the classic sense. Industrial practice and “modes of representation” are put in the place where human authors used to be. The film that results is the product of economic, social and cultural intentions rather than personal ones. More than that, the argument of consistency is ultimately an appeal to a logical and static model of the past, allowing little room for innovation – which, after all, is what Jacobs and the other advocates of the Edison Catalogue and Cross-Cut Versions claim Porter did.
But the point is not whether Musser and Gaudreault are right or wrong or even whether their arguments are sound or not. The point is first, the nature of the arguments that such textual criticism employs, arguments which are idealistic and conservative. The goals of most textual critics and editors are recuperative. Traditionally they have defined the changes wrought by time upon texts as “contamination”, and have understood themselves, as art restorers often do, to be cleansing cultural objects. In some sense they are always striving to hold on to the past. But the past uncontaminated by the present, by some context which interprets it, cannot be grasped. Artifacts of the past, like prints of films or catalogue descriptions, have persisted into the present, more or less physically changed, but their continued existence at best only guarantees the recuperation of a plane of expression from the past, not of texts or sign-functions intact. Inevitably textual critics write history rather than merely preserve the past; and the history written by the practice of textual criticism is a story of searching for lost objects and hidden motives – what cannot be recovered: original meaning.
Perhaps the most intriguing theoretical statement to invoke the Copyright Version of Life of an American Fireman was advanced by Noël Burch in Screen at about the same time that Musser and Gaudreault’s articles appeared.  He too castigates “the adherents of linear history” for their simplicity and the inaccurate picture they present. For Burch, Porter’s work represents “the first stages of the process” of the establishment of what has come to be accepted as the right way of communicating cinematically, or what he has called “the Institutional Mode of Representation”. He ties the practices of early cinema to its function as entertainment for the working classes and relates the “Primitive Mode of Representation” of those films to other examples of popular working class art from the same period (pp. 103-105). When he comes to describe Life of an American Fireman, he uses the Copyright Version and speculates that:
Porter felt, probably with reason, that the audience was not yet ready to accept [cross-cutting], and he (or someone else, it doesn’t really matter) decided to show the two actions successively… Once again one of Porter’s `steps forward’ in fact ends by accentuating some features of the primitive cinema even more strongly than before… here its non-linearity. (p. 103)
The Copyright Version is essential for the historical argument here, but Burch uses none of the traditional methods of textual criticism to validate the print he has chosen to accept. Instead he asserts that “around 1940… someone (who?) simply re-edited the film, leaving out parts that quite obviously originally belonged, and Porter was turned into the modest inventor of one of the basic syntagms of the institutional mode… some ten years before any such figure began to appear in the standard syntax”. (p. 104)
The last phrase, I think, invokes a linear model of history as the grounds for disbelieving those who claim Porter as “the modest inventor”. The date given for what Burch calls the “re-editing” of the film is, so far as I am aware, entirely unwarranted, although it is the date of the Cross-Cut print’s acquisition by the Museum of Modern Art.  Rhetorically however, Burch’s claim of re-editing has the significance of asserting ever more strongly the authenticity of the Copyright Version, as of course do the implications of deliberate fraud in his statement.
This strategy is pursued further in the same paragraph when Burch turns his attention to the Edison Catalogue description, which he treats as a problem of incompatibility between verbal description and what was actually shown in the film. He suggests that “the aspiration to this linear ubiquity which was to become characteristic of the institutional mode had reached a level where it could be written down but not yet realised on film”. (p. 104)
The idea that the Edison Catalogue description might refer to an actual print from the era is never entertained for a moment. Although Burch explicitly intends to demonstrate that Porter’s work challenged certain aspects of the Primitive Mode of Representation, like Musser he believes that a Cross-Cut Version from 1903 is not just improbable, but formally and historically impossible. 
Life of an American Fireman and the idea of an original text
Burch’s piece is titled “Porter, or ambivalence”, yet there is nothing ambivalent about his attitude towards the authenticity of the Copyright Version. In a Postscript he asserts unequivocally that the “distribution copy found in Maine is identical with [the] copyright version” (p. 105).
But this is not the case, according to Gaudreault, who details the differences. This “Recovered Version” is identical for Scene 7, as I have mentioned, but it is not for the way in which the shots are joined together elsewhere in the print. There are no dissolves (which Gaudreault at one point calls “superimposed fades”) in the Recovered Version, whereas most of the shots in the Copyright Version are joined by dissolves, except for the last two: Gaudreault explains that “the next-to-the-last shot ends with a fully densified image and the final shot begins with a fade-in”. (p. 55) These are the very shots which are the focus of all the controversy: an interior showing the rescue of a mother and child and an exterior apparently repeating that rescue.
Gaudreault, like Burch, attaches no importance to these differences, yet they are significant. They show that some release prints of the film were not identical to the Copyright Version. Moreover, they involve what one historian at least, believes to have been a stylistic feature of the Porter films of the period: the dissolve. In his pioneering Film Style and Technology: History and Analysis, Barry Salt asserts unequivocally that “in Porter’s The Life of an American Fireman (1903) all the shots were joined with dissolves” and that dissolves were (very briefly) “the standard form of transition” for Porter’s films in the early years of this century.  .
Gaudreault points out that the title that precedes the Recovered Version of Life of an American Fireman bears the date, September 30, 1902. (p. 56) The Copyright Version is supposed to have been deposited in 1903. The Edison Catalogue is also from 1903. Clearly the Recovered Version is much the closest to an “original” – in point of time, at least – of any existing version.
Presumably between the time the Recovered Version was sold and the time that the Copyright Version was given to the Library of Congress someone in the Edison Company changed the way in which most of the Scenes of Life of an American Fireman were joined. At the same time, that person did not alter the way in which the shots of Scene 7 were joined.
The Edison Catalogue posits dissolves between every Scene or shot of the film. Both the Copyright and Cross-Cut Versions show dissolves between the first six Scenes, as the Catalogue describes. Here the correspondence between these three versions suggests that by 1903 the Edison Company had determined that “its” version of the film was to join the first six Scenes by dissolves, although this had not been the case in 1902.
The Edison Catalogue is quite unambiguous in its language about how the last two shots in Scene 7 are joined. “We now dissolve”, it says. But, according to Gaudreault, neither the Recovered Version nor the Copyright Version shows a dissolve. They both cut shot 8 (the interior) cleanly and fade up shot 9 (the exterior). The Cross-Cut Version does not dissolve at all during Scene 7. Here the Edison Catalogue Version is internally consistent, whereas all other versions are inconsistent. It is also, of course, inaccurate about all existing prints. The Recovered Version is also internally consistent, insofar as the modes of transition between shot/Scenes in it are systematically opposed to those described in the Edison Catalogue.
Both Musser and Burch explicitly recognize the role exhibitors played in editing films. Indeed this is a central contention of Musser’s position. The best known example of this control, cited by everyone everywhere it seems, is the closeup of a cowboy firing at the audience from The Great Train Robbery, which exhibitors were told could be attached to the head or the tail of the film according to their preference. Musser emphasises that this was merely an acknowledgement of common exhibition practice, which extended to the construction of narratives (including narratives about putting out fires) made from single shot films acquired from any number of different sources, although Burch attributes this “extraordinary intuition of the possibilities offered by montage” to Porter alone. (p. 103) Musser does not say that exhibitors bought outright the films that they showed at this period and thus were entitled by law to do almost anything they wished with them, including making “new” films of them, but this too was part of the industrial circumstances surrounding film production and distribution in 1902. 
Salt adds a further wrinkle. He implies that Porter may have been producing footage for Kinetoscopes as well as for films intended for projection. (p. 60) Kinetoscopes were “peep show” machines that rarely displayed more than a single shot for viewers. When multi-shot films were displayed in Kinetoscopes the practice was to put each section (often just one shot long) of the film into a separate viewing machine. The viewer went from machine to machine, paying each time, in order to see the whole thing. If Life of an American Fireman were displayed in this way, there would certainly have been no dissolves between the sections of the film intended for separate machines.
In these conditions it seems pointless to attempt to establish which version of a film is the “original”. As Walter Benjamin was at some pains to argue, modern means of reproduction make hash of traditional notions of originality in art.  Benjamin’s insight has not been taken fully into account in these writings on different versions of films, which all seem to have as their aim the establishment of one and only one correct edition, like the lost original versions sought after by classical textual critics.
I suggest that the instance of Life of an American Fireman requires understanding on a somewhat different level than it has prompted heretofore and that the confusion between existing and hypothesised versions should be recognised as a sign that textual plurality is a condition of the study of film.  It is possible to explain the variations detailed by Gaudreault and others by claiming that what was sold to exhibitors at this time was no more than a register from which one could choose what one wished to show to the public.
The Life of an American Fireman does indeed, as Burch and Musser argue, illustrate a transition between modes of representation. Dissolves are used to connect shot/Scenes in all of the versions of the film that can be dated after 1902. Dissolves would make clean re-editing difficult: heads and tails of shot/Scenes would have to be chopped out. This does suggest an attempt by the producer (the Edison Company) to control possible re-editing of the film by exhibitors. In this regard it may be significant that the Recovered Version is the one most likely to have actually been shown in or 1902, for that version contains no dissolves at all, and may by that circumstance illustrate the extent of exhibitor control over the content of the films they bought during that year, at least.
The lack of a dissolve between the last shots in all but the Catalogue Version coupled with the undeniable existence of prints in which all but the last shot/Scenes are connected by dissolves, suggests it was at that point in the film that exhibitors commonly exercised their control. If that was the case, then a number of different edited versions of Scene 7 would have existed in 1903 and afterwards – and it is just remotely possible that some of those might have resembled the Cross-Cut print, if only because it does seem remotely possible that the Edison Catalogue description may have functioned something like a recommendation from the film’s producers, a suggestion of what one can do with one’s print.
The very existence of the Copyright and Recovered Versions, both definitely made from some sorts of negatives (which cannot be so certainly asserted for the Cross-Cut Version), can be taken as evidence for the existence of a plurality of film texts in this instance. One need not argue that an actual Edison Catalogue Version existed nor that the existing Cross-Cut Version dates from 1902 to embrace the position I am outlining here, and I am not making either contention. What I am contending is that many versions of Life of an American Fireman existed, even in 1903. Furthermore, the Cross-Cut Version, no matter when it was constructed, is no less legitimate than any other. It is a version, as definitive and final as other versions of the film. It is an instance of the film’s textual plurality. It even has a claim to an important place in a history of film for the nationalistic polemics and the attempts at scholarship it has inspired.  It should be clear that this position is not at all at odds with the revisionism of Musser and Burch. On the contrary, I would contend that it supports their understanding of the film past in some ways better than the arguments they have themselves adduced.
Films as plural texts
At this point it seems appropriate to return to Red River. Many – perhaps all – of the existing prints of Red River are missing a scene in which Dunson kills the last Indian of a band who have attacked him and Groot. The absence of this scene causes the sequence to which it is appended to end on an enigmatic note and it seems likely that its omission is the result of normal wear-and-tear on reels (it occurs at the end of the first reel) or, perhaps, of a species of vigilante censorship that is not uncommon, rather than being integral to the final cut of either the Book or the Voice Versions. No matter. Its absence constitutes the circumstance for another legitimate (definitive and final) version of the film. To go to the other extreme, if the rushes of Red River were to be recovered, they too would make up a version of the film. 
It should be clear that textual plurality, as I understand it, is not an oddity for cinema. Recently it has even been acknowledged overtly as normal and accepted practice. When Francis Ford Coppola re-edited The Godfather (1972) and The Godfather, part II (1974) for television he added scenes not in the released film versions as well as deleting others.  Charles Barr has published an interesting comparison of the silent and sound versions of Blackmail (Alfred Hitchcock, 1929) that illustrates another, not uncommon, instance of a plural text.  Napoléon vu par Abel Gance, like many films of the twenties and early thirties, was shot in separate versions – for domestic distribution and for export – that is, at least two negatives were made. One explanation for the different versions of Life of an American Fireman would be that more than one negative was made, whether or not more than one version was shot.  Certainly different negatives would have been made for the different versions of Red River, and very possibly both were made from the same original shot material.
A negative is not the original of a film, although perhaps it comes close to that ideal. A film has no original. A film is always a copy. Strictly speaking, the same is true for printed literature, and a case for textual plurality as a condition of literary studies (although not under that name) has been argued by Hans Zeller and, less categorically, by Jerome McGann.  In a telling phrase, McGann describes the textual editor as “dealing with a both/and, not an either/or situation”. (p. 104) Ultimately neither Zeller nor McGann is quite comfortable with the idea of a multiplicity of texts, and both mull over ways of deciding which one to use. But the direction of their thinking is, it seems to me, that there is not one to use, but many, that a text is not singular, but plural. That this is even more certainly the case for film has, I hope, been adequately demonstrated in what has gone before.
In literary textual criticism recently G. Thomas Tanselle has had occasion to return to Fredson Bowers’ concept of “radiating texts”.  “Radiating texts” are those, like the differing published versions of a syndicated newspaper article, in which it may be impossible to establish for certain which of many variants should be given editorial priority. The situation seems much the same for prints of films. Tanselle, however, makes a distinction between “works” and “documents” in order to argue for the necessity of making reconstructed, critical, editions:
If we wish to experience the texts of works (or versions) and not simply the texts of documents, we must leave the certainty (or relative certainty) of the documentary texts for the uncertainty of our reconstructions… The process of critical editing is the ineluctable, if unending, effort to surmount the limitations of artifacts in the pursuit of works from the past. (p. 6)
A hypothetical follower of Tanselle, in other words, would have some warrant to argue for the establishment of critical, authoritative film texts in spite (or because) of the circumstance of film’s textual plurality. That warrant, however, depends upon the acceptance of the idea of a “work”, an intangible object “which can be stored only through conversion to another form, which in effect becomes a set of instructions for reconstituting the work”. (p. 5)
Many contemporary film academics would not be comfortable with Tanselle’s argument even when their own practice suggests something like its unthought acceptance. For example, referring to a recent version of Intolerance (David W. Griffith, 1916) that attempted to reconstruct the print shown at the film’s New York premiere, Miriam Hansen acknowledges the film’s plural existence:
Given the instability of the text… such a reconstruction necessarily privileges one, albeit a relatively less arbitrary moment in the film’s history, it cannot – nor does it pretend to – “restore” the work in a supposedly timeless originality. (p. 39) 
Hansen’s remarks preface a long and sustained textual analysis of the film in which she has occasion to mention variants in the versions she has studied. The variants she cites modify details of the analysis but have no effect on the overall argument. Indeed, the newly-reconstructed version was not among the versions she was able to use (330 n. 17), and her analysis does not noticeably suffer from its absence. Finally, in other words, her analysis, like virtually all film analysis, is rather more applicable to an imaginary “work”, than to any existing version of that work.
Paul Eggert’s argument, in an article printed in the same issue of Studies in Bibliography as Tanselle’s, resembles what may be implied, but not followed through, in Hansen’s comments on the variations of Intolerance.  Eggert, writing from the point of view of an art “restorer”, argues that each existing version of a work attests to historical circumstances of that version’s production and reception. Thus the variants of Red River or Life of an American Fireman might be used as evidence in arguments about how people thought about motion pictures at different historical moments.  I have no strong quarrel with this way of thinking, but I do think that one should recognise that the key to its persuasive power is the concept of use value. The variants are not worth our while “for themselves” but only for the uses to which they can be put.
Instead I think that the most powerful case for recognising textual plurality and for paying attention to textual variation is very like an aesthetic one. It seems to me that the specificity and multifariousness of aesthetic experience validate McGann’s contention that “far from representing an ‘alien’ condition for messages, it seems to me that ‘the physical’ (whether oral or written) is their only condition”,  and Peter Shillingsburg’s assertion that “works are known through proliferation of texts, not through their refinement or concentration”.  The variants in plural or radiating texts are, first of all, worth attention for themselves, because of their variety, their specificity, their difference. These are matters of tangible surface detail, not of ideal generating forces, whether imagined from beneath (as for an original or copy text) or from above (as for an author’s intended text). But refocussing attention to the surface of actual prints is refocussing on what is unarguably there as well as on what is usually taken as the aesthetic aspect of a work. The grandest claim that can be made for textual plurality is that it is true, and by this I mean not only that it corresponds to specific industrial and historical practice, but also that it reflects inescapable and important conditions of communication that are often held to one side in the analysis of cinematic texts.
 I am indebted to the two referees of this manuscript, whose comments have helped to bring work begun in the 80s into the 90s. The late Gerald Mast read portions of an earlier draft and generously commented on it, and some of his suggestions have influenced certain sections. Conversations with Enno Patalas and Ronald Haver, held in 1983, and FIAF 42 (13-19 April 1986), hosted by the National Film and Sound Archive of Australia, were important in helping me to understand certain aspects of film restoration as it is practiced today.
I have made unrelenting use of Richard J. Thompson’s wonderful files and library, as well as of his ears and mind. My interest in textual criticism and editing originates, as all else worthwhile in my life originates, with Diane.
 Kevin Brownlow, Napoleon: Abel Gance’s Classic Film (London: Jonathan Cape, 1983). The information in the foregoing paragraph is all taken from this book.
 Tim Lucas, “The Watchdog barks”, Video Watchdog, issue 1 (1990), p. 3.
 Originally published in French in 1964 as F. W. Murnau, the English translation of the book nine years later had been significantly revised and enlarged (Murnau, London: Secker and Warburg, 1973). Still, the text of chapter 7, “The riddle of Nosferatu”, is only some 6-7 pages long. I think it is significant that it has only been in the last few years that Eisner’s assessment has been seriously discussed – and that a “popular” or “fan” magazine, Video Watchdog, rather than an academic or serious film journal, has provided the platform for that discussion (see David Walker, “Nosferatu: the unauthorized undead”, Video Watchdog, issue 19 [September/ October 1993], pp. 48-61; and the same author’s “Nosferatu update”, Video Watchdog , issue 24 [August 1994], pp. 4-6).
 The two pieces prior to 1995 were: Tony Pipolo, “The spectre of Joan of Arc: textual variation in the key prints of Carl Dreyer’s film”, Film History, Vol. 2 No. 4 (1988), pp. 301-324; and Joseph Garncarz, “Fritz Lang’s M: a case of significant film variation”, Film History, Vol. 4 No. 3 (1990), pp. 219-226. The two since then are both exemplary of the “new understanding of scholarship” to which I refer: Giorgio Bertellini, “Restoration, geneology and palimpsets. On some historiographical questions”, Film History, Vol. 7 No. 3 (1995 ), pp. 277-290; and Yuri Tsivian, “The wise and wicked game: re-editing and Soviet film culture of the 1920s”, Film History, Vol. 8 No. 3 (1996), pp. 327-343.
 Patrice Petro, Joyless Streets: Women and Melodramatic Representation in Weimar Germany (Princeton: Princeton University Press, 1989), pp. 216-218.
 Too much can be made of the dangers of interpreting the wrong print. The most glaring recent instance of misinterpretation arising from a “corrupt” print of which I am aware occurs in an article on sexuality in Metropolis (Fritz Lang, 1927), in which Andreas Huyssen has made the classic error of categorically asserting authorial intention on the basis of something which is only true for most of those prints in circulation at the time of writing (“Fritz Lang does not feel the need to explain the female features of Rotwang’s robot”: “The vamp and the machine: technology and sexuality in Fritz Lang’s Metropolis “, New German Critique, No. 24-25 [1981/82], p. 225). However, Huyssens’ overall argument is not really affected by this gaffe.
 Lawrence Alloway, Violent America: The Movies 1946-1964 (New York: Museum of Modern Art, 1971), pp. 29-34.
 Two who may feel comfortable with textual instability are Richard Maltby and Ian Craven, who discuss the issue in Hollywood Cinema: An Introduction (Oxford: Blackwell, 1995), pp. 292-293.
 For the latter see Fredson Bowers, Textual and Literary Criticism, (Cambridge: Cambridge University Press, 1959), pp. 77-101. Further references to this work appear as page numbers in brackets.
 Gerald Mast, Howard Hawks, Storyteller (New York: Oxford University Press, 1982), pp. 337-346. This is one of the few attempts at textual criticism in film to have appeared in a book-length study, and is certainly the most detailed to have appeared in any textual analysis of film. Further references to this work appear as page numbers in brackets. Mast’s approach has been influenced by Philip Gaskell’s introduction to From Writer to Reader: Studies in Editorial Method (Oxford: The Clarendon Press, 1978), p. 1-10. Further references to this work appear as page numbers in brackets.
 John Belton, “Letters,” Movietone News, issue 52 (1976), pp. 24-25. Further references to this work appear as page numbers in brackets. Belton’s letter follows on from a discussion initiated in the previous issue’s “Letters” section which contains some pertinent remarks by Richard Jameson (Movietone News, issue 51 , p. 29).
 Mast’s evidence for Hawks’ lack of preference does not seem to support such strong wording. He cites a letter from David Shepard that says merely that Hawks “did not seem to have strong feelings” on the point. (p. 381, n. 38) Later in his discussion he admits that Hawks told United Artists to “follow the original release” (i.e., the Voice Version) when the film was re-released. Richard Jameson’s letter in Movietone News, referred to in note 12, says twice that Hawks preferred the Voice Version, and is apparently citing some personal and recent communication, possibly an interview (an interview with Hawks was conducted on July 12, 1976 by Kathleen Murphy, a portion of which was published in Movietone News, issue 54). Mast clearly thinks that the preferences Hawks expressed were motivated by legal concerns rather than by aesthetic ones.
 There is no necessary linkage between the attenuated ending and the voice-over narration which together characterise the Voice Version. The circumstances which dictated the changes in the ending were quite different from those which determined the change to voice-over narration. According to Mast, Howard Hughes forced the changes in the ending by threatening a suit for plagiarism based on similarities between this ending and the ending of The Outlaw (Howard Hughs, 1943), a film he had produced and Hawks had (in part) directed in 1940 (pp. 343-344).
 The position is argued in Bowers’ essay, “Some principles for scholarly editions of nineteenth-century American authors”, Studies in Bibliography, Vol. 17 (1964), pp. 223-228, and memorably condensed in that writer’s image in Textual and Literary Criticism, of textual critics trying to “restore the shape of the lost manuscript as [they] strip away … the veil of print” (p. 81).
 This is called “the Lachmann Method”, after Karl Lachmann (1798-1859), who first developed the complicated genealogical procedures associated with it. It is perhaps most succinctly set forth in Paul Maas’ Textual Criticism (Oxford: Clarendon Press, 1958). My understanding of the various positions has been much clarified by their presentation and discussion in Jerome McGann, A Critique of Modern Textual Criticism (Chicago: University of Chicago Press, 1983) and James Thorpe, Principles of Textual Criticism (San Marino: Huntingdon Library, 1972). Further references to this work appear as page numbers in brackets.
 For a critique of the scientific pretense of modern textual criticism see Thorpe (p. 57-68). Mast refers to a personal letter from Belton proposing a “composite version” which would be the Voice Version plus all of the material not currently included in that version (pp. 381-382, n. 42). Such a proposal tacitly recognizes some merit in the Book Version, but eclectic versions are considered “dangerous” by Thorpe. (p. 190) A preferred soundtrack would have to be selected for such a composite version, for Mast mentions (presumably “minor”) differences in the soundtracks of the two versions (p. 339). Mast does not comment on Belton’s proposal.
 See Thorpe (pp. 37-47, pp. 186-193). Some textual critics believe that printed editions better embody an author’s intentions than manuscripts. Gaskell (p. 6) proposes that the arguments for considering “the first printed edition” the most authoritative are “because the author passed it in proof and it is what the original audience read”. These would also be the arguments for the Voice Version, were anyone to make them.
 Lewis Jacobs, The Rise of the American Film: A Critical History (New York: Harcourt, Brace and Company, 1939), pp. 40-41.
 Andre Gaudreault, “Detours in film narrative: the development of cross-cutting,” Cinema Journal, Vol. 19 No. 1 (1979), pp. 40-41 (further references to this work appear as page numbers in brackets); Charles Musser, “The early cinema of Edwin Porter”, Cinema Journal, Vol. 19 No. 1 (1979), pp. 1-38 (further references to this work appear as page numbers in brackets). The French publication which contained Musser and Gaudreault’s articles was Les cahiers de la Cinémathèque (Winter 1979). The two pieces are not always in complete agreement. For example, Gaudreault details a total of thirteen shots for Scene 7 in the more elaborate version (pp. 40-41), while Musser says it contains only nine shots in that Scene (p. 30).
 These are Musser’s words (p. 30). Gaudreault merely implies that the Copyright and/or the Recovered Version is closest to “Porter’s original” (p. 39).
 Musser’s assertion that the Edison catalogue description reflects the Copyright Version holds only if the alternative that it might be describing is the Cross-Cut Version. It is possible, in other words, that the Edison Catalogue may describe some other version actually in release, but not (as will be seen) the print actually found in the late 1970s in Maine.
 Not entirely forgotten however, for repeated representation of the same action is one of the minor markers of the Soviet “montage” style of the mid- to late-twenties.
 Apparently Jacobs re-ordered the stills on the plate he used from the Museum of Modern Art to make the “Jamison Continuity” that illustrates his section on Life of an American Fireman. Gaudreault (and Burch, see the discussion which follows) stop short of accusing him of having a hand in the editing of the Cross-Cut Version, which appeared shortly after his book did.
 Noël Burch, “Porter, or ambivalence,” Screen, Vol. 19 No. 4 (1978-79), pp. 91-105 (further references to this work appear as page numbers in brackets). There is a good reason for the coincidence of dates: all three had attended the 1978 FIAF Conference at Brighton. Much of the substance of Burch’s argument is reproduced in his stimulating book on early cinema, Life to those Shadows (London: The British Film Institute, 1990), pp. 204-206.
 Gaudreault cites a 1977 letter from Eileen Bowser saying that the Museum of Modern Art acquired the Cross-Cut print in 1944 from Pathe News (p. 56, n.3). Ms. Bowser has told me that no other dating is possible on this print (personal conversation, 18 April 1986). Indeed in Life to those Shadows, Burch rephrases this portion of his argument: “it is clear that, at some point in its history, some unknown distributor had felt the need to tamper with a film whose syntax was no longer acceptable” (p. 205).
 In Life to those Shadows Burch calls what constrains Porter “primitive autarchy ” and claims that its effects can still be seen as late as 1935 in La belle équipe (Julien Duvivier, 1936) (p. 206).
 Barry Salt, Film Style and Technology: History and Analysis, 2nd ed. (London: Starword, 1992), p. 52. Salt does not say to what print he is referring. Further references to this work appear as page numbers in brackets.
 Gaudreault quotes the title on the Recovered Version already referred to, which indicates an attempt to restrict the actions of print owners at the time. The restriction is of some interest today, for Gaudreault explains that it enjoins exhibitors not to copy the film on pain of infringing “the patents under which it is made and sold”. (p. 56) I take this title as evidence that exhibitors did copy films, a circumstance that further complicates the quest for “original versions” of the films of this period.
 Walter Benjamin, “The Work of Art in the Age of Mechanical Reproduction”, Illuminations (New York: Harcourt, Brace and World, 1968), pp. 219-253.
 It is possible that an inappropriate model has been accepted in thinking of film texts. Jazz record collectors have for years gloried in the plurality of texts that is a condition of recorded jazz. Different ‘takes’ of the same song recorded by the same musicians at the same recording date are positively valued and accepted as the normal circumstance, at the same time that one take may be considered superior to others.
 I am not saying that the date when a version of a film appears is unimportant or insignificant – merely that the significance of that date is, as all significance, determined by context, and is subject to historical conditions. Discovering the earliest version, the longest version, the version most likely to have been authorised by the director, or the producer, or the exhibitor or the office-boy may all be worthwhile activities, as well as others I have not named or even thought of. Each such version only reasserts the basic condition of textual plurality, a condition which I believe mandates that different ideas should govern textual editing in film than guide traditional textual editing in literature.
 At this point I am drawing the line at non-filmic versions of films, but there are arguments, some of them quite persuasive, for extending a film’s textual plurality to include written documents and production stills.
 See Carlos Clarens, Crime Movies (New York: W. W. Norton & Co., 1980), pp. 280-282.
 Charles Barr, “Blackmail: silent & sound”, Sight & Sound, Vol. 52 No. 2 (1983), pp. 123-126.
 John Fell suggests this in A History of Films (New York: Holt, Rinehart and Winston, 1979: p. 47). Popular films of this period sometimes wore out their negatives and were re-shot. I do not know whether the various prints of Life of an American Fireman have been examined to ascertain whether different versions were shot, and I have deliberately refrained from suggesting anything of the kind since it is not germane to the main argument.
 Hans Zeller, “A new approach to the critical constitution of literary texts,” Studies in Bibliography 28 (1975), pp. 231- 264; Jerome McGann, A Critique of Modern Textual Criticism, especially pp. 95-109 and pp. 117-123. Further references to this work appear as page numbers in brackets.
 G. Thomas Tanselle, “Editing without a copy-text”, Studies in Bibliography, Vol. 47 (1994), pp. 1-22. Further references to this work appear as page numbers in brackets.
 Miriam Hansen, Babel and Babylon: Spectatorship in American Silent Film (Cambridge: Harvard University Press 1991), p. 133. Further references to this work appear as page numbers in brackets.
 Paul Eggert, “Editing painting/conserving literature: the nature of the ‘work'”, Studies in Bibliography, Vol. 47 (1994), pp. 65-78.
 Indeed, this is exactly what the recent articles in Film history by Bertellini and Tsivian, cited in note 5, do.
 Jerome McGann, correspondence, 20 December 1989, quoted in D. C. Greetham, “[Textual] criticism and deconstruction”, Studies in Bibliography, Vol. 44 (1991), p. 9, n. 14. Greetham himself is a champion of multiple meanings, but not so clearly of multiple works.
 Peter Shillingsburg, “Text as matter, concept, and action”, Studies in Bibliography, Vol. 44 (1991), pp. 31-82. Shillingsburg’s main argument revolves around a single, imaginary, work nonetheless.