Electronic Textual Editing: Authenticating electronic editions [Phill Berrie, Paul Eggert, Chris Tiffin, and Graham Barwell]


‘The scholarly editor's basic task is to present a reliable text’ (undated draft revision, MLA Guidelines for Editors of Scholarly Editions)

The book is generally seen as a trustworthy carrier of text because, once printed, text cannot be changed without leaving obvious physical evidence. This stability is accompanied by a corresponding inflexibility. Apart from handwritten marginal annotation, there is little augmentation or manipulation available to the user of a printed text. Electronic texts are far more malleable. They can be modified with great ease and speed. This modification may be careful and deliberate (e.g. editing, adding markup for a new scholarly purpose), it may be whimsical or mendacious (e.g. forgery), or it may be accidental (e.g. mistakes made while editing, or minor mistranslations by a software system). And the nature of the medium makes the potential impact of these modifications greater because the different versions of the text can be quickly duplicated and distributed, beyond recall by the editor. Does the electronic future, then, hold in store something akin to medieval scribal culture? If this is the risk, will scholars be prepared to put several years of their lives into the painstaking creation of electronic editions of important historical documents or works of literature and philosophy?

How can textual reliability be maintained in the electronic environment? There is a major question here of authority and integrity which, if not more acute than that in the print domain, at least has different characteristics. Especially where it is crucial that a text be stable and long-lasting—in the case, for example, of legal statutes, cumulative records or scholarly editions—a non-invasive method of authentication is required. Following a discussion of various problems associated with the markup (encoding) of electronic texts and the danger to ongoing textual reliability that markup poses, we describe a potential model.

The subjectivity of markup

Verbal texts being prepared in a scholarly manner for electronic delivery and manipulation need to be marked up for structure and the meaning-bearing aspects of presentation. In the electronic domain, the features of text that, in the print domain, have long been naturalised by readers demand explicit categorising and interpretation. This is not straightforward. The most trivial things can ask tricky questions. What, for instance, is the meaning of small capitals or italics in a nineteenth-century novel? Traditionally, italics are seen either as a form of emphasis (and therefore a substantive aspect of meaning), or as presentational (as in the name of a ship or a painting). Neither can be rendered in the ASCII character set. As they cannot responsibly be ignored, a decision about their function (and therefore their presentation by the software) has to be provided by the human editor. Under the current paradigm, the instruction must be entered into the text file.

Similarly, electronic-text editors are forced to decide whether line-breaks are meaningful, whether a line of white space is a section break within a chapter or whether it was only a convenience dictated by the size of the printed page and the desire to avoid widows and orphans. Editors have to decide whether a wrong-font comma, a white space prior to a mark of punctuation or a half-inked character is meaningful—should it be tagged, or not? The instruction (recorded in markup) will be an editorial interpretation, made, probably, in the context of what is currently known about contemporaneous print-workshop practice and convention. In making explicit what in the physical text was implicit, the editor is inevitably providing a subjective interpretation of the meaning-bearing aspects of text. Of course a later editor, or the same editor returning with new information, may disagree with the earlier interpretation.

The arduous business of entering, proof-reading, amending and consequently re-proofing a transcription containing the new interpretation (the print-edition paradigm) can seemingly be avoided in the electronic medium; but in fact a new state of the text will have been created. Accidental corruption of the verbal text is very possible, so collation and careful checking of the new state against the old will be necessary. The same thing will be true if interpretation of other features of text is added, for example linguistic features, historical annotations or cross-references. Even though markup is usually separated from text by paired demarcators, as its density increases, so does the practical difficulty in proofing the text accurately.

Consider the following scenario. No-one expects any two scribal copies of the same work to be textually identical since scribes will almost certainly have changed or added things, large or small. This instability is, however, not restricted to pre-1455 or even the pre-1800 period, before the age of the steam-driven machine press. Optical collation in scholarly editing projects has proved again and again that no two copies of the same edition are precisely identical, even if printed in the industrial age. Printing involves change, as well as wear and tear; inking varies, and paper has imperfections. While recent editorial theory has shown that the physical carrier can itself affect the meaning of text, the prospect of marking up text to record every physical variation in every known copy of a work would create a file of bewildering complexity whose reliability would be in serious doubt. No editor can foresee all the uses to which an electronic scholarly edition can be put, nor all the interpretative markup that will be required. The more the attempt to provide it is pursued via a more and more heavily marked-up file, the more the reliability of the text is put at risk.

This situation argues the need for an automated authentication technique that separates verbal text from markup while retaining all the functionality of a computer-manipulable file. The proposal that we describe below involves such ‘stand-off’ markup. It also addresses another problem of markup that has often been observed. The current standard for the markup of humanities texts, that of the Text Encoding Initiative, requires an objective textual structuring to be declared on the assumption that if computers are to manipulate parts of text powerfully, then text needs to be seen as an ordered hierarchy of content objects with its various divisions and parts appropriately identified. The difficulty with this assumption is that texts are not just objective or ideal things. They incorporate a stream of perhaps only lightly structured human decision-making, of which traces have been left behind as part of the production process. Nor can we, as readers, help participating in the business of making meaning as we read and interpret what we see on the page. The advantage of our participation is that we, unlike computers and logic systems, can handle structural contradictions and overlaps with relative ease and safety. But if we then attempt to codify the texts for use with systems that cannot handle contradictions, the systems reveal their inadequacies. At present, only fudges—partly satisfactory work-arounds—are possible to deal with this problem.

Authentication technologies

Authentication technologies were developed by information scientists to provide a reliable basis for sending verifiable messages over networks. These technologies are based on the mathematical routines of cryptography, but are designed to work with clear-text messages. (The subtle forms of meaning-bearing presentation, discussed above, are not normally relevant here.) The goal of such technologies is not to obscure the information contained in the message, but to verify that it was sent by the person claiming to have sent it and has not been altered in the course of transmission. Meeting these requirements has allowed the development of e-commerce with such services as Internet banking.

These services require a large amount of infrastructure to support them. Changes deemed necessary to the authentication protocols and procedures must be carried out quickly because of the potential risk of criminal exploitation of a weakness. While financial institutions have the money to pay for these high maintenance costs, such resources are not available to an academic community interested in authenticating their electronic editions. 1 Authenticated financial transactions over the Internet have a lifetime of minutes if not seconds, whereas full-scale electronic editions must have a life of decades if they are to justify the investment of the editor's time and energy. The chance of an electronic edition becoming unusable because of the obsolescence of its authentication system rules out the use of proprietary and invasive solutions. 2

Fortunately, the authentication requirements for electronic editions are not as exacting as those for e-commerce where it is a requirement that the creator of a message be verifiable. In electronic editions, the detection of textual corruption is the primary concern. An authentication system must protect the reliability of the encoded text, by indicating if and where a file has been corrupted, thus allowing it to be replaced from a trusted master-copy.

The best authentication method is bit-by-bit comparison of the working copy of the file against a locked master-copy. Some electronic editions at present provide their master files on non-volatile media; working files are always generated afresh from the master files. Unfortunately, this solution is very weak for long-term storage as the master files are bound to a particular storage technology. And the system does not allow for the possibility of revised or additional interpretative markup.

Most authentication methods involve the use of hashing algorithms. 3 In its simplest form, a hashing algorithm steps through the characters of a piece of text using a mathematical formula to calculate a hash value that is dependent on the sequence of the characters in the text. The nature of the mathematical formula used in the algorithm means that the resulting hash value is highly representative of the text because small changes in the text produce large changes in the calculated value. Authentication is achieved by comparing the stored hash value of the master copy with the calculated hash value of a working file. If they are the same, it is extremely likely that the two files are identical. This technique prevents from going undetected corruptions of a file that are otherwise easy to overlook.

Stand-off markup and authentication

‘I want to discuss what I consider one of the worst mistakes of the current software world, embedded markup; which is, regrettably, the heart of such current standards as SGML and HTML’ (Nelson 1).

The problem of maintaining the authenticity of a text file across platforms is not a trivial one. In addition, it is desirable to prevent the proliferation of different versions of a text that would otherwise be brought about by (future) developments in, or additions to, markup, annotation, and cross-referencing. The use of stand-off markup, within an electronic-text environment possessing strong authentication characteristics, potentially allows these desiderata to be met.

To illustrate how such authentication might be achieved, let us take the case of a literary work extant in several typesettings. After the base transcription file of each typesetting was prepared, each such file would be a lexical transcription of the original, but only minimally marked up—since the editor's interpretative responsibilities could be reported within stand-off markup files. The verbal text in the base file would need to be contained within uniquely identifiable text elements. This could be done at the level of the paragraph in the case of prose, or at the level of, say, the line in the case of verse. The identifiers would need to be inserted in the text to act as markers, and the text proofed against the original. After proofing, the file's authenticity could be maintained by an authentication mechanism based on a simple hashing algorithm. Ideally, authentication would be done at the text-element level, so that a change to even a single character would be immediately discernible when the hash value for the text element was checked. Authentication at the text-element level would allow possible corruptions to be quarantined while leaving the rest of the text useable.

Once the base transcription file had been prepared and proofed, markup (e.g. SGML using the TEI DTD 4 ) would be inserted, its operation tested and then removed into a separate, stand-off file. Stand-off files would also store the hash value of the text element to which the markup could be validly applied. The result of this structuring is that the tags would carry a test of the authenticity of that portion of the text, and any attempt to reapply the tags to a corrupted version of the text element would result in a notified error.

A model developed along lines such as these would offer a number of advantages. First, by supporting the standard TEI-compliant SGML it could be used within an SGML environment giving access to all the available browsers and tools. However, the base transcription file would not be dependent on SGML and the separate markup files could be easily manipulated to comply with whatever markup schemes were required. 5 Second, it would enable the text to be annotated or augmented with analytical markup, in parallel and continuously, while still retaining its textual integrity. Third, the levels of markup could be developed independently for different purposes and applied selectively to meet different user-requirements. This would provide a means of future-proofing the edition against the obsolescence brought about by subjective markup since any edition deemed unsuitable for a particular application is liable to spawn a text competitor that would ultimately vie with the original one for maintenance resources.

To date, only one implementation of the proposed model has been developed for electronic editions: the Just In Time Markup (JITM) system. It has utilities for inserting tags, subsequently removing them, and running the verification process. The embedding into the base file of the markup from the stand-off files creates a virtual document—a ‘perspective’—which is inserted into a template conforming to the appropriate Document-Type Definition. 6 Because any markup added to the base file is extracted into stand-off markup files, and the base file authenticated ‘just in time’ when a call is made to create the new perspective that incorporates the added markup, an automatic proofreading of the base file is in effect being continually carried out. 7 This has the potential to significantly reduce the time involved in the creation of an electronic edition while maintaining the academic rigor required for such a project. The same authentication system continues to ensure the reliability of the edition after publication; and the same textual resources do not need to be newly transcribed or re-proofread for each new editorial or other study.

There are further advantages. First, in the original creation of the base transcription file, proofing can, if desired, be simplified by separate checking of the markup on the one hand and the words and punctuation on the other. Second, different or conflicting structural markups can be applied to the same base file because they are in different files and can be applied to the base file selectively. Finally, because the JITM system separates the transcriptions from the markup, the question of copyright is simplified. Since the markup is interpretative (as explained above, and more obviously in the case of added explanatory and textual notes), a copyright in it can be clearly identified and defended. In all of this, the base transcription file remains as simple as possible (thereby greatly easing its portability into future systems) and the authentication mechanism remains non-invasive. JITM is, in other words, an open rather than a proprietary system. 8


Ensuring the continuing reliability of electronic editions is a bigger issue than for print editions. The creator's responsibility to the users of the edition does not end with its publication, and steps must be taken to ensure that it is protected against corruption by the very processes and medium that gave it life. Authentication technologies can provide the required reliability, but must be applied in such a way that they do not affect the long-term availability and reliability of the edition through obsolescence.

The use of stand-off mark up and abstracted authentication techniques potentially allows editions to have their markup revised, reinterpreted or enhanced, and their protection mechanisms easily upgraded or replaced, when future developments require it. This will be able to be done without compromising the base transcription files or wasting the editorial labor that has gone into establishing them.

The Canadian National Archives has decided to archive only clear text in their born-digital archives (Brodie). The extra costs involved in archiving the authentication technologies necessary to authenticate the original, cryptographically secure files is considered too great a burden.
These authentication solutions are largely based on the idea of the digital signature where the file to be authenticated has attached to it a cryptographic signature calculated from the contents of the file and a unique private key registered to the owner. The user of the file uses a public key provided by the message originator to authenticate the file, and the correspondence between the public key and the private key guarantees that the file was sent by the owner of the private key. The infrastructure involved in this system involves the calculation, registration and distribution of the authentication keys. Currently these keys are unique prime numbers, at least one hundred digits in length. The distribution of these keys is handled by sophisticated servers that are expensive to maintain.
E.g. the National Library of Australia records in its on-line catalogue a Message Digest version 5 (MD5) hash value for its digital assets in the Picture Australia service so that the authenticity of the files downloaded by users can be checked.
The stand-off markup paradigm would readily support the use of other, normally embedded markup systems in parallel if this were a requirement.
When writing for this chapter began the P4 version of the TEI DTD had just been released. Now, the technology has progressed such that XML is the requisite language and the P5 version of the TEI DTD is almost upon us. Trusting in the stability of embedded markup for long-lived etexts is short-sighted at best.
While the base transcription file does not in itself adequately represent a historical state of the work being edited, the default perspective in JITM for new users is the one that records the physical presentation of the original. More experienced users, and scholars seeking to interpret the base file or turn it to new purposes, can work with the base file.
Each tag markup instruction incorporates a hash value for the text element into which it is to be inserted. It is the comparison of this stored value against the value calculated for the text element which provides the automatic proofreading of the JITM system.
The algorithms for the Just In Time Markup system and the hashing algorithm it uses are to be made public in due course. For papers about the project (which remains in development as of late 2002), go to the Australian Scholarly Editions Centre webpage http://www.unsw.adfa.edu.au/ASEC/JITM/publications.html ; and for the development website itself go to http://jitm.its.adfa.edu.au/JITM/HNL/JITM_frames.html . The trial used files created for the (print) critical edition of Marcus Clarke's novel of the 1870s, His Natural Life, and a separate testing was done by Peter Shillingsburg. Just In Time Markup is copyright 2002 Graham Barwell, Phill Berrie, Paul Eggert and Chris Tiffin.

Last recorded change to this page: 2007-10-31  •  For corrections or updates, contact webmaster AT tei-c DOT org