When texts are considered random combinations of strings, this can entertain interesting predictions about the completion of the complete works of William Shakespeare by a(n army of) hypothetical monkey(s) randomly hitting the keys on a typewriter (see the infinite monkey theorem). Indeed, to a computer, the textual universe is a library of Babel, in which every possible text is as likely (or rather unlikely) to exist as any other text. From this perspective, perfect duplicates may perfectly co-exist with gazillions of close approximations and texts that have nothing in common at all. In an intellectual, human context, where characters are ordered along (arbitrary) rules, two sensible texts will most likely have nothing in common at all, rather than being related in any way. In fact, the chances of identical texts can almost be ruled out to either perfect mechanical photocopies, or blatant cases of plagiarism, and as such be considered uninteresting from an intellectual point of view. However, especially in the context of greatly valued literary, cultural, and/or historical works, the odd chance that such a text has a closely resembling counterpart becomes quite interesting. It may at least indicate some kind of relationship between both texts, even provide insight in its transmission through time, shed light on its history and conception, perhaps tell us something about the creative process of its author, or by extension provide insights in The Creative Process in the working of the human mind. These domains of knowledge inform different kinds of theories of textual criticism, each with their own research interests, principles and practises. What they all have in common, however, is an attempt to represent related texts found in different physical witnesses as different versions of the same abstract work.

As we have seen already, in order to make this world of meaning accessible to/via computers, text encoding with TEI provides a sensible approach. Moreover, besides the general provisions for text encoding, the TEI Guidelines define a range of specific elements and mechanisms to represent textual variation in a sensible way for further analysis. The TEI Guidelines devote a complete chapter, 12. Critical Apparatus, to the documentation of specific elements that are grouped into the textcrit module for the encoding of textual variation. In order to use the elements covered in this tutorial module, you are thus required to include the dedicated textcrit TEI module in your TEI schema.

Similar to all other TEI modules, the elements and attributes defined in the TEI textcrit module can be used for the encoding of existing source materials (be they in print or digital form), or the encoding of electronic documents from scratch. However, the use of this module in the context of electronic critical editing, adds another perspective to this traditional authorial/editorial angle (see [1]). Electronic critical editions can be created from scratch either by encoding different primary source materials straight as a critical edition, or by generating the edition from previously encoded electronic transcriptions of those materials as independent texts in their own right. Therefore, the tags defined in the TEI textcrit module can be used to:

digitise	an existing print edition
create	an electronic edition, e.g. by recording some or all of the known variations among different witnesses to the text in a critical apparatus of variants
generate	an electronic edition from encoded transcriptions of the documentary source material

In the examples in this TBE module, critical editing with TEI will be understood as the act of encoding material sources in a TEI representation that allows for the creation or generation of a digital edition in some form (using any output format in the digital medium, e.g. HTML pages, PDF, flash movies,...), rather than digitising an existing critical edition. In this sense, the authorial/editorial angle of this TBE module differs from that of the other modules (focusing on the digitisation of a material source text in a certain genre). However, the strategies discussed in this tutorial for representing textual variation can equally be applied to the digitisation of existing critical editions. Where there are differences, these will be pointed out explicitly.

For example, consider following texts:

Some of these images may look more or less familiar to you: they are facsimiles from the first page of chapter 2 of the printed TEI Guidelines throughout their different incarnations, from version P2 (1992) to the latest version, P5 (2009). As you can imagine, the technological evolutions of these 17 years have prompted considerable changes to this chapter that introduces the technological background of text encoding with TEI, ranging from rephrasing, addition or deletion of notes, changes in italicisation, restructuring of paragraphs, etc. One way of approaching this textual variation could consist of encoding these text versions as physically distinct TEI documents, in which corresponding text structures could be aligned by a common identification mechanism. For example, the first couple of paragraphs in these 4 text witnesses could be encoded in different TEI documents as follows:

This would allow for maximal representation of the distinct material sources, and leave the identification of the actual variation either to further processing or human inspection. A variant of this approach could integrate the transcriptions of the text in all material witnesses in a single TEI document, and make use of appropriate linking attributes to point out the alignment between the different text structures. In their naivety, such systems are both redundant and crude. While providing all text of all text witnesses, and aligning the corresponding text structures, they provide little insight in the places where the different witnesses actually differ.

In order to encode the actual textual variation between the different text versions in a meaningful way, the TEI Guidelines provide a specialised module of elements and attributes that allow you to encode textual variation at word level. This TBE tutorial will first discuss how to describe the different text witnesses represented in the critical edition; next deal with the encoding of textual variants between these witnesses in isolation; then treat different ways of integrating such records of variation within the encoding of the critical edition; and finally point out potential problems and pitfalls when creating a critical edition with TEI.

When creating, generating or digitising a critical edition, it is of crucial importance to document the text witnesses whose transcriptions it contains. This can be done in a <listWit> (list of witnesses) element, which can be put either in the <sourceDesc> section of the TEI header (when creating or generating a critical edition), or a division (<div>) of the edition's text, usually in the <front> section (when digitising an existing critical edition). The <listWit> element should describe each text witness in its own <witness> element. This element can contain a prose description of the witness in one or more paragraphs (<p>), or using a more specialised element for bibliographic description (<bibl>, <biblStruct>, or <biblFull>). The witness definitions should provide a unique identification code in the @xml:id attribute. This code is used as a sigil in the critical edition, in order to connect the textual variants with the respective witnesses in which they occur (see 4. Encoding Textual Variants). For example, the witness list for our critical edition of the TEI Guidelines could look as follows:

Such bibliographic descriptions of course are easier for printed works than for manuscripts; for the latter type of witnesses, some kind of description inside <listWit> is advised, preferably with a pointer (using <ptr/> or <ref>) to a full description of the manuscript inside <msDescription>. Since this element is not discussed in these TBE tutorials, you are referred to TEI Guidelines, 10.2 The Manuscript Description Element for a discussion of this element, and TEI Guidelines, 12.1.4.3 The Witness List for examples of describing manuscript witnesses in a digital edition.

In a critical edition, it may make sense to discern groups of witnesses, that have many text variants in common in comparison to other witnesses, and can often be conveniently summarised in one sigil. In the witness list, witnesses can be grouped by wrapping their <witness> descriptions in nesting <listWit> structures. The common sigil then can be provided as the value for an @xml:id attribute of the group's <listWit> element. The nested witness groups can be labelled with a <head> element. For example, in our sample text witnesses it may make sense to discern those versions of the TEI Guidelines dealing with SGML, and those dealing with XML. This could look as follows:

After this discussion of encoding textual variants themselves, it is time to have a look at the bigger picture: how do you integrate these variants into an electronic critical edition? The TEI Guidelines provide 3 different mechanisms for integrating apparatus entries in the encoding of texts (don't let the names intimidate you):

In this overview, the [I] and [E] labels indicate where an apparatus encoded with that method can be physically located with regards to the transcription of the (base) text it is linked to:

• [E]: external apparatus: the apparatus is located outside the transcription of a base text, either in some other part of the TEI document containing the transcription, or in a physically distinct document
  → location-referenced, double-end-point-attached
• [I]: internal apparatus: each apparatus entry is located inline in the transcription of a (base) text, at the place where the variant occurs
  → location-referenced, double-end-point-attached, parallel segmentation

The method chosen and the physical location of the apparatus must be encoded in the TEI Header, in the <variantEncoding/> element inside the <encodingDesc> section. This is an empty element with two mandatory attributes (see TBE Module 2: The TEI Header -- The Variant Encoding):

@method	: indicates the method of linking the critical apparatus to the text: either location-referenced, double-end-point, or parallel-segmentation.
@location	: indicates the location of the critical apparatus with regards to the text: either external or internal.

While the <app> element provides a powerful and efficient means for representing textual variation, some caveats must be pointed out. First off, the content model for the <lem> and <rdg> elements is limited to phrase-level elements. This may pose a problem for variants that involve larger structural units, like the addition or deletion of a paragraph. For example, if we consider the first block of text in the #p4 version:

 As originally published in previous editions of the Guidelines, this
 chapter provided a gentle introduction to 'just enough' SGML for anyone to
 understand how the TEI used that standard. Since then, the Gentle Guide
 seems to have taken on a life of its own independent of the Guidelines,
 having been widely distributed (and flatteringly imitated) on the web. In
 revising it for the present draft, the editors have therefore felt free to
 reduce considerably its discussion of SGML-specific matters, in favour of a
 simple presentation of how the TEI uses XML. 

All other versions start with the phrase The encoding scheme defined by these Guidelines, which is the start of the second paragraph in version #p4 as well. Therefore, this block of text can be considered an addition compared to the earlier versions, while it was deleted in version #p5. Intuitively, one might wish to encode this as follows:

However, <p> being a member of a 'chunk-level' model class of elements, it is not allowed as contents of <rdg>. There are two ways of solving such problems:

• changing the encoding: if the content allows it, you can look for alternative ways to encode the contents (without resorting to tag abuse, however!)
• changing the TEI scheme: by redefining the content model of <rdg>, you can make sure that the encoding validates in a TEI-conformant way. For details on how to customise a TEI scheme, see TBE module 8. Customising TEI, ODD, Roma.

In this example, the contents of the variant text block permits an interpretation as a note that could be characterised as a disclaimer. In TEI, the <note> element is a member of the global model class, and thus may occur inside <rdg>. A valid alternative to the previous encoding could be:

A harder problem, however, occurs when the variation occurs on a structural level. For example, if you look closely at the facsimiles of the TEI Guidelines above, you'll notice that there is a paragraph shift at the sentence starting with Historically, the word markup has been used :

• in the #p2 and #p3 versions, this sentence starts the third paragraph
• in the #p4 and #p5 versions, this sentence is part of the second paragraph

This poses a harder problem to the representation, as it involves markup itself (i.e. the end and start tag of the third paragraph are the subject of variation). As XML requires proper nesting, this is a problem in any XML representation of this kind of structural variation. Again, two strategies could be followed (none of which is ideal, however):

• encode structural variants as variant structures. However, this may obscure their alignment.
• encode structural variants using milestone elements instead of full-blown XML structures. However, depending on your view on the texts, this could be considered less orthodox from an encoding point of view, as it implies some notion of a base text that determines the encoding of the others.

The first option would compare the individual transcriptions of these text witnesses, some of which spread more or less the same textual contents over 3 paragraphs, while others use only 2 paragraphs. In a parallel segmented apparatus, this might look as follows:

This approach treats the shifting paragraph as a variant in its own right, that is present in some witnesses (#p2 and #p3), while absent in the others (#p4 and #p5). The second apparatus entry then omits the text of #p2 and #p3, while including the (corresponding) text of #p4 and #p5. However, as this example illustrates, the alignment of the corresponding text fragments between both groups of witnesses (those starting a new paragraph and those that don't) is lost: there is no way of telling how the phrases SGML is an international standard [...] .More exactly, SGML [...] (in #p2 and #p3) and XML is an extensible markup language [...]. More exactly, XML [...] correspond. This kind of encoding could be less problematic when generating an electronic critical edition (in which case the more complicated apparatus encoding could be generated by an automatic collation routine). When creating an electronic edition, the construction of such a more complex apparatus entry could be less desirable.

The other solution would be to encode the paragraph break in the #p2 and #p3 versions using an empty milestone marker: an empty element that indicates some kind of structural boundary in the text where it occurs, as in this parallel segmented example:

Since the milestone paragraph boundary marker (<milestone type="p"/>) removes the intrusive XML boundaries, this allows us to compare the text between all versions. However, this implies that the encoding of the third paragraph in the #p2 and #p3 versions is suppressed, in contrast to the other paragraphs in these text versions. This could be less a problem when creating an electronic critical edition, rather than generating one. In the latter case, the milestone encoding would reflect a dependency on a base text (that does not have the paragraph break). Moreover, it presupposes some kind of structural alignment prior to the encoding of the individual texts.

This tutorial module has focused on the encoding of textual variation in different text witnesses. Although the determination of textual variation itself can depend on the editorial theories for the critical edition, and the TEI Guidelines offer many possibilities to encode textual variation, we'll conclude with a possible encoding as a critical edition of the text samples we used in this tutorial module. In this example, we chose for a parallel segmented internal apparatus, which could look as follows: