An Encoding Model for Genetic Editions

Work Plan

The planned evolution of this document and the encoding model it describes may be summarized as follows:

• Initially, we worked on a pseudo-encoding model, expressed using our own terminology. This stage was completed by July 2009.
• This pseudo-encoding model was re-expressed using the TEI P5 encoding framework and revised extensively.
• The same step-by-step approach will be taken for the standardisation process:
  • First we develop an application profile, that is, a well-documented TEI customisation.
  • This will be presented to the TEI Council as a proposed modification of the TEI
  • If the TEI Council approves it, the Council will decide how it should be integrated into the current Guidelines.

This draft document is publicly available for discussion and feedback from the community. The document source is maintained in the TEI subversion repository at http://tei.svn.sourceforge.net/viewvc/tei/trunk/genetic/ ; background information about the development of the proposals and associated materials, is hosted on the TEI Wiki at http://wiki.tei-c.org/index.php/Category:Genetic_Editions .

Conventions used

Although the entire document is a draft and therefore susceptible of changes, some sections are less stable than others. In particular, when a section or a particular element requires further discussion or is considered an open problem, such a section or element is marked by a * mark.

As required for TEI conformance, non-TEI elements are defined in a distinct non-TEI namespace. In the usage examples and throughout this document that namespace is mapped to the prefix ge:, while TEI elements are not marked by any namespace prefix.

1.1 Fact vs. Interpretation

In German editorial theory there is a well known opposition between what is there in the source document, the record (Befund), and the interpretation of this phenomenon (Deutung). This opposition implies that there is a way to talk about the record without any interpretation. Yet at some possibly simplistic level, everything we say about a text is based on interpretation, particularly in the realm of genetic criticism. ¹ At the same time, there is an obvious difference between the interpretation that some trace of ink is indeed a specific letter and the assumption that a change in one line of a manuscript must have been made at the same time as a change in another line because their effects are textually related (for example, the first change was to a rhyming word, which necessitated the second change). Therefore we propose to talk about differing levels of interpretation, thus differentiating between ‘what’s there’ (document/fact) and ‘how does it relate’ (text/interpretation).

1.2 Status vs. Process

When a scholar examines a written text, especially a manuscript, the object of the investigation is usually to discover the final result of that writing and rewriting process. The examination can take various forms: it can be approached from a codicological or documentary point of view, ‘photographing’ the resulting product, from a textual point of view, or from a genetic point of view, by trying to describe the flow of authoring.

The present proposal addresses both approaches: sections 3.1 Transcription of a document and 3.2 Textual Alterations are presented from the codicological or documentary point of view, while the other sections investigate the process of writing and re-writing of the text, thus constituting the purely genetic parts of the present proposal.

1.3 Document vs. Text

In Manuscript Studies (Editing, Codicology, Palaeography, Art History, History) the first level of enquiry is always the document, the physical support that lies in front of the scholar’s eyes.

To understand the text that is contained in the manuscript, a deep study of the manuscript itself is fundamental: the layout, the type of script, the type of support, the binding, and many other aspects are able to tell us about when, where, and why this particular text was composed. The text however represents a different level of enquiry: it is a construct, derived from the reading of the documents.

In the case of modern draft manuscripts scholars must give detailed consideration to the layout, the different stratifications of writing and the disposition of these in the physical space; all of these, together with an understanding of the text, are required to gain insight about the composition, time of revisions, and flow (flux) of the text. Furthermore, in some cases, we know that the kind of physical support used to record it not only influences but may also actually determine the text itself. For instance, the content and the length of letters are often determined by the size and quantity of the paper available to the writer; even more so for items such as postcards.

The TEI has traditionally prioritised the text level. Of the two possible views available to someone transcribing a primary source (text and document), the TEI privileges the text (hence Text Encoding Initiative). Such physical or topographical information as a typical TEI encoding provides is subordinate to the main structural encoding, whether because it is represented by empty elements (<pb/>, <lb/>, <cb/>) or attributes (<add place="">, <note place="">, or rend). The TEI thus reflects the not uncommon view that, while relevant, documents are somehow less relevant than the texts they embody; to use a bibliographical metaphor, texts are ‘substantial’ while documents are ‘accidental’.

However, for genetic editions, a focus on the document is crucial. In many cases, the only way to reconstruct the process of writing and re-writing which leads to a new text is to examine a specific document. We therefore propose to complement the existing text-focussed approach with a new encoding scheme focussed instead on the document.

We should then clarify the way we will use the following words:

• Document the physical object, the manuscript or other primary source, comprising one or more written surfaces
• Text the intellectual content of the document, the words written in their logical (non physical) sequence
• Facsimile a surrogate representation (not necessarily digital) of the document (a photograph, a microfilm, a digital image)
• Document level transcription of the content of a document according to its physical layout
• Text level transcription of the content of a document according to its logical, semantic meaning

1.4 Writing Acts vs. Text Stages

As noted in the previous section, the focus on purely documentary aspects in a genetic edition often serves to reconstruct the writing process, which in turn enables allow the editor to justify a genetic analysis, for example to identify which text passage has been altered first or which textual variant predates another. The encoding of the writing process, it must be observed, is not simply a different point of view about the text in hand, but rather regards as a relevant a set of information fundamentally different from that captured by a purely textual encoding.

When talking about alterations on a textual level, it is the stages of the text which vare of major interest. These stages are the results of various alterations applied to the draft; each addition, deletion, substitution etc. is viewed from the perspective of the possibly new text state it yields. For example the following substitution is well-defined from this perspective:

This example stresses the writing process and provides two additional pieces of information: firstly the fact that the author did not strike through the whole of the first word and secondly that a change in writing material happened just before the word scenery was written. In current encoding practice, which tends to favour the textual view, the two perspectives are often integrated, subordinating aspects of the writing process to their textual outcome or regularizing them so they do not interfere with a text-driven evaluation of the encoding.

As the example shows, this integrative approach can be a viable one in many cases, but taking this pragmatic shortcut blurs a specific editorial distinction, maybe exactly the distinction one wants to emphasize (e.g. to separate Befund from Deutung). Furthermore, it makes it particularly hard to be as precise about the documentation of ‘what’s there’ as one is about the interpretation of ‘what’s the possible outcome’. Ideally in a genetic edition the editor would prefer to encode the two perspectives separately and align them later on instead of trying to integrate them right away, and thereby potentially neglect information of importance in their genetic analysis.

3.1 Transcription of a document

Like a facsimile, a <ge:document> contains information about the written surfaces constituting a document. Because of this similarity, we would like to use the same elements (surface and zone) as proposed in the existing TEI scheme, although these place limits on what can be described. Specifically, the zone element as currently defined can represent only a rectangular area; it also lacks any way of stating the baseline applicable to any writing contained within it

The size of the writing surface is defined by a set of cartesian coordinates measured from the top left corner. The co-ordinates of all zones identified within the writing surface are given in terms of the same co-ordinates, as further discussed in the TEI proposals for facsimile. It will often be the case that explicit dimensions for a manuscript page (expressed in mm for example) are also supplied in a msDesc element in the TEI Header, but this is not a requirement; in particular there is no assumption that the co-ordinate system defined by a surface maps to any particular external dimensions, nor that the co-ordinate systems of different documents necessarily correspond.

A surface element may contain any number of zone, graphic, <ge:line> or tableelements. The graphic element is used to point to any graphic (non textual) component forming part of the page, in the usual TEI manner. The zone element is used to delimit any contiguous section of writing which the encoder wishes to identify for some purpose.

Zones can be nested and grouped, and can also overlap. Their positioning with respect to the surface element is defined by coordinate values taken from the same co-ordinate system as the surface itself, measured from the top left corner. The element carries a rotate attribute which describes (in degrees) the orientation of the surface with respect to the content (writing, images) in that zone, with respect to its normal orientation. Note that the mechanism aims to describes the process by which the content of a specific zone has been supplied (i.e. the author has physically rotated the writing surface) rather than the orientation of the writing.

Zones are arbitrarily defined by the encoder according to the layout of the writing surface and can make use of a standardised vocabulary (e.g. the top margin).

To overcome the inherent limitations of the existing zone and surface elements, we need to extend their capability to include the definition of arbitrary polygons, using an attribute such as svg:points, from the Standard Vector Graphics (SVG) XML namespace. This would also provide a way of defining a baseline for the writing. This work is not yet fully elaborated.

The attribute stage is used to indicate the stage in a writing campaign to which this zone has been assigned by the encoder, as further discussed in 6.1 Stages/ Revision campaigns below.

Within a zone, individual lines of writing are usually, but not necessarily, distinguished using the <ge:line>. Zones can also include table elements, as lines and tables represent the principal ways of organising texts on a surface. In the case of an interlineated text, interlinear writing can be treated as a line on its own (perhaps characterised by a type attribute) or as textual addition, encoded with an add (see below 3.2 Textual Alterations).

In the following imaginary example, there are two main areas of writing, the diary entry (black ink) and another (supposedly later) annotation in blue ink.

Note that the diary entry forms a zone which itself contains two zones: one containing the date, and the other containing three lines about birds. The five lines of annotation in blue ink form another zone, to record which the diary page has been rotated 90° clockwise by the author. Using the elements discussed so far, the page might be transcribed as follows:

Is it possible to combine both perspectives (documentary and textual views) within a single encoding? In general a document-based transcription, which is done page-by-page and possibly line-by-line, is almost certain to overlap with some part of a the text-based structure. The cleanest solution may be to encode both structures separately, providing both a <ge:document> and a distinct text solution, perhaps using some form of external pointing to link the two, and minimizing redundancy of encoding by using XInclude. This option is further discussed below and also in the TEI Guidelines.

The written surfaces of which a document is composed are not always homogenous. In the following example, taken from the Walt Whitman archive, two pieces of newsprint have been glued to a piece of blue paper on which a poem is being drafted:

Figure 1. Image from http://www.whitmanarchive.org/resources/sleepers/duk.00258.001.jpg

The two pieces of newsprint might perhaps be regarded as special kinds of zone, but they are effectively new surfaces, since they might contain additional written zones themselves (such as the numbers in this case). We therefore propose a distinct element, <ge:patch>, which can appear within a surface, and behaves effectively like one, except that it contains specialized attributes to provide additional information.

• patch contains a part of a written surface which was originally physically distinct but became attached to it at the time that one or more written zones were created.

  binder	Describe the method by which a patch is or was connected to the main surface
  type	characterizes the element in some sense, using any convenient classification scheme or typology.
  height	height of the patch in mm
  width	width of the patch in mm

Using this element, the Whitman draft above might be encoded as follows:

<surface>
 <zone>
  <ge:line>Poem</ge:line>
  <ge:line>As in Visions of — at</ge:line>
  <ge:line>night —</ge:line>
  <ge:line>All sorts of fancies running through</ge:line>
  <ge:line>the head</ge:line>
 </zone>
 <ge:patch
   type="newsprint"
   binder="glue"
   height="40"
   width="90"> Spring has
   just set in here, and the weather.... a steamer <zone>
   <ge:metaMark function="sequence">2</ge:metaMark>
  </zone></ge:patch>
 <ge:patch
   type="newsprint"
   binder="glue"
   height="35"
   width="90"> "The shores
   on either side of the Sound are... The In- <zone>
   <ge:metaMark function="sequence">3</ge:metaMark>
  </zone></ge:patch>
</surface>

The <ge:metaMark> element used in this example is further discussed below ( 3.2.3 Metamarks)

3.2 Textual Alterations

Traces of authorial alteration (correction, addition, deletion, etc.) are frequently found within a single document, and may also be inferred when different documents are compared. It is however an open question as to whether inter-document discrepancies at the dossier level should be regarded in the same way as intra-document alterations. If two witnesses are collated, we may observe that a word present in one is missing from the other: does it necessarily follow that this is an addition or a deletion, which we would not hesitate to mark with an add or del tag if we are transcribing a single manuscript? We return to this question below.

Most of these elements imply a certain level of semantic interpretation; for instance the usage of the add element to encode, say, interlinear insertions involves a decision that the interlinear text has been deliberately inserted rather than simply misplaced. As discussed above (see 1.4 Writing Acts vs. Text Stages), the use of these elements when transcribing from a documentary perspective is a pragmatic shortcut. In cases where it is felt desirable to keep the recording of ‘what is on the page’ entirely separate from ‘what is the editor’s interpretation’ (see 1.1 Fact vs. Interpretation), we provide two generic elements, <ge:mod> and <ge:modSpan>, which can be used to record any kind of modification identified in the document. These elements can be categorised by means of their type attribute, and visual aspects of their appearance can be described by means of the rend attribute, but they provide no further interpretation of the function or intention of the passage so marked up.

Whether such a modification, for instance a struck-out passage, is to be interpreted as a deletion or as some other phenomenon (e.g. as being already used) should be be expressed using the other, more semantically motivated, elements, which function at a different level of description.

3.2.1 Additions, fixations and clarifications

A writer may sometimes rewrite material a second time without significant change and in the same place. We consider this a distinct activity from addition as usually defined because no new textual material results but the status of existing material changes. We distinguish two variants of this: fixation where the first version was a tentative draft which is subsequently fixed, for example by inking it over; and clarification, where the first version was badly written and has been rewritten for clarity. The element <ge:rewrite> is provided to cover both cases.

In this simple example, taken from the papers of Henrik Ibsen, the writer wrote the word skuldren hastily, and then returned to it to make the letter l larger and clearer:

Figure 2. Image from a ms of Peer Gynt, Collin 2869, 4°, I.1.1, the Royal Library of Copenhagen

We might transcribe this word as follows:

<ge:line>... Sku<ge:rewrite cause="unclear">l</ge:rewrite>dren </ge:line>

The following example, taken from a manuscript of Jane Austen's Sanditon, shows a rewriting where a pencilled passage has been fixed with ink, with some modification:

Figure 3. Image from page 70 of the Sanditon manuscript

In this example, Austen sees in the fixation an opportunity to manipulate the text previously written, and thus changes the pencilled could but get a young to the inked could get a young, writing the inky get on top of pencilled but and striking over with black ink the pencilled but. A simple way of encoding this might be as follows:

<ge:rewrite cause="fix" hand="#ja2" stage="#s1">Now, if we could <subst stage="#s1">
  <add>get</add>
  <del>but</del>
 </subst>
 <del stage="#s1" rend="overstrike">get</del> a young Heiress</ge:rewrite>

where the stage attribute groups the operations that belong to the rewriting stage, assuming an implied previous stage (the first layer of writing). If the first layer need to be addressed independently, another pointer can be added to the stage attribute:

<ge:line stage="#s0">
 <ge:rewrite cause="fix" hand="#ja2" stage="#s1">Now...</ge:rewrite></ge:line>

where #s0 and #s1 both point to the declaration of such stages in the header, the first one conventionally representing the first layer of writing and the second the rewriting stage.

A single rewrite may not be sufficient, and it may be that the document becomes almost unreadable as a result of repeated clarification. In the following example, we can distinguish at least two attempts to write the letters er in the word bægerklang:

Figure 4. Image from http://www.emunch.no/tei-mm-2008/ms.html

We might encode this by nesting the rewrite element as follows:

<ge:line>ved Bæg<ge:rewrite cause="unclear" stage="#stage2">
  <ge:rewrite cause="unclear" stage="#stage1">er</ge:rewrite></ge:rewrite> ...</ge:line>

The stage attribute used here is discussed further below ( 6.1 Stages/ Revision campaigns).

Metamarks and other markup-like strokes can also be inked over with the same purpose as the fixation or clarification of text passages. For instance, in a draft version of Goethe’s Faust, a passage was struck through once in pencil during one revision and then again with ink during a later revision, supposedly to fixate the deletion.

Figure 5. Fixation of a deletion in Goethe’s Faust

We propose an element redo (the opposite of undo, see 3.2.7 Undoing alterations), which can be used to encode this process as follows:

<ge:line>
 <ge:redo
   xml:id="redo_3"
   hand="#g_t"
   target="#mod_1"
   cause="fix"/>
 <ge:modSpan
   xml:id="mod_1"
   rend="strikethrough"
   spanTo="#anchor_1"
   hand="#g_bl"/>Ihr hagren, triſten, krummgezog<ge:mod rend="strikethrough">nen</ge:mod>ener Nacken</ge:line>
<ge:line>Wenn ihr nur piepſet iſt die Welt ſchon matt.<anchor xml:id="anchor_1"/></ge:line>

3.2.2 Deletions and marked as used

In general, deletion in a source is marked using the del or delSpan element. However, it is useful to distinguish cases where a passage has been ‘indicated as superfluous or spurious in the copy text by an author, scribe, annotator, or corrector’ (TEI P5, s.v. del) from cases where a passage has been struck through or otherwise marked as having been used or copied to another location. In this latter case, the author does not intend to suppress the content, but only to mark that it has been transferred or reused. The element <ge:used> is provided to mark this kind of ‘deletion’.

The following page from the Walt Whitman archive has been crossed through to indicate used material:

Figure 6. Page from http://www.whitmanarchive.org/resources/sleepers/20051105_0650.jpg

This page contains many internal deletions, but these should be distinguished from the ‘deletion’ signalled by the large cross, which actually shows that the page has been transferred or re-used, not deleted.

Material marked as re-used in this way often spans more than one zone or line. For that reason, the <ge:used> element is a spanning element, indicating the end of the used area by means of a spanTo attribute. We might encode the above page as follows:

<surface>
 <ge:used rend="cross" spanTo="#X2"/>
 <zone>
  <ge:line rend="underline">The Poet</ge:line>
  <ge:line>
   <del rend="strikethrough">I think</del> His sight is
     the</ge:line>
  <ge:line> sight of the ? and</ge:line>
  <ge:line>has sent the instinct of the</ge:line>
  <ge:line>? dog</ge:line>
 </zone>
 <zone>
  <ge:line>I think <ge:rewrite>ten</ge:rewrite> million</ge:line>
<!-- ... -->
  <ge:line>well; those <subst>
    <del rend="strikethrough">supple-fingered gods</del>
    <add>journeymen divine.</add>
   </subst></ge:line>
  <anchor xml:id="X2"/>
 </zone>
</surface>

3.2.3 Metamarks

By metamark we mean marks such as numbers, arrows, crosses, or other symbols introduced by the writer into a document expressly for the purpose of indicating how the text is to be read. Such marks thus constitute a kind of markup of the document, rather than forming part of the text.

Unlike marginal notes or other additions to the text, meta-marks indicate a deliberate alteration of the writing (e.g. ‘move this passage over there’). We also consider as metamarks dates introduced to mark the beginning of a manuscript or a revision, but not forming part of it.

The <ge:metaMark> element carries a function attribute which specifies the function of the meta-mark and a target attribute which points to the element or elements concerned.

The following example is taken from Kundige bok 2, a 15th century legal book from the city of Göttingen, containing regulations of everyday life issued by the city council

Figure 7. Malte's example

In the second paragraph, the sentence beginning ‘Ock en schullen de bruwere...’ was first written along with the word lege ("read") in the left hand margin, functioning as a metamark to indicate that this sentence forms part of the regulations. A further sentence was then added, while at some later stage or stages the text and also the metamark were deleted. We might encode this as follows:

<del>
 <ge:metaMark function="flag" target="#s1">lege</ge:metaMark>
 <s xml:id="s1">Ock en schullen de bruwere des hilgen dages nicht over
 <lb/>setten noch uppe den stillen fridach bruwen.</s>
 <add>
  <s>Noch nymande <lb/>over setten, se en sehin denne erst, dat uppe den
     bonen <lb/>neyn stro noch, huw noch flaß ligghe, by pine eyner
     halven <lb/>roden, deme bruwere so wol alse dem bruwheren to
     murende.</s>
 </add>
</del>

Here are some further examples showing the use of this element, taken from the manuscript drafts of Thomas Moore's Lalla Rookh (1817). The first shows a simple use of a metamark used to take stock of the progress of composition:

Figure 8. Lalla Rookh

At regular points throughout the various drafts of the work, a number occurs, usually in the right margin (in this instance, "100"). These numbers result from the author counting the number of verse lines he has composed to the given point, and are not part of the text, but represent a stage at which Moore is taking stock of the progress of his composition.

<surface>
 <zone>
<!-- main zone -->
  <ge:line>Be this she cried &amp; wing’d her flight</ge:line>
  <ge:line>My offering at the Gates of Bliss</ge:line>
  <ge:line>
   <del>Fully to know the odours <gap extent="1" unit="word" reason="illegible"/>
   </del></ge:line>
  <ge:line>
   <del>Tho foul to heaven the vapour went</del></ge:line>
  <ge:line>
   <del>From vulgar</del>
   <add>common</add>
   <del>victors</del>, blood like this</ge:line>
  <ge:line> Shed out for freedom, flows so bright. <ge:metaMark function="count">100</ge:metaMark></ge:line>
  <ge:line> It would not stain the purest <subst>
    <del>fount</del>
    <add>rill</add>
   </subst> .</ge:line>
  <ge:line>“That sparkles thro the fields of light. <del>
    <ge:metaMark function="count">100</ge:metaMark>
   </del></ge:line>
  <ge:line> Behold her in the skies again —</ge:line>
  <ge:line>
   <subst>
    <del>But</del>
    <add>And</add>
   </subst>, tho so fleet her pinions bore</ge:line>
  <ge:line> The spirit of the Warriors slain,</ge:line>
  <ge:line>Now reach’d &amp; pass’d the gates before her</ge:line>
 </zone>
 <zone>
<!-- left zone -->
  <ge:line> Tho foul too oft the</ge:line>
  <ge:line>tears that still</ge:line>
  <ge:line>Tho foul the droppings <add>weepings</add></ge:line>
  <ge:line>that distil</ge:line>
  <ge:line>Tho foul the tears that</ge:line>
  <ge:line>oft distil</ge:line>
  <ge:line>From glory’s faulchion’</ge:line>
 </zone>
</surface>

Figure 9. Lalla Rookh 2

This example demonstrates the use of a common proof-correction mark: in the left margin of the page, adjacent to a group of three cancelled lines of verse, the word "stet" is written. "Stet", a Latin word meaning "let it stand" is commonly used by authors, editors and proofreaders where a previous action should be disregarded. In this instance, Moore is indicating that the three deleted verse lines should be let stand, as evidenced by their appearance in the first printed edition of Lalla Rookh. The word, "stet" does not form part of the text, but rather declares that a certain function be performed upon the text.

<surface>
 <zone>
  <ge:line>
   <gap extent="1" unit="word" reason="illegible"/>
   <del>in his light</del>
   <add>within</add> eyelids, within the spray</ge:line>
  <ge:line>From Eden’s fountain, when it lies</ge:line>
  <ge:line>
   <hi rend="underline">On that blue</hi>
   <del>before that</del> flower, which, Brahmins say</ge:line>
  <ge:line> Can only <add>blooms nowhere but</add> bloom in
     Paradise,</ge:line>
  <ge:line>
   <del xml:id="del1">“Nymph of a bright, fair but erring line!</del></ge:line>
  <ge:line>
   <ge:metaMark function="undo" target="#del1 #del2 #del3">stet</ge:metaMark>
   <del xml:id="del2">(He gently <add>gently</add> he said) one hope
       is thine</del></ge:line>
  <ge:line>
   <del xml:id="del3">One hope (he gently said) is thine</del></ge:line>
 </zone>
</surface>

Other examples of <ge:metaMark> can be seen in marked-up proofs such as the following, taken from the Walt Whitman archive:

Figure 10. http://www.whitmanarchive.org/resources/sleepers/loc.00295.jpg

3.2.4 Alternative Readings

Figure 11. Lalla Rookh 3

In this example two alternative readings are provided, without either one being prioritised or subordinated. While the author apparently first composed the line "Alone before his native river -", at some later point, he entertained the possibility of using the word "beside" instead of "before." In the context of this manuscript, there is no indication of which word the Moore favours, so the status of these words as possible alternative readings needs to be encoded. The evidence of the first edition of Lalla Rookh shows that the word "beside" was chosen, but for the purposes of encoding this manuscript, the facility to encode two equally-possible alternative readings needs to be available.

<zone>
 <ge:line>Alone <seg type="alternative" xml:id="alt1">before</seg>
  <add place="above" type="alternative" xml:id="alt2">beside</add> his
   native river ­—</ge:line>
 <alt targets="#alt1 #alt2" mode="excl" weights="0 1"/>
</zone>

3.2.5 Transpositions

Metamarks are commonly used in the context of transposition, that is, the moving of words or blocks by the author to a different position using arrows, asterisks or numbers or other metamarks. One possible approach (used, for instance in HNML) would be to regard such transpositions as a special kind of substitution, and actually to represent the result of the transposition indicated by the metamarks in the encoding, for example by considering the segment previous to the transposition as deleted, and substituted by the one after the transposition.

Our recommendation is to record the actual state of the witness, but in such a way as to facilitate its reorganization as a distinct processing step. We propose to represent the re-alignment of transposed blocks or segments by means of a stand-off mechanism. The elements <ge:transposeGrp> and <ge:transpose> are provided for this purpose. For example, in the following extract from an Ibsen manuscript

Figure 12. Extracted from http://www.emunch.no/tei-mm-2008/ms.html

, the underlined numbers 1 and 2 indicate that, although the word bör precedes the word hör in the text, the order of the two words should be reversed. We may encode this as follows:

<ge:line>
 <seg xml:id="ib01">bör</seg>
 <ge:metaMark
   rend="underline"
   function="transposition"
   target="#ib1"
   place="above">2.</ge:metaMark>
og <seg xml:id="ib02">hör</seg>
 <ge:metaMark
   rend="underline"
   function="transposition"
   target="#ib02"
   place="above">1.</ge:metaMark></ge:line>
<ge:transposeGrp>
 <ge:transpose>
  <ptr target="#ib02"/>
  <ptr target="#ib01"/></ge:transpose></ge:transposeGrp>

Note the use of the generic seg element to identify the sections of text being transposed. When (as in the following example) the whole of line is to be transposed, there is no need to delimit the sections concerned:

Figure 13. Extracted from http://www.emunch.no/tei-mm-2008/ms3.html

<ge:line xml:id="ib3">
 <ge:metaMark function="transposition" place="margin-left">2.)</ge:metaMark> thi da er du med Himmelen i
Pagt; — </ge:line>
<ge:line xml:id="ib4">
 <ge:metaMark function="transposition" place="margin-left">1.)</ge:metaMark> da kan du Folkets Jøkelhjerter tine;</ge:line>
<ge:transposeGrp>
 <ge:transpose>
  <ptr target="#ib4"/>
  <ptr target="#ib3"/></ge:transpose></ge:transposeGrp>

When transposition is made, the whole element indicated is understood to be moved, not just its contents. In the above example, the metamarks are thus understood to be moved along with the lines to which they apply.

In case the area to be transposed is overlapping with some other kind of markup, the generic milestone can be used instead of seg or any other existing elements.

One or more transposeGrp elements may be supplied either embedded within the text or in the profileDesc of the header, depending on local preference. Each transposeGrp can contain one or more transpose element, each of which defines a single transposition.

3.2.6 Substitution

In the current model for the TEI subst element, one or more additions and deletions may be combined if they are considered as representing a single editorial act, a substitution. Without extension, this model could not therefore include cases such as the following example taken from Thomas Moore's Lalla Rooke

Here the word pondering is deleted, and the phrase she mus'd are added, while the word thus remains unchanged. It seems appropriate to treat all of this as a single substitution. This would require a modification to the content model of subst so as to permit text along with other members of model.pPart.transcriptional, so that this example could be encoded as follows:

<ge:line>While <subst>
  <del>pondering</del> thus <add>she
     mus'd</add>
 </subst>, her pinions fann'd</ge:line>

3.2.7 Undoing alterations

In some cases an author indicates that an alteration is itself to be altered: for example, a struck through passage may be restored via a dotted underlining, or the underlining of a passage may be deleted by a wavy line.

The TEI provides an element restore for one specific kind of alteration to an alteration, namely the undoing of a deletion. We propose a more general element, <ge:undo>. The element <ge:undo> usually encloses the element (e.g. the add, del etc.) to be undone. If it appears within such an element, the implication is that only this part of the parent element has been cancelled. For example, in this passage taken from Giacomo Leopardi's Zibaldone (p. 3595), the phrase si rechi á was underlined word by word, and then the underlining of the word si was cancelled.

This could be encoded as follows:

<ge:line> che e’ <hi rend="underline">
  <ge:undo spanTo="#x2"/>si <anchor xml:id="x2"/> rechi a’</hi>
 <del rend="overstrike">dotti</del>
 <hi rend="underline">denti</hi> l’un d’essi cibi</ge:line>

To make explicit the relation between the undoing act and the initial act, we propose an alternative way of encoding these scribal acts, namely an empty <ge:undo> element, which points to the element the effect of which is being cancelled. If more than one (not coherent) part of the deletion is undone, more than one undo element will be needed, and each part undone must be given an identifier.

In the following example, three text stages can be identified:

• s1 (initial): This is just some sample text, we need a real example.
  
• s2: This is not a real example.
• s3: This is just some text, not a real example.

This can be encoded using empty <ge:undo> elements as follows:

<ge:line>This is <del stage="#s2" xml:id="del_1" rend="overstrike">
  <ge:undo
    target="#del_1"
    spanTo="#X02"
    rend="dotted"
    stage="#s3"/>just some
 <anchor xml:id="X02"/>sample <ge:undo
    target="#del_1"
    spanTo="#x4"
    rend="dotted"
    stage="#s3"/>text,<anchor xml:id="x4"/> we need
 </del>
 <add stage="#s2">not </add>a real example.</ge:line>

The target attributes on undo point to the elements representing the initial acts (the deletions) which are undone. The rend attribute indicates the way this reversion is indicated. The spanTo attributes points to the anchor which marks the point in the text where this reversion finishes. Since two non-contiguous parts of the deletion are undone, there are two <ge:undo> elements, each with the appropriate attribute values. If spanTo is not supplied, the <ge:undo> is understood to refer to the whole of the text contained by the element indicated by its target attribute.

3.2.8 Instant corrections

The use of tags such as del and add necessarily implies that the modification concerned was made at some time after the original writing. An exception to this is where a false start or ‘instant’ correction has been identified: the author starts to write, and then immediately corrects what has been written. A special mechanism is provided for this case: an instant attribute has been introduced to att.editLike, whose datatype is data.xTruthValue and false is the default value. When the value of instant is true this indicates that the addition or deletion is considered to belong to the same writing stage as the rest of the unmodified document, while false means some stage later than the current stage.

An example of false start can be seen in the following line:

Figure 14. http://www.whitmanarchive.org/resources/sleepers/uva.00256.001.jpg

in which we can detect the following sequence of events:

1. The letter T is written and then immediately deleted
2. The word The is written, deleted, and replaced by the word His
3. The added word His is then deleted
4. The initial letter i of the words iron necklace is overwritten with a capital I

To indicate that the first of these acts must have taken place before the others, we might encode this revision campaign as follows:

<ge:line>
 <del instant="true">T</del>
 <subst>
  <del>The</del>
  <add place="above">
   <del rend="overstrike">His</del>
  </add>
 </subst>
 <subst>
  <del rend="overwritten">i</del>
  <add place="superimposed">I</add>
 </subst>ron necklace</ge:line>

4.1 Assembling a dossier

Since a dossier needs to be assembled from many documents, which will most probably be encoded in distinct TEI documents, one way to assemble a dossier’s contents for further processing would be to use the existing teiCorpus element in combination with XML Inclusions (XInclude). The teiCorpus construct provides for a common place to record metadata regarding the organization of the dossier itself, independently of the metadata regarding each particular document contained within it, which would be held in a discrete TEI Header attached to that particular encoded document. The XInclude mechanism provides a convenient means of managing the many separate resources which are likely to be needed to constitute a complete dossier. For example, supposing we have a dossier comprising three documents, each of which has been encoded in its own document:

(Note that each of the documents referenced (document1.xml etc) should contain a complete TEI element.)

4.2 Genetic Graphs

Looking at the documents which constitute a given dossier, there are many types of relationships which can be identified, both amongst complete documents, and amongst parts of those documents, including alterations, revisions, stages and other compositional phenomena. A further complexity arises if for example an author chooses to correct two different versions at the same time, ³ . We may thus need to express that two or more documents are related in different ways; for instance, one document may be the sequel of another, one may have been drafted at the same time as another, one may contain material or treat topics related to those of another, for example a newspaper article may inspire or be quoted by a given work.

We propose to model the ‘genetic relations’ one can deduce from such penomena by means of a graph. ⁴ A graph is a mathematical representation composed of nodes and arcs. Nodes represent objects, which are related: Typical objects in our case are documents, document components, text stages (as defined in 6.1 Stages/ Revision campaigns below) or even single phenomena (acts of writing or a textual alterations). Each arc between two nodes then represents a relation, in our case a genetic relation, between the two objects it links. Arcs may be typed and directed to further specify the kind of relationship they represent. For our purposes one can assume that a typical genetic graph is directed and acyclic, because we generally expect the paths through a genetic graph (formally defined as an ordered set of nodes traversed by following its arcs) to represent genetic processes with an implied chronological order. In this respect a genetic graph resembles a family tree, in that there is a single terminal node, representing the final or published version of a text with many preceding nodes linking to it, either directly or via other nodes, each of which represents draft versions of various parts. Alternatively, there may be more than one final state, e. g. in the case of multiple published texts or unpublished fragments.

A visual representation of an exemplary genetic graph might look as follows:

What can be seen here is on a very abstract level: The nodes (drawn as rectangular shapes with letters) represent parts of the dossier, the directed arcs (drawn as lines with arrows) represent relationships between the parts (probably of an evolutionary type). Thus the graph can be read as A and B being precursors of C, C in turn leading to F, F to G, G to H, and H leading to the final state Z, which additionally draws influences from E and D.

Such an abstract graph structure can be easily imported into or exported from a graph database, transformed into other XML-based representations of graph-based data models like RDF/XML, or it can be serialized to SVG for interactive visualization.

4.3 Genetic analysis

With the graph’s abstract structure defined, one can now go on and specify, what the nodes and arcs actually represent. This is done by assigning properties to them via the value attribute. In the example the nodes A-H point to individual witnesses, D and E being an exception as they point to different parts of the same witness. The node Z (the final state in our example) supposedly points to the edited text and each arc’s value points to a resource, whose contents comprise the genetic analysis, which underlies the assumed genetic relations expressed in the graph.

[...]

4.4 Constructing genetic graphs

It is highly unlikely, that editors will construct and encoded genetic graphs by hand, unless the dossier is as simply structured as the given example. Rather a typical workflow might be:

1. The editor transcribes each witness in a separate TEI document.
2. During the transcription process phenomena of genetic relevance are annotated with identifiers, that classify these phenoma as belonging to a certain genetic process or that simply relates these phenomena in a well-defined way.
3. By automatically collecting and evaluating all those annotations from all transcribed documents, a preliminary graph is constructed, that can be visualized and navigated.
4. Steps 1-3 are repeated up to the point, where the graph cannot be further refined via this procedure.
5. Optionally, the automatically constructed graph can be annotated manually.

For example in the genetic edition of Goethe’s Faust, the final text is known and can therefore serve as a canonical reference system for constructing genetic relations. Each verse has a unique number:

This makes it possible to express genetic relations by using these same verse numbers to index corresponding verses or ranges of verse-numbers throughout their various witnesses, stages or variants. Although a genetic graph generated from the coindexed verses will not be of a very fine granularity, it can be a useful preliminary base for subsequent refinement. By incorporating metadata for the witnesses (especially dating information) and results of collation eith other relevant information, the graph can be enriched in an iterative manner and periodically checked for consistency. Finally, together with the transcriptions, collation results, and a critical apparatus it can be archived as an integral part of the genetic edition.

6.1 Stages/ Revision campaigns

The existing element creation (within the TEI Header profile description) is defined as the appropriate location for all information relating to the genesis or production of a text. We modify it slightly to permit a new stageNotes element which contains a number of stageNote elements, one for each identified stage:

In the following example taken from the genetic edition of Goethe’s Faust, the editor has identified four distinct stages:

The stageNotes elements carries an attribute ordered, which can take the values true or false (the default). The attribute specifies whether the order of child elements signifies a temporal order for the revision campaigns which they document. In the Faust example above, the editor has asserted that the four stages distinguished are ordered chronologically according to the order of the stageNote elements. Note that asserting a specific order early on, though probably one of the hardest tasks in a genetic analysis, can considerably reduce the encoding effort in assigning textual alterations to stages during the transcription, as we will see below. For instance deletions can only be assigned to a stage that follows the one in which the passage being deleted was written down. Hence, having a certain order of stages put in place before transcription begins, will allow the encoder to reduce verbose tagging, where default assumptions based on the natural order of actions can be made.

If necessary, stageNotes elements can be nested hierarchically. This may be helpful in two cases. Firstly one can build up hypotheses about related revisions step-by-step, starting with stages of smaller coverage, whose members are certainly related, and then in a subsequent pass grouping these stages in turn, thereby extending their reach.

Another use case for nested stageNote elements would be the need to express a partial ordering of revision campaigns.

In addition to the possibility of ordering text stages in relation to each other, stageNote elements may carry a number of attributes from the att.datable class (period, when, notBefore, notAfter, from, and to) which allow each stage to be dated as exactly or inexactly as necessary, in the same way as is currently possible for the TEI date element.

Each stageNote element, apart from declaring a text stage, may also contain references to other annotations contained within the teiHeader or in the document (as shown in the previous example). Such references, along with the textual content are purely documentary and do not affect the textual stage associated with any element thus referred to. The association of a textual component with a writing stage is always made explicitly, either by pointing from the stageNote attribute @target to one or more elements, or (for preference) by pointing from the element concerned to the stageNote element by means of its stage attribute:

This simple example shows the latter of the two options: The relevant stages are declared in the header; then textual alterations and acts of writing are assigned to them. So the whole sentence was realized in the first stage, while the substitution of “house” with “mouse” happened at the second stage.

A more complex and complete example:

Note first, that a stage, once assigned to an element, is inherited by all descendents of that element, unless overridden by a subsequent assignment. So in the example above the three verses are assigned to the first stage initially. The writing of Nun (as part of the substitution in the first verse) takes place in the second stage and is repeated or fixated in the third. Also the substitution in the second verse is done repeatedly: initially it takes place in the second stage, but is fixated as a whole in the third.

As one can see, the interpretation of what the stage assignments mean for a particular text passage and the actions upon it can be based on a number of implicit assumptions and constraints which have the effect of minimizing the amount of tagging necessary. If it is desired to make these presuppositions more explicit, one can differentiate between acts of writing and textual alterations and accordingly between their assignment to stages as follows:

Here two transcriptions of the same passage are given, one from a documentary perspective stressing the writing process, and one from the textual perspective emphasizing textual alterations. The assignment to stages is also done differently here, pointing from the stage notes to the text passages and alterations in question. From the documentary perspective, the stage assignments describe the writing process, in that they specify, which segment has been written when and how often. From the textual perspective, the markup concentrates on the order of textual alterations and makes no assumptions about the order of writing.