From TEI to HRIT and back again

Since we are designing a software suite to more or less replace embedded markup there has to be some way to import legacy texts. At first I thought the problem was insurmountable. Even if the original encoders had stuck to recommended guidelines such as the TEI (Text Encoding Initiative) they would have been forced to customise their encoding in two ways:

1. By adding custom tags and attributes, and
2. By making a selection of tags from the large number of available ones

In the second case it is clear that any general solution that embraced an arbitrary subset of TEI would have to support all of it. Since there are currently 519 tags in the scheme, and (probably) thousands of attributes, that is a daunting prospect for any programmer. And we are talking about meaningful conversion into an entirely different software system, not a simple one-for-one mapping. And in respect to point 1 any customised tags would either have to be left out, or their function would need to be specified by the user.

Solving the problem

When forced to perform the task, however, I soon realised that any customised tags must have already been specified by a user who understood XML. So that same user could supply a customised table of conversion in XML to say what should be done with them. If they didn't follow the Guidelines then they have to do a little extra work, but they're not shut out.

And in the second case only a small subset of TEI is regularly used by digital humanists. For the purposes of defining versions, for example, only a small number of tags come into play, and even customised ones would have to follow one of only a couple of basic patterns, which could be programmed in as general functions. The customisations could be handled by a 'recipe', or set of instructions on how to convert the files. A default recipe would be provided for standard files, which the user could extend or change at will.

Why do this at all?

Because HRIT format is much more powerful than TEI:

1. It allows arbitrary overlap of properties.
2. It does not mandate any standard tag names
3. It supports versions natively including transpositions
4. It allows mixing and matching of markup sets in the one text

That's more than enough reasons to move from TEI to HRIT. Another way of looking at it is to say that rather than replacing TEI it seeks to enhance it, and use it as an interchange format between HRIT and non-HRIT users. It depends on what kind of 'spin' you prefer.

Two-way conversion

Any conversion applied to legacy files (or, if you prefer, current files) would have to be reversible. Those who had imported their files into HRIT and changed their minds later on would feel 'locked in' if they couldn't back out, and those who hadn't made the switch would likewise be frightened off by that very prospect. So the overall process looks like this. Red/green arrows indicate as yet unavailable/available paths:

'TEI' refers to any TEI-encoded file. The two-way process works like this:

• Splitter splits the TEI file into N versions. By default it splits <app><rdg>...</rdg></app> structures as well as nested <del> and <add> and <choice> structures into versions. Unsplitter, not yet written, will take the versions (possibly modified) and try to put them back into one file, although this may be difficult. The recipe file is used by splitter to direct the splitting. It can be customised by the user to control which elements are split and how.
• Stripper removes the remaining markup from each separate version in TEI format. A different recipe file specifies simplifications of elements intended to be rendered as formats in the final HTML. One simplification might be the reduction of <hi rend="italic"> to the property 'italics'. The output of stripper is the HRIT standoff XML format. (But stripper is written in such a way that another format can be added if required). It expresses every TEI element as a potentially overlapping property with possible 'annotations' or attributes. These attributes are ignored by the formatter but are not lost. Elements like the TEI-header, which contain metadata about the text, are entirely hidden but also not lost. This is to enable later reversal of the stripping process. Each version produces a pair of markup and plain text files that are separately merged into a single CorTex and a single CorCode file. It is these files that are edited and read by the HRIT system.
• Formatter takes the properties of the CorCode and combines them with the information from the CSS file into HTML. The CSS is used not only to change the appearance of the text on a web page but also to transform the markup. For example the CSS rule span.italics can be used to change the appearance of italics, but also to convert properties called 'italics' into spans of class 'italics'. In this way we can avoid use of XSLT. But what about the 'annotations' that were originally attributes in the TEI-XML? They are simply ignored (although not lost). If you want to convert an element plus some attribute(s) into a HTML element using formatter, you must first specify a rule to simplify them to a plain property using splitter's recipe file.

Multi-lingual MVDs

There are plenty of cases where the concept of 'work' spans more than one basic version in one language. Just think of the multi-lingual laws of the EU, the Romulo of Virgilio Malvezzi translated into several languages, each having its own textual history, or the Chronicles of Eusebius, in Latin, Greek and Armenian. The question is, how can you align the same text written in a different language? Can one align Latin and Greek, or French and German? In my opinion, no, or at least not automatically. Quite apart from the language dissimilarity, translations often have quite different structures, making alignment particularly difficult. But a tiny change to the definition of an MVD makes it possible to align such texts manually and to use the MVD format as a storage facility.

Tweaking the groups

MVDs have always had a simple grouping mechanism. You can group versions by type. For example, versions of a particular recension, or internal versions (corrections or revisions of a single manuscript) can be grouped together to keep them separate from versions in other physically different documents. Now if we assign one of these groups a simple attribute, called 'merge' and set it to 'true' or 'false', then we can control how an MVD is built up. For example, imagine we have French, German and Italian translations of some work, each in several versions. We could group all the Italian versions together, and similarly for the German and French ones. And we could set each group's attribute 'merge' to 'true'. But each such group would belong to a higher group, whose 'merge' attribute would be 'false'. So the merging program would know, on being given version 23 (French) to add to the MVD, not to merge it with version 16 (German) because their shared parent group is not merged. Here's how it would look schematically inside the resulting MVD:

This might also be a good strategy whenever the same 'work' is substantially rewritten, like the Morte d'Arthur and other medieval tales. Versions of each rewrite would get their own group and we wouldn't attempt to align them automatically because it just gets too messy.

Linking the translations

Now we can extend the standoff markup mechanism described in the previous post to link the texts of the different languages manually. We add a view that displays two versions of an MVD side by side:

Selecting some text on the right or left highlights it independently (you can do this in Javascript). Now select something in the opposite version and press the 'link' button. This creates an annotated property that specifies a link between the two selected ranges and records it via standoff markup. The view could then give the user graphical feedback by formatting the two selected blocks rigidly side-by-side:

They could also scroll together in sync, as they currently do in compare view. If blocks are transposed between languages (as often happens) the text might jump around a bit as you scroll, but so long as we align on the most central block it should work OK. Also, the alignment would hold for all the aligned versions on either side, not merely for the ones currently selected. If you had 12 German versions and 16 French ones, they would all be aligned at the same point of their shared text. You could even display an apparatus at the bottom of each side so the user could see the variants of the versions in each language.

How much work is that?

Although a special view would have to be designed, there is not much else needed to make it work. It might even be a good idea to add such a view to the MVD-GUI suite and see what people can do with it – but only once the standoff mechanism is up and running, because this solution depends on it.

Standoff Properties explained

I've been asked for a more detailed explanation of how CorCode works as a set of standoff properties. I'll try but it won't be all that brief.

Embedded markup

Since at least the 1980s humanities texts have been described using embedded markup codes, but this leads to several problems:

1. The structure imposed on the text is a tree and maybe the structure we want to describe is not.
2. The embedded codes need to be standardised because otherwise we can't share texts or create shared software. But there are so many codes we need to define that the standard soon becomes unwieldy.
3. Embedded markup lacks flexibility. We can't easily exchange one set of markup for another, or merge two sets.
4. Users who edit the texts have to read it through the smoke-screeen of the tags and their attributes. And they have to learn a complex system that is becoming ever more complex.

Standoff markup is a partial solution to these problems. Removing tags from the text clarifies it for the reader, and allows the exchange of one set of tags for another. But with standoff markup we still can't combine two tag sets or define non-tree structures. And because the standoff codes depend on the inviolability of the text, we can't edit it.

What I was trying to explain in the previous post is that we can in fact overcome all of these problems by defining markup as a simple set of overlapping named properties. I'm not the first to suggest this by any means: in fact it resembles to varying degrees George's valency idea, LMNL, Thaller's extended string model, eComma, LORE and other annotation systems, and even TexMECS to some extent. But I'd like to describe my implementation because I think it offers some advantages over previous attempts.

Properties

Properties have a name, an offset and a length, which describe a range in the text. That's one string and two numbers. An example of a property is 'italics' at offset 23 with length 5. Let's just consider the offsets first.

Absolute versus relative offsets

With absolute offsets (as used in JITM and every other standoff system I know) the offsets increase for properties as we move through the text. So if we had properties at offsets 2, 10, 23, 45, 106, 230, 1022, 1100, 1495, 1567 and we added 121 characters at the start of the text, the first offset would have to change to 123, AND we would have to add 121 to all subsquent ones: 123, 131, 144, 166, 227, 351, 1143, 1221, 1616, 1688.

With relative offsets we only record the differences between an offset and the previous one. So the same sequence would read: 2, 8, 13, 12, 61, 124, 792, 78, 395, 72. (That's obtained by subtracting 2 from 10, then 10 from 23, then 23 from 45 etc.) Now when we add 121 characters at the start, the sequence changes to 123, 8, 13, 12, 61, 124, 792, 78, 395, 72. Only the first one needs to change because the relative distances between the other properties haven't altered.

Property lengths

If, instead of just inserting text outside of a property we altered the length of the property itself, say by extending a paragraph labelled with a 'p' (if we use TEI), then with relative offsets only the length of that property and the offset of the following one would change. Let's say that the length of the text covered by property 3 was 5 characters and we extended it to 12, then we'd change the length of property 3 from 5 to 12 and the offset of property 4 from 12 to 19 by adding 7 (i.e. 12-5). The other properties preceding property 3 AND following property 4 would not change.

Property names

Now let's consider the names of the properties. We can make them multi-lingual and go beyond what TEI can do. Europeans see TEI as based on English texts. (e.g. look at Domenico's objections in Scrittura e filologica nell'era digitale p. 170). Why should we not call 'italics' 'kursiv' if we are Germans? The entire standard encoding scheme is based on English words and concepts. Why do we have to standardise them when we can just let the users choose what they want to call them? Or they can provide translations for their property names so others can read their markup. So instead of explicit names I propose that we have a table of properties at the start of the list:

1 italics
2 paragraph
3 stage
etc.

Then when we want to use the italics property we just say #1 23 5 – which means 'property 1 (italics) at relative offset 23 of length 5'. Of course the computer handles all this. We never see these values directly, only through their representation on the screen via formatted text, not even when we edit them via the GUI.

Having got the properties into this form we can write a table that provides translations of all the properties in the file into any other languages we choose. And texts marked with '#1' will show up as 'kursiv' for Germans and 'corsivo' for Italians (or المائل for Arabs). TEI can't do this because the English names are burned into the standard.

Editing the text in this form

Each time we edit the underlying text we have to adjust the standoff properties so that they still correspond. But thanks to the use of relative offsets this is easy. After editing the base text the user commits it to the server. The server computes the differences between the old version and the new one. From this we obtain a set of insertions and deletions.

Insertions

If we insert text outside of an property or within a property we just follow the rules described above in adjusting the relative offsets and property lengths of any lists of properties that describe that text.

Deletions

If we delete a bit of text that completely contains a property we delete that property and adjust the relative offset of the next one in the list. If the property's range is only partly deleted (at the start or at the end) we simply adjust its length and also the offset of the next property in the list.

So, in both cases we can edit the text and its underlying properties quite cleanly.

Publishing digital editions

If we publish version 1 of a text and someone writes a property list for it, and then we change the text and issue edition 2, then their properties can easily be adjusted using these procedures. So, on requesting a copy of King Lear, the server informs the user that his edition of King Lear is out of date and would he/she like to update it. The updates are performed automatically and the old properties now refer to edition 2.

Merging property lists into CorCode

Yet another advantage of relative offsets is the ability to merge lists of properties belonging to different versions. Let's say we have 5 versions of Shakespeare's King Lear. We could define properties like stage, speaker, speech, paragraph, line, italics etc for ranges within each version, but like the text these properties would mostly be the same. Tedious. If we had used absolute offsets the lists of properties would all be different because they would contain different offsets throughout. Just one extra character would change all the absolute offsets from then on, and it would fail to merge. But with relative offsets most of the properties, like the text they describe, will be exactly the same. So we can merge all the property lists into one CorCode to correspond to the one CorTex. And when we apply a new property to one version it will automatically be adopted by all other versions - should we so desire - without having to redefine it for each version separately.

Turning overlapping properties into HTML

To make all these advantages practical we will have to convert a text marked up in this way into HTML for the browser. But how to do it? There is no hierarachical structure left, it's not XML, we can't use XSLT and the target language IS a hierarchy. In fact all this has already been done in eComma. How eComma works I don't know so I'll explain how I would do it.

We can scan the text and its property lists and deduce the hierarchical structure. If a property of type line is always inside a property called speech we can deduce that we might render that in HTML as lines inside speeches, say as <span> inside <p>, i.e. as <p><span>...</span></p>. But often in Shakespeare a line is divided between two speakers. Then we can simply break the line up into two lines, because it is most often contained by the speech property. (If it had been the other way around we would have had to split the <p>...</p> instead). So we can resolve all cases of overlap and discover hierachical structure from a simple analysis of the properties.

Translating the properties into HTML tags is also easy. Each HTML file these days comes with a CSS file that tells the browser how to format elements. So if we want to format speech properties specially we provide a css rule:

p.speech { text-indent: 5px }

This tells the browser to indent paragraphs of class 'speech' by five pixels. The neat thing about CorCode is that we can reuse this definition to convert speech properties into paragraphs. We just follow the recipe contained in the CSS rule: speeches are p's of class 'speech'. So we don't need XSLT.

Simplication of XML to properties

When we convert legacy XML files to CorCode format we have to simplify XML elements with their attributes into property names. So <hi rend="italics">word</hi> becomes #1 23 5 (remember we defined '#1' to be italics). So the CSS rules can all be simple and don't have to take account of complex XML attributes. Not all XML properties are as simple as the italics example, but if we provide a list of recipes of how to convert them I think it can be done. For example the TEI coding for a page number looks like this: <pb n="42" ed="1"/>. The "42" is really text and should be represented in the main text with the property of 'page-number'. The 'ed' attribute is really a version specification and should be expressed in the CorText and its versions. So there's nothing left. TEI markup is complex because it is a mixture of every kind of information we want to include that involves text. But we really need to separate out things like anchors to external images into a separate CorCode file that is handled specially in a GUI. When we focus on actual properties that the text really has, as opposed to programming information, there is not much left to represent.