Introduction

This document describes how to produce a customization of the TEI P5 schema. From the start, the TEI was intended to be used as a set of building blocks for creating a schema suitable for a particular project. This is in keeping with the TEI philosophy of providing a vocabulary for describing texts, not dictating precisely what those texts must contain or might have contained. This means that it is likely, not just possible, that you will want to have a tailored view of the TEI.

What do we mean by a customization? It is important to understand that there is no single DTD or schema which is the TEI; you always choose from the available modules (there are currently 22 of them, listed in ) those that you want, with the caveat that the three modules core, header and textstructure (and tei, when using RELAX NG) should always be chosen unless you are certain you know what you are doing. Elements in these modules are referred to throughout the other modules, and hence these modules cannot be eliminated without careful adjustments.

There are three ways of customizing the TEI: Writing a high-level specification for a view of the TEI, and generating an ad hoc DTD or schema; this is the preferred method. Using the DTD modules, and specifying in the document DTD subset which features you want activated. Using the RELAX NG modules, and writing a wrapper schema. Note that it is not possible at present to use W3C Schema modules for customization.

Although there is no default schema, TEI P5 does provide a number of example customizations which may very well meet your needs, which can be downloaded from the TEI web site or from within the Roma interface: tei_bare: TEI Absolutely Bare teilite: TEI Lite tei_corpus: TEI for Linguistic Corpora tei_ms: TEI for Manuscript Description tei_drama: TEI with Drama tei_speech: TEI for Speech Representation tei_odds: TEI for authoring ODD tei_allPlus: TEI with maximal setup, plus external additions tei_svg: TEI with SVG tei_math: TEI with MathML tei_xinclude: TEI with XInclude (experimental) tei_dictionaries: TEI for Dictionaries (experimental)

Choosing the basic set of modules may be sufficient, but it's also possible that you may want to tailor your TEI schema more tightly. For instance, once you have decided that your application will make use of the msdescription and linking modules, you may also want to remove elements from some of the modules which you do not expect to use, to reduce confusion and avoid the accidental use of elements you don't need rename elements (see for more discussion of this) add, delete or change attributes for existing elements, perhaps to make the datatype stricter add new elements, and insert them into the TEI class system We will be seeing examples of each of these in the following sections.

Below is a table of all of the TEI modules. More information about each one is given in the TEI Guidelines; each module corresponds to a single chapter.

The TEI modules. analysis Simple analytic mechanisms certainty Certainty and uncertainty core Elements common to all TEI documents corpus Header extensions for corpus texts declarefs Feature system declarations dictionaries Printed dictionaries drama Performance texts figures Tables, formulae, and figures gaiji Character and glyph documentation header The TEI Header iso-fs Feature structures linking Linking, segmentation and alignment msdescription Manuscript Description namesdates Names and dates nets Graphs, networks and trees spoken Transcribed Speech tagdocs Documentation of TEI modules tei Declarations for datatypes, classes, and macros available to all TEI modules textcrit Text criticism textstructure Default text structure transcr Transcription of primary sources verse Verse structures

Writing ODD specifications

The TEI is written in a source format called ODD (One Document Does it All) which includes the schema fragments, prose documentation, and reference documentation for the TEI Guidelines in a single document.The concepts of ODD were devised and implemented by Lou Burnard and Michael Sperberg-McQueen early in the development of the TEI. The language developed over time as the TEI was put together, and one form of it was documented in the TEI Guidelines (versions 3 and 4); unfortunately, that version of the markup was not what was actually used to write the TEI Guidelines, which diverged into a more complex scheme. For version 5 of the TEI, the entire ODD language was heavily revised and simplified by a working group led by Sebastian Rahtz, and the Guidelines themselves brought into conformance with it.) An ODD specification is a normal TEI XML document which makes use of the tagdocs module. This adds a series of elements which are used to specify a new schema, and modifications to the TEI element structure. It is described in detail in the TEI Guidelines chapter , so only a brief summary will be given here.

The recommended way to customize the TEI is to create a formal specification expressing your customizations, as an XML document using TEI ODD markup; this can then be compiled into a suitable DTD, RELAX NG schema or W3C Schema (together with the appropriate reference documentation), using the Roma program. Roma is a web-based interface for creating TEI customizations, which allows you to fill in simple forms to choose modules, add and delete elements, change attributes, and make other customizations. Advanced users can also create the ODD by hand using normal XML editing tools.

If, however, you intend to make extensive use of the TEI in conjunction with other schemas written in RELAX NG, working directly with the RELAX NG modules is probably the best skill to learn. Typical TEI users are more likely to work solely within the confines of the TEI, and need to use DTDs and W3C Schema as well as RELAX NG, and so writing customizations in the TEI's own language is usually better.

There are several important reasons why this high-level method is recommended: It is independent of the schema type (DTD, RELAX NG schema, W3C schema) and the resulting specification can be used to generate a schema in any of these schema languages. It lets you document your work using the familiar TEI markup. It provides full access to the TEI class system. The Roma utilities generate a single, portable, schema file which you can transfer to other people without worrying about link dependencies.

A TEI schema is defined by a schemaSpec element containing an arbitrary mixture of explicit declarations for objects (i.e. elements, classes, or macro specifications) and references to other objects containing such declarations. In simplified form, the data model is schemaSpec = (moduleRef | elementSpec | macroSpec | classSpec )* where elementSpec, macroSpec and classSpec contain definitions of TEI objects. moduleRef references groups of existing definitions, in one of two ways: If the key attribute is provided, it refers to the TEI name for a module, and details of that are accessed from the TEI web service database (which may be a local installation). If the url attribute is provided, it refers to an external file of schema definitions in the RELAX NG language (this is used to pull in non-TEI schemas)

In the simplest case, a user-defined schema might simply combine all the declarations from some nominated modules: <TEI> <teiHeader> <fileDesc> <titleStmt> <title>TEI with simple setup</title> <author>Sebastian Rahtz</author> </titleStmt> <publicationStmt><p>freely available</p></publicationStmt> <sourceDesc> <p>Written from scratch.</p> </sourceDesc> </fileDesc> </teiHeader> <text> <body> <schemaSpec ident="oddex1" start="TEI"> <moduleRef key="header"/> <moduleRef key="core"/> <moduleRef key="tei"/> <moduleRef key="textstructure"/> </schemaSpec> </body> </text> </TEI> Note that this is a normal TEI document, with a metadata header. In the other examples that follow, we will usually omit the outer TEI wrapper and just show the schemaSpec element.

An ODD processor, given such a document, will combine the declarations which belong to the named modules, and deliver the result as a schema of some requested type. It might also generate documentation for all (and only) the elements declared by those modules. The start attribute of schemaSpec is used to specify in a RELAX NG schema which elements are valid entry points.

You can address individual elements or classes of modules by the adding elementSpec, classSpec or macroSpec elements after moduleRef. Each of these must have a mode attribute on it, which can take four values: addthe object is entirely new. replacethe object entirely replaces the existing object with the same ident. deleteall references to the original object with the same ident are removed from the schema. changechild elements of the object which appear in the original specification are replaced by the versions in the new specification. This may be at any level, as we will see in examples below. It is an error to provide replace, delete or change versions for objects which do not already exist in the TEI, and an error to add something with the same ident attribute as an existing object in the TEI.

Working with RELAX NG schema modules

If you want to use the RELAX NG schema modules,Examples of RELAX NG in this section are presented using the compact syntax; when you write TEI customizations in the ODD system, it is necessary to use the XML syntax. you must always write a wrapper schema, selecting the appropriate modules. Thus a minimal TEI schema might look like this: namespace ns1 = "http://www.tei-c.org/ns/1.0" namespace rng = "http://relaxng.org/ns/structure/1.0" include "http://www.tei-c.org/schema/relaxng/header.rnc" inherit = ns1 include "http://www.tei-c.org/schema/relaxng/core.rnc" inherit = ns1 include "http://www.tei-c.org/schema/relaxng/tei.rnc" inherit = ns1 include "http://www.tei-c.org/schema/relaxng/textstructure.rnc" inherit = ns1 start = TEI This is clearer than the DTD method, as it loads files containing definitions from explicit URLs. It is then possible to override any patterns in the included files; so the following schema namespace ns1 = "http://www.tei-c.org/ns/1.0" namespace rng = "http://relaxng.org/ns/structure/1.0" include "http://www.tei-c.org/schema/relaxng/header.rnc" inherit = ns1 [ define [ name = "mentioned" notAllowed [ ] ] ] include "http://www.tei-c.org/schema/relaxng/core.rnc" inherit = ns1 include "http://www.tei-c.org/schema/relaxng/tei.rnc" inherit = ns1 include "http://www.tei-c.org/schema/relaxng/textstructure.rnc" inherit = ns1 start = TEI loads the header module, but then redefines the meaning of mentioned to be the special RELAX NG pattern notAllowed. This is a powerful and elegant mechanism; the only downside is that you must understand the inner structure of the TEI modules.

RELAX NG patterns are defined for the TEI as follows: Each element, macro and class identifies the module it is part of; this determines which schema file its definition is written to. Every macro (defined by a macroSpec in the source) has a RELAX NG pattern of the same name. e.g. macro.glossSeq = altIdent?, equiv*, gloss?, desc? This can be redefined as desired. Class specifications generate a number of patterns, depending on their type: An attribute class generates a pattern which references the definition of each of the class attributes. Each attribute generates a pattern. A model class generates a pattern with an initial value of notAllowed Thus the att.timed attribute class generates tei.timed.attributes = tei.timed.attribute.start, tei.timed.attribute.end, tei.timed.attribute.dur tei.timed.attribute.start = attribute start { datatype.uri }? tei.timed.attribute.end = attribute end { datatype.uri }? tei.timed.attribute.dur = attribute dur { xsd:duration }? while the model.listLike model class generates model.listLike = notAllowed Every element generates at least three patterns; the first defines the element itself, the second defines its content, and the third its attributes. For example, the top-level element TEI is defined with: TEI = element TEI { TEI.content, TEI.attributes } TEI.content = tei.teiHeader, tei.teiText TEI.attributes = [ a:defaultValue = "5.0" ] attribute version { xsd:decimal }?, [ a:defaultValue = "TEI" ] attribute TEIform { text }? Each of these can be redefined separately. In addition, for each model class of which the element is a member, it generates an addition to the class pattern. Thus biblItem is a member of the model.biblLike, att.declarable, and att.typed classes, so it produces: tei.bibl |= biblItem tei.declarable |= biblItem tei.typed |= biblItem so that any reference to model.biblLike will now allow for biblItem too.

Working with the DTD subset

It is also possible to work with DTD modules, although the TEI does not recommend this any more. You specify which modules of the TEI you want to use by means of the DTD internal subset. A minimal TEI document using this method might start as follows: ]> ]]> This loads the obligatory modules header, core, and textstructure by setting the corresponding parameter entity to INCLUDE.

There is a parameter entity for each module(created by prefixing the module name with TEI., so we could request the linking module to be loaded by adding ]]> to the DTD subset. It is also possible to disable particular elements from the modules by setting a parameter corresponding to the element. So ]]> would remove ab from the list of allowed elements. Although this type of customization is useful, it is not possible to use the method to add new elements, change attributes, or manipulate classes. That sort of change requires a deeper understanding of writing DTD extensions, beyond the scope of this introduction.

Roma (command line)

An ODD specification can be processed in a scripting environment by using the roma command-line script. This takes the form: Usage: roma [options] schemaspec [output_directory] options, shown with defaults: --xsl=/usr/share/xml/tei/stylesheet --teiserver=http://tei.oucs.ox.ac.uk/Query/ --localsource= # local copy of P5 sources options, binary switches: --doc # create expanded documented ODD (TEI Lite XML) --lang=LANG # language for names of attrbutes and elements --doclang=LANG # language for documentation --dochtml # create HTML version of doc --patternprefix=STRING # prefix relax patterns with STRING --docpdf # create PDF version of doc --nodtd # suppress DTD creation --norelax # suppress RELAX NG creation --noxsd # suppress W3C XML Schema creation --noteic # suppress TEI-specific features --debug # leave temporary files, etc. By default the script creates DTD, XSD and RELAX NG schemas, each of these can be suppressed if needed, and a set of summary documentation can be created. The xsl and teiserver options point to resources which roma needs to do its job; if you have a local copy of the TEI XSL stylesheets, or a local TEI eXist database, you can make the script independent of web access.

For information on using the web-based interface to roma, see Creating Customizations with Roma.

Making use of non-TEI schemas

The TEI was designed to capture all the vagaries of literary and linguistic text; it does not attempt to describe other specialised descriptive languages, such as those for chemistry, mathematics, and vector graphics, or the technical vocabulary of fields like law, health care and computer science. Some of the areas have been addressed as thoroughly as the TEI in their own standards. But what if we want to write a composite document mixing material from two fields? Since all the TEI elements are in their own XML namespace, it is easy to write a document which interleaves TEI markup with markup from another namespace, as in this example of TEI and Docbook: <p> The button on our web page shows the the date of the manuscript: <guibutton xmlns:dbk="http://docbook.org/docbook-ng"> <date calendar="Julian" when="1732-02-22">Feb. 11, 1731/32, O.S.</date> </guibutton> Note that the representation is as found in the text, not normalized. </p> But what about validating this XML against a schema? Using the Namespace-based Validation Dispatching Language (see ), we can validate the two languages separately, but we also want a TEI customization which checks where insert of foreign elements is permitted. This means importing another schema, and changing one or more TEI classes to allow for the new element(s). If it is also required that TEI elements be allowed inside the elements of the other namespace, we also have to modify the other namespace.

Two common cases which do not require interleaving are: redefining the content of formula to allow for MathML markup. redefining the content of figure to allow SVG markup. In each case, we first need a moduleRef which loads the external schema in RELAX NG format: ]]> These schemas can be downloaded from and ; note that they may each need a small fix to remove the RELAX NG start pattern, as this causes a conflict with the TEI definition. These define respectively two patterns called mathml.math and svg.svg, which we can proceed to add to TEI content models. For MathML, we can redefine an existing macro which is already provided as a hook inside the content of formula: <macroSpec xmlns:rng="http://relaxng.org/ns/structure/1.0" type="pe" ident="datatype.Formula" mode="change" xml:base="examples/addmath.xml"> <content> <rng:ref name="mathml.math"/> </content> </macroSpec> For SVG, we need to change the model of figure, simply adding a reference to svg.svg at the end of a choice list: <elementSpec ident="figure" mode="change" xml:base="examples/addsvg.xml"> <content> <rng:zeroOrMore xmlns:rng="http://relaxng.org/ns/structure/1.0"> <rng:choice> <rng:ref name="model.Incl"/> <rng:ref name="figure"/> <rng:ref name="figDesc"/> <rng:ref name="graphic"/> <rng:ref name="head"/> <rng:ref name="p"/> <rng:ref name="svg.svg"/> </rng:choice> </rng:zeroOrMore> </content> </elementSpec>

Internationalisation

A common requirement for changing existing elements is to make the visible names suit a local language. If we want to use the TEI in an entirely Spanish-speaking environment, it can be useful to have a copy of the TEI schema with all the names converted to Spanish. Documents can be created and edited using this schema, and then translated back to the canonical form for long-term archiving or distribution.

These translations are possible because the TEI defines names in English for elements and attributes, but does not use these names directly in content models for other elements. This means that the names can be changed without breaking the rest of the system. For example, the content model for series is series.content = (text | model.gLike | title | editor | respStmt | biblScope | model.global)* but the title here refers to the pattern called title; this is defined with: title = element title { title.content, title.attributes } If we change it to title = element titulo { title.content, title.attributes } the definition for series will still work, and the pointers to the content and attributes of title remain correct.

If we create documents using this schema, how can we be sure the back translation is easy? Because we can always go back to the source of the customization to find the original name.

The translation process in ODD is simple. Each element or attribute affected must be supplied in change mode, with simply an altIdent provided. For example, here are some translations into Spanish: <schemaSpec xml:base="examples/spanish.xml"> <elementSpec ident="quote" module="core" mode="change"> <altIdent type="lang">cita</altIdent> </elementSpec> <elementSpec ident="cit" module="core" mode="change"> <altIdent type="lang">citaCompl</altIdent> </elementSpec> <elementSpec ident="mentioned" module="core" mode="change"> <altIdent type="lang">mencionado</altIdent> </elementSpec> <elementSpec ident="when" module="linking" mode="change"> <altIdent type="lang">cuando</altIdent> <attList> <attDef mode="change" ident="unit"> <altIdent type="lang">unidad</altIdent> </attDef> </attList> </elementSpec> </schemaSpec> Notice that each attDef element must also specify change mode, as well as the parent elementSpec.

Constructing specifications like this by hand is both tedious and error-prone, and it would be unwise for each separate project to make its own translations. The TEI Consortium therefore maintains a set of translated names,A project initiated by Alejandro Bia, and extended by Sebastian Rahtz and Arno Mittelbach. and a utility to generate the appropriate ODD code for elements from all the modules you have selected. The Roma application automates this to choosing from a drop down list as shown in the figure below.

Choosing language for names in Roma

The effect of using a translated schema is shown in the image below; the oXygen editor is shown editing with Spanish element and attribute names.

Editing TEI text using a schema translated to Spanish