P5: Guidelines for Electronic Text Encoding and Interchange

The example below presents the TEI encoding of the name-value pair <part of speech, common noun>, where the name (key) ‘part of speech’ is abbreviated as ‘POS’, and the value, ‘common noun’ is symbolized by ‘NN’. The entire name-value pair is encoded by means of the element f. In TEI XML, that element acts as the container, labeled with the name attribute. Its contents may be complex or simple. In the case at hand, the content is the symbol ‘NN’.

The datcat attribute relates the feature name (i.e., the key) to the data category ‘part of speech’, while the attribute valueDatcat relates the feature value to the data category common noun. Both these data categories should be defined in an external and preferably open reference taxonomy or ontology.

‘NN’ is the symbol for common noun used e.g. in the CLAWS-7 tagset defined by the University Centre for Computer Corpus Research on Language at the University of Lancaster. The very same data category used for tagging an early version of the British National Corpus, and coming from the BNC Basic (C5) tagset, uses the symbol ‘NN0’ (rather than ‘NN’). Making these values semantically interoperable would be extremely difficult without a human expert if they were not anchored in a single point of an established reference taxonomy of morphosyntactic data categories. In the case at hand, the string ‘http://hdl.handle.net/11459/CCR_C-1256_7ec6083c-23d4-224d-6f94-eecbe6861545’ is both a persistent identifier of the data category in question, as well as a pointer to a shared definition of common noun.

While the symbols ‘NN’, ‘NN0’, and many others (often coming from languages other than English) are implicitly members of the container category ‘part of speech’, it is sometimes useful not to rely on such an implicit relationship but rather use an explicit identifier for that data category, to distinguish it from other morphosyntactic data categories, such as gender, tense, etc. For that purpose, the above example uses the datcat attribute to reference a definition of part of speech. The reference taxonomy in this example is the CLARIN Concept Registry.

The assumption here is that the relevant feature values are collected in a place that the annotation document in question has access to — preferably, a single document per linguistic resource, for example an fsdDecl that is XIncluded as a sibling of text or a child of encodingDesc; a taxonomy available resource-wide (e.g., in a shared header) is also an option.

Note that these Guidelines do not prescribe a specific choice between datcat and valueDatcat in such cases. The former is the generic way of referencing a data category, whereas the latter is more specific, in that it references a data category that represents a value. The choice between them comes into play where a single element — or a tight element complex, such as the f/symbol complex illustrated above — make it necessary or useful to distinguish between the container data category and its value.

In the context of dictionaries designed with semantic interoperability in mind, the following example ensures that the pos element is interpreted as the same information container as in the case of the example of <f name="POS"> above.

This mechanism creates implications that are not always wanted, among others, in the case at hand, suggesting that the identifiers ‘pos’ and ‘adj’ belong to a namespace associated with the CLARIN Concept Repository (CCR), whereas that is solely a shorthand mechanism whose scope is the current resource. Documenting this clearly in the header of the dictionary is therefore advised.

Yet another possibility is to associate the information about the relationship between a TEI markup element and the data category that it is intended to model already at the level of modeling the dictionary resource, that is, at the level of the ODD, in equiv element that is a child of elementSpec or attDef.

The targetDatcat attribute is designed to be used in, e.g., feature structure declarations, and is analogous to the targetLang attribute of the att.pointing class, in that it describes the object that is being referenced, rather than the referencing object.

Above, the fDecl uses targetDatcat, because if it were to use datcat, it would be asserting that it is an instance of the container data category part of speech, whereas it is not — it models a container (f) that encodes a part of speech. Note also that it is the f that is modeled above, not its values, which are used as direct references to data categories; hence the use of datcat in the symbol element.

The TEI Abstract Model can be expressed as a hierarchy of attribute-value matrices (AVMs) of various types and of various levels of complexity, nested or grouped in various ways. At the most abstract level, an AVM consists of an information container and the value (contents) of that container.

A simple example of an XML serialization of such structures is, on the one hand, the opening and closing tags that delimit and name the container, and, on the other, the content enclosed by the two tags that constitues the value. An analogous example is an attribute name and the value of that attribute.

In a TEI XML example of two equivalent serializations expressing the name-value pair <part-of-speech,common-noun>, namely <pos>commonNoun</pos> and pos="common-noun", one would classify the element pos and the attribute pos as containers (mapping onto the first member of the relevant name-value pair), while the character data content of pos or the value of pos would be seen as mapping onto the second member of the pair.

The att.datcat class provides means of addressing the containers and their values, while at the same time providing a way to interpret them in the context of external taxonomies or ontologies. Aligning e.g. both the pos element and the pos attribute with the same value of an external reference point (i.e., an entry in an agreed taxonomy) affirms the identity of the concept serialised by both the element container and the attribute container, and optionally provides a definition of that concept (in the case at hand, the concept part of speech).

The value of the att.datcat attributes should be a PID (persistent identifier) that points to a specific — and, ideally, shared — taxonomy or ontology. Among the resources that can, to a lesser or greater extent, be used as inventories of (more or less) standardized linguistic categories are the GOLD ontology, CLARIN CCR, OLiA, or TermWeb's DatCatInfo, and also the Universal Dependencies inventory, on the assumption that its URIs are going to persist. It is imaginable that a project may choose to address a local taxonomy store instead, but this risks losing the advantage of interchangeability with other projects.

Historically, datcat and valueDatcat originate from the (the now obsolete) ISO 12620:2009 standard, describing the data model and procedures for a Data Category Registry (DCR). The current version of that standard, ISO 12620-1, does not standardize the serialization of pointers, merely mentioning the TEI att.datcat as an example.

Note that no constraint prevents the occurrence of a combination of att.datcat attributes: the fDecl element, which is a natural bearer of the targetDatcat attribute, is an instance of a specific modeling element, and, in principle, could be semantically fixed by an appropriate reference taxonomy of modeling devices.