Linking, Segmentation, and Alignment

This chapter discusses a number of ways in which encoders may represent analyses of the structure of a text which are not necessarily linear or hierarchic. In this chapter, tag sets and global attributes are provided for the following common requirements: to link disparate elements in a single document using the id attribute (section ); to link disparate elements in a single document without using the id attribute or to link elements in different documents (section ); to segment text into elements convenient for the encoder and to mark arbitrary points within documents (section ); to represent correspondence or alignment among groups of text elements, both those with content and those which are empty (section );We use the term alignment as a special case of the more general notion of correspondence. Let A stand for an element with the attribute id=A, and suppose elements A1, A2 and A3 occur in that order and form one group, while elements B1, B2 and B3 occur in that order and form another group. Then a relation in which A1 corresponds to B1, A2 corresponds to B2 and A3 corresponds to B3 is an alignment. On the other hand, a relation in which A1 corresponds to B2, B1 to C2, and C1 to A2 is not an alignment. to synchronize elements of a text, that is to represent temporal correspondences and alignments among text elements (section ) and also to align them with specific points in time (section xref target=SAsymp); to specify that one text element is identical to or a copy of another (section ); to aggregate possibly noncontinguous elements (section ); to specify that different elements are alternatives to one another and to express preferences among the alternatives (section ); to associate segments of a text with interpretations or analyses of their significance (section ).

These facilities all use the same basic set of techniques, which depend on the ability to point to an element which has some form of identifier. The most convenient such identifier, and that which is recommended by these Guidelines wherever possible, is provided by the global id attribute, as defined in section . An extension to this mechanism is provided, for elements which are located in different SGML documents, or to which identifiers cannot be attached (perhaps because they are held on read-only media), known as the TEI extended pointer mechanism in section . For many of the topics discussed in this chapter, a choice of methods of encoding is offered, ranging from simple but less general ones, which use attribute values only, to more elaborate and more general ones, which use specialized elements.

The following DTD fragments show the overall organization of the additional tag set discussed in the remainder of this chapter. The file teilink2.ent begins by declaring a set of additional attributes available globally when this tag set is enabled. This is followed by declarations for the attribute classes pointer and pointerGroup to which most of the elements discussed in this chapter belong; these attributes are all further described in the remainder of the chapter. ]]>

The element declarations for this tag set are contained in the file teilink2.dtd: ]]>

This tag set is made available by the mechanisms described in section ; this implies that the document type subset for a document using any of the tags or attributes described in this chapter must define a parameter entity TEI.linking with the value INCLUDE. For example, a document using this additional tag set and the prose base would begin with a series of declarations like the following: ]> ]]> Pointers

We say that one element points to others if the first has an attribute whose value is a reference to the others: such an element is called a pointer element, or simply a pointer. Among the pointers that have been introduced up to this point in these Guidelines are note, ref and ptr. These elements all indicate an association between one place in the document (the location of the pointer itself) and one or more others (the elements whose identifiers are specified by the pointer's target attribute). This element set defines a variation on this basic kind of pointer, known as a link which specifies both ends of an association. In addition, we define a syntax for representing locations in a document by a variety of means not dependent on the use of SGML identifiers. Pointers and Links

In section we introduced the simplest pointer elements, ptr and ref. Here we introduce additionally the link element, which represents an association between two (or more) locations by specifying each location explicitly. Its own location is irrelevant to the intended linkage. defines a pointer to another location in the current document in terms of one or more identifiable elements. Attributes include: specifies the destination of the pointer as one or more SGML identifiers defines a reference to another location in the current document, in terms of one or more identifiable elements, possibly modified by additional text or comment. Attributes include: specifies the destination of the reference as one or more SGML identifiers defines an association or hypertextual link among elements or passages, of some type not more precisely specifiable by other elements. Attributes include: specifies the SGML identifiers of the elements or passages to be linked or associated. The ptr element may be called a pure pointer, because its primary function is simply to point. A pointer sets up a connection between an element (which, in the case of a pure pointer, can be thought of simply as a location in a document), and one or more others, known collectively as its target.

The ptr and ref elements bear a target attribute (in the singular), because they point, conceptually, at a single target, even if that target may be discontinuous in the document. The link element bears a targets attribute, with a plural name, because it specifies at least two targets, each of which is a unitary object. It may be thought of as a representing a double link between the objects specified.

As members of the class pointer, these elements share a common set of attributes: categorizes the pointer in some respect, using any convenient set of categories. specifies the creator of the pointer. specifies when the pointer was created. specifies the kinds of elements to which this pointer may point. where more than one identifier is supplied as the value of the target attribute, this attribute specifies whether the order in which they are supplied is significant. Legal values are: Yes: the order in which IDREFs are specified as the value of a target attribute should be followed when combining the targeted elements. No: the order in which IDREFs are specified as the value of a target attribute has no significance when combining the targeted elements. Unspecified: the order in which IDREFs are specified as the value of a target attribute may or may not be significant. specifies the intended meaning when the target of a pointer is itself a pointer. Sample values include: if the element pointed to is itself a pointer, then the target of that pointer will be taken, and so on, until an element is found which is not a pointer. if the element pointed to is itself a pointer, then its target (whether a pointer or not) is taken as the target of this pointer. no further evaluation of targets is carried out beyond that needed to find the element specified in the pointer's target.

The targType and targOrder attributes may be used to constrain the scope of a link to certain element types. For example: ]]> This is a complete unconstrained link, of type echo. It assumes only that there is an element with identifier P1 and another with identifier P2 somewhere in the current document. ]]> This is a slightly more constrained link of the same type. P1 and P2 must now both identify a p, a seg, or a note, but there is no requirement as to which is which. (This may be useful if, as is often the case, different elements may participate in the same kind of link.) ]]> In this variation, not only must the link targets be either p or note elements, but the one with identifier P1 must be a p, and that with identifier P2 must be a note. Note that the present Guidelines provide no direct way of saying that P1 may identify either a seg or a p and P2 must identify a note.

These attributes are most useful if applied to a group of links, when additional constraints may also be specified, as further discussed in section below.

Double connection among elements could also be expressed by a combination of pointer elements, for example, two ptr elements, or one ptr element and one note element. All that is required is that the value of the target (or other pointing) attribute of the one be the value of the id attribute of the other. What the link element accomplishes is the handling of double connection by means of a single element. Thus, in the following encoding: ..... ]]> P1 points to P2, and P2 points to P1. This is logically equivalent to the more compact encoding: ]]>

As noted above, all elements pointed to or linked by these elements must be identifiable using the global id attribute. This implies that they must be present in the same document, and that they must bear unique id values. Pointing or linking to external documents and pointing or linking where SGML identifiers are not available is implemented by the external pointing mechanisms discussed in section , where the xptr and xref elements are discussed. External links and links involving elements without identifiers do not require a special element; they may be represented using the standard link element, but an intermediate xptr element must be provided within the current document, to bear the id attribute used in the target of the link. Using Pointers and Links

As an example of the use of these mechanisms which establish connections among elements, consider the practice (common in 18th century English verse and elsewhere) of providing footnotes citing parallel passages from classical authors.

The figure shows the original page of Pope's Dunciad which is discussed in the text.
Such footnotes can of course simply be encoded using the note element (see section ) without a target attribute, placed adjacent to the passage to which the note refers:The type attribute on the note is used to classify the notes using the typology established in the Advertisement to the work: The Imitations of the Ancients are added, to gratify those who either never read, or may have forgotten them; together with some of the Parodies, and Allusions to the most excellent of the Moderns. In the source text, the text of the poem shares the page with two sets of notes, one headed Remarks and the other Imitations. (Diff'rent our parties, but with equal grace The Goddess smiles on Whig and Tory race, Virg. Æn. 10. Tros Rutulusve fuat; nullo discrimine habebo. —— Rex Jupiter omnibus idem. 'Tis the same rope at sev'ral ends they twist, To Dulness, Ridpath is as dear as Mist) ]]>

This use of the note element can be called implicit pointing (or implicit linking). It relies on the juxtaposition of the note to the text being commented on for the connection to be understood. If it is felt that the mere juxtaposition of the note to the text does not make it sufficiently clear exactly what text segment is being commented on (for example, is it the immediately preceding line, or the immediately preceding two lines, or what?), or if it is decided to place the note at some distance from the text, then the pointing or the linking must be made explicit. We now consider various methods for doing that.

First, a ptr element might be placed at an appropriate point within the text to link it with the annotation: (Diff'rent our parties, but with equal grace The Goddess smiles on Whig and Tory race, 'Tis the same rope at sev'ral ends they twist, To Dulness, Ridpath is as dear as Mist) Virg. Æn. 10. Tros Rutulusve fuat; nullo discrimine habebo. —— Rex Jupiter omnibus idem. ]]> The note element has been given an arbitrary identifier (N3.284) to enable it to be specified as the target of the pointer element. Because there is nothing in the text to signal the existence of the annotation, the rend attribute has been given the value unmarked.

Second, the target attribute of the note element can be used to point at its associated text, provided that an id attribute has been supplied for the associated text. Since, in this case, the note itself contains a pointer to the place in the text which it is annotating, this has also been encoded, using a ref element, which bears a target attribute of its own and contains a (slightly misquoted) extract from the text marked as a quote element: Verse 283-84. ——. With equal grace Our Goddess smiles on Whig and Tory race. Virg. Æn. 10. Tros Rutulusve fuat; nullo discrimine habebo. —— Rex Jupiter omnibus idem. ]]>

Combining these two solutions gives us the following associations: a pointer within one line indicates the note the note indicates the line a pointer within the note indicates the line Note that we do not have any way of pointing from the line itself to the note: the association is implied by containment of the pointer. We do not as yet have a true double link between text and note.

Thirdly, therefore, we supply identifiers for both verse line and annotation, and use a link element to associate the two. Note that the ptr element and the target attribute on the note may now be dispensed with: (Diff'rent our parties, but with equal grace The Goddess smiles on Whig and Tory race, 'Tis the same rope at sev'ral ends they twist, To Dulness, Ridpath is as dear as Mist) Verse 283-84. ——. With equal grace Our Goddess smiles on Whig and Tory race.] Virg. Æn. 10. Tros Rutulusve fuat; nullo discrimine habebo. —— Rex Jupiter omnibus idem. ]]>

The targets attribute of the link element here bears the identifiers of the note followed by that of the verse line. The targType and targOrder attributes may be used to enable application programs to check that the identifiers in fact pick out a note element and an l element and in that order. If targOrder has the value N, then the elements indicated by the targets attribute have to be either note or l elements, but are otherwise unconstrained. If neither attribute is present, then the only constraint is that the identifiers given must apply to some element within the current document.

For completeness, we could also allocate an identifier to the reference within the note and encode the association between it and the verse line in the same way: ]]> Indeed, the two links could be combined into one, as follows: ]]> Groups of Links

Clearly, there are many reasons for which an encoder might wish to represent a link or association between different elements. For some of them, specific elements are provided in these Guidelines; some of these are discussed elsewhere in the present chapter. The link element is a general purpose element which may be used for any kind of association. The element linkGrp may be used to group links of a particular type together in a single part of the document; such a collection may be used to represent what is sometimes referred to in the literature of Hypertext as a web, a term introduced by the Brown University FRESS project in 1969. defines a collection of associations or hypertextual links. As a member of the class pointerGroup, this element shares the following attributes with other members of that class: optionally specifies the identifiers of the elements within which all elements indicated by the contents of this element lie. describes the function of each of the values of the targets attribute of the enclosed link, join or alt tags. It is also a member of the pointer class, and therefore also carries the attributes specified in section above, in particular the type attribute: categorizes the pointer in some respect, using any convenient set of categories.

The linkGrp element provides a convenient way of establishing a default for the type attribute on a group of links of the same type: by default, the type attribute on a link element has the same value as that given for type on the enclosing linkGrp.

Typical software might hide a web entirely from the user, but use it as a source of information about links, which are displayed independently at their referenced locations. Alternatively, software might provide a direct view of the link collection, along with added functions for manipulating the collection, as by filtering, sorting, and so on. To continue our previous example, this text contains many other notes, of a kind similar to the one shown above. To avoid having to repeat the type=imitation on each note, we may specify it once for all on a linkGrp element containing all links of this type. The targType and targOrder attributes can also be specified for a linkGrp element: A place there is, betwixt earth, air and seas Where from Ambrosia, Jove retires for ease. ... Sign'd with that Ichor which from Gods distills. ... (Diff'rent our parties, but with equal grace The Goddess smiles on Whig and Tory race, 'Tis the same rope at sev'ral ends they twist, To Dulness, Ridpath is as dear as Mist) Ovid Met. 12. Orbe locus media est, inter terrasq; fretumq; Cœlestesq; plagas — Alludes to Homer, Iliad 5 Virg. Æn. 10. Tros Rutulusve fuat; nullo discrimine habebo. —— Rex Jupiter omnibus idem. ]]>

Additional information for applications that use linkGrp elements can be provided by means of special attributes. First, the domains attribute can be used to identify the text elements within which the individual targets of the links are to be found. Suppose that the text under discussion is organized into a body element, containing the text of the poem, and a back element containing the notes. Then the domains attribute can have as its value the identifiers of the body and the back, to enable an application to verify that the link targets are in fact contained by appropriate elements, or to limit its search space: A place there is, betwixt earth, air and seas Where from Ambrosia, Jove retires for ease. ... Sign'd with that Ichor which from Gods distills. ... (Diff'rent our parties, but with equal grace The Goddess smiles on Whig and Tory race, 'Tis the same rope at sev'ral ends they twist, To Dulness, Ridpath is as dear as Mist) Ovid Met. 12. Orbe locus media est, inter terrasq; fretumq; Cœlestesq; plagas — Alludes to Homer, Iliad 5 Virg. Æn. 10. Tros Rutulusve fuat; nullo discrimine habebo. —— Rex Jupiter omnibus idem. ]]>

Note that there must be a single parent element for each domain; if some notes are contained by a section with identifier dunnotes, and others by a section with identifier dunimits, an intermediate pointer must be provided (as described in section ) within the linkGrp and its identifier used instead.

Next, the targFunc attribute can be used to provide further information about the role or function of the various targets specified for each link in the group. The value of the targFunc attribute is a list of names (formally, SGML name tokens), one for each of the targets in the link; these names can be chosen freely by the encoder, but their significance should be documented in the encoding declaration in the header.No special element is provided for this purpose at present: the information should be supplied as a series of paragraphs at the end of the encodingDesc element described in section . In the current example, we might think of the note as containing the source of the imitation and the verse line as containing the goal of the imitation. Accordingly, we can specify the linkGrp in the preceding example thus: ]]>

The link and linkGrp elements are formally defined as follows: ]]> Intermediate Pointers

In the preceding examples, we have shown various ways of linking an annotation and a single verse line. However, the example cited in fact requires us to encode an association between the note and a pair of verse lines (lines 284 and 285).

There are a number of possible ways of correcting this error: one could use the target and targetEnd attributes of the note element to delimit the span to which the note applies (see further section ). Alternatively one could create an element to encode the couplet itself and assign it an id attribute, which can then be linked to the note and ref elements. This could be done either explicitly by means of a lg element, as defined in section , or a seg element, as defined in section , or implicitly, by means of the join element discussed in section .

A third possibility however, is to use an intermediate pointer as follows: (Diff'rent our parties, but with equal grace The Goddess smiles on Whig and Tory race, ]]> When the target attribute of a ptr or ref element specifies more than one element, the indicated elements are always understood to be combined or aggregated in some way to produce the object of the pointer.

In this example, the targOrder attribute should be specified to indicate that the order in which identifier values are supplied in the target attribute is significant. The id attribute provides an identifier which can then be linked to the note and ref elements: ]]>

The evaluate=all attribute value is used on the link element to specify that any pointer encountered as a target of that element is itself evaluated. If evaluate had the value none, the link target would be the pointer itself, rather than the objects it points to.

Where a linkGrp element is used to group a collection of link elements, any intermediate pointer elements used by those link elements should be included within the linkGrp.

Intermediate pointers of this kind are particularly important when extended pointers (discussed in the next section) are in use. Extended Pointers

Where the object of a link or pointer element is not contained within the current document, or where it does not bear an id attribute, it is not possible to point at it with a ptr or ref element, nor to link it directly with a link element, because no IDREF value can be supplied for the target or targets attribute of these elements. In such cases, the encoder must indicate the intended element indirectly by means of the elements discussed in this section. These elements identify their target using a special TEI-defined extended pointer notation, defined in section below and designed for compatibility with HyTime. HyTime is an international standard (ISO 10744) built on SGML. It provides facilities for representing both static and dynamic information for processing and interchange by hypertext and multimedia applications. See ISO/IEC 10744 Information Technology --- Hypermedia/Time-based Structuring Language (HyTime) ([Geneva]: International Organization for Standardization, 1992). Extended Pointer Elements

To point or refer to locations in the current or some other document without requiring that the target bear an SGML identifier, the following elements should be used: defines a pointer to another location in the current document or an external document. defines a reference to another location in the current document, or an external document, using an extended pointer notation, possibly modified by additional text or comment. These elements are both members of the element class pointer, and therefore carry the same attributes as other members of that class, listed above (see section ). They are also members of the class xPointer, from which they inherit the following attributes: specifies the document within which the desired location is to be found. specifies the start of the destination of the pointer, as an expression in the TEI extended-pointer notation described in section . specifies the endpoint of the destination of the pointer, as an expression in the TEI extended pointer notation.

Unlike the pointer elements discussed in the previous section, these elements do not specify their target by means of a target attribute. Instead these elements use one or both of the attributes from and to to delimit a portion of some document specified by the doc attribute. In all other respects, these elements correspond with the elements ptr and ref discussed in sections , and . Note that there is no element xlink corresponding with the link element; links can be made both within and between documents using the same syntax, as further discussed below.

The values of the from and to attributes on the xptr and xref elements indicate the point or passage being referred to by showing how to locate it, using one or more special keywords, as defined below in section . Examples are given there.

The xptr and xref elements are formally defined as follows: ]]> Extended Pointer Syntax

As noted above, the elements xptr and xref are used to represent a link between their own location (the link origin) and some other location (the destination), which may or may not be in the same document. Software supporting intra- and inter-document links (e.g. hypertext systems) should provide access from the location of such an element to the destination.

This section defines the allowable values for the attributes from, to, and doc of the xptr and xref elements.

An xptr or xref element with no attributes at all is, by definition, a link to the root (i.e. the document element --- by default, this is the TEI.2 element) of the document in which it appears.

The doc attribute value must be the name of an entity declared in the SGML document type declaration. If only the doc attribute is given a value, then by definition the destination is the entire entity named by the doc value. A more specific location within another entity must be specified with the from and the to attributes, as described below.

The from and the to attributes indicate the specific location pointed at, within the entity named by the doc attribute (or within the current document, if no doc value is given). Their values are referred to below as location pointer specifications. When both attributes are specified, the span pointed at by the element runs from the starting point of the span indicated by from to the ending point of the string specified by to. If the latter precedes the former in the document, then the pointer is in error and fails. If only the from attribute is specified, the to attribute defaults to the same value; the effect is that the element as a whole points to the span indicated by the from attribute. It is a semantic error to specify a value for to but not for from. Location Ladders

Each location pointer specification consists of a sequence of location terms, each of which consists of a keyword specifying a location type followed by one or more parenthesized parameter lists, each of which specifies a location value via a list of parameters. Location types and values, and the parameters within a location value, must be separated by white space characters.

Using terms borrowed from HyTime, we say that each TEI location term in a specification provides the location source for the next, and the entire specification is equivalent to a location ladder. By specifying the entire ladder in a single attribute value, the TEI extended pointer mechanism greatly reduces the syntactic and processing complexity of hypertextual pointers.

In formal terms: The notation used for this formal grammar is that defined in chapter . Location Terms

The keywords used in location terms are these; references to the tree mean the tree representing the SGML document hierarchy. points at the root of the target document points at the location of the pointer points at an ID within the target document gives a canonical reference to a location in the target document indicates an element found by descending one level in the tree indicates an element found by descending one or more levels in the tree indicates an element found by ascending one or more levels in the tree indicates an element found by traversing the older siblings of the current location source indicates an element found by traversing the younger siblings of the current location source indicates an element found by traversing the entire portion of the document preceding the current location source indicates an element found by traversing the entire portion of the document which follows the current location source specifies a regular expression to be located within the existing location source points at one or more tokens in the character content of the location source points at one or more characters in the character content of the location source points at a location using coordinates in some (application-defined) n-dimensional space points at a location using some non-SGML method, and gives the name of the method points at a location using the HyQ query language defined by ISO 10744 (HyTime) (in the to attribute only) points at the same span as was indicated by the from attribute

In formal terms: Note that the keywords, though shown here quoted in uppercase, are not case sensitive.

Each location term specifies a location in the target document; this location may be a single point, more often a span of text (often the span of a single element) within the target document. The location ladder as a whole is interpreted from left to right, and each location term specifies a location relative to the location specified by the sequence prior to that point (i.e. to its location source). Unless here or id is specified as the first location term, the beginning location source is always root. An empty location sequence thus is the same as root and specifies the entire destination entity.

In general, the search for the location specified by a location term will be conducted only within its location source (i.e. within the location already identified by preceding location terms). There are however several exceptions. The terms root, here, and id all ignore the location source defined by any preceding terms and therefore make sense only as the first items in the ladder. The terms ancestor, next, and previous do not ignore the location source, but select a new span from the adjacent or enclosing portions of the text, and not from within the location source. Finally the location terms foreign, space, and HyQ are not defined fully here; they may or may not ignore the existing location source.

Some of the location terms make sense only in SGML documents; these are id, child, ancestor, descendant, previous, next, preceding, and following. The latter six involve traversing the tree representing the SGML document hierarchy and are most easily understood when their location source is a single SGML element. If the location source is not a single SGML element, the tree-traversal keywords operate upon its beginning end-point, its front end (in English, this will be the leftmost point of the location source; in Arabic or Hebrew it will be the rightmost point). In this case child and descendant have no meaning, since character data has no descendants in the document tree; the first ancestor of such a location source is the element immediately containing the character data in question, and the siblings referred to by next and previous are the other children of that immediately containing element.

The details of each keyword are given below, along with definitions of their syntax and semantics of their results. Examples are also provided. It is strongly recommended that when IDs are available, they should be used in preference to the other methods for pointing defined here.

For all keywords, the description assumes that the target document does in fact contain a span or element which matches the description; otherwise, the location term has no referent and is said to fail. If any location term fails, the entire pointer fails. No backtracking or retrying is performed (and indeed for the most part the location terms are defined as having only one matching location, so backtracking would in most cases lead to no better result). The ROOT Keyword

The location term root selects the root of the destination document tree; in SGML terms, this is the document element. Since it ignores any existing location source, the root keyword makes sense only as the first location term in the ladder. Since root is assumed as the implicit first term in any ladder, the following two location ladders have the same meaning: The HERE Keyword

The keyword here designates the location at which the pointer element itself is situated; it allows extended pointers to select items like the paragraph immediately preceding the one within which this pointer occurs. Since it ignores any existing location source, this keyword typically makes sense only as the first location term in a location specification.

To designate the paragraph preceding the current one, the following location ladder could be used: HERE ANCESTOR (1 P) PREVIOUS (1 P) (See below for descriptions of the keywords ancestor and previous.) The ID Keyword

The resulting location is the element within the destination entity whose ID attribute has the value specified as the location value. The ID location type typically makes sense only as the first location pair in a location specification, but there is no syntactic requirement that it be so.

For example, the location specification ID (a27) chooses the necessarily unique element of the destination entity which has an attribute of declared value of type ID, whose value is a27. The REF Keyword

The resulting location is an element which can be found by interpreting the location value in accordance with document-specific rules for a canonical reference. Such reference systems, particularly common in documents of interest to classical and biblical scholars, must also be defined in the TEI header, using the refsDecl element (see section ). If more than one element matches the canonical reference, the first one encountered is chosen.

For example, the location specification REF (MT.2.1) chooses the first element of the destination entity which is identified by the canonical reference MT.2.1 The CHILD Keyword

The child location type specifies an element or span of character data in the document hierarchy using a location value which functions as a domain-style address. The value is a series of parenthesized steps, separated by white space. Each such step represents one level of the hierarchy within the location source. Each step may contain one or more parameters separated by white space and interpreted in order as follows: an instance indicator, which is a signed or unsigned integer or the special value ALL optionally, an expression matching an SGML generic identifier optionally, one or more pairs of expressions, the first matching an SGML attribute name and the second matching an SGML attribute value

In formal terms, the location value of child is a series of steps:

Location values of the same form are also used by the keywords descendant, ancestor, previous, and next; details of the interpretation may vary from keyword to keyword.

If an instance indicator alone is specified, as a number n, it selects the nth child of the location source. If the special value ALL is given, then all the children of the location source are selected. If the instance indicator is specified with following parameters, it selects all, or the nth, among those children of the location source which satisfy the other parameters. If a negative number is given, the nth child is counted from the last child of the location source to the first. The location source must contain at least n children;Strictly speaking, |n| (absolute value of n) children. if it does not, the child term fails.

In formal terms, the first parameter of a step is an instance indicator, which in turn is either the special value ALL or a signed integer:

If a second parameter is given, it is interpreted as an SGML generic identifier, and only elements of the type indicated will be selected. For example, the location specification CHILD (3 DIV1) (4 DIV2) (29 P) chooses the 29th paragraph of the fourth sub-division of the third major division of the initial location source. The location specification CHILD (3 DIV1) (4 DIV2) (-2 P) chooses the next-to-last paragraph of the fourth div2 of the third div1 in the location source.

Constraint by generic identifier is strongly recommended, because it makes links more perspicuous and more robust. It is perspicuous because humans typically refer to things by type: as the second section, the third paragraph, etc. It is robust because it increases the chance of detecting breakage if (due to document editing) the target originally pointed at no longer exists.

The generic identifier may be specified as a normal SGML name, as a (parenthesized) regular expression, or using the reserved values #CDATA or *. Regular expressions take the form described below; the location term CHILD (3 (DIV[123]) matches the third element which has a generic identifier of div1, div2, or div3. If the generic identifier is specified as *, any generic identifier is matched; this means that CHILD (2 *) is synonymous with CHILD (2). If the second parameter is #CDATA, the location term selects only untagged sub-portions of an element having SGML mixed content.

The location ladder CHILD (3 #CDATA) thus chooses the third span of character data directly contained by the current location source. If the location source is a paragraph containing a sentence (A) an embedded quotation, marked as a q another sentence (B) an embedded note, marked as a note another sentence (C) a second embedded quotation, marked as a q where the three sentences A, B, and C are character data enclosed by no element smaller than the paragraph itself, then CHILD (3 #CDATA) selects sentence C, while CHILD (3) selects sentence B.

If specified as a name (i.e. without parentheses), the generic identifier is case sensitive if and only if the SGML declaration specifies that generic identifiers are case sensitive (by default they are not). If specified as a regular expression, the expression given is always case sensitive; in the usual case this means the regular expression should be in uppercase, as in the examples here.

In formal terms the second parameter of a step is defined thus:

The third and fourth parameters, if given, are interpreted as an attribute-value pair, and only elements which match that pair in the way described below will be selected; the fourth and fifth parameters, and all following pairs of parameters, are interpreted in the same way. When more than one pair is given, all must be matched.

The third, fifth, seventh, etc., parameters are interpreted, if specified, as attribute names. Like generic identifiers, attribute names may be specified as * in location ladders in the (unlikely) event that an attribute value constitutes a constraint regardless of what attribute name it is a value for. The attribute name parameter may also be specified as a parenthesized regular expression.

For example, the location term CHILD (1 * TARGET *) selects the first child of the location source for which the attribute target has a value. The location term CHILD (1 * (TARGET(S?)) *) will select the first child of the location source for which an attribute called either target or targets has a value.

As with generic identifiers, attribute names are case sensitive if and only if the SGML declaration says they are; regular expressions are always case sensitive and should usually be uppercased, as shown here.

In formal terms, the attribute-name parameter of a tree-traversal step is defined thus:

If a fourth, sixth, eighth, etc., parameter is specified, it is interpreted as an attribute value, and only elements satisfying the other constraints and also bearing an attribute of the specified name and value will be selected. The attribute value may be specified exactly as in an SGML document; as a consequence, if the attribute value to be specified contains white space characters, it must be enclosed in quotation marks. The attribute value may also be specified as a regular expression, enclosed in parentheses, or using the two special values #IMPLIED and *.

For example, the location specification CHILD (1 * N 2) (1 * N 1) chooses an element using the global n attribute. Beginning at the location source, the first child (whatever kind of element it is) with an n attribute having the value 2 is chosen; then that element's first direct sub-element having the value 1 for the same attribute is chosen.

The location specification CHILD (1 FS RESP ((lanc|LANC)(s|S|ashire|ASHIRE))) selects the first child of the location source which is an fs element bearing a resp attribute with the value lancs, lancashire, LANCS, or LANCASHIRE (as well as other possible combinations which are left to the reader's ingenuity). If specified with quotation marks or as a regular expression, the attribute-value parameter is case-sensitive; otherwise not.

The location specification CHILD (1 FS RESP #IMPLIED) selects the first child of the location source which is an fs element for which the resp attribute has been left unspecified. The location ladder ROOT DESCENDANT (1 (DIV[01234567]) TYPE chapter N 2) selects the second chapter of a text, regardless of whether chapters are tagged using div, div1, div2, or some other text-division element. It does so by selecting the first text-division element in the document which is of type chapter and has the n value 2.

In formal terms, the attribute-value parameter of a tree-traversal step is defined thus: The DESCENDANT Keyword

If the descendant keyword is used, the location term selects an element or character-data string which is a descendant of the current location source. Like child, descendant takes as a value a series of one or more parenthesized steps, which may contain the same four parameters described above. The set of elements and strings which may be selected, however, is the set of all descendants of the location source (i.e. the set of all elements contained by it), rather than only the set of immediate children.

The location specification ID (a23) DESCENDANT (2 TERM LANG DE) thus selects the second term element with a lang of de occurring within the element with an id of a23. The search for matching elements occurs in the same order as the SGML data stream; in terms of the document tree, this amounts to a depth-first left-to-right search.

If the instance number is negative, the search is a depth-first right-to-left search, in which the right-most, deepest matching element is numbered -1, etc. The location specification DESCENDANT (-1 NOTE) thus chooses the last note element in the document, that is, the one with the rightmost start-tag. The ANCESTOR Keyword

The ancestor location term selects an element from among the direct ancestors of the location source in the document hierarchy. The location value is of the same form as defined for the child and descendant location types. However, the ancestor keyword selects elements from the list of containing elements or ancestors of the location source, counting upwards from the parent of the location source (which is ancestor number 1) to the root of the document instance (which is ancestor number -1).

The location source must have at least as many ancestors as the absolute value of the instance number specified as the first parameter of the step. The ancestor type thus may not be specified as the first component of a location specification, because the initial location source in effect at that point is the root, which has no ancestors.

For example, the location term ANCESTOR (1 * N 1) (1 DIV) first chooses the smallest element properly containing the location source and having attribute n with value 1; and then the smallest div element properly containing it. The location term ANCESTOR (1) chooses the immediate parent of the location source, regardless of its type or attributes. The location term ANCESTOR (1 * LANG fr) selects the smallest ancestor for which the lang attribute has the value fr. The term ANCESTOR (-1 * LANG fr) selects the largest ancestor for which the lang attribute has the value fr. Without the attribute specification, the term ANCESTOR (-1) selects the largest ancestor of the location source and is thus normally synonymous with the keyword ROOT. If the instance indicator is given as ALL, then all the ancestor elements which match the later parameters are selected; since the largest of these will necessarily include all the others, the value ALL is thus synonymous with the value (-1) when used with ANCESTOR. Finally, the term ANCESTOR (1 (DIV[0123456789]?)) chooses the smallest div element of any level which contains the location source. The PREVIOUS Keyword

The previous keyword selects an element or character-data string from among those which precede the location source within the same containing element. We speak of the elements and character-data strings contained by the same parent element as siblings; those which precede a given element or string in the document are its elder siblings; those which follow it are its younger siblings.

The instance number in the location value of a previous term designates the nth elder sibling of the location source, counting from most recent to less recent. The location ladder ID (a23) PREVIOUS (1) thus designates the element immediately preceding the element with an id of a23. Negative instance numbers also designate elder siblings, counting from the eldest sibling to the youngest. The location source must have at least as many elder siblings as the absolute value of the instance number. If the location source has at least one elder sibling, then the location term PREVIOUS (-1) designates its eldest sibling and is thus synonymous with the ladder ANCESTOR (1) CHILD (1) The value ALL may be used to select the entire range of elder siblings of an element: the location ladder ID (a23) PREVIOUS (ALL) thus designates the set of elements which precede the element with an id of a23 and are contained by the same parent. The NEXT Keyword

The keyword next behaves like previous, but selects from the younger siblings of the location source, not the elder siblings. The location ladder ID (a23) NEXT (1) thus designates the element or string immediately following the element which has an id of a23. Negative instance numbers also designate younger siblings, counting from the youngest sibling to the location source. The location source must have at least as many younger siblings as the absolute value of the instance number. If the location source has at least one younger sibling, then the location term NEXT (-1) designates its youngest sibling and is thus synonymous with the ladder ANCESTOR (1) CHILD (-1) The PRECEDING Keyword

The preceding keyword selects an element or character-data string from among those which precede the location source, without being limited to the same containing element. The set of elements and strings which may be selected is the set of all elements and strings in the entire document which occur or begin before the location source. (For purposes of the keywords PRECEDING and FOLLOWING, elements are interpreted as occurring where their start-tag occurs.) The PRECEDING keyword thus resembles PREVIOUS but differs in searching a larger set of strings and elements; its result is not guaranteed to be a subset of its location source.

The instance number in the location value of a preceding term designates the nth element or character-data string preceding the location source, counting from most recent to less recent. The location ladder ID (a23) PRECEDING (5) thus designates the fifth element or string before the element with an id of a23. Negative instance numbers also designate preceding elements or strings, counting from the eldest to the youngest; the ladder ID (a23) PRECEDING (-5) thus selects the fifth element or string in the document overall, assuming that it precedes the element with an id of a23. It is thus normally synonymous with ROOT DESCENDANT (5) differing only in that it fails if four items or fewer precede element A23. The location source must have at least as many elder siblings as the absolute value of the instance number; otherwise, the preceding term fails. The value ALL may be used to select the entire portion of the document preceding the beginning of the location source: the location ladder ID (a23) PRECEDING (ALL) designates the entire portion of the document preceding the start-tag for element A23. The FOLLOWING Keyword

The keyword following behaves like preceding, but selects from the portion of the document following the location source, not that preceding it. The location ladder ID (a23) FOLLOWING (1) thus designates the element or string immediately following the element which has an id of a23. Negative instance numbers select elements or strings counting from the end of the document to the location source. There must be at least as many elements or strings following the location source as the absolute value of the instance number. If the location source has at least one following element or string, then the location term FOLLOWING (-1) designates the youngest of these and is thus synonymous with the ladder ROOT DESCENDANT (-1) The PATTERN Keyword

The pattern keyword selects the first place within the location source which matches a pattern-matching expression included as the location value. If more than one location matches that expression, there is no error, but the second and later matches are ignored.

Matching is defined to be case-sensitive, i.e. abc is not the same as ABC. The pattern is expressed as a regular expression in which the following characters have special meanings, similar to those of many Unix programs (such as grep) which handle regular expressions: match any single character (including white space characters). match any character from the set enclosed within the brackets. If, however, the first enclosed character is ˆ, then match any character not from the set enclosed within the brackets. For example, [ˆaeiou] would match any character except a, e, i, o, or u. If the next character is a, d, n, or s, the expression matches any character from a pre-defined group, as shown below; otherwise, the next character is to be taken literally, even if it would otherwise have a special meaning. The special character classes are: Note that although \n for newline is provided, its use is discouraged. match zero or more occurrences of the previous regular expression. match one or more occurrences of the preceding regular expression. match zero or one occurrences of the preceding regular expression. match the following regular expression only at the beginning of the location source. match the preceding regular expression only at the end of the location source. match either the regular expression on the left, or the one on the right. match the regular expression within the parentheses. (Parentheses are used to control application of the *, ?, +, and | operators, etc.)

For example, the location specification PATTERN (Chapter.8) chooses the first instance of the content string Chapter which is followed by any single character and then the digit 8, within the location source. Various elements which contain that location could be selected by following the pattern location term with one or more of other types such as ancestor (see above).

It is recommended practice to use structure-oriented location types to specify the destination element as narrowly as possible, and then to specify a pattern only within that element context. If element boundaries are encountered within the location source, however, they are ignored and have no effect on the pattern matching operation.

In formal terms, the location value of the pattern keyword is defined thus: The TOKEN Keyword

The token keyword selects a sequence of one or more tokens chosen from within the character content of the location source, where tokens are counted exactly as for the corresponding HyTime tokenloc form. The location value must be either a single positive integer, or a pair of positive integers separated by white space, representing the first and the last token numbers to be included in the resulting location. If two integers are specified, the second must not be less than the first. The location source must contain at least as many tokens as are specified in the location value.

This location type should not be used to count across element boundaries. It is recommended practice to use structure-oriented location types to specify the destination element as narrowly as possible, and then to specify a token location only within that element context. If element boundaries are encountered within the location source, they are ignored.

This location type behaves intuitively only for strings containing an alternating sequence of SGML name-characters and white space; this is the type of string found, for example, in SGML attribute values of type IDREFS, such as a21 z a13. For compatibility with the HyTime standard, all characters not included in the class of name characters by the current SGML declaration (by default this includes all punctuation other than the hyphen and full stop) are treated as white space characters.

For example, the location specification ID (a27) TOKEN (3 5) chooses the 3rd, 4th, and 5th tokens from the content of the element whose identifier is a27. If this element contained the string This is _not_ a very good idea, the target selected would be not_ a very.

In formal terms the location value of the token and str keywords is defined as a range: The STR Keyword

The str keyword identifies a sequence of one or more characters chosen from within the character content of the location source, where characters are counted exactly as for the HyTime dataloc form with quantum=str, which has a corresponding meaning and usage. The location value must be either a single positive integer, or a pair of positive integers separated by white space, indicating the first and the last characters to be included in the resulting location. If two integers are specified, the second must not be less than the first. The location source must have at least as many characters as are specified in the larger of the integers.

This location type should not be used to count across element boundaries. The recommended practice is to use structure-oriented location types to specify the destination element, and then to specify a character location only within that element context. If element boundaries are encountered, however, within the location source, they have no effect.

Character offsets in an SGML document must be counted not from the original source file, but from the output of the SGML parser, (the element structure information set or ESIS). This is because the rules of SGML allow certain characters to be deleted or expanded transparently.

For example, the location specification ID (a27) STRLOC (3 5) chooses the 3rd 4th and 5th characters of the content of the element having identifier a27. If this element contained the string This turned out to be an even worse idea, the result would be the string is (i, s and a space).

In multi-byte character sets it is characters which are counted, not bytes. However, in the case of diacritics coded by sequences of bit combinations rather than having separate code points for every combination of letter and diacritic, the diacritics are counted. This means that the following location ladder may retrieve different strings, depending on the system character set in use and on the entity declarations in effect: In some character sets, where ö and ä are encoded as single characters, it will select the string Götterdämmerung; in others, where they are encoded with distinct characters for umlaut, a, and o, it will select the string Götterdämmeru, truncating the last two letters. If a system-dependent definition is used (containing e.g. a printer escape sequence), the results are even less predictable. For this reason, the str keyword must be used with caution and should be avoided where possible. The SPACE Keyword

The space location term applies to entities which represent graphical or spatio-temporal data; typically such entities are not encoded in SGML, but in one of many specialized graphical formats. SGML provides standard mechanisms (the NOTATION declaration and related constructs) for specifying what format such an entity uses.

The location value for space consists of two or three parenthesized parameter lists. The first contains the name of the co-ordinate space in use. The second and third each consist of any number of signed integers. The numbers in a parameter list represent locations along each dimension of a Cartesian co-ordinate space with all axes orthogonal; the length of the list equals the number of dimensions/axes of the space (usually, but not inevitably, 2, 3, or 4).

If the third parameter list is not specified, the location is the single point in the co-ordinate space specified by the second parameter list. If all three parameter lists are specified, the location is the rectangular prism defined by treating corresponding items of the second and third lists as inclusive bounds along each dimension in turn.

The mapping from co-ordinates to physical or display space, and the meaning and ordering of the axes, are not defined by these guidelines. They should be specified in the TEI header unless they can be determined by definition from the format in which the referenced entity is known to be encoded (for example, many graphics formats can only encode locations in units of pixels, counted in a 3 dimensional left-handed co-ordinate space).

Time may be construed as an axis in addition to any others; when it is, it is TEI recommended practice that it be positioned last. The units used must be defined in the TEI header; it is acceptable in certain media (such as videodiscs) to use frame numbers as a surrogate axis for time.

For example, SPACE (2D) (0 0) (1 1) specifies the location of the unit square tangent to the origin in quadrant 1 of a common graph.

The location value for a space location term is a NAME enclosed in parentheses, followed by a point pair: The FOREIGN Keyword

The foreign keyword takes any number of parenthesized parameter lists, and is terminated by the end of the attribute value, or by the next non-parenthesized token, whichever comes first.

The meaning of the foreign location term is not defined by these Guidelines. It is intended for use in pointing to special kinds of non-SGML, non-coordinate space data. That is, it should be used for making links to data which cannot be specified using the other mechanisms. The meaning of any foreign location types must be specified in the TEI header, as a series of paragraphs at the end of the encodingDesc element defined in section . If more than one such type is used, it is TEI recommended practice that the first parameter list to foreign be a name associated with the particular type by documentation in the TEI header.

For example, assume that some program uses a proprietary data format called XFORM, and that the program has supplied an identifier 06286208998 for some piece of data it owns. Then the location specification FOREIGN (XFORM) (06286208998) would be one way of expressing a link to that piece of data. The HYQ Keyword

The HyQ keyword takes a single parenthesized parameter lists, which contains an expression in the HyQ query language defined by the HyTime standard. See documentation on HyTime and HyQ for definitions of HyQ expressions. The DITTO Keyword

The ditto keyword is valid only as the first location term in a ladder, and only within the to attribute of an extended pointer element. It designates the location result of the from attribute on the same element. Thus in the pointer ]]> the from attribute designates the first occurrence of the string Wagnerian in the div containing the element with an id of a23. The to attribute designates the first occurrence of the string Liebestod which occurs after Wagnerian, within the same div. Without the ditto keyword, it would be necessary to repeat the entire location ladder of the from attribute in the to attribute, which would be error-prone for complex expressions. Using Extended Pointers

As noted above, when only the from attribute is specified, the xref or xptr element points at the span indicated by from. When both from and to are specified, the element points at the span running from the beginning of the span indicated by the former to the end of the span indicated by the latter. To point at the second, third, and fourth paragraphs of the second chapter (div1) in the body of the current document, therefore, one may specify either of the following: ]]>

To point to the term occurring in the current termEntry with attribute n = 2, only the from attribute would be required: ]]>

The following example demonstrates how elements from two different documents may be combined ]]> The first xptr indicates the element in doc1 which has identifier d1.1. The second indicates the second subelement of the element in doc2 which has identifier d2.1. These two elements are pointed to as a single item by the ptr element and given the identifier p1. This aggregation, finally, is linked with two other elements both in the current document, with identifiers s1 and s2.

An extended pointer, as described above, may specify as its target only a single destination. Where the intended destination of a link is an aggregation or alignment of destinations, possibly in separate documents, an intermediate pointer of some kind must be used, as described in section elsewhere in this chapter. Like any other element, an xref and xptr may be given a unique id within the document that contains them. This id value can then be supplied as one of the target values for an intermediate ptr or link element, to represent aggregation or linkage respectively. The join element discussed in section may also be used.

For example, a modern commentary on an older text must frequently refer to that text, which might well be encoded in a separate SGML document. Some discussions will refer to set of discrete passages in the older text, and will thus require multi-headed pointers. In such a case, the document type declaration must contain a declaration for an SGML entity containing the older text, which might look something like this: ]]> In the commentary itself, reference will be made to this external document, using xptr and xref elements. When the commentary refers to aggregates of discontiguous passages, xptr elements are used to point to the individual passage, and a ref element may refer to these passages as a group by pointing to the xptrs: ...

In the references to Theobald, Pope's satire characteristically ...

]]> If the same discontiguous target is to be referred to repeatedly, it may be convenient to give it a single identifier, thus: ...

In the references to Theobald, Pope's satire characteristically ...

]]>

A hypertext web might associate passages of the text and notes with the individuals mentioned, the ancient authors imitated, or thematic content, thus: ...

Individuals Named in the Text A bookseller and publisher ... ... Attorney, active also as editor and reviewer ... ... ...
Ancient Authors Imitated in the Text Virgil Homer Ovid ... ... ... ... ]]> Segments and Anchors

In this section, we define two general purposes elements which may be used to mark and categorize both a span of text and a point within one. These elements have several uses, most notably to provide elements which can be given identifiers for use when aligning or linking to parts of a document, as discussed elsewhere in this chapter. They also provide a convenient way of extending the semantics of the TEI markup scheme in a theory-neutral manner. contains any arbitrary phrase-level unit of text (including other seg elements). Attributes include: provides a sub-categorization of the segment is marked. specifies whether or not the segment is complete. Legal values are: the segment is incomplete either the segment is complete, or no claim is made as to its completeness the initial part of an incomplete segment a medial part of an incomplete segment the final part of an incomplete segment attaches an identifier to a point within a text, whether or not it corresponds with a textual element. These elements are both members of the class seg, and inherit from it the attribute type: characterizes the type of segment. characterizes the function of the segment.

The seg element may be used at the encoder's discretion to mark almost any segment of the text of interest for processing. One use of the element is to mark text features for which no appropriate markup is otherwise defined, i.e. as a simple extension mechanism. Another use is to provide an identifier for some segment which is to be pointed at by some other element, i.e. to provide a target, or a part of a target, for a ptr or other similar element.

Several examples of uses for the seg element are provided elsewhere in these Guidelines. For example: as a means of marking segments significant in a metrical or rhyming analysis (see section ) as a means of marking typographic lines in drama (see section ) or title pages (see section ) as a means of marking prosody- or pause-defined units in transcribed speech (see section ) as a means of marking linguistic or other analyses in a theory-neutral manner (see chapter passim)

In the following simple example, the seg element simply delimits the extent of a stutter, a textual feature for which no element is provided in these Guidelines. Don't say I-I-I'm afraid, Melvin, just say I'm afraid. ]]> The seg element is particularly useful for the mark-up of linguistically significant constituents such as the phrases that may be the output of an automatic parsing system. This example also demonstrates the use of the id attribute to carry an identifier which other parts of a document may use to point to, or align with: Literate and illiterate speech in a language like English are plainly different. ]]>

As the above example shows, seg elements may be nested directly within one another, to any degree of analysis considered appropriate. This is taken a little further in the following example, where the type and subtype attributes have been used to further categorise each word of the sentence (the id attributes have been removed to reduce the complexity of the example): Literate and illiterate speech in a language like English are plainly different . ]]>

(The example values shown are chosen for simplicity of comprehension, rather than verisimilitude). It should also be noted that specialized segment elements are defined in section to facilitate this particular kind of analysis. These allow for the explicit mark up of units called s-units, clauses, phrases, words, morphemes and characters, which may be felt preferable to the more generic approach typified by use of the seg element. Using these, the first phrase above might be encoded simply as Literate and illiterate speech ]]> Note the way in which the type attribute of these specialized elements now carries the value carried by the subtype attribute of the more general seg element. For an analysis not using these traditional linguistic categories however, the seg element provides a simple but powerful mechanism.

In language corpora and similar material, the seg element may be used to provide an end-to-end segmentation as an alternative to the more specific s element proposed in chapter for the mark-up of orthographic sentences, or s-units. However, it may be more useful to use the s element for this purpose, since this means that the seg element can then be used to mark both features within s-units and segments composed of s-units, as in the following example: See section , where the text from which this fragment is taken is analyzed. Sigmund, the son of Volsung, was a king in Frankish country. Sinfiotli was the eldest of his sons. ]]>

Like other elements, the seg tag must be properly enclosed within other elements. Thus, a single seg element can be used to group together words in different sentences only if the sentences are not themselves tagged. The first of the following two encodings is legal, but the second is not. a dozen. Or two or three. Give me a dozen. Or two or three. ]]>

The part attribute may be used as one simple method of overcoming this restriction: Give me a dozen. Or two or three. ]]> Another solution is to use the join element discussed in section ; this requires that each of the seg elements be given an identifier. For further discussion of this generic encoding problem see also chapter .

The seg element has the same content as a paragraph in prose: it can therefore be used to group together consecutive sequences of inter class elements, such as lists, quotations, notes, stage directions etc. as well as to contain sequences of phrase-level elements. It cannot however be used to group together sequences of paragraphs or similar text units such as verse lines; for this purpose, the encoder should use intermediate pointers, as described in section or the methods described in section . It is particularly important that the encoder provide a clear description of the principles by which a text has been segmented, and the way in which that segmentation is represented. This should include a description of the method used and the significance of any categorization codes. The description should be provided as a series of paragraphs within the segmentation element of the encoding description in the TEI header, as described in section .

The remainder of this chapter contains a number of examples of the use of the seg element simply to provide an element to which an identifier may be attached, for example so that another segment may be linked or related to it in some way. We conclude this section by introducing the anchor element which serves an identical purpose, but has no content. It may be thought of as an empty seg, or as an artifice enabling an identifier to be attached to any position in a text.

Like the milestone element discussed in section , the anchor element is useful where multiple views of a document are to be combined, for example, when a logical view based on paragraphs or verse lines is to be mapped on to a physical view based on manuscript lines. It differs from the milestone and related elements in that the anchor element should not be used to mark the start or end of an arbitrary zone within a text, but only to mark arbitrary points used for alignment.

For example, suppose that we wish to mark the end of the fifth word following each occurrence of some term in a particular text, perhaps to assist with some collocational analysis. This can most easily be done with the help of the anchor tag, as follows: English language. Except for not very English at all at the time English was still full of flaws English. This was revised by young ]]> In the next section we discuss ways in which these anchor points can be used to represent an alignment, for example such as one might get in a keyword-in-context concordance.

These elements are formally defined as follows: ]]> Correspondence and Alignment

In this section we introduce the notions of correspondence, expressed by the corresp attribute, and of alignment, which is a special kind of correspondence involving an ordered set of correspondences. Both cases may be represented using the link and linkGrp elements introduced in section . We also discuss the special case of alignment in time or synchronization, for which special purpose elements are proposed in section . Correspondence

A common problem in text analysis is to determine correspondences between two or more parts of a single document, or between places in different documents. Provided that SGML elements are available to represent the parts or places to be linked, then the global linking attribute corresp may be used to encode such correspondence, once it has been identified. points to elements that correspond to the current element in some way. This is one of the attributes made available by the mechanism described in the introduction to this chapter (). Correspondence can also be expressed by means of the link element introduced in section .

Where the correspondence is between spans, the seg element should be used, if no other element is available. Where the correspondence is between points, the anchor element should be used, if no other element is available.

The use of the corresp attribute with spans of content is illustrated by the following example: Shirley, which made its Friday night debut only a month ago, was not listed on NBC's new schedule, although the network says the show still is being considered. ]]> Here the anaphoric phrases the network and the show have been associated directly with the elements to which they refer by means of corresp attributes. This mechanism is simple to apply, but has the drawback that it is not possible to specify more exactly what kind of correspondence is intended. Where this attribute is used, therefore, encoders are encouraged to specify their intent in the associated encoding declarations in the TEI Header.

Essentially, what the corresp attribute does is to specify that the element that has the attribute and the element(s) the attribute points to are doubly linked. The corresp attribute is thus distinct from the target attribute in that it is understood to create a double, rather than a single, link. It is also distinct from the targets attribute in that the latter lists all the identifiers of the elements that are doubly linked, whereas the corresp doubly links the element that bears the attribute with the element(s) that make up the value of the attribute. Therefore, we can also use the link and linkGrp elements defined in section to indicate correspondence among elements. Moreover, the use of these elements provides a convenient place to indicate what kind of correspondence is intended as in the following retagging of the preceding example. Shirley, which made its Friday night debut only a month ago, was not listed on NBC's new schedule, although the network says the show still is being considered. ]]>

In the following example, we use exactly the same mechanism to express a correspondence amongst the anchors introduced following the fifth word after English in a text: English at all at the time English was still full of flaws English. This was revised by young ]]> Alignment of Parallel Texts

One very important application area for the alignment of parallel texts is multilingual corpora. Consider, for example, the need to align translation pairs of sentences drawn from a corpus such as the Canadian Hansard, in which each sentence is given in both English and French. Concerning this problem, Gale and Church write:See William A. Gale and Kenneth W. Church, Program for aligning sentences in bilingual corpora, Computational Linguistics 19 (1993): 75-102, from which the example in the text is taken. Most English sentences match exactly one French sentence, but it is possible for an English sentence to match two or more French sentences. The first two English sentences [in the example below] illustrate a particularly hard case where two English sentences align to two French sentences. No smaller alignments are possible because the clause ...sales...were higher... in the first English sentence corresponds to (part of) the second French sentence. The next two alignments ... illustrate the more typical case where one English sentence aligns with exactly one French sentence. The final alignment matches two English sentences to a single French sentence. These alignments [which were produced by a computer program] agreed with the results produced by a human judge.

The alignment produced by Gale and Church's program can be expressed in four different ways. The encoder must first decide whether to represent the alignment in terms of points within each text (using the anchor element) or in terms of whole stretches of text, using the seg element. To some extent the choice will depend on the process by which the software works out where alignment occurs, and the intention of the encoder. Secondly, the encoder may elect to represent the actual encoding using either corresp attributes attached to the individual anchor or seg elements, or using a free standing linkGrp element.

We present first a solution using anchor elements bearing only corresp attributes:

According to our survey, 1988 sales of mineral water and soft drinks were much higher than in 1987, reflecting the growing popularity of these products. Cola drink manufacturers in particular achieved above-average growth rates. The higher turnover was largely due to an increase in the sales volume. Employment and investment levels also climbed. Following a two-year transitional period, the new Foodstuffs Ordinance for Mineral Water came into effect on April 1, 1988. Specifically, it contains more stringent requirements regarding quality consistency and purity guarantees.

Quant aux eaux minérales et aux limonades, elles rencontrent toujours plus d'adeptes. En effet, notre sondage fait ressortir des ventes nettement supérieures à celles de 1987, pour les boissons à base de cola notamment. La progression des chiffres d'affaires résulte en grande partie de l'accroissement du volume des ventes. L'emploi et les investissements ont également augmenté. La nouvelle ordonnance fédérale sur les denrées alimentaires concernant entre autres les eaux minérales, entrée en vigueur le 1er avril 1988 après une période transitoire de deux ans, exige surtout une plus grande constance dans la qualité et une garantie de la pureté. ]]>

There is no requirement that the corresp attribute be specified in both English and French texts, since (as noted above) this attribute is defined as representing a mutual association. However, it may simplify processing to do so, and also avoids giving the impression that the English is translating the French, or vice versa. More seriously, this encoding does not make explicit the fact that it is in fact the entire stretch of text between the anchors which is being aligned, not simply the points themselves. If for example one text contained material omitted from the other, this approach would not be appropriate.

We now present the same passage using the alternative linkGrp mechanism and marking explicitly the segments which have been aligned:

According to our survey, 1988 sales of mineral water and soft drinks were much higher than in 1987, reflecting the growing popularity of these products. Cola drink manufacturers in particular achieved above-average growth rates. The higher turnover was largely due to an increase in the sales volume. Employment and investment levels also climbed. Following a two-year transitional period, the new Foodstuffs Ordinance for Mineral Water came into effect on April 1, 1988. Specifically, it contains more stringent requirements regarding quality consistency and purity guarantees.

Quant aux eaux minérales et aux limonades, elles rencontrent toujours plus d'adeptes. En effet, notre sondage fait ressortir des ventes nettement supérieures à celles de 1987, pour les boissons à base de cola notamment. La progression des chiffres d'affaires résulte en grande partie de l'accroissement du volume des ventes. L'emploi et les investissements ont également augmenté. La nouvelle ordonnance fédérale sur les denrées alimentaires concernant entre autres les eaux minérales, entrée en vigueur le 1er avril 1988 après une période transitoire de deux ans, exige surtout une plus grande constance dans la qualité et une garantie de la pureté. ]]>

Note that use of the seg element allows us to mark up the orthographic sentences in both languages independently of the alignment: the first translation pair in this example might be marked up as follows: According to our survey, 1988 sales of mineral water and soft drinks were much higher than in 1987, reflecting the growing popularity of these products. Cola drink manufacturers in particular achieved above-average growth rates. Quant aux eaux minérales et aux limonades, elles rencontrent toujours plus d'adeptes. En effet, notre sondage fait ressortir des ventes nettement supérieures à celles de 1987, pour les boissons à base de cola notamment. ]]> A Three-way Alignment

The preceding encoding of the alignment of parallel passages from two texts requires that those texts and the alignment all be part of the same SGML document. If the texts are in separate documents, then additional xptr elements must be supplied, as discussed in section . These external pointers may appear anywhere within the document, but if they are created solely for use in encoding links, they may for convenience be grouped within the linkGrp (or other grouping element that uses them for linking).

To demonstrate this facility, we consider how we might encode the alignments in an extract from Comenius' Orbis Sensualium Pictus.

The figure shows the page from the Orbis pictus of Comenius which is discussed in the text.
Each topic covered in this work has three parts: a picture, a prose text in Latin describing the topic, and a carefully-aligned translation of the Latin into English, German or some other vernacular. Key terms in the two texts are typographically distinct, and are linked to the picture by numbers, which appear in the two texts and within the picture as well.

Our example uses the English translation of Charles Hoole (1659), and is taken from John E. Sadler, ed., John Amos Comenius Orbis Pictus: a facsimile of the first English edition of 1659 (Oxford: Oxford University Press, 1968) (The Juvenile Library).

First, we present the text portions. The English and Latin portions have been encoded as distinct div elements. Identifiers have been attached to each typographic line, but no other encoding added, to simplify the example.

The Study The Study is a place where a Student, a part from men, sitteth alone, addicted to his Studies, whilst he readeth Books,
Muséum Museum est locus ubi Studiosus, secretus ab hominibus, solus sedet, Studiis deditus, dum lectitat Libros,
]]>

Next we assume that we have stored a digitized image of the picture itself in some external entity we will call com98 (for further discussion of the handling of external images and graphics, see section ). We further assume that we can address portions of this image as a two-dimensional co-ordinate space. The SPACE location method of the xptr element (discussed in section above) can now be used to point to the whole picture and to two portions of it, one containing the picture of a student and the other of a book, as follows: ]]> Note that each external pointer has its own unique identifier, in addition to the n attribute, which last holds the visible label (or explainer) used for this image portion in the original.

As printed, the text exhibits three kinds of alignment. The English and Latin portions are printed in two parallel columns, with corresponding phrases, (represented above by seg elements), more or less next to each other. Particular words or phrases are marked as terms in the two languages by a change of rendition: the English text, which otherwise uses black letter type throughout, has the words The Study, a Student, Studies, and Books in a roman font; in the Latin text, which is printed in roman, the corresponding words (Museum, Studiosus, Studiis, and Libros) are all in italic. Numbered labels appear within the text portions, linking keywords to each other and to sections of the picture. These labels, which have been left out of the above encoding, are attached to the first third and last segment in each language quoted below, and also appear (rather indistinctly) within the picture itself. If it is desired to transcribe them in the text, they might be encoded using as ref elements, anchor elements, or xptrs to the picture; the number itself would be transcribed as the value of the n attribute (or as the content of the ref).

The first kind of alignment might be represented by using the corresp attribute on the seg element. The second kind might be represented by using the gloss and term mechanism described in section . The third kind of alignment might be represented using pointers embedded within the texts, although this would involve some duplication. We choose however to use the link element, since this provides an efficient way of representing the three-way alignment between English, Latin and picture without redundancy. ]]>

This map, of course, only aligns whole segments and image portions, since these are the only parts of our encoding which bear identifiers and can therefore be pointed to. To add to it the alignment between the typographically distinct words mentioned above, new elements must be defined, either within the text itself or externally by using the extended pointer mechanism. Encoding these word pairs as term and gloss, although intuitively obvious, requires a non-trivial decision as to whether the Latin text is glossing the English, or vice-versa. Tagging all the marked words as term avoids the difficult decision, but might be thought by some encoders to convey the wrong information about the words in question. Simply tagging them as additional embedded seg elements with identifiers that can be aligned like the others is also a possibility. All of these require the addition of further markup to the text. This may pose no problems, or it may be infeasible (e.g. if the text is held on a read-only medium). If it is not feasible to add more markup to the original text, the extended pointer mechanism is likely to be the best choice. For example, to indicate that the words Studies and Studiis correspond, two external pointers might be defined and aligned as follows: ]]> Synchronization

In the previous section we discussed two particular kinds of alignment: alignment of parallel texts in different languages; and alignment of texts and portions of an image. In this section we address another specialized form of alignment: synchronization. The need to mark the relative positions of text components with respect to time arises most naturally and frequently in transcribed spoken texts, but it may arise in any text in which quoted speech occurs, or events are described within a time frame. The methods described here are also generalizable for other kinds of alignment (for example, alignment of text elements with respect to space), and may thus be regarded as providing a simplified version of the HyTime system of finite space co-ordinates. Aligning Synchronous Events

To mark synchronous elements, the synch attribute, which is one of the linking attributes that are available for all text elements, may be used. points to elements that are synchronous with the current element. Alternatively, the link and linkGrp elements may be used to make explicit the fact that the synchronous elements are aligned.

To illustrate the use of these mechanisms for marking synchrony, consider the following representation of a spoken text:

This representation uses numbers in brackets to mark the points at which speakers overlap each other. For example, the [1] in A's first speech is to be understood as coinciding with the [1] in B's second speech.This sample is taken from a conversation collected and transcribed for the British National Corpus.

To encode this we use the base tag set for spoken texts, described in chapter , together with the additional tag set described in the present chapter. First, we transcribe this text, marking the synchronous points with anchor elements, and providing a synch attribute on one of each of the pairs of synchronous anchors. As noted in the example given above (section ), correspondence, and hence synchrony, is a symmetric relation; therefore the attribute need only be specified on one of the pairs of synchronous anchors. So you're It will be nice in a way, but, be strange. Yeah , yeah, cos it, its the not ]]>

Next, we encode the same example using link and linkGrp elements to make the temporal alignment explicit; the id attributes are provided for the link and linkGrp elements for a reason that is given in the next section, . The first time in twenty five years, we've cooked Christmas for a blooming great load of people. So you're It will be nice in a way, but, be strange Yeah, yeah, cos it, it's the not

]]>

As with other forms of alignment, synchronization may be expressed between stretches of speech as well as between points. When complete utterances are synchronous, for example, if one person says What? and another No! at the same time, that can be represented without anchor elements as follows. What? No! ]]>

A simple way of expressing overlap (where one speaker starts speaking before another has finished) is thus to use the seg element to encode the overlapping portions of speech. For example, So you're It will be nice in a way, but, be strange. Yeah , yeah, cos it, its the not ]]> Note in this encoding how synchronization has been effected between an empty unclear element and a seg, and between an entire u element and another seg, using the synch attribute. Alternatively, a linkGrp could be used in the same way as above. Placing Synchronous Events in Time

A synchronous alignment specifies which points in a spoken text occur at the same time, and the order in which they occur, but does not say at what time those points actually occur. If that information is available to the encoder it can be represented by means of the when and timeline elements, whose description and attributes are the following: indicates a point in time either relative to other elements in the same timeline tag, or absolutely. Attributes include: supplies an absolute value for the time. specifies the numeric portion of a time interval specifies the unit of time corresponding to the interval value. identifies the reference point for determining the time of the current when element, which is obtained by adding the interval to the time of the reference point. supplies an identifier, unique to the document, for each when element. provides a set of ordered points in time which can be linked to elements of a spoken text to create a temporal alignment of that text. Attributes include: designates the origin of the timeline, i.e. the time at which it begins. specifies the numeric portion of a time interval specifies the unit of time corresponding to the interval value of the timeline or of its constituent points in time.

Each when element indicates a point in time, either directly by means of the absolute attribute, whose value is a string which specifies a particular time, or indirectly by means of the since attribute, which points to another when. If the since is used, then the interval and unit attributes should also be used to indicate the amount of time that has elapsed since the time specified by the element pointed to by the since attribute; the value -1 can be given to indicate that the interval is unknown.

If the when elements are uniformly spaced in time, then the interval and unit values need be given once in the timeline, and not repeated in any of the when elements. If the intervals vary, but the units are all the same, then the unit attribute alone can be given in the timeline element, and the interval attribute given in the when element.

The origin attribute in the timeline element points to a when element which specifies the reference or origin for the timings within the timeline; this must, of course, specify its position in time absolutely.

The following timeline might be used to accompany the marked up conversation shown in the preceding section: ]]> The information in this timeline could now be linked to the information in the linkGrp which provides the temporal alignment (synchronization) for the text, as follows: ]]>

To avoid the need for two distinct link groups (one marking the synchronization of anchors with each other, and the other marking their alignment with points on the time line) it would be better to link the when elements with the synchronous points directly: ]]>

Finally, suppose that a digitized audio recording is also available. The extended pointer syntax described in section could be used to address positions on or portions of this recording directly. Assuming that xptr elements with identifiers X1, X2, etc., have been defined to do this, these identifiers could also be included as a fourth component in each of the above link elements, thus providing a synchronized audio track to complement the transcribed text.

For further discussion of this and related aspects of encoding transcribed speech, refer to chapter .

The when and timeline elements are defined as follows: ]]> Identical Elements and Virtual Copies

This section introduces the notion of a virtual element, that is, an element which is not explicitly present in a text, but the presence of which an application can infer from the encoding supplied. In this section, we are concerned with virtual elements made by simply cloning existing elements. In the next section (), we discuss virtual elements made by aggregating existing elements.

It is useful to be able to represent the fact that one element of text is identical to others, for analytical purposes, or (especially if the elements have lengthy content) to obviate the need to repeat the content. For example, consider the repetition of the date element in the following material: In small clumsy letters he wrote: April 4th, 1984.

He sat back. A sense of complete helplessness had descended upon him.

His small but childish handwriting straggled up and down the page, shedding first its capital letters and finally even its full stops: April 4th, 1984. Last night to the flicks.

]]> Suppose now that we wish to encode the fact that the second date element above has identical content to the first. The sameAs attribute is provided for this purpose. Using it, we can recode the last line of the above example as follows: April 4th, 1984. Last night to the flicks. ]]>

The sameAs attribute may be used to document the fact that two elements have identical content. It may be regarded as a special kind of link. It should only be attached to an element with identical content to that it indicates, or to one the content of which clearly designates it as a repetition, such as the word repeat or bis in the representation of the chorus of a song, the second time it is to be sung. The relation specified by the sameAs attribute is symmetric: if a chorus is repeated three times and each repetition bears a sameAs attribute indicating the first occurrence of the element concerned, it is implied that each chorus is identical, and there is no need for the first occurrence to specify any of its copies.

The copyOf attribute is used in a similar way to indicate that the content of the element bearing it is identical to that of another. The difference is that the content is not itself repeated. The effect of this attribute is thus to create a virtual copy of the element indicated. Using this attribute, the repeated date in the first example above could be recoded as follows: ]]>

An application program should replace whatever is the actual content of an element bearing a copyOf attribute with the content of the element specified by it. If the content of the element specified includes other elements, these will become embedded within the element bearing the attribute. Care must be taken to ensure that the document is a legal SGML document both before and after this embedding takes place. If, for example, the element bearing a copyOf attribute requires a mandatory sub-component, then this component must be present (though possibly empty), even though it will be replaced by the content of the targetted element.

The following example demonstrates how the copyOf attribute may be used in conjunction with the seg element to highlight the differences between almost identical repetitions: My object all sublime I shall achieve in time‐ To let the punishment fit the crime, ; And make each pris'ner pent Unwillingly represent A source of innocent merriment, ! His He will ]]>

For further examples of the use of this attribute, see chapters and , where it is used to reduce the complexity of formal analytic representations of structure. Aggregation

Because of the strict hierarchical organization of an SGML document, or for other reasons, it may not always be possible or desirable to include all the parts of a possibly fragmented text segment within a single element. In section we introduced the notion of an intermediate pointer as a way of pointing to discontinuous segments of this kind. In this section we first describe another way of linking the parts of a discontinuous whole, using a set of linking attributes, which are made available for any tag by following the procedure described at the beginning of this chapter. We then describe how the link element may be used to aggregate such segments, and finally introduce the join element, which is a special-purpose linking element specifically for representing the aggregation of parts, and the joinGrp for grouping join tags.

The linking attributes for aggregation are next and prev; each of these attributes has a single identifier as its value: points to the next element of a virtual aggregate of which the current element is part. points to the previous element of a virtual aggregate of which the current element is part.

The join element is also a member of the class of pointer elements, and so may carry any of the attributes of that class; for the list, see section .

Here is the material on which we base our first illustration of the use of these mechanisms. Our problem is to represent the S-units identified below as qs3 and qs4 as a single (but discontinuous) whole: Monsieur Paul, after he has taken equal parts of goose breast and the finest pork, and broken a certain number of egg yolks into them, and ground them very, very fine, cooks all with seasoning for some three hours. But, she pushed her face nearer, and looked with ferocious gloating at the pâté inside me, her eyes like X rays, he never stops stirring it! Figure to yourself the work of it — stir, stir, never stopping! ]]>

Using the prev and next attributes, we can link the s-units with identifiers s1 and s2, either singly or doubly as follows: But, he never stops stirring it! But, he never stops stirring it! But, he never stops stirring it! ]]> Double linking of the two S-units, as illustrated by the last of these encodings, is equivalent to specifying a link tag: ]]>

Such a link element must carry type=join attribute value to specify that the link is to be understood as joining its targets into a single aggregate.

The join element is equivalent to a link element of type join; unlike a link, the default value for the targOrder attribute which this element also inherits from the pointer class is Y. Also unlike the link element, the join element can additionally specify information about the virtual element which it represents, by means of its result attribute. And finally, unlike the link element, the position of a join element within a text is significant: it must be supplied at a position where the element indicated by its result attribute would be contextually legal. identifies a possibly fragmented segment of text, by pointing at the possibly discontiguous elements which compose it. Attributes include: specifies the name of an element which this aggregation may be understood to represent. specifies the SGML identifiers of the elements or passages to be joined into a virtual element. specifies whether or not the order in which components of the join are listed on the targets attribute is significant. Legal values are: Yes: the order should be followed when combining the targeted elements. No: the order has no significance when combining the targeted elements. Unspecified: the order may or may not be significant. groups a collection of join elements and possibly pointers. Attributes include: describes the result of the joins gathered in this collection. To conclude the above example, we now use a join element to represent the virtual sentence formed by the aggregation of s1 and s2: ]]> As a further example, consider the following list of authors' names. The object of the join element here is to provide another list, composed of those authors from the larger list who happen to come from Heidelberg: Authors Figge, Udo Heibach, Christiane Heyer, Gerhard Philipp, Bettina Samiec, Monika Schierholz, Stefan ]]>

The following example shows how join can be used to reconstruct a text cited in fragments presented out of order. The poem being remembered (an unusual translation of a well known poem by Basho) runs When the old pond / gets a new frog, / it's a new pond.

How does it go? da-da-da gets a new frog ...

When the old pond ...

... It's a new pond.

]]>

As with other forms of link, a grouping element joinGrp is available for use when a number of join elements of the same kind co-occur. This avoids the need to specify the result attribute for each join if they are all of the same type, and also allows us to restrict the domain within which their target elements are to be found, in the same way as for linkGrp elements (see ). Like a join, a joinGrp may appear only where the elements represented by its contents are legal. Thus if we had created many join tags of the sort just described, we could group them together, and require that their components are all contained by an element with the identifier MFKFhungry as follows: ]]>

The join element is useful as a means of representing non-hierarchic structures (as further discussed in chapter ). It may also be used as a convenient way of representing a variety of analytic units, like the span and interp elements discussed in chapter . As an example, consider the following passage:

Zui-Gan called out to himself every day, Master.

Then he answered himself, Yes, sir.

And then he added, Become sober.

Again he answered, Yes, sir.

And after that, he continued, do not be deceived by others.

Yes, sir; yes, sir, he replied.

Suppose now that we wish to represent an interpretation of the above passage in which we distinguish between the various voices adopted by the character Zui-Gan. In the following encoding, the who attribute has been used for this purpose; id attributes have also been added:

Zui-Gan called out to himself every day, Master.

Then he answered himself, Yes, sir.

And then he added, Become sober.

Again he answered, Yes, sir.

And after that, he continued, do not be deceived by others.

Yes, sir; yes, sir, he replied.

]]>

The id values specified now allow us to link the material spoken by each voice:

Zui-Gan called out to himself every day, Master.

Then he answered himself, Yes, sir.

And then he added, Become sober.

Again he answered, Yes, sir.

And after that, he continued, do not be deceived by others.

Yes, sir; yes, sir, he replied.

]]>

However, by using the join element, we can directly represent the complete speech attributed to each voice: ]]>

Note the use of the global n attribute to supply a descriptive name to distinguish the two virtual q elements represented by the join elements; this is necessary because the current proposals do not allow for any way of specifying the attributes to be associated with a virtual element, and hence we cannot specify a who value for them.

Suppose now that id attributes, for whatever reasons, are not available. Then xptr elements may be created using any of the methods described in section . The id attributes of these elements may now be specified by the targets attribute on the join elements.

Zui-Gan called out to himself every day, Master.

Then he answered himself, Yes, sir.

And then he added, Become sober.

Again he answered, Yes, sir.

And after that, he continued, do not be deceived by others.

Yes, sir; yes, sir, he replied.

]]>

For a definition of the syntax used by the xptr element, see section above. The extended pointer with identifier rzuiq2 (for example) may be read as the first q in the first p, inside the sixth div1 within the second div0 element of the current document.

As mentioned above, there is no need for the join and xptr elements to be held in the same SGML document as the text; indeed, if, for example, the text is held on a read-only medium, this may not be possible. The doc attribute of the xptr element may be used to specify the name of the SGML entity within which its target is to be found. ]]>

Here are the formal declarations of the join and joinGrp elements. ]]> Alternation

This section proposes elements for the representation of alternation. We say that two or more elements are in exclusive alternation if any of those elements could be present in a text, but one and only one of them is; in addition, we say that those elements are mutually exclusive. We say that the elements are in inclusive alternation if at least one (and possibly more) of them is present. The elements that are in alternation may also be called alternants.

The need to mark exclusive alternation arises frequently in text encoding. A common situation is one in which it can determined that exactly one of several different words appears in a given location, but it cannot be determined which one. One way to mark such an exclusive alternation is to use the linking attribute exclude. Having marked an exclusive alternation, it can sometimes later be determined which of the alternants actually appears in the given location. To preserve the fact that an alternation was posited, one can add the linking attribute select to a tag which hierarchically encompasses the alternants, which points to the one which actually appears. To assign responsibility and degree of certainty to the choice, one can use the certainty tag described in chapter . Also see that chapter for further discussion of certainty in general.

The exclude and select attributes may be used with any element assuming that they have been declared following the procedure discussed in the introduction to this chapter. points to elements that are in exclusive alternation with the current element. selects one or more alternants; if one alternant is selected, the ambiguity or uncertainty is marked as resolved. If more than one alternant is selected, the degree of ambiguity or uncertainty is marked as reduced by the number of alternants not selected.

A more general way to mark alternation, encompassing both exclusive and inclusive alternation, is to use the linking element alt. The description and attributes of this tag and of the associated grouping tag altGrp are as follows. These elements are also members of the pointer class and therefore have all the attributes associated with that class. identifies an alternation or a set of choices among elements or passages. Attributes include: specifies the SGML identifiers of the alternative elements or passages. If mode=excl, each weight states the probability that the corresponding alternative occurs. If mode=incl each weight states the probability that the corresponding alternative occurs given that at least one of the other alternatives occurs. groups a collection of alt elements and possibly pointers.

To take a simple hypothetical example, suppose in transcribing a spoken text, we encounter an utterance that we can understand either as We had fun at the beach today. or as We had sun at the beach today. We can represent the exclusive alternation of these two possibilities by means of the exclude attribute as follows. We had fun at the beach today. We had sun at the beach today.

]]>

If it is then determined that the speaker said fun, not sun, the encoder could amend the text by deleting the alternant containing sun and the exclude attribute on the remaining alternant. Alternatively, the encoder could preserve the fact that there was uncertainty in the original transcription by retaining the alternants, and assigning the select=we.fun attribute value to the div tag that encompasses the alternants, as in: We had fun at the beach today. We had sun at the beach today. ]]>

The above alternation (including the select attribute) could be recoded by assigning the exclude attributes to tags that enclose just the words or even the characters that are mutually exclusive, as in: See section for discussion of the w and c tags that can be used in the following examples instead of the seg type=word and seg type=character tags. We had fun sun at the beach today.

We had f s un at the beach today.
]]>

Now suppose that the transcriber is uncertain whether the first word in the utterance is We or Lee, but is certain that if it is Lee, then the other uncertain word is definitely fun and not sun. The three utterances that are in mutual exclusion can be encoded as follows. We had fun at the beach today. We had sun at the beach today. Lee had fun at the beach today. ]]>

The preceding example can also be encoded with exclude attributes on the word segments We, Lee, fun and sun: We Lee/seg> had fun sun at the beach today. ]]>

The value of the select attribute is defined as a list of identifiers (IDREFS); hence it can also be used to narrow down the range of alternants, as in: We had fun at the beach today. We had sun at the beach today. Lee had fun at the beach today. ]]> This is interpreted to mean that either the first or the third u tag appears, and is thus equivalent to just the alternation of those two tags: We had fun at the beach today. Lee had fun at the beach today. ]]>

The exclude attribute can also be used in case there is uncertainty about the tag that appears in a certain position. For example, the occurrence of the word May in the S-unit Let's go to May can be interpreted, in the absence of other information, either as a person's name or as a date. The uncertainty can be rendered as follows, using the exclude attribute. Let's go to May . ]]>

Note the use of the copyOf attribute discussed in section ; this avoids having to repeat the content of the element whose correct tagging is in doubt.

The copyOf and the exclude attributes also provide for a simple way of indicating uncertainty about exactly where a particular element occurs in a document.An alternative way of representing this problem is discussed in chapter . For example suppose that a particular div2 element appears either as the third and last of the div2 elements within the first div1 element in the body of a document, or as the first div2 of the second div1. One solution would be to record the div2 in its entirety in the first of these positions, and a virtual copy of it in the second, and mark them as excluding each other as follows: ]]> In this case, the select attribute, if used, would appear on the body tag.

Mutual exclusion can also be expressed using a link; the first example in this section can be recoded by removing the exclude attributes from the u tags, and adding a link as follows: In this example, we have placed the link next to the tags that represent the alternants. It could also have been placed elsewhere in the document, perhaps within a linkGrp. We had fun at the beach today. We had sun at the beach today. ]]>

Now we define the specialized linking element alt,making it a member of the pointer class of elements, and assigning it a excl (for mutually exclusive) attribute, which can have either of the values Y or N. Then the following equivalence holds: = ]]>

It is in the nature of alternation that the order of the targets is irrelevant; hence the targOrder attribute of the alt defaults to the value N. The preceding link may therefore be recoded as the following alt tag. ]]>

Other attributes that are defined specifically for the alt element are weights and percent. The weights attribute is to be used if one wishes to assign probabilistic weights to the targets (alternants). Its value is a list of numbers, corresponding to the targets, expressing the probability that each target appears. The percent attribute is used to indicate whether the weights are stated as percentages (percent=Y, the default) or as the actual probabilities (percent=N). If the alternants are mutually exclusive, then the weights must sum to 100% (or 1, if percent=N is specified).

Suppose in the preceding example that it is equiprobable whether fun or sun appears. Then the alt that represents the alternation may be stated as follows: ]]>

The assignment of a weight of 100% to one target (and weights of 0% to all the others) is equivalent to selecting that target. Thus the following encoding is equivalent to the second example at the beginning of this section. We had fun at the beach today. We had sun at the beach today. ]]> The sum of the weights for alt excl=N tags ranges from 0% to (100 x k)%, where k is the number of targets. If the sum is 0%, then the alternation is equivalent to exclusive alternation; if the sum is (100 x k)%, then all of the alternants must appear, and the situation is better encoded without an alt tag.

If it is desired, alt tags may be grouped together in an altGrp tag, and attribute values shared by the individual alt tags may be identified on the altGrp tag. The targFunc attribute defaults to the value 'first.alternant next.alternant'. Thus, specifying the extendTarg=2 attribute value permits the alternants to be extended indefinitely.

To illustrate, consider again the example of a transcribed utterance, in which it is uncertain whether the first word is We or Lee, whether the third word is fun or sun, but that if the first word is Lee, then the third word is fun. Now suppose we have the following additional information: if we occurs, then the probability that fun occurs is 50% and that sun occurs is 50%; if fun occurs, then the probability that we occurs is 40% and that Lee occurs is 60%. This situation can be encoded as follows. We Lee/seg> had fun sun at the beach today. ]]>

From the information in this encoding, we can determine that the probability is about 28.5% that the utterance is We had fun at the beach today, 28.5% that it is We had sun at the beach today, and 43% that it is Lee had fun at the beach today.

Another very similar example is the following regarding the text of a Broadway song. In three different versions of the song, the same line reads Her skin is tender as a leather glove, Her skin is tender as a baseball glove, and Her skin is tender as Dimaggio's glove.The variant readings are found in the commercial sheet music, the performance score, and the Broadway cast recording.

If we wish to express this textual variation using the alt element, we can record our relative confidence in the readings Dimaggio's (with probability 50%), a leather (25%), and a baseball (25%).

Let us extend the example with a further (imaginary) variation, supposing for the sake of the argument that the next line is variously given as and she bats from right to left (with probability 50%) or now ain't that too damn bad (with probability 50%). Using the alt element, we can express the conviction that if the first choice for the second line is correct, then the probability that the first line contains Dimaggio's is 90%, and each of the others 5%; whereas if the second choice for the second line is correct, then the probability that the first line contains Dimaggio's is 10%, and each of the others is 45%. This can be encoded, with an altGrp tag containing a combination of exclusive and inclusive alt tags, as follows. Her skin is tender as Dimaggio's a leather a baseball glove, and she bats from right to left. now ain't that too damn bad. ]]>

Here are the formal declarations of the alt and altGrp elements. ]]> Connecting Analytic and Textual Markup

In chapters and and elsewhere, provision is made for analytic and interpretive markup to be represented outside of textual markup, either in the same document or in a different document. The elements in these separate domains can be connected, either with the pointing attributes ana (for analysis) and inst (for instance), or by means of link and linkGrp elements. Numerous examples are given in these chapters, particularly in sections , and .