Introduction

The first deliverable was the selection of TEI P5 elements suitable for encoding ESDS materials, the recommendation for how best to store the required metadata and structure the encoded documents. In addition a Relax NG schema was produced along with customised TEI documentation for this schema. This schema may be modified later in response to the evolving needs of the project. This file is authored as a TEI ODD XML file which then is passed through TEI Roma to produce schemas (RNG/RNC/XSD/DTD) and documentation (PDF/HTML). All changes should be made in the TEI ODD file rather than in the generated schemas or documentation.

Scope of Schema

In deciding the metadata scheme and forms of markup that should be used to encode the interviews, it was agreed that a very light encoding should be used. However, to allow for later expansion and re-purposing of the materials, and re-use of the schema, a number of extra elements have been included. While these certainly can be cut out now, it was thought that it might be more straightforward to include them now, rather than introduce them at a later date. The elements included number around 130, and while this may seem like a lot, it is a great reduction from the almost 500 elements the TEI currently contains. It could easily be further constrained to an extremely minimal schema if desired.

Processing of Files

As the project files are already producing them in a set XML format, it was recommended that they continue to produce these files in that format. The resulting files could then be converted en masse at an appropriate time in the future. Of the sample files provided, none were found to have structures that would not be able to be easily converted to the proposed TEI format. The recommendation was also that this should be done through XSLT. This also impacted the decision to not worry as much about the tightness of the schema, since the conversion will be under the control of ESDS.

Later Deliverables

It is believed that this inital proposed schema is robust enough to deal with the majority of the other deliverables, and the TEI ODD file will be edited to document these in due course and the documentation and schemas regenerated at that point.

Elements Included Text Structure: TEI, body, div, group, text. TEI Header: authority, availability, change, distributor, editorialDecl, equiment, extent, fileDesc, funder, idno, keywords, langUsage, language, noteStmt, principal, profileDesc, projectDesc, publicationStmt, recording, recordingStmt, revisionDesc, sourceDesc, sponsor, state, teiHeader, titleStmt. Core: abbr, addrLine, address, author, bibl, corr, date, desc, distinct, divGen, editor, emph, equiv, expan, foreign, gap, gloss, graphic, head, measure, mentioned, milestone, name, note, orig, p, pb, ptr, pubPlace, q, ref, reg, resp, respStmt, rs, sic, teiCorpus, term, title, unclear. Corpus: activity, firstLang, locale, particDesc, setting, settingDesc, textDesc. Linking: anchor, join, joinGrp, link, linkGrp, seg. Names&Dates: addName, affiliation, age, birth, death, education, faith, floruit, forename, genName, langKnowledge, langKnown, listPerson, nameLink, nationality, occasion, occupation, orgDivn, orgName, orgTitle, orgType, particLinks, persEvent, persName, persState, persTrait, person, personGrp, placeName, relation, residence, roleName, sex socecStatus, surname. Spoken: event, kinesic, pause, shift, u, vocal, writing.

Structure of Files

The recommendation is to use a single file per interview and any grouping of these to be done at a later stage if necessary. The basic structure should be:

]]>

Here the TEI element is in the TEI namespace, and XML comments are used to indicate missing data or notes. Each file should have a teiHeader (discussed below), followed by a text element. This should contain a body element and one div for the entire interview or for each section if it is segmented. Each utterance (in this kind of interview meant to mean speech by one person in usual turn-taking), is marked with a u element. This element must have a who attribute which should be a URI pointer pointing back up to the header where a person element has an xml:id defined for that person. Either generic xml:id's such as interviewer and subject can be used (and referenced in the u element's who attribute as #interviewer since it is a URI fragment) or specific ones relating solely to that file could be used.

These elements can be defined as:

The teiHeader

The teiHeader element should contain all the metadata for that particular interview. There are two major child elements of teiHeader for the purpose of this project:

A template fileDesc might look like: ESDS Qualidata

As much information as is available should be included. There are places for a lot more detail not shown in this simple template, consult the TEI guidelines for further information. The most important pieces of information here are the title and the idno, since these are what identify the file.

The profileDesc for this project will usually contain two important child elements:

The particDesc element should contain a listPerson element with at least two person elements defined, one for the subject and one for the interviewer. The person element for the subject should contain as much information as is available for the interviewee. Some pieces of information which can be recorded here include:

For the purposes of this project, where known age, birth, sex, and occupation will be most commonly recorded. A template person element for a subject might look like this:

There should always be a person element for both subject and interviewer even if there is no data available for the interviewer. This can consist merely of:

Interviewer Unknown

The settingDesc element is a place to record information concerning the setting of the interview. A template of this might look like: 3 May 1979

In addition to the date and locale, the activity a subject is undertaking whilst being interviewed (e.g. 'sewing', 'driving', 'skiing') amongst other information can be recorded if known.

Transcription of Speech

As mentioned above each utterance should be in a u element. This should look something like:

The important aspects are: the who attribute (which should point to the xml:id of a person in the header), and the xml:id which should be consecutively numbered from u1.

There are various optional phrase-level elements available if the project's needs eventually call for them.

These include phenomena of interest in spoken corpora such as:

As well as those commonly used in such transcriptions:

Name and Dates

There are a variety of elements available for recording names and dates both in the header and outside it. In general the date element should be given W3C or ISO standard values, these also have an existing standard formulation for date ranges and durations. Some of the elements available include:

ID references with xml:id

As mentioned previously, the who attribute on the u element should refer back to an xml:id reference in the header. If generalised xml:id's are used (e.g. interviewer and subject), then a possible problem occurs if all the interviews are later treated as a corpus and collected into a separate file under a teiCorpus structure. (If this is done, the TEI recommends that this should be done using XInclude, rather than physically copying the files into one large file.) However this is done, the various xml:id's would no longer be unique, and thus the master file would fail to validate.

A simple transformation would be recommended before any such gathering together of the files. What this would do is replace all xml:id's with a compound based on the unique idno element in the header. Thus, if an interview's idno might be int0004, then the xml:id's in this file might be transformed simply to such things as int0004-interviewer and int0004-subject. Care must also be taken to transform any references to these, most notably in the who attribute of each utterance. This entire process is quite easily scriptable with XSLT, so should be of little concern, however I felt it worth mentioning here.