The Spoken Corpus of the Survey of English Dialects: Language variation and oral history

The Spoken Corpus of the Survey of English Dialects:
Language variation and oral history

Judith
Beare

Routledge

Brad
Scott

Routledge
Scott@routledge.co.uk

1999

University of Virginia

Charlottesville, VA

ACH/ALLC 1999

editor

encoder

Sara
A.
Schmidt

The Survey of English Dialects: The spoken corpus, recorded in
England 1948-1973 is an SGML-based linguistic corpus which includes
part of speech tagging, in which the text is linked to its corresponding audio
file, and will be delivered within an extended DynaText application. It will be
published in the winter of 1999. This paper will describe the context of the
creation of such a resource, indicating the benefits and disadvantages of these
methodologies for researchers in a number of disciplines compared with existing
resources, and will discuss practicalities of the novel mark-up and other
technical features that the project engenders. We will also highlight the
practical questions that relate to the design and development of such a resource
within a commercial publishing environment.
The Survey of English Dialects was started in 1948 by Harold Orton at the
University of Leeds and has been producing unique research data ever since. The
initial work comprised a questionnaire-based survey of traditional dialects
based on extensive interviews from 318 locations all over rural England. These
were published as the Survey of English Dialects: Basic
Material (13 vols) in the 1960s, and have formed the core data for a
number of other publications since then.Such as Clive Upton et al.
(1990) Survey of English Dialects: Dictionary and
Grammar, and H. Orton et al. (1978) Linguistic
Atlas of England.
During the course of the survey a number of recordings were made as well as the
detailed interviews. These unique recordings are between eight and twenty
minutes duration, equalling about 60 hours of dialogue. They are invariably of
elderly people talking about life, work and recreation and have not been widely
available, but are potentially an important resource, not merely for
dialectologists, but for linguists more generally, those studying variation in
world Englishes and many historians. The recordings were made from the 1950s
through to the 1970s. The early ones are on 78 rpm disks in four minute chunks,
and the later ones are on reel-to-reel tapes and may extend to over twenty
minutes of free conversation.
During 1997, Juhani Klemola at the University of Leeds received a grant from the
Leverhulme Trust to transcribe the recordings, which work has now been
completed. From the mark-up scheme deriving from the transcription process the
data has been converted to TEI-conformant SGML, and then part of speech (POS)
tagging added by Tony McEnery and colleagues at the University of Lancaster
using their CLAWS tagger. The results of this process, given that the target
text comprises some very irregular words and grammar, will be reported. In
addition, to facilitate the use of the resource, the text and audio will be
linked, so that the user can easily hear the audio relating to a given segment
of the text. The technical issues in creating this will be discussed.Some existing resources have already combined SGML linguistic data with a
linked audio (e.g., the Map Task Corpus developed
by Henry S. Thompson and colleagues at the University of Edinburgh).
However, as far as we are aware, the SED Spoken
Corpus will be the first to extend such functionality in a
Windows application using a commercially available SGML browser.
For the study of the dialects of England, currently available resources tend to
focus on unusual words in isolation, rather than dialectal variation in natural
speech.Most of the major resources in this area are only
available in print form; alternatively, there are of course numerous
collections of recordings made for more specific studies, though very little
is widely available in electronic form. However, W. Elmer and G. Schiltz at
the University of Basel are in the process of creating an impressive
electronic version of the SED Basic Material, which includes a facility to
search by phonetic string and generate a map of its geographical
distribution. However, the development of the application of
computing methodologies to linguistic data such as the BNC and ICE-GB has
generated a number of corpora which are available for researchers in linguistics
and language engineering, and which many specialist researchers now use.
Building on these developments, the SED Spoken Corpus
has been conceived as a resource that should be accessible to a user group with
a wide spectrum of technical literacy, not only in linguistics, but also in
history. This necessitates the design of a simple interface and presumes no
knowledge of SGML to effectively use the resource, but which also allows and
supports research by those in the corpus linguistics community. Any difficulties
and compromises which were necessary to achieve this will be outlined.
A number of features of the development process will be described and discussed.
These include:
the delivery of a combined text and audio dataset within a (modified)
DynaText application;
designing a tool that does not necessarily assume great knowledge of
using computers in linguistics or history, though which can support
those users who want access to all the tagged data and the audio files,
whether this be for interrogating the data using the SGML in
sophisticated ways, or to load data into other applications (given that
one cannot predict every conceivable use to which people may want to put
the data);
linguistic and organisational issues in implementing the POS
tagging;
text-audio linking.

Given the numerous competing requirements in the development of such an
application, the paper will also address those factors which may be seen as
(necessary) limitations of the functionality of the application.
The SED Spoken Corpus was never conceived as an oral
history project and the recordings were never constructed with such a purpose in
mind. Nevertheless, even with the currently limited awareness of the material
within the oral history community, these recordings are known as a unique and
important collection for which there is strong research interest. This of course
raises an additional challenge in the creation of the electronic resource, as
the data and the application have to be designed to support the interrogation of
rather disparate academic groups. The issues so raised and solutions adopted
will be outlined and discussed. Other themes specifically relating to the oral
historical dimension of the data will be included, in particular the question of
interviewee identity.