Digital Audio Archives, Computer-Enhanced Transcripts, and New Methods in Sociolinguistic Analysis

Introduction
Traditional methods in sociolinguistic analysis have often relied on the repeated close listening of a set of audio recordings counting the number of times
particular linguistic variants occur in lieu of other
variants (a classic sociolinguistic example is the tabulating
of words using final –in’ for final –ing; cf. Fischer 1958, Trudgill 1974, etc.). These tabulations are normally
recorded into a spreadsheet using a program such as
Microsoft Excel, or even just into a hard-copy tabulation
sheet. The results are then presented as summaries in publications or conference papers as the “data” used for description, explanation, and theory building. Some
approaches in linguistics, such as discourse analysis, rely heavily on the development of transcripts of the audio recordings and often the focus of analysis is on the
transcript itself and not the original recording or interview
event. However, scholars following a wide variety of
sociolinguistic approaches have repeatedly highlighted
the confounds that arise from these treatments of
“pseudo-data” (i.e., analysts’ representations of the data) as data. Linguists such as Blake (1997) and Wolfram
(e.g., 1993) have discussed problems relating to the
tabulation and treatment of linguistic variables and raised
the issue that individual scholars’ methods are often not comparable. In discussing transcription theory, Edwards
has repeatedly pointed out that “transcripts are not
unbiased representations of the data” (Edwards 2001: 321). In general, the understanding that linguistic data is more elusive than traditional “hard science” data is
widespread but not acted upon. In this paper, we present
a project underway at North Carolina State University to argue that computer-enhanced approaches can propel sociolinguistic methodology into a new, more rigorous era.
The North Carolina Sociolinguistic Archive and Analysis Project
The North Carolina Language and Life Project (NCLLP) is a sociolinguistic research initiative at North Carolina State University (NCSU) with one of the largest audio collections of sociolinguistic data on
American English in the world. It consists of approximately
1,500 interviews from the late 1960s up to the present, most on analog cassette tape, but some in formats ranging
from reel-to-reel tape to digital video. The collection
features the interviews of Walt Wolfram, Natalie
Schilling-Estes, Erik Thomas, and numerous other
scholars. The NCLLP has partnered with the NCSU
Libraries on an initiative titled the North Carolina
Sociolinguistic Archive and Analysis Project (NC SLAAP). NC SLAAP has two core goals: (1) to preserve the NCLLP’s recordings through digitization; and (2) to enable and explore new computer-enhanced techniques
for interacting with the collection and for conducting
sociolinguistic analysis.
NCSU Libraries has as one of its chief goals the
long-term preservation of the recordings made by the NCLLP,
and it regards digitization as an appropriate means of preservation. Academic libraries may still be less expert than some commercial organizations when it comes to digitizing and storing audio, but they may be even less equipped to maintain analog audio collections properly (cf. Brylawski 2002, Smith, Allen, and Allen 2004). Archivists and librarians also sometimes point out that digitization and storage of audio may not be worth the expense and difficulty if the sole goal is preservation (cf. Puglia 2003). However, when scholarly digital projects
can contribute significantly to the advancement of a
discipline, as in the case of NC SLAAP, surely significant
investments are called for.
The NC SLAAP project has from the beginning planned to integrate sociolinguistic analysis tools into the archive. This has been achieved to a large degree by integrating the open source phonetic software application Praat
(http://www.praat.org) into the web server software. In
brief overview, the NC SLAAP system is an Apache
web server currently housed on a Macintosh G5 computer
running Mac OS 10.4. Data are stored in a MySQL database
and application pages are written in PHP. The web server
communicates with third-party open source applications
to do most of its “heavy” processing. Most importantly,
the web server communicates with Praat to generate
real-time phonetic data (such as the pitch data and the
spectrogram illustrated in Figure 1).
Figure 1: Transcript Line Analysis Example
While certain feature sets are still under development,
NC SLAAP, even in its current state, provides a range of
tools that greatly enhance the usability of the audio data.
These features include an audio player with an annotation
tool that allows users to associate notes with particular
timestamps, an audio extraction feature that allows users
to download and analyze particular segments of audio
files, sophisticated transcript display options (as partly
illustrated in Figure 1, above), and extensive search and
query tools. Importantly, the NC SLAAP software helps
to address concerns about the treatment of “pseudo-data”
as data, because it enables scholars to better access,
check, and re-check their (and their colleagues’) variable
tabulations, analyses, and conclusions. In short, the
NC SLAAP software is an attempt to move us one step
– hopefully, a large step – closer to the “real” data.
The features of the NC SLAAP software have potentially
tremendous implications for a wide range of linguistic
approaches. We focus on only one such feature here: the
implications relating to transcription theory.
Transcription Method and Theory Improvements to the traditional text transcript are
extremely important because the transcript is often
the chief mediating apparatus between theory and data in
language research. Language researchers have long been
concerned with the best method and format for transcribing
natural speech data (cf. Edwards 2001). Researchers
frequently incorporate a number of different transcription
conventions depending on their specific research aims.
Discourse analysts (e.g., Ochs 1979) traditionally focus
most heavily on transcription as theory and practice, but
researchers studying language contact phenomena (as in
Auer 1998) also have their own transcription conventions
for analyzing and presenting their data. At the other end
of the spectrum are variationists and dialectologists, who
also use transcripts, even if often only for presentation
and illustration.
Despite the importance of the transcript for most areas
of linguistics, little work has been done to enhance the
usability and flexibility of our transcripts. Yet the way a
researcher builds a transcript has drastic effects on what
can be learned from it (Edwards 2001). Concerns begin
with the most basic decision about a transcript: how
to lay out the text. Further decisions must be made
throughout the transcript-building process, such as
decisions about how much non-verbal information
to include and how to encode minutiae such as
pause-length and utterance overlap. Furthermore, the
creation of a transcript is a time- and energy-intensive
task, and researchers commonly discover that they must
rework their transcripts in mid-project to clarify aspects
of the discourse or speech sample.
The NC SLAAP software seeks to improve the linguistic
transcript by moving it closer to the actual speech that
it ideally represents (Kendall 2005). In the NC SLAAP
system, transcript text is treated as annotations on the audio
data: transcripts are broken down into utterance-units that
are stored in the database and directly tied to the audio
file through timestamping of utterance start and end
times. Transcript information can be viewed in formats
mimicking those of traditional paper transcripts, but can
also be displayed in a variety of dynamic ways – from
the column-based format discussed by Ochs (1979) to a
finer-level focus on an individual utterance complete with
phonetic information (as shown in Figure 1, above). Conclusion
NC SLAAP is a test case for new ways of
approaching linguistic analysis, using computers to maintain a strong tie between the core audio data and the analysts’ representations of it. In many senses the project is still in a “proof of concept” stage. However, we feel that it has made large steps towards new and more rigorous methods for sociolinguistic analysis and data management. In addition, it can serve as a model for academic libraries as a project that incorporates digital preservation with significant scholarly advancement.
References
Auer, P. (ed.) (1998). Code-Switching in Conversation, London: Routledge.
Blake, R. (1997). Defining the Envelope of Linguistic Variation: The Case of “Don’t Count” Forms in the Copula Analysis of AAVE. Language Variation and Change 9: 57-79.
Brylawski, S. (2002). Preservation of Digitally Recorded
Sound, in Building a National Strategy for
Preservation: Issues in Digital Media Archiving, Council on Library and Information Resources Publication 106.http://www.clir.org/pubs/abstract/pub106abst.html
Edwards, J. (2001). The Transcription of Discourse,
in Handbook of Discourse Analysis, eds. Deborah Tannen, Deborah Schiffrin, and Heidi Hamilton: 321-348, Oxford and Malden, Ma: Blackwell.
Fischer, J. (1958). Social Influences on the Choice of a Linguistic Variant. Word 14: 47-56.
Kendall, T. (2005). Advancing the Utility of the
Transcript: A Computer-Enhanced Methodology,
paper presented at the Twelfth International
Conference on Methods in Dialectology: Moncton, New Brunswick, Canada. August 2005.
Puglia, S. (2003). Overview: Analog vs. Digital for
Preservation Reformatting, paper presented at the 18th Annual Preservation Conference, March 27, 2003, at University of Maryland College Park. http://www.archives.gov/preservation/conferences/papers-2003/puglia.html
Smith, A., D. Allen, and K. Allen (2004). Survey of the State of Audio Collections in Academic Libraries. Council on Library and Information Resources Publication 128. http://www.clir.org/pubs/abstract/pub128abst.html
Ochs, E. (1979). Transcription as Theory, in Developmental Pragmatics, eds. Elinor Ochs and Bambi Schieffelin: 43-72, New York: Academic Press.
Trudgill, P. (1974). The Social Differentiation of English in Norwich, Cambridge: CUP.
Wolfram, W. (1993). Identifying and Interpreting
Variables. American Dialect Research. Ed. Dennis R.
Preston. Amsterdam: John Benjamins Publishing Company. 193-221.