PAX - an annotation based concordancing toolkit

Dafydd Gibbon & Thorsten Trippel

{gibbon,trippel}@spectrum.uni-bielefeld.de 


We demonstrate the PAX - "Portable Audio Concordance System" - proof-of-concept
prototype of a multilingual audio concordance toolkit, which was designed to
support the efficient documentation of endangered languages (here mainly the
Ivory Coast languages Ega, Kwa, Koulango).

The main requirements were (1) Standardisation (coding requirements should
be standardised to ensure archive exchangeability and to avoid
incompatibilities e.g. with proprietary fonts, (2) Interoperability (the
implementation should be interoperable across the main platforms (Linux,
Windows, Mac), and (3) Low-cost, low-end (it should not depend on recent
software versions or high-end hardware in order to ensure usability under
local conditions, and with offline capability to reduce networking expense
as well as online capability). Full specification documents and a release
version can be found at:
	http://www.spectrum.uni-bielefeld.de/LangDoc/EGA/

PAX concordance design is based on a function f: CORPUS -> <KWIC,SIGTRANS>,
where CORPUS is a set of annotated signals partitioned for different languages,
KWIC is the keyword in context concordance, and SIGTRANS is a set of
representations of signal transformations (audio, waveform, F0, spectrogram).
The corpus consists of digital signals, mainly of spoken narratives, which were
annotated at word level. The user interface has three panels which are called
sequentially: (1) corpus options, (2) word options, (3) KWIC output with
options for hearing audio, and viewing waveform, F0, spectrogram.

The input data are time-aligned in SAMPA standard ASCII IPA coding (slightly
modified for the efficient coding of tone languages), using Praat, Transcriber
or esps/waves+ software. These operational formats are translated (with Perl
format converter modules) to an XML archive format using the TASX DTD (see
Milde, this Workshop), retaining the SAMPA coding. TASX-XML and SAMPA were
selected in order to fulful the archive exchange requirement.

PAX concordance architecture is modular, with wordlist extraction and KWIC
concordance construction modules (Perl), and signal extraction and
processing modules (C and Praat scripts). These modules feed two independent
GUI server modules, (1) a Perl CGI application with HTML and WAV output for
online use with a web server, (2) a Perl/Tk application for offline use without
a local CGI server. This hybrid implementation strategy was chosen in order to
to fulfil the low-cost, low-end, on/offline and interoperability requirements. 

The PAX toolkit is currently being extended to time-aligned multimodal data.
After the conclusion of the project, the tools will be made generally available
as GPL (or similar) free software.