Creating a database and query-tools for a large, multi-speaker
linguistic corpus

Jeff Good and Ronald Sprouse
The Turkish Electronic Living Lexicon (TELL)
University of California, Berkeley

The Turkish Electronic Living Lexicon (TELL) represents the first
large-scale effort to collect recordings of a very large number of
morphological paradigms (more than 17,000) from several speakers of
Turkish. The nature of this data presents several challenges for the
creation of the database itself and the query tools that access it. The
primary dimensions along which the database is organized include (1)
speaker, (2) paradigm, and (3) utterance. Each of these will be
discussed in turn.

Speaker: Specifying that an uttered paradigm is associated with a
particular speaker is straightforward. However, to allow comparison
between speakers, the database must also include a way of stating that
paradigms from different speakers can be from the same abstract "word"
since different speakers may have different phonological representations
for a given word.

Paradigm: A paradigm is a complex object that cannot simply be taken to
be a set of utterances. Instead, it is a set utterances, all having the
same part of speech and each of which is labelled for a particular
morphological category. In addition, TELL associates orthographic,
semantic, and etymological information with a paradigm as well as
marking paradigms for a specific set of phonological irregularities
which may be of interest to those researching Turkish phonology.

Utterance: Utterances are also complex objects consisting of both a
transcription and the recorded utterance itself. One of the goals of
TELL is to make both the transcriptions and the digitized recordings
available online.

TELL has implemented an SQL relational database which solves the
representational issues of the database. Using a system of lexical data
tables, containing actual paradigms, and a system of pointers and
relational tables, it is possible not only to represent complex internal
relationships, like those in a paradigm, but also possible to link
related paradigms from different speakers and to link a given paradigm
to orthrographic and etymological information.

An additional challenge for TELL is the design of an online user
interface. Possible user-specified search variables include: speaker,
phonological string, morphological form, etymology, and phonological
irregularity. Since a detailed user interface would be unnecessarily
complex for the occasional user of the database, we have implemented
both simple and advanced interfaces. We expect the majority of casual
users to prefer the simple search interface, while serious researchers
will appreciate the enhancements of the advanced interface, which allows
full access to linguistically-important variables, including phonemic
representations, syllabic templates, speaker, and etymology.