Speaker
Description
Computing processes typically require input data to perform actions
that generate output data. While input data can sometimes be generated
computationally, it often originates from external sources. In Natural
Language Processing and Digital Humanities, this input is typically
sourced from human activities, including spoken or written language
and music.
In the current era of Large Language Models (LLMs) that provide
practically usable tools, access to appropriate training data is
essential. These models generally perform better when larger data
collections are available for training, making data accessibility
crucial.
For most South African languages, only limited amounts of digitally
accessible data are available. Many existing data collections are
sourced from government websites, providing texts in highly specific
genres. Texts from diverse genres --- including newspaper articles,
literary works, and social media data --- are not openly accessible in
digital formats.
The SADiLaR (South African Centre for Digital Language Resources)
repository hosts data collections that are as openly accessible as
possible. The repository currently contains 357 directly downloadable
items and 56 metadata-only items (indicating the existence of data
collections).
The underlying principle of SADiLaR's repository is that providing a
centralized space for data collections makes them more easily findable
and accessible. (Additionally, submitted data collections are
requested to be as interoperable and reusable as possible to ensure
adherence to FAIR principles.)
This abstract serves as a call for action with two main objectives.
First, we encourage researchers to submit their digital language data
to the SADiLaR repository. Contributing to the repository increases
the availability of South African language data, making it more easily
findable and accessible. This data can then be used for training models,
ultimately benefiting language users, for example, through the
development of LLMs for these languages. (Note: copyright remains with
the original copyright owner; contributing to the repository does not
transfer copyright ownership.)
Each contribution to SADiLaR's repository receives a persistent
identifier, enabling consistent referencing of data collections. These
identifiers can be used as citations in publications, ultimately
benefiting researchers associated with the data collections.
Second, we encourage researchers to utilize the data collections
available in SADiLaR's repository. The repository contains a wealth of
useful data collections, and searching there first can streamline
research processes. SADiLaR's repository exists to facilitate work in
Natural Language Processing and Digital Humanities; collectively, we
can leverage this cyber-infrastructure to advance our fields.
| Institute | South African Centre for Digital Language Resources |
|---|---|
| Presenting Author | Menno van Zaanen |
