Representational Adequacy in Bantu Language Resources: Tone as a Case for Relational Language Databases
DOI:
https://doi.org/10.55492/dhasa.v7i02.7687Keywords:
Tone, Bantu Languages, Property-Graph, Language ResourcesAbstract
Language models typically treat orthographic text as sufficient, but for Bantu languages like Sesotho, this text omits linguistic features such as tone and vowel quality, leading to ambiguities. We argue this is a representational, not data, problem. We propose shifting from static corpora to property-graph, multimodal, versioned databases that link orthography with speech, phonology, morphology, syntax, semantics, variation, and provenance, enabling task-specific dataset generation. Developing such databases requires linguistic expertise; we outline a curriculum to build capacity for creating phonological and phonetic representations, including tone, for computational use.
Downloads
Published
Issue
Section
License
Copyright (c) 2026 Mpho Primus, Sabine Zerbian

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.