A human-in-the-loop pipeline for cleaning the South African MultilingualLearner Corpus of Academic Texts (SAMuLCAT) with XML tagging
DOI:
https://doi.org/10.55492/dhasa.v7i02.7677Keywords:
data cleaning, mojibake, South African Multilingual Learner Corpus of Academic Texts (SAMuLCAT), student writing, XML taggingAbstract
Empirical analyses of student assignments require clean, structured data for human and machine handling, but raw digital submissions can contain unforeseen character-encoding artefacts, formatting inconsistencies, and boilerplate. The paper summarises the approach and outcomes of a data-cleaning exercise for a corpus of South African student writing. Python regular expressions, rule-based heuristics, and audit logs were employed to distinguish student prose and repair character corruptions. The automated stage achieved over 80% accuracy in XML tagging, with post-editing resolving edge cases. Encoding repairs account for 1.03% of the characters in the corpus, and extraneous data removed amounts to 13% of the original text.
Downloads
Published
Issue
Section
License
Copyright (c) 2026 Bronwyn J. Bowles-King

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