Spell Error Corpus for Minang Language

The Spell Error Corpus for Minang Language (SPEML) is a dataset designed to support spelling error detection and correction for Minangkabau, a low-resource language of Indonesia. The base corpus was derived from a publicly available Indonesian–Minangkabau parallel corpus in the minangNLP GitHub repository. From the aligned sentence pairs, only the Minangkabau column was extracted, yielding 16,372 sentences. The texts were preprocessed through normalization, cleaning, and deduplication, resulting in 16,334 unique Minangkabau sentences. SPEML contains 164,662 misspelled word forms organized into seven error categories. Spelling errors were generated using rule-based procedures covering character insertion (1–3 characters), character deletion (1–3 characters), character substitution (1–3 characters), character transposition (single adjacent swap), punctuation errors, real-word errors, and loanword errors, followed by expert validation for linguistic plausibility. This dataset can be reused to train, test, and benchmark NLP models, including spell checkers and language models, for Minangkabau spelling error detection and correction, as well as to evaluate methods by error type and edit length.