phuthangdev/ncb_nlp_dataset

Dataset containing noun classes for some Niger Congo Bantu languages # NCB Noun Class Dataset A structured dataset of noun class and concord features for Niger-Congo B (NCB) languages, developed to support the **bpMetric** bootstrapping potential metric — a tool for predicting how easily NLP resources can be adapted between related Bantu languages. This dataset was created as part of the MSc thesis *"Developing a Metric for Assessing the Bootstrapping Potential of Natural Language Processing Resources Between Bantu Languages"* (Phuthang Makhupane, University of Cape Town, 2026). ## What's in this dataset Each row describes one noun class for one language, along with its associated concord forms. Four languages are comprehensively documented — isiZulu, Sesotho, Setswana, and isiXhosa — with partial coverage for additional NCB languages including Shona, isiNdebele, Swahili, and Silozi. ### Columns | Column | Meaning | |---|---| | `LANGUAGE` | The language the row describes (e.g. `zulu`, `sotho`, `tswana`, `xhosa`) | | `NOUNCLASS` | The noun class number (e.g. `1`, `1a`, `2`, `2a`, ...) | | `FULL_PREFIX` | The full noun class prefix for that class | | `QC_ALL` | All-inclusive quantitative concord ("all of") | | `RC` | Relative concord | | `QC_EXISTS` | Existential quantitative concord ("some/there exists") | | `SC` | Subject concord | | `NEG_SC` | Negative subject concord | | `PC` | Possessive concord | | `PLURAL_CLASS` | The noun class this class pluralises into (`null`/`--` where not applicable) | Missing or inapplicable values are encoded consistently: `null` denotes an unknown/undocumented value, and `--` denotes a confirmed absence. Both are excluded when computing feature-set overlap, to avoid skewing similarity calculations. ## How it's used These nine features (`NOUNCLASS`, `FULL_PREFIX`, `QC_ALL`, `RC`, `QC_EXISTS`, `SC`, `NEG_SC`, `PC`, `PLURAL_CLASS`) are each treated as a set per language, and compared pairwise using the Jaccard similarity index. The average across all nine features produces the bpMetric score for a …