Evaluating the effectiveness of Xitsonga named entity recognition
Bibliographic Data
| ID | 20253296 |
|---|---|
| Authors | Kulani Excellent Mhlongo (0000-0002-5845-8861, North-West University), Respect Mlambo (0000-0002-0348-6726, African Union, corresponding author) |
| Year | 2026 |
| Volume | 13 |
| Issue | 1 |
| Publication date | 2026-12-31 |
| Peer Reviewed | Yes |
| Open Access | Yes |
| Type | ARTICLE |
| Venue | Cogent Arts and Humanities (JOURNAL) |
| Journal identifiers | ISSN: 2331-1983 • E-ISSN: 2331-1983 |
| Publisher | Informa UK Limited (PUBLISHER • GB) |
| DOI | 10.1080/23311983.2026.2683929 |
| OpenAlex | W7165752056 |
| Language | EN |
| References cited | 23 |
The development of robust Named Entity Recognition (NER) systems for African languages such as Xitsonga remains constrained by limited data availability, complex morphological structures, and orthographic variation. These challenges underscore the need for systematic evaluation of existing models to assess their effectiveness and adaptability within indigenous linguistic contexts. This study evaluates the performance of the CTexT NCHLT Web Services Named Entity Recogniser for Xitsonga (NCHLT-NERXI) by employing standard evaluation metrics, including Precision, Recall, and F1-score. A 3,242-token dataset was web-scraped from the online Xitsonga newspaper Titimuleni Nthavela and subjected to quantitative analysis. The findings reveal variable performance of the NCHLT-NERXI model across different entity types, highlighting both its strengths and limitations in processing Xitsonga text. The model attained its highest accuracy in identifying location entities, with a Precision of 0.79, Recall of 0.76, and an F1-score of 0.78. Miscellaneous entities displayed a clear imbalance, achieving high Precision (0.88) but low Recall (0.48), yielding a moderate F1-score of 0.62. The person category showed more consistent results, with Precision (0.62), Recall (0.64), and an F1-score of 0.63. Conversely, the organisation category recorded the weakest performance, with Precision (0.45), Recall (0.17), and an F1-score of 0.25. These results collectively highlight the NCHLT-NERXI’s strength in structured categories such as locations and persons, while revealing ongoing challenges in complex or infrequent entities like organisations
Entity linking · Named entity · Data Quality and Management · Natural Language Processing Techniques · Topic Modeling
| Citation velocity | historical |
|---|---|
| Highly cited | No |