Naija POS-Tagged Data

The Naija POS-Tagged QA Data dataset contains interlinear text annotated with word-level part-of-speech (POS) labels and English glosses where available. The filename and examples in the file suggest that the language is Naija/Nigerian Pidgin; however, the XML does not explicitly identify the language variety, source texts, sampling or collection procedure, or annotation guidelines. These details should be confirmed from the project records where possible. The supplied file contains 5,275 numbered phrase/line entries and 78,321 base-language form entries, including punctuation. English glosses are present for 69,591 entries, and 69,586 entries have POS labels. The file contains 16 distinct POS labels: v, n, pro, adp, adj, det, subordconn, q, coordconn, adv, nprop, verbprt, interj, num, pro-form, and prt. The labels are preserved as they appear in the source file. Because the XML does not provide a formal tagset specification, the intended meanings of some labels—particularly project-specific labels—should be checked against the original annotation guidelines if available. The data file is XML 1.0 encoded as UTF-8, but its structure uses Microsoft Word 2003 WordML with interlinear content represented through paragraph styles and Visual Markup Language (VML) text boxes. Base-language forms, English glosses, and POS labels are aligned in these text boxes rather than encoded as a simple set of explicit token elements and attributes. Users developing scripts to process the data should account for this WordML representation. Some entries have no POS label or English gloss: 8,735 entries lack a POS label and 8,730 lack an English gloss. Punctuation and other entries may account for some of these omissions; their absence should not automatically be treated as annotation errors. These counts describe the supplied file and do not constitute an independent assessment of annotation quality. The accompanying README, data dictionary, and POS-tag inventory provide further information about the file structure, observed labels, and coverage. The XML itself does not specify the corpus sources, authorship or annotators, quality-assurance procedure, or licence for redistribution. Users should consult the Zenodo record and project documentation for provenance, citation, and applicable reuse terms, and should confirm that they have permission to redistribute any underlying source text.