Datasetpublic

abeeranajam31/urdu-emergency-corpus

Urdu Emergency Communication Corpus (UEC) A small, balanced, annotated pilot corpus of simulated Urdu emergency-communication utterances, built to study which linguistic features distinguish low- from high-urgency communication using corpus-linguistic methods (frequency, keyness, collocation analysis). Full project, code, executed analysis notebooks, and research report: github.com/abeeranajam31/urdu-emergency-corpus ⚠️ Important: this is a simulated… See the full description on the dataset page: https://huggingface.co/datasets/abeeranajam31/urdu-emergency-corpus.

sourceHugging Facemitupdated 15d agoView on Hugging Face

0likes35downloads

Dataset Card

Urdu Emergency Communication Corpus (UEC)

A small, balanced, annotated pilot corpus of simulated Urdu emergency-communication utterances, built to study which linguistic features distinguish low- from high-urgency communication using corpus-linguistic methods (frequency, keyness, collocation analysis).

Full project, code, executed analysis notebooks, and research report: github.com/abeeranajam31/urdu-emergency-corpus

⚠️ Important: this is a simulated, researcher-constructed corpus

Not real emergency-call transcripts. All 60 utterances were authored by the researcher to be typologically representative of Urdu emergency communication, balanced across 5 situation types and 4 urgency levels. Real emergency-call data raises consent and privacy issues out of scope for this pilot — see the research report for the full rationale. Do not cite this as a naturalistic-speech corpus.

Composition

60 utterances, perfectly balanced: 15 per urgency level (low/medium/high/ critical) × 12 per emergency type (medical/accident/fire/crime/other), 3 per (type × urgency) cell.

Schema

Field	Type	Description
`id`	string	`UEC-###`
`text`	string	Urdu utterance
`english_gloss`	string	Rough English gloss (not a certified translation)
`emergency_type`	string	`medical`, `accident`, `fire`, `crime`, `other`
`urgency`	string	`low`, `medium`, `high`, `critical` — manually annotated
`speech_act`	string	`request`, `information`, `warning`, `distress`, `command`
`person_reference`	string	`first`, `third`, `both`, `none`
`repetition`	bool	Manually annotated: is a word/phrase repeated for emphasis (e.g. `جلدی جلدی`)?
`temporal_expression`	object	`{present, matches}` — lexicon-tagged (ابھی، فوراً، جلدی، کل، کچھ دیر پہلے)
`distress_marker`	object	`{present, matches}` — lexicon-tagged (خدا، پلیز، مدد، بچاؤ، بچائیں)
`intensifier`	object	`{present, matches}` — lexicon-tagged (بہت، انتہائی، بالکل، فوراً)
`negation`	object	`{present, matches}` — lexicon-tagged (نہیں، نہ)

Full annotation methodology: `data/annotation_schema.md` in the GitHub repo.

Loading

python

from datasets import load_dataset

ds = load_dataset("abeeranajam31/urdu-emergency-corpus", split="corpus")
print(ds[0])

Key finding (see full report for details)

Lexical repetition (e.g. جلدی جلدی "quickly quickly") is exclusive to high/critical-urgency utterances (0% at low/medium, 40% at critical) — an emergent pattern the annotation scheme did not assume in advance.

Limitations

Simulated (not real) data; single annotator, no inter-annotator agreement measured; small sample (n=60). Full discussion: research report §8.

Citation

bibtex

@misc{uec_corpus_2026,
  author = {Najam, Abeera},
  title = {Linguistic Signals of Urgency in Urdu Emergency Communication: A Corpus-Based and Computational Study},
  year = {2026},
  howpublished = {Independent research project},
  url = {https://github.com/abeeranajam31/urdu-emergency-corpus}
}

License

MIT

Contact

Abeera Najam — abeeranajam@gmail.com