HausaNLP at SemEval-2026 Task 7: Prompt-based Hausa Cultural Question Answering

Résumé

We describe HausaNLP's submission to SemEval-2026 Task 7 Track 1 (short-answer cultural question answering).Our system is a training-free, prompt-based pipeline targeting native Hausa (ha-NG).Two design decisions distinguish it from a generic zero-shot baseline.We use locale-conditional prompting: ha-NG questions receive a system prompt instructing concise standard Hausa output with explicit Boko-script characters (á, â, Î, ű).Second, we use a two-model fallback pipeline: GPT-4o handles the primary pass, and Gemini 1.5 Flash retries any rows where the primary call returned an error or empty output, separating modelknowledge failures from API-availability failures.On the official development leaderboard, our best run reached 36.4 accuracy.Error analysis shows that a non-trivial fraction of failures are placeholder strings caused by API errors rather than incorrect generations, and that surface-level mismatches (verbosity, orthographic variation) account for many of the remaining errors.Code, prompts, and processing scripts are released for reproducibility.

Citer ce document

Accès au document

Texte intégral en lecture en ligne, réservé aux abonnés SPHAERO et aux membres de l'institution. Se connecter

Auteur(s)

Statistiques

Consultations : 1

Téléchargements : 0