Low-Resource Indonesian Health Conversations

The dataset consists of 5,000 question–answer (QA) pairs collected and annotated for low-resource Indonesian health conversations in a child education context. Each QA pair represents a natural interaction between elementary school students and a conversational health system, covering everyday health topics such as hygiene, nutrition, exercise, and disease prevention.

The queries are characterized by short sentence structures, simple vocabulary, and direct questioning patterns, reflecting the linguistic and cognitive profiles of children. The corresponding responses are designed to be concise, informative, and pedagogically appropriate, providing clear explanations that support understanding rather than merely delivering factual answers.

All data instances are annotated using a hierarchical intent schema, consisting of coarse-grained categories and fine-grained sub-intents. This structure enables the modeling of semantic relationships between health concepts and supports more accurate intent classification in ambiguous or overlapping cases.

Despite its moderate size, the dataset reflects a realistic low-resource scenario, where annotated conversational data in Indonesian remains limited. At the same time, the scale of 5,000 QA pairs provides sufficient diversity and coverage to support robust training and evaluation of transformer-based conversational models.

Overall, the dataset captures both the simplicity of child-oriented language and the semantic complexity of health-related queries, making it suitable for research on hierarchical intent modeling, context-aware dialogue systems, and explainable NLP in educational settings.