CareCorpus+ for Environment: Extended Classification and Augmentation of Caregiver Strategies to Create Enabling Environments for Young Children with Disabilities

ACL ARR 2026 January Submission8253 Authors

Keywords: caregiver strategies, pediatric rehabilitation, data annotation, data augmentation, low-resource NLP

Abstract: Caregivers play a central role in supporting the participation of children with disabilities in everyday activities. Environmental strategies, such as rearranging routines or spaces, are central to this support, yet they remain scarce and difficult to model systematically due to their free-text nature. To address this gap, we present 1,848 environment-focused caregiver strategies from pediatric rehabilitation contexts, manually annotated into five clinically grounded subcategories capturing activity demands, social contexts, physical environments, policies, and resources. Using this dataset, we benchmark a multi-class classification task and evaluate both manual and synthetic data augmentation. BERT-based models achieve improvements of up to 32% in macro-F1, supporting systematic evaluation of NLP methods for modeling caregiver strategies in low-resource rehabilitation settings.

Paper Type: Long

Research Area: Resources and Evaluation

Research Area Keywords: corpus creation, benchmarking, NLP datasets, evaluation methodologies

Contribution Types: Approaches to low-resource settings, Data resources, Data analysis

Languages Studied: English

Submission Number: 8253