Video clips from the head camera and transcripts of adults speaking to the infant were fed to the authors’ model, which used a ‘contrastive’ approach for both visual and language learning.