Fig. 13

Boxplots showing the F1 scores obtained from different classifiers: a distributions of the F1 scores in the picture description and story recall tasks, b distributions of the F1 scores for web and phone-based recordings, and c distributions of the F1 scores in the linguistic and acoustic features. These boxplots show that the picture description task, phone-based recordings, and acoustic features provide better performance than the story recall task, web-based recordings, and linguistic features