Table 4.

Machine learning classification results of models trained on manually corrected transcripts without pauses compared to results of models trained on manually corrected transcripts (with pauses).

Task and model type		Transcripts without pauses AUROCa	Transcripts with pauses AUROC	Change in AUROCb
Picture description
	RFc	0.666	0.687	0.021
	GNBd	0.730	0.725	−0.005
	LRe	0.755	0.743	−0.012
	BERTf	0.686	0.691	0.005
Experience description
	RF	0.631	0.636	0.005
	GNB	0.676	0.677	0.001
	LR	0.692	0.674	−0.018
	BERT	0.622	0.649	0.027

^aAUROC: area under the receiver operating characteristic curve.

^bPositive change in AUROC indicates that the pause model outperformed the no-pause model.

^cRF: random forest.

^dGNB: Gaussian naive Bayes.

^eLR: logistic regression.

^fBERT: Bidirectional Encoder Representations from Transformers.