MIT Press

Skip to Main Content

Table 6:

Alternative models’ evaluation performances with different LM base models and sentence embedding models (Sent Emb). All models are trained with the same ITI+SF+SI objective.

Base Model	Sent Emb	Clustering	Disambiguation		Span Detection
Base Model	Sent Emb	Homogeneity	F1	Acc	Seq Acc	Tkn Recall	Tkn Acc
BART	MPNet	0.6450	95.73	93.25	76.01	90.75	98.17
BART	BART	0.4671	95.75	93.29	74.55	88.66	98.02
BERT	MPNet	0.4879	91.42	86.36	56.05	78.19	97.34

Base Model	Sent Emb	Clustering	Disambiguation		Span Detection
Base Model	Sent Emb	Homogeneity	F1	Acc	Seq Acc	Tkn Recall	Tkn Acc
BART	MPNet	0.6450	95.73	93.25	76.01	90.75	98.17
BART	BART	0.4671	95.75	93.29	74.55	88.66	98.02
BERT	MPNet	0.4879	91.42	86.36	56.05	78.19	97.34