A Bichannel Transformer with Context Encoding for Document-Driven Conversation Generation in Social Media

Table 3

Human evaluation of baselines and our proposed model.

Models	Fluency	Dialogue coherence	Diversity
SEQ [24]^§	1.27	0.81	0.42
SEQS [24]^§	1.13	0.96	0.71
D3G [26]^§	1.29	1.12	0.84
Transformer [25]^¶	1.34	1.17	0.90
Incremental transformer (our impl)‡	1.35	1.27	0.93
BCTCE	1.34	1.29	0.95

The marks in this table are the same as those in Table 1.