DialoGPT on DailyDialog++: NLI Diversity increases from 4.11 to 10.24 with 6.3 samples

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

• DialoGPT on EmpatheticDialogues: NLI Diversity increases from 3.68 to 10.11 with 7.1 samplesfinding0.899
  DTG result for DialoGPT on EmpatheticDialogues using NLI metric
• BlenderBot on EmpatheticDialogues: NLI Diversity increases from -8.90 to -1.72 with 16.5 samplesfinding0.824
  DTG result for BlenderBot on EmpatheticDialogues; requires most resampling of all conditions
• Diversity Threshold Generation results in average 137% increase in NLI Diversity compared to standard generationfinding0.785
  Key headline result of the DTG procedure across all conditions
• NLI Diversity is limited by the scope of the NLI model's training dataclaim0.780
  Limitation acknowledged in discussion section
• DialoGPT generates more diverse utterances across all conditions than BlenderBot 1.0claim0.775
  Model comparison derived from Diversity Threshold Generation experiment results
• Confidence NLI Diversity achieves ρ=0.64 correlation with human diversity judgments on conTestfinding0.774
  Highest human correlation for semantic diversity metric
• Beam search DTG achieves ending NLI Diversity of 5.35 vs nucleus sampling's higher performance, starting from -5.05finding0.773
  Confirms nucleus sampling produces more semantically diverse outputs than beam search
• Confidence NLI Diversity achieves state-of-the-art performance on measuring semantic diversityclaim0.762
  Main performance claim of the paper