finding
active
finding:beam-search-dtg-achieves-ending-nli-diversity-of-5-35-vs-nucleus-sampling-s-higher-performance-starting-from-5-05Beam search DTG achieves ending NLI Diversity of 5.35 vs nucleus sampling's higher performance, starting from -5.05
Confirms nucleus sampling produces more semantically diverse outputs than beam search
Source paper
extracted_from(2022) · Katherine Stasaski · Marti A. Hearst
Neighborhood — ranked by edge-count
Papers (1)
paper
Related by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- New semantic diversity dimension added to prior finding that nucleus sampling is more lexically diverse
- DTG result for DialoGPT on DailyDialog++ using NLI metric
- Key headline result of the DTG procedure across all conditions
- Confidence NLI Diversity achieves ρ=0.64 correlation with human diversity judgments on conTestfinding0.743Highest human correlation for semantic diversity metric
- BlenderBot on EmpatheticDialogues: NLI Diversity increases from -8.90 to -1.72 with 16.5 samplesfinding0.740DTG result for BlenderBot on EmpatheticDialogues; requires most resampling of all conditions
- Probing limited.
- Quantifies how much of the base model's diversity VS can recover compared to baseline prompting
- Probes fire on 36.9%/31.1% (Kimi) and 34.0%/29.9% (GLM) of τ³-bench/ShoppingBench vs monitor 3.9–8.4%finding0.738Out-of-distribution firing.