Skip to main content

Table 5 Semantic role induction results achieved using different contextualized representations of verb arguments

From: Semantic frame induction through the detection of communities of verbs and their arguments

  d Edges CC Clusters Purity F1 BCubed F1
BERT 0.37 348,761 0.81 2.00 ±0.00 57.27 ±0.00 45.69 ±0.00
Context-Free 0.90 157,154 0.55 4.47 ±0.88 65.30 ±0.69 52.28 ±0.76
Syntactic Context 0.78 146,978 0.57 3.57 ±0.88 68.44 ±5.34 55.85 ±4.13
Semantic Context 0.73 97,442 0.56 5.93 ±1.00 71.88 ±1.61 56.91 ±1.94
Free + Syntactic 0.84 140,768 0.56 3.80 ±0.98 65.41 ±1.05 52.89 ±0.77
Syntactic + Semantic 0.73 101,180 0.57 5.63 ±0.80 70.66 ±3.29 56.36 ±2.11
All 0.78 143,273 0.57 4.40 ±0.61 67.97 ±3.50 54.41 ±2.82
  1. The first row refers to BERT representations, the second block to the different levels of ELMo representations, and the third block to combinations of those levels. d refers to the neighboring threshold. CC refers to the clustering coefficient