Skip to Main Content
Table 3

Comparison of Phase 1 (P1) classes and clusters extracted from Phase 2 (P2) distance matrices for classes 10, 15, and 3 against FrameNet fine-grained frames (frame) and parent frames (parent), and VerbNet top-level (top) and first-level (1st) classes. All scores are B-Cubed F-scores, measuring the overlap between P1 classes/P2 clusters and the FrameNet and VerbNet classes for the shared verbs. For Phase 2, optimal columns show scores obtained for the optimal clustering solution in terms of F1 score, determined iteratively over values of k = {1, …, N}, where N is the size of each class (#10–100, #15–87, #3–30); gold columns show scores for clustering solutions with k = Kgold, where Kgold is the number of classes in FrameNet or VerbNet in which the shared verbs participate. We do not report gold values where the number of gold classes was larger than the number of verbs in a given P2 sample (Kgold > N), due to multiple class membership of individual verbs in gold resources.

 FrameNet frame  VerbNet 1st  
parenttop
P1 0.247   –   0.302   –   
  
P2 koptimal gold optimal gold optimal gold optimal gold 
  
#10 0.527 0.247 0.407 – 0.666 0.259 0.481 0.337 
#15 0.470 0.289 0.448 – 0.407 0.324 0.449 0.388 
#3 0.546 0.501 0.578 – 0.642 0.566 0.618 0.616 
 FrameNet frame  VerbNet 1st  
parenttop
P1 0.247   –   0.302   –   
  
P2 koptimal gold optimal gold optimal gold optimal gold 
  
#10 0.527 0.247 0.407 – 0.666 0.259 0.481 0.337 
#15 0.470 0.289 0.448 – 0.407 0.324 0.449 0.388 
#3 0.546 0.501 0.578 – 0.642 0.566 0.618 0.616 
Close Modal

or Create an Account

Close Modal
Close Modal