Skip to Main Content
Table 8: 
Qualitative error analysis: The number (percentage) of error patterns of Arc-Zeager in pipeline/joint scenarios, on a sample of 100 sentences from the dev set, in uninfused lattices settings.
SystemStrategyMorphology ZeagerPipelineGold ZeagerPipelinePredicted ZeagerArcGreedyJoint 
Total Number of Errors 390 641 546 
TYPE 1:    
Could be Considered Correct 64 (16.4) 62 (9.6) 17 (3.1) 
Difficult Clause Attachment 62 (15.8) 103 (18.8) 109 (19.9) 
Difficult PP Attachment 27 (6.9) 41 (6.4) 30 (5.4) 
TYPE 2:    
Wrong arc due to Seg/TagError in the focus word 0 106 (16.5) 56 (10.2) 
Wrong arc due to Seg/TagError in other words 0 58 (9.0) 37 (6.7) 
Wrong arc label due to tag error 23 (3.6) 31 (5.6) 
TYPE 3:    
Gold Standard is Wrong 51 (13.0) 48 (7.5) 44 (8.0) 
Trainset is Inconsistent 68 (17.4) 60 (9.3) 69 (12.6) 
Prediction is Underspecified 65 (16.6) 76 (11.8) 83 (15.2) 
TYPE 4:    
Other 53 (13.6) 64 (9.9) 70 (12.8) 
SystemStrategyMorphology ZeagerPipelineGold ZeagerPipelinePredicted ZeagerArcGreedyJoint 
Total Number of Errors 390 641 546 
TYPE 1:    
Could be Considered Correct 64 (16.4) 62 (9.6) 17 (3.1) 
Difficult Clause Attachment 62 (15.8) 103 (18.8) 109 (19.9) 
Difficult PP Attachment 27 (6.9) 41 (6.4) 30 (5.4) 
TYPE 2:    
Wrong arc due to Seg/TagError in the focus word 0 106 (16.5) 56 (10.2) 
Wrong arc due to Seg/TagError in other words 0 58 (9.0) 37 (6.7) 
Wrong arc label due to tag error 23 (3.6) 31 (5.6) 
TYPE 3:    
Gold Standard is Wrong 51 (13.0) 48 (7.5) 44 (8.0) 
Trainset is Inconsistent 68 (17.4) 60 (9.3) 69 (12.6) 
Prediction is Underspecified 65 (16.6) 76 (11.8) 83 (15.2) 
TYPE 4:    
Other 53 (13.6) 64 (9.9) 70 (12.8) 
Close Modal

or Create an Account

Close Modal
Close Modal