MIT Press

Table 14

Ablation study results for the axiom parsing component. We remove features of the axiom parsing component one by one as listed in Table 6 and observe the fall in performance in terms of the axiom parsing performance as well as the overall performance to gauge the value of the various features.

	F1		SAT Scores
	Literals	Full Parse	Practice	Official	Textbook
Span Similarity	71.8	64.6	51	40	42
No. of Relations	82.3	70.5	60	51	49
Span Lengths	86.0	72.0	63	54	50
Relative Position	83.9	69.2	60	52	47
Discourse Markers	77.4	68.4	55	48	47
Punctuations	73.5	65.0	52	45	45
Text Organization	74.4	66.2	52	47	46
RST Parse	84.6	70.8	62	52	49
Soricut & Marcu	83.2	69.8	61	52	50
Head Node, etc.	85.3	71.6	62	54	49
Syntax	75.5	66.6	54	47	46
Dominance	73.9	66.1	53	47	44
XML Structure	77.6	68.0	59	51	46
Unablated	87.5	73.2	64	55	51

	F1		SAT Scores
	Literals	Full Parse	Practice	Official	Textbook
Span Similarity	71.8	64.6	51	40	42
No. of Relations	82.3	70.5	60	51	49
Span Lengths	86.0	72.0	63	54	50
Relative Position	83.9	69.2	60	52	47
Discourse Markers	77.4	68.4	55	48	47
Punctuations	73.5	65.0	52	45	45
Text Organization	74.4	66.2	52	47	46
RST Parse	84.6	70.8	62	52	49
Soricut & Marcu	83.2	69.8	61	52	50
Head Node, etc.	85.3	71.6	62	54	49
Syntax	75.5	66.6	54	47	46
Dominance	73.9	66.1	53	47	44
XML Structure	77.6	68.0	59	51	46
Unablated	87.5	73.2	64	55	51

View Large

This Feature Is Available To Subscribers Only