Table 12 depicts the contingency table of the classifications by the experts (rows) and one randomly chosen run of the Attribute Bagging classifier with i = 100 (columns). The table shows that there are two major sources of errors: First, the confusion between the qualitative and event classes, which is responsible for 14 errors (see dark-gray shaded cells in the table; also note that the related Q–QE and E–QE misclassifications account for another 14 errors). To compare, note that the confusion between the qualitative and relational classes only accounts for six of the errors, and there are no cases of confusion between event and relational adjectives.

Table 12 

Contingency table comparing the gold standard (rows) against run 2 of the Attribute Bagging classifier with i = 100 (columns). Dark-gray cells highlight the confusion between the qualitative and event classes; medium-gray cells highlight the overgeneration of polysemous adjectives; light-gray cells highlight the opposite case, that is, the generation of monosemous adjectives that should have been tagged as polysemous.

.	Q .	E .	R .	QR .	QE .	ER .	Total .
Q	90		2			0	107
E		17	0	1			37
R	4	0	20		0		30
QR		0		13	0	1	23
QE			0	0	5	0	7
ER	0	0		1	0	3	6
Total	110	22	28	22	19	9	210

.	Q .	E .	R .	QR .	QE .	ER .	Total .
Q	90		2			0	107
E		17	0	1			37
R	4	0	20		0		30
QR		0		13	0	1	23
QE			0	0	5	0	7
ER	0	0		1	0	3	6
Total	110	22	28	22	19	9	210

View Large