Verb-sense Classifier

This system addresses the verb sense disambiguation (VSD) problem, a sub-problem of word sense disambiguation (WSD), for English. It predicts the sense that a verb features in a given sentence, from among other, potential different meanings.

This system was developed as a part of Semantic Role Labeling system.

Performance

Here is the performance of this system after the most recent training:

Label	Correct	Excess	Missed	Precision	Recall	F1
01	4230	205	113	95.38	97.4	96.38
02	378	88	147	81.12	72	76.29
03	178	30	60	85.58	74.79	79.82
04	51	16	14	76.12	78.46	77.27
05	16	10	16	61.54	50	55.17
06	10	8	10	55.56	50	52.63
07	3	2	3	60	50	54.55
08	6	4	0	60	100	75
09	3	2	3	60	50	54.55
10	0	3	3	0	0	0
11	10	1	0	90.91	100	95.24
12	3	1	3	75	50	60
13	1	2	1	33.33	50	40
14	5	1	3	83.33	62.5	71.43
15	0	2	0	0	0	0
16	0	1	0	0	0	0
17	1	1	0	50	100	66.67
18	0	1	0	0	0	0
19	0	0	1	0	0	0
20	0	0	1	0	0	0
All	4895	378	378	92.83	92.83	92.83

Usage

While you can dig through the code to use it directly, we suggest you use it through our pipeline. More details in pipeline’s instructions.

Citation

If you use this system, and want to give credits to our system, please cite the following work:

@inproceedings{PRYZT04,
    author = {V. Punyakanok and D. Roth and W. Yih and D. Zimak and Y. Tu},
    title = {Semantic Role Labeling via Generalized Inference over Classifiers Shared Task Paper},
    booktitle = {CoNLL},
    pages = {130--133},
    year = {2004},
    comment = {Semantic Parsing; Structure Learning with Expressive Constraints; Constraint Optimization; Integer Linear Programming},
}