Results

SANCL 2012 - Home‎ > ‎Shared Task‎ > ‎

Constituency Parsing Results:


	Domain A (answers)				Domain B (newsgroups)				Domain C (reviews)				Domain D (wsj)				Average (A-C)
Team	LP	LR	F1	POS	LP	LR	F1	POS	LP	LR	F1	POS	LP	LR	F1	POS	LP	LR	F1	POS
BerkeleyParser*	75.86	75.98	75.92	90.20	77.87	78.42	78.14	91.24	77.65	76.68	77.16	89.33	88.34	88.08	88.21	97.08	77.13	77.03	77.07	90.26
OHSU	73.21	74.60	73.90	90.15	73.22	75.48	74.33	91.14	76.31	76.22	76.27	90.05	83.17	83.79	83.48	96.84	74.25	75.43	74.83	90.45
Vanderbilt	75.09	76.78	75.93	91.76	78.10	79.05	78.57	92.91	77.74	78.18	77.96	91.94	87.82	88.00	87.91	97.49	76.98	78.00	77.49	92.20
IMS	79.46	78.10	78.78	90.22	80.85	80.12	80.48	91.09	81.31	78.61	79.94	89.93	89.83	88.96	89.39	97.31	80.54	78.94	79.73	90.41
Stanford	78.79	77.91	78.35	91.21	81.41	80.49	80.95	91.62	81.95	80.32	81.13	92.45	90.00	88.93	89.46	97.01	80.72	79.57	80.14	91.76
Alpage-1	80.67	80.36	80.52	91.17	84.22	83.12	83.67	93.22	82.01	81.04	81.52	91.58	90.20	89.62	89.91	97.20	82.30	81.51	81.90	91.99
Alpage-2	80.77	80.43	80.60	91.14	84.71	83.36	84.03	92.58	82.28	81.24	81.76	91.63	90.19	89.56	89.87	97.22	82.59	81.68	82.13	91.78
DCU-Paris13-2	80.02	79.22	79.62	91.61	83.13	82.18	82.65	93.60	82.92	82.12	82.52	92.96	88.43	88.29	88.36	97.29	82.02	81.17	81.60	92.72
DCU-Paris13-1	82.96	81.43	82.19	91.63	85.01	83.65	84.33	93.39	84.79	83.29	84.03	92.89	90.75	90.32	90.53	97.53	84.25	82.79	83.52	92.64

Dependency Parsing Results:


	Domain A (answers)			Domain B (newsgroups)			Domain C (reviews)			Domain D (wsj)			Average (A-C)
Team	LAS	UAS	POS	LAS	UAS	POS	LAS	UAS	POS	LAS	UAS	POS	LAS	UAS	POS
Zhang&Nivre*	76.60	81.59	89.74	81.62	85.19	91.17	78.10	83.32	89.60	89.37	91.46	96.84	78.77	83.37	90.17
UPenn	68.54	82.28	89.65	74.41	86.10	90.99	70.17	82.88	89.02	81.74	91.99	96.93	71.04	83.75	89.89
UMass	72.51	78.36	89.42	77.23	81.61	91.28	74.89	80.34	89.90	81.15	83.97	94.71	74.88	80.10	90.20
NAIST	73.54	79.89	89.92	79.83	84.59	91.39	75.72	81.99	90.47	87.95	90.99	97.40	76.36	82.16	90.59
IMS-2	74.43	80.77	89.50	79.63	84.29	90.72	76.55	82.18	89.41	86.88	89.90	97.02	76.87	82.41	89.88
IMS-3	75.90	81.30	88.24	79.77	83.96	89.70	77.61	82.38	88.15	86.02	88.89	95.14	77.76	82.55	88.70
IMS-1	78.33	83.20	91.07	83.16	86.86	91.70	79.02	83.82	90.01	90.82	92.73	97.57	80.17	84.63	90.93
Copenhagen	78.12	82.91	90.42	82.90	86.59	91.15	79.58	84.13	89.83	90.47	92.42	97.25	80.20	84.54	90.47
Stanford-2	77.50	82.57	90.30	83.56	87.18	91.49	79.70	84.37	90.46	89.87	91.95	95.00	80.25	84.71	90.75
HIT-Baseline	80.75	85.84	90.99	85.26	88.90	92.32	81.60	86.60	90.65	91.88	93.88	97.76	82.54	87.11	91.32
HIT-Domain	80.79	85.86	90.99	85.18	88.81	92.32	81.92	86.80	90.65	91.82	93.83	97.76	82.63	87.16	91.32
Stanford-1	81.01	85.70	90.30	85.85	89.10	91.49	82.54	86.73	90.46	91.50	93.38	95.00	83.13	87.18	90.75
DCU-Paris13	81.15	85.80	91.79	85.38	88.74	93.81	83.86	88.31	93.11	89.67	91.79	97.29	83.46	87.62	92.90

* Baseline models trained only on the Ontonotes WSJ training corpus. For constituents this is the publicly available BerkeleyParser (Petrov et. al ACL 2006). For dependencies this is a reimplementation of the transition-based parser of Zhang&Nivre ACL 2011 with the TnT (Brants ANLP 2000) part-of-speech tagger.

POS tag accuracies differ due to rounding errors and tiny discrepancies in evalb and eval.pl.

Comments

	Search this site

Results

Shared-Task Links