Parsing clinical text using the state-of-the-art deep learning based parsers: a systematic comparison

Table 1 Performance of deep learning based dependency parsers on the MiPACQ corpus (%)

Parser	Corpus	Word embeddings	UAS	LS	LAS
Stanford parser	Penn TreeBank	Gigaword	80.62	89.09	77.59
	MiPACQ	Gigaword	90.49	94.95	89.00
	MiPACQ	MIMICIII	90.30	94.84	88.75
Bist-parser	Penn TreeBank	Gigaword	81.08	89.35	78.20
	MiPACQ	Gigaword	90.72	95.18	89.25
	MiPACQ	MIMICIII	90.62	95.16	89.16
Dependency-tf	Penn TreeBank	Gigaword	79.14
	MiPACQ	Gigaword	88.65
	MiPACQ	MIMICIII	88.80
jPTDP-parser	Penn TreeBank	Gigaword	79.47	85.76	74.62
	MiPACQ	Gigaword	88.50	92.36	85.53
	MiPACQ	MIMICIII	88.94	92.69	86.10

Highest performance in terms of each evaluation criterion is highlighted in boldface

ISSN: 1472-6947