PERLEX: A Bilingual Persian-English Gold Dataset for Relation Extraction

Table 8

F1-scores for the nine classes of PERLEX for all six models.

Class	Baseline (%)	CNN (%)	Att-BLSTM (%)	BLSTM-LET (%)	R-BERT (%)	BERTEM-MTB (%)
Cause-Effect	77.41	80.53	82.13	81.06	83.51	86.07
Component-Whole	53.22	62.56	63.59	63.74	66.87	67.09
Content-Container	70.31	75.84	73.82	75.90	77.24	78.53
Entity-Destination	73.53	75.82	79.24	81.05	84.77	86.00
Entity-Origin	60.95	66.80	66.80	66.92	74.95	76.08
Instrument-Agency	64.81	57.78	54.72	59.56	62.25	69.51
Member-Collection	58.66	70.25	67.56	69.55	75.17	76.00
Message-Topic	59.07	71.94	74.90	75.33	81.34	85.19
Product-Producer	53.07	61.99	63.74	64.02	71.67	74.46