Skip to main content

Table 4 Monolingual experiments

From: Advances in monolingual and crosslingual automatic disability annotation in Spanish

System

Precision

Recall

F-measure

A−

A+

A−

A+

A−

A+ (stdev)

2018 Shared task best system [49]

 

75.00

 

81.00

71.46

78.60

Best published result [14]

79.00

83.00

69.00

79.00

74.00

81.00

BETO

77.73

78.41

66.57

74.22

71.64

75.98 (±2.15)

\(FLAIR_{SkipNG\_EHR}\)

83.23

83.05

73.07

80.64

77.82

81.82 (±1.51)

\(FLAIR_{FT}\)

84.90

83.67

71.18

81.22

77.43

82.43 (±1.28)

\(FLAIR_{Wiki2V}\)

85.63

84.27

76.27

80.78

80.67

82.64 (±0.37)

\(FLAIR_{LM\_EHR}\)

84.53

85.66

72.20

83.26

77.87

84.43 (±0.93)

Combined approaches

\(FLAIR_{B2}\) (\(FLAIR_{LM\_EHR}\)

   + \(FLAIR_{Wiki2V}\))

87.61

87.68

76.13

85.88

81.47

86.77 (±0.50)

\(FLAIR_{B3}\) (\(FLAIR_{LM\_EHR}\)

   + \(FLAIR_{Wiki2V}\)

   + \(FLAIR_{FT}\))

87.67

87.66

73.51

82.83

79.96

85.16 (±1.59)

  1. Results of the different approaches used for automatic disability annotation in Spanish (the best results are presented in bold). A−: without Acronym and abbreviation module. A+: with the Acronym and abbreviation module. The upper part of the table shows the results using a single source of pre-trained embeddings, while the lower part presents the combinations of the best two (B2) and three (B3) embedding types