Skip to main content

Table 3 Experimental results on GEC

From: A multi-task learning framework for efficient grammatical error correction of textual messages in mobile communications

Work

Method

CoNLL-14

BEA-19

P

R

\(\user2{F}_{{{\mathbf{0}}{\mathbf{.5}}}}\)

P

R

\(\user2{F}_{{{\mathbf{0}}{\mathbf{.5}}}}\)

Zhao et al. [7]

NMT

65.5

33.2

54.9

-

-

-

LaserTagger [31]*

SL

50.9

26.9

43.2

53.4

38.5

49.6

GECToR [12]**

SL

66.8

33.7

55.8

64.2

47.0

59.8

Chen et al. [32]

SL+NMT

66.0

24.7

49.5

62.7

38.6

55.7

Previous (BERT)[15]

SL

55.1

32.3

48.3

50.6

40.5

48.2

Previous (SpanBERT)[15]

SL

57.2

34.1

50.4

54.5

44.3

52.1

Our approach (BERT)

SL

56.3

35.0

50.2

54.8

48.2

53.4

Our approach (SpanBERT)

SL

59.8

36.5

53.0

57.8

50.4

56.1

  1. * The result is implemented by Chen et al. [32]
  2. ** The result is implemented by us with official released code with BERT