Base

L

% of correct prediction of text fragment sequence

Advantage of NFIM (%)

By unconditional probability

By 1^st fragment

By 2^nd fragment

By mean probability

Method 1

Method 2

Method 3

Method 4

NFIM

“Russian”

1

37.74

25.71

24.19

19.62

19.66

16.51

17.26

19.75

11.48

43.77

2

21.88

14.63

14.59

11.80

11.81

11.22

10.90

11.48

8.98

21.34

“German”

1

24.10

18.11

13.11

12.47

12.89

11.84

10.72

10.86

7.60

41.09

2

20.02

14.17

12.11

11.18

11.20

10.48

10.10

9.84

7.52

30.90

“Ukrainian”

1

41.62

27.53

25.18

23.81

23.91

18.62

19.76

17.05

10.07

69.32

2

3.27

2.07

2.15

1.84

1.84

1.75

1.73

1.63

1.50

8.71

“English”

1

30.99

19.32

18.52

14.78

14.88

13.48

12.42

13.27

8.74

42.06

2

63.06

42.61

41.05

32.79

32.79

31.43

29.96

31.91

24.62

21.66