previous next Title Contents

3 Results


Below we give the results of testing. A number of tables are provided to show the performance of the two products with respect to the various coverage sub-attributes and error types. For each of the three functionality attributes, however, we have also provided one general table giving overall measures of both products for comparison purposes.

3.1 Lexical coverage

Common word coverage: open classes

Product: System A





Class interval

(no. of words)

Relative frequency

Total no. of words

No. of words recognised
Percentage of words recognised
1-189
48
189
186
98
190-1732
21
1543
1527
99
1733-6780
13
5048
4698
93

Using the relative frequencies of each class as weights, the weighted mean score is:

48 * 98 + 21 * 99 + 13 * 93 / 48 + 21 + 13 = 97

Common word coverage: open classes

Product: System B





Class interval

(no. of words)

Relative frequency

Total no. of words

No. of words recognised
Percentage of words recognised
1-189
48
189
189
100
190-1732
21
1543
1533
99
1733-6780
13
5048
4783
95

Using the relative frequencies of each class as weights, the weighted mean score is:

48 * 100 + 21 * 99 + 13 * 95 / 48 + 21 + 13 = 99

Common word coverage: closed classes

Product: System A



Total no. of words

No. of words recognised
Percentage of words recognised
748
736
98

Common word coverage: closed classes

Product: System B



Total no. of words

No. of words recognised
Percentage of words recognised
748

746

100

Coverage of loan words

Product: System A



Total no. of words

No. of words recognised
Percentage of words recognised
135

53

39

Coverage of loan words

Product: System B



Total no. of words

No. of words recognised
Percentage of words recognised
135

69

51

The results obtained for lexical coverage are summed up below in a more concise form, disregarding information on the number of words contained in the various lists, and giving the weighted mean as total percentage score for common word coverage:

Lexical coverage





Product

Common words

(open classes)

weighted mean %

Common words

(closed classes)

simple %

Loan words

simple %

System A

97

98
39
System B

99

100
51

3.2 Error coverage

The first two tables below show the results obtained relative to the various corruption rules. Note that, for the reasons explained earlier, results for rule 2 have been omitted. In the third table, the rules have been grouped according to more general error types, so that a more concise view of the two products' performance can be provided. The error types referred to here are defined in D12: note, in particular, that the category "errors in loan words" subsumes a variety of different spelling errors occurring in relation to certain letter combinations that are especially difficult for Danish writers. The error type should not be confused with the coverage sub-attribute "loan words".

Error coverage

Product: System A




Error types

No. of errors

generated

No. of errors

signalled

% of errors

signalled

1. rer > re
16
16
100
3. et > ed
235
235
100
4. VrV > VV
445
445
100
5. VrC > VC
907
902
99
6. nd > n
682
671
98
7. r > rd
1142
1142
100
8. k > g
1063
1061
100
9. ce > se
58
58
100
10. ch > sj
14
14
100

Error coverage

Product: System B




Error types

No. of errors

generated

No. of errors

signalled

% of errors

signalled

1. rer > re
16
12
75
3. et > ed
235
235
100
4. VrV > VV
445
440
99
5. VrC > VC
907
857
94
6. nd > n
682
630
92
7. r > rd
1142
1130
99
8. k > g
1063
1025
96
9. ce > se
58
58
100
10. ch > sj
14
14
100

Error coverage






Product

Errors related to letter r

(1) %

Other suffixation errors (3) %
Silent letters

(4-7)

%

Letter substitution

(8) %

Errors in loan words

(9-10) %

System A

100

100
99
100
100
System B

75

100
96
96
100

3.3 Suggestion adequacy

The tables below give figures on suggestion adequacy. The first two show the behaviour of the two products with respect to the various corruption rules. The third one is a more concise table, which abstracts away from error type.

Suggestion adequacy

Product: System A






Error types


No. of errors

signalled

1st sugg. correct

Correct sugg. among 2nd-4th


Correct sugg. not among first 4


No sugg. offered




No.

%
No.
%
No.
%
No.
%
1. rer > re
16
3
19
13
81
0
0
0
0
3. et > ed
235
45
19
188
80
2
1
0
0
4. VrV > VV
445
329
74
78
18
6
1
32
7
5. VrC > VC
902
731
81
107
12
18
2
46
5
6. nd > n
671
447
67
142
21
53
8
29
4
7. r > rd
1142
1052
92
28
3
3
0
59
5
8. k > g
1061
806
76
138
13
49
5
68
6
9. ce > se
58
36
62
16
28
2
3
4
7
10. ch > sj
14
1
7
1
7
3
22
9
64

Suggestion adequacy

Product: System B






Error types


No. of errors

signalled

1st sugg. correct

Correct sugg. among 2nd-4th


Correct sugg. not among first 4


No sugg. offered




No.

%
No.
%
No.
%
No.
%
1. rer > re
12
2
17
7
58
3
25
0
0
3. et > ed
235
3
1
79
34
153
65
0
0
4. VrV > VV
440
309
70
63
14
47
11
21
5
5. VrC > VC
857
610
71
132
15
77
9
38
5
6. nd > n
630
433
69
104
16
68
11
25
4
7. r > rd
1130
960
85
93
8
36
3
41
4
8. k > g
1025
849
83
80
8
64
6
34
3
9. ce > se
58
48
83
6
10
1
2
3
5
10. ch > sj
14
10
72
1
7
0
0
3
21

Suggestion adequacy






Product

No. of errors

signalled

1st sugg. correct (%)
Correct sugg. among 2nd-4th (%)
Correct sugg. not among 2nd-4th (%)
No sugg. offered
System A

4544

76
16
3
5
System B

4401

73
13
10
4

previous next Title Contents