Punctuation/recase module performs badly... has to be redesigned

[proycon]

No description provided.

[proycon]

New module is implemented but has to be tested more thoroughly and parameters have to be tweaked. Initial evaluation of Valkuil on CITO data show recasing and deletions are not/hardly working and precision/recall of missing punctuation insertion is still very low.

OVERALL RESULTS
=================
 Documents                                  :  520
 Total number of corrections in output      :  4390
 Total number of corrections in reference   :  10831
 Matching output corrections (tp)           :  1926
 Missed output corrections (fp)             :  2464
 Missed reference corrections (fn)          :  9118
 Virtual total (tp+fn)                      :  11044
 Precision (micro)                          :  0.44
 Recall (micro)                             :  0.17
 F1-score (micro)                           :  0.25

Aggregated corrections when they are on the same words:
 Aggregated average corrections in output              :  1.04
 Total number of aggregated corrections in output      :  4072
 Total number of aggregated corrections in reference   :  10831
 Matching output aggregated corrections (tp)           :  1713
 Missed output aggregated corrections (fp)             :  2359
 Missed reference aggregated corrections (fn)          :  18236
 Virtual total (tp+fn)                                 :  19949
 Aggregated precision (micro)                          :  0.42
 Aggregated recall (micro)                             :  0.09
 Aggregated F1-score (micro)                           :  0.14

PER-MODULE RESULTS
====================
Precision for confusible_de_het :  0.45     (89/196)
Precision for confusible_deze_dit :  0.37     (7/19)
Precision for confusible_hard_hart :  1.0     (1/1)
Precision for confusible_hun_zij :  0.48     (10/21)
Precision for confusible_licht_ligt :  1.0     (3/3)
Precision for confusible_me_mijn :  0.8     (53/66)
Precision for confusible_u_uw :  0.85     (45/53)
Precision for confusible_word_wordt :  0.9     (112/125)
Precision for confusible_zei_zij :  0.8     (4/5)
Precision for confusiblesuffix_d_dt :  0.69     (9/13)
Precision for errorlist :  0.71     (256/359)
Precision for hunspell :  0.5     (957/1929)
Precision for puncrecase :  0.11     (124/1114)
Precision for runon :  0.51     (71/138)
Precision for splits :  0.6     (185/310)


PER-CLASS RESULTS
====================
archaic :  P=0  R=0.0   F=0.0
capitalizationerror :  P=0.0    R=0.0   F=0.0
confusion :  P=0.66     R=0.18  F=0.28
missingpunctuation :  P=0.11    R=0.07  F=0.08
missingword :  P=0      R=0.0   F=0.0
nonworderror :  P=0.52  R=0.55  F=0.54
redundantpunctuation :  P=0     R=0.0   F=0.0
redundantword :  P=0    R=0.0   F=0.0
runonerror :  P=0.56    R=0.45  F=0.5
spliterror :  P=0.6     R=0.22  F=0.32
uncertain :  P=0        R=0.0   F=0.0

REFERENCE CLASS DISTRIBUTION
================================
archaic :  1 0.0%
capitalizationerror :  2374 21.9%
confusion :  1832 16.9%
missingpunctuation :  1849 17.1%
missingword :  960 8.9%
nonworderror :  1747 16.1%
redundantpunctuation :  306 2.8%
redundantword :  490 4.5%
runonerror :  333 3.1%
spliterror :  831 7.7%
uncertain :  108 1.0%

OUTPUT CLASS DISTRIBUTION
================================
capitalizationerror :  5 0.1%
confusion :  507 11.5%
missingpunctuation :  1114 25.4%
nonworderror :  2184 49.7%
runonerror :  270 6.2%
spliterror :  310 7.1%