[PetersWerkWiki] [TitleIndex] [WordIndex

Werkoverleg met GvN.

Taken:

  1. EarleyParser

    1. Parsen met POS-tags door Alpino.
      • POS-tags met waarschijnlijkheden gebruiken. Gedaan.

Gewoon:

   Precision+        Recall+          Precision-        Recall-        Crossing brackets
 Min.   :0.1667   Min.   :0.03774   Min.   :0.2222   Min.   :0.07547   Min.   :0.00000  
 1st Qu.:0.4355   1st Qu.:0.43070   1st Qu.:0.5897   1st Qu.:0.57468   1st Qu.:0.00000  
 Median :0.5455   Median :0.54258   Median :0.6859   Median :0.68860   Median :0.03226  
 Mean   :0.5446   Mean   :0.53697   Mean   :0.6628   Mean   :0.64832   Mean   :0.05365  
 3rd Qu.:0.6595   3rd Qu.:0.66549   3rd Qu.:0.7500   3rd Qu.:0.75361   3rd Qu.:0.08333  
 Max.   :0.8889   Max.   :0.88889   Max.   :0.8889   Max.   :0.88889   Max.   :0.34483  

Gewogen POS-tags:

 Min.   :0.2083   Min.   :0.1818   Min.   :0.2222   Min.   :0.1818   Min.   :0.00000  
 1st Qu.:0.4418   1st Qu.:0.4395   1st Qu.:0.5788   1st Qu.:0.5816   1st Qu.:0.00000  
 Median :0.5455   Median :0.5441   Median :0.6875   Median :0.6882   Median :0.03297  
 Mean   :0.5498   Mean   :0.5461   Mean   :0.6624   Mean   :0.6573   Mean   :0.05674  
 3rd Qu.:0.6667   3rd Qu.:0.6667   3rd Qu.:0.7532   3rd Qu.:0.7543   3rd Qu.:0.08693  
 Max.   :0.8889   Max.   :0.8889   Max.   :0.8889   Max.   :0.8889   Max.   :0.34483  

Vergelijk:

cd /net/aistaff/kleiweg/Earley/2013-08-29
../pairsview ../2013-08-14/clef_part0001_multi_000.parse multi_000.parse

Verbeteringen niet(?) door gebruik van scores, maar doordat POS-tagger geen complete zinnen meer als Proper_name geeft.

Test voor vergelijking met/zonder weging op dezelfde set POS-tags loopt nog.


CategoryParsing