Open pfefferniels opened 2 years ago
further mistakes
<w id="w2a2" t="on" b="4849 2"><xlit isLatinExt="1" isLatin1="1" latin1Text="on"/><moot tag="FM.fr" word="ohne" lemma="ohne"/>on</w>
)Not sure if relevant, but CAB seems to make use of an "exception lexicon"
Could we report those errors for improving the language model?
Yes absolutely – I'd suggest to collect as many as possible and report them in one go. Not sure where exactly to report though … (Bryan Jurish perhaps?)
While generally text normalization using DTA's CAB service works fine, it introduces some mistakes, e.g. Schleiffer gets normalized to "Schluffer" and some more. Probably we should collect those and correct them after normalization again.