README.mbma

Toad/trunk/README.mbma

Toad: Trainer Of All Data, the Frog training collection

Antal van den Bosch Ko van der Sloot. March 21 2012
ILK Research Group, Tilburg University

Describes the steps to create a Timbl tree for the MBMA module in Frog
anew, from the ground up.

Assumes the presence of /corpora/FrogData/mbma-merged.lex, the lexicon
based on CELEX (not e-Lex: it is not the same, or does not agree
necessarily with what is in mblem.lex, the lexicon on which the
lemmatizer MBLEM is trained on).

The lexicon contains lines with <word> <morphological-analysis>, where
the analysis is a space-delimited list of analysis codes that map
one-to-one to the letters of the word: 

aanblazend P 0 0 V 0 0 0+Ras>z pt 0 0
aanblazende P 0 0 V 0 0 0+Ras>z pt 0 0 E
aanblazing P 0 0 V 0 0 0+Ras>z N_V* 0 0/e
aanbleef P 0 0 V 0 0+Rij>ee 0 0/ve
aanbleven P 0 0 V 0 0+Rij>e 0+Rf>v vm 0

The steps:

1. Create instances from the merged lexicon. These are 6-1-6 letter
   windows, comma separated, each labeled with the class associated
   with the middle letter.

   > makembma -i mbma-merged.lex -o mbma.data

2. Create the Timbl tree, of the IGTree variety.

   > timbl -a1 -w2 -f mbma.data -I mbma.igtree