Train a weighted finite-state transducer

Synopsis

wfst_train [WFSTFILE] [input file0] ... [-o output file] [-wfst ifile] [-data ifile] [-o ofile] [-heap int]

This takes an existing WFST and data and splits states in an entropy reduce way to produced a new WFST that better models the given data.

Options

• -wfst: ifile The WFST to start from

• -data: ifile Sentences in the language recognised by WFST

• -o: ofile Output file for trained WFST

• -heap: int " {210000}" Set size of Lisp heap, needed for large rulesets