README.cmatrix

Since version 0.9.8alpha, foma allows attaching a confusion matrix specification to a network.  The command "read cmatrix <filename>" read a confusion matrix and attaches is to the top network on the stack.  Subsequent "apply med" commands will use this matrix in determining the minimum cost approximate match to a word.  If no confusion matrix is specified, Levenshtein distance is used (insert = substitute = delete = 1). 

The command "print cmatrix" also prints out the confusion matrix attached to the top network in tabular format.

The format of the confusion matrix should be clear from the following example confusion matrix file:

--CUT HERE--
Insert 1
Substitute 2
Delete 1
Cost 1
a:b c:d
Cost 3
:x x: x:y
--CUT HERE--

The above snippet specifies a matrix where the default insertion cost is 1 unit, the default substitution cost is 2 units, and the default deletion cost is 1 unit.  Also, substituting an "a" with a "b" costs 1 unit, as does substituting a "c" for a "d".  Inserting an "x" costs 3 units, as does deleting an "x" and substituting an "x" for a "y".

All costs must be positive integer values.  A cost specification that involves symbols not found in the alphabet of the top network are not included in the matrix, but are warned about.