mappings.adjmap

Overview

Adjudication maps are used to ensure that the output of data is normalized when testing, such that known/expected oddities can be filtered.

The mappings are applied to both ASR hypotheses and to NLU output (intent names and entity output).

Below is an example of a baseline adjudication map, showing various substitutions applied to align with transcribed text.

Sample mappings.adjmap

#
# General
#

s/^/ /ig;
s/$/ /ig;

s/ en / & /ig;
s/ and / & /ig;
s/ och / & /ig;

s/,/ /ig;
s/’/'/ig;
s/\-/ /ig;
s/，/ /ig;

s/ \$PI / PI /ig;
s/ \$PIs / PI /ig;

s/œ/oe/ig;
s/é/e/ig;
s/É/E/ig;
s/í/i/ig;
s/ S.A. / SA /ig;

# put spaces at boundaries of ASCII text and Asian text
s/([a-zA-Z0-9\-]+)(?=[一-鿿])/$1 /ig;
s/(?<=[一-鿿])([a-zA-Z0-9\-]+)/ $1 /ig;

s/^ +//g;
s/ +$//g;

#
# NLU
#

# specify that the builtin NO_MATCH intent should map to a user-defined intent
s/NO_MATCH/userDefinedRejectIntent/ig;