Running SLU on collected transcriptions¶
To run SLU on collected transcriptions and obtain more accurate interpretation (to be used for the NLG CrowdFlower task), you need to perform these steps:
Extract texts from the asr_transcribed.xml files in the call log directories:
./extract_texts.py call_log_dir > extracted.tsv
Reparse using SLU for the given language:
./reparse_<en|cs>.py extracted.tsv > reparsed.tsv
(Optionally) filter out just abstracted versions of the interpretations:
./abstract.sh reparsed.tsv abstract.tsv
These can then be analyzed/sorted etc. and/or fed to the generate_reply_tasks.py script for the NLG CrowdFlower task.
More information can be found in documentation strings in the respective script files.