I am currently documenting the behavior of the Genesys Cloud NLU engine during intent training. I have noticed that when I have two intents with very similar training utterances, the engine sometimes returns a 400 error during the ‘Publish’ phase instead of just giving a low confidence score. Is there an undocumented limit on the ‘Utterance Overlap’ or a specific threshold where the engine refuses to publish a bot flow due to ambiguity?
Hello Ter91. I am an IVR designer and I have seen this happen when there is more than ninety percent similarity between utterances in different intents. The engine is trying to prevent you from creating a bot that will be impossible to manage. You should use the ‘Intent Health’ tool in the Architect UI to find these overlapping utterances. It will highlight exactly which phrases are causing the ambiguity so you can move them to a more appropriate intent or combine the intents entirely.
I have built several custom bot interfaces. Ter91, the 400 error during publication usually means you have an utterance that is exactly the same in two different intents. The NLU engine requires each utterance to be unique to a single intent within the same bot flow. If you have any duplicates, the publication will fail. I recommend writing a small script to scan your bot flow JSON for duplicate utterances before you try to publish via the API. It saved us a lot of time!