Hello, I have a few questions related to lookup table usage. I am trying to detect an entity value in an intent but those values that are only listed in lookup and not in the training examples are not detected. The intent is detected correctly.
- Should this work using the default config and pipeline (supervised_embeddings)? I see most examples explicitly tune the ner_crf component.
- Does the lookup name in the nlu.md config file matter? Should it match the entity name? In the official docs we have a ‘currency’ entity but lookup tables named ‘currencies’ and ‘additional_currencies’. Training Data Format
- Are lookup tables treated similar to regexps? For the latter the docs explicitly state that the name does not matter, that it’s just a human readable description.
- Is there a rule of thumb regarding the numbers or the ratios between entity values occurring in training examples and lookup table? For example should 6 training examples with explicit food names and a lookup table of 10 food names result in all 10 foods being correctly found?