Sorry this is a long post, it’s mostly code blocks so you can see what I’m doing.
TL;DR
My inform intent is only picking up on one example phrase, unless I specifically use training data values in my chat, then it works properly
The Long explanation of what I’ve done so far
I’m having difficulty getting my inform to work properly so I decided to look at some of the Rasa projects because I’m a learn by example kind of guy.
NLU Data
Examining the finance-bot NLU data’s inform intent, it looks like I have the proper formatting.
Here’s a random snippet from the inform intent in nlu.md from the finanical-demo
- at [starbucks](vendor_name)
- [target](vendor_name)
- [Amazon](vendor_name)
- [Starbucks](vendor_name)
- [Target](vendor_name)
- I want to pay the [current balance](payment_amount)
My nlu.md data (that subsequently is giving me a headache) looks like this (same formatting):
## intent:inform
- my first name is [Steve](customer_first_name)
- my last name is [Franklin](customer_last_name)
- my email is [kayla@gmail.com](customer_email)
- please change my first name to [Jeffrey](customer_first_name)
- my first name should be [Jennifer](customer_first_name)
- that's wrong, my first name is not [John](customer_incorrect_first_name) it is [Jon](customer_first_name)
- my last name is [Johnson](customer_last_name) not [Jonson](customer_incorrect_last_name)
- please change my last name to [Simpson](customer_last_name)
- my last name should be [Marks](customer_last_name)
- that's wrong, my last name is not [Stevens](customer_incorrect_last_name) it is [Stephens](customer_last_name)
- can you hyphenate my last name [Franklin-Marshall](customer_last_name)
- my married name is now [Ford](customer_last_name)
- my maiden name is [Steppen](customer_last_name)
Stories
What is interesting to me (and quite confusing) is that there are no story examples that reference the inform intent in the finance-demo or the helpdesk-assistant either
I however, have two stories that look like this:
## Profile change contact first name
* inform{"customer_first_name": "Steven"}
- utter_tell_new_customer_name
## Profile change contact last name
* inform{"customer_last_name": "Myers"}
- utter_tell_new_customer_name
The issue is - I can say my first name is Jon - and that works and sets the entity customer_first_name properly as seen here:
Next message:
my first name is Jon
{
"intent": {
"name": "inform",
"confidence": 0.9971489310264587
},
"entities": [
{
"entity": "customer_first_name",
"start": 17,
"end": 20,
"value": "Jon",
"extractor": "DIETClassifier"
}
],
BUT… if I type my last name is Wheat, it picks up on the inform intent, however, doesn’t properly set the entity, as seen here:
Next message:
my last name is Wheat
{
"intent": {
"name": "inform",
"confidence": 0.9683238863945007
},
"entities": [],
The kicker is, if I use the exact phrase from my training data (my last name is Franklin), it works:
Next message:
my last name is Franklin
{
"intent": {
"name": "inform",
"confidence": 0.9954349994659424
},
"entities": [
{
"entity": "customer_last_name",
"start": 16,
"end": 24,
"value": "Franklin",
"extractor": "DIETClassifier"
}
],
For the record, if I use the exact phrasing of any of my NLU data it response properly in all cases.
It is interesting that the first name phrase works with a random first name yet the last name does not. If I was doing things 100% wrong, I would expect nothing to work.
maybe my config needs to get tweaked?
language: en
pipeline:
- name: WhitespaceTokenizer
- name: RegexFeaturizer
- name: LexicalSyntacticFeaturizer
- name: CountVectorsFeaturizer
analyzer: "char_wb"
min_ngram: 1
max_ngram: 4
- name: DIETClassifier
epochs: 100
- name: EntitySynonymMapper
- name: ResponseSelector
epochs: 100
policies:
- name: AugmentedMemoizationPolicy
- name: FormPolicy
- name: MappingPolicy
- name: TEDPolicy
max_history: 5
epochs: 100
- name: MemoizationPolicy
Am I missing the larger point of the training data or how inform is supposed to work? Maybe I need more “identical” lines with different last names? or maybe I’m doing it completely wrong?
Thanks for reading 
and I’m open to suggestions
from my side, keep the 
