TED Policy not really generalizing

Hello, I have an issue with TED Policy and it may be that I do not understand it well. Lets say I have one intent and 3 entity

- intent_A
- entity_alpha
- entity_beta
- entity_gamma

Now I create 2 stories as follow:

steps:
intent_A
entity_alpha
answer_1

steps:
intent_A
entity_beta
answer_2

If I would write a user message now in the form of intent_A+entity_alpha+entity_gamma , I would expect the answer 1 as it is the closest match (2 out 3 matches, same intent and one similar entity) but I am now getting a fallback. My understanding was that TED was capable of generalizing to such case. Any idea why this doesnt work ?

TED Policy. The Transformer Embedding Dialogue (TED) Policy is a multi-task architecture for next action prediction and entity recognition . The architecture consists of several transformer encoders which are shared for both tasks. MyCCPay Login

Thank you for your answer but it doesnt help me in the case I explained in my original post :frowning: edit: the previous post seems to be SPAM. Re upping my question as I still did not find a solution to it over the weekend.

After testing, I can confirm again that using TED Policy, the generalization over entities does not seem to happen. Any idea why ?