Question - How to design so that I can feed language grammar as part of conversation?

[levirtevs]

[sanjayaksaxena]

Hello @levirtevs

A question answering system can get very complex. At one end you have a simple system close-ended system capable of handling limited questions from a specific domain to complex systems like ChatGPT. Depending on the requirement one can create a simple rule-based system to a ML driven one involving large corpus & training steps.

Assuming you have a closed-ended system with a small corpus from where the answers have to be extracted, you can experiment with the following simple approach:

1. Based on the question, use a BM25 based search to first identify document(s) that are likely to contain the answer.
2. From the selected identify the highest scoring sentence in context of the question using similarity methods; start with a simple BoW model... better efficacy may be achieved using word vectors.

Beyond this, please refer to recent research on the subject.

Best,
Sanjaya