Interesting classification results on a Word document between Conversational Search and Document Retrieval projects in Discovery v2. When applying the pre-trained SDU to the identical DOCX document in the different types of projects, it identifies tables correctly and completely in the Document Retrieval project but not in Conversational Search. The latter provides bunched up text only.

How come ? 

As a result, we are pressing-on with DR-types in Discovery for use with Watson Assistant. Would be great to get feedback from IBM on whether there is an underlying problem with Conversational Search, tho ? Shots below. Otherwise, we are starting to get decent results with DR-types .... early days for our adoption.

DR:

CS:

------------------------------
Simon
(Technologist and Consultant - new to IBM Watson in August 23 !)
------------------------------

Hi Simon, underlying representation and the results should be exactly same. It's just tables are not rendered in the Conversational Search UI. Good way to test would be to see what gets returned via the API.