Feature request: smarter reported_speech about edsnlp

[bdura]

Hello @MohamedBsh, thanks for the input! There might be a slight confusion there: the eds.reported_speech pipeline's aim is to detect when the caregiver quotes the patient.

I think the behaviour you're looking for is achieved by the eds.family pipeline, although it only looks for family members. It will be renamed and modified to account for other types of experiencers (eds.experiencer?). This has been a long time coming, I'll create a feature request to reflect that.

Using a parse tree is definitely a good idea, and has been in the depths of our backlog for some time. An efficient parser would be a great way to provide very good baselines to many other problems (eds.negation and other qualifiers pop to mind). The one limitation with this approach would be training. Do you have any idea how a parser pre-trained on non-clinical documents fares in our setting? I'll open another issue to discuss this further!

[bdura]

I'm closing this issue, feel free to re-open it if something's missing!