How to map IPC roles (ipc_a/ipc_b) to actual participant IDs?

[puckikk1202]

Have you read the Contributing Guidelines?

• I have read the Contributing Guidelines.

Issue Type

Missing files or data

Dataset Label

improvised

Dataset Split

train

Affected Files/Data and Issue Description

Are they simply based on the order in filelist.csv? For example,
4:V00_S0809_I00000309_P0947,improvised,dev,0,0,1,0,0 381:V00_S0809_I00000309_P0383,improvised,dev,0,40,1,0,0
then ipc_a = P0947, and ipc_b = P0383.
Or do they following other rules which have not been mentioned?

Steps to Reproduce

simply run the web demo and search :V00_S0809_I00000309

Additional Context

No response

Self-service

• I'd be willing to help investigate this data issue. Add comment

[JasonZheng20]

Hi @puckikk1202 !

Unfortunately this is a known limitation of the current metadata that is called out in the paper.

In some cases, the prompt text does not align with the spoken material indicating that the ordering of prompts was altered (likely due to off-by-one errors in the script ordering).

IPC roles do not have a deterministic mapping to their respective participants in each dyad. This is because we allowed session moderators a degree of freedom in assigning prompts based on their observations about which participants would respond more or less favorably. A downside of this approach is that we did not record which participant responded to which prompt in an interaction.

There is an effort currently being undertaken to obtain the correct attribution of IPC roles for each participant in each dyad.

[puckikk1202]

Thank for your great contribution! IPC value make a lot of sense in this type of conversational datasets! Hope then IPC role annotate can done soon!