Good Morning,
I’m tutoring a student who is developing a “counseling simulation scenario” using ConvAI. They have limited technical knowledge, and we’ve been trying to use the Narrative Design feature to build a guided “therapy session” scenario involving a character that takes on the patient role.
We’ve tried defining a very simple linear narrative tree to try to limit the “information” the character is supposed to share during the various phases of the appointment, but we’re having severe difficulties getting the system to pick up on really basic decisions.
The documentation available for this feature is limited, and the provided examples are very basic. Are any guidelines, technical documents, or more complex examples available for this specific feature?
If you can help me, I also have two more on-point questions:
Which information does the Narrative Design system use to check whether a transition between two sections should happen?
We’re working with scripts written in Italian. Could this be an issue?
We’re sorry to hear you’re having difficulties implementing your counseling scenario with Narrative Design. To better assist you, could you please share the Character ID you’re working with, along with:
A brief description of your Narrative Design setup (what the sections and triggers look like),
What response you are expecting,
And what actual response you are receiving?
This information will help us diagnose the issue more accurately and guide you accordingly.
Hello, I took a bit of time to try to define a complete “sensible representation” of the specific scenario we would like to implement.
This is the ID of a character configured with said narrative script (29d12808-1613-11f0-818e-42010a7be01a) and since we’re working in Italian I’m attaching a linearized version of this narrative tree in PDF format both in Italian and English (automatically translated, I quickly checked it but there may be some mistranslations).
So, as I said in the previous post, what we would like to present is a clinical counseling scenario, with the ConvAI character playing the patient role. Most of the clinical info is set in its profile or knowledge bank and we’re trying to use the Narrative Design feature to partially guide the development of the counseling visit.
Each section includes section specific objectives and instructions (e.g. the “Welcome and Introductions” section has as an objective to “to establish initial contact with the therapist”).
We’re trying to use decisions to check if the player, who interprets the therapist role, has gathered a set of needed information from the patient before proceeding to the subsequent section (e.g. to close the first section “Welcome and Introductions”, we would like the therapist to have presented themselves and asked a set of question related to the patient’s generalities).
What I’m having an issue with is expressing decisions effectively:
In the provided script I’ve tried to include the most “semantic” decisions possible. Using the ones provided I’m at least able to, at some point, trigger the section transition, which however can happen even if the actual condition is not properly met or is only partially met (e.g sometimes the first decision gets triggered as soon as the therapist presents themselves).
If the decisions are too simple (e.g. “The therapist has asked and received from the patient her name” to exit the first section), the transition just does not happen.
And another issue I’m having (language related) is that the Narrative design system strips all special symbols from the text inputted in the decision fields (e.g. accented letters that are used in our target language). Is there any known workaround?
Now, considering the documentation on this feature is very limited and the examples provided are very simple, I would like to ask:
What info is actually used to check if a decision is met? The current conversation? The current conversation + section content?
Are there any rules/guidelines to write effective decisions?
Considering this specific use case, is there any way to combine (in logical AND) multiple, simpler decisions?