Prepare Bear
Don't BEAR it alone!
Team
they/them
he/him
he/him
he/him
he/him
Problem and Design Overview
Communicating emotions to others is something that all people have to deal with in life in order to maintain their relationships and get support for their issues, but for people struggling with mental illness, this can be simultaneously especially important and especially difficult. Due to the strength and confusing nature of what they feel, it can seem like the world is unsupportive or doesn’t understand, preventing emotions from being expressed.
In many cases, people fear receiving a bad response from others and don’t share. It’s something that is hard to build up the courage to do, and people can feel unprepared, unskilled, or anxious in a situation where they want or need to express how they feel. It’s a problem that has very few dedicated resources to help a person overcome their anxieties and feel prepared.
Prepare Bear: a sympathetic companion that helps you process and prepare through speech
Prepare Bear is a robotic animal device that communicates with users through speech, similar to Alexa, helping them work through emotions and practice conversations. It practices conversations with users and gives them practical experience speaking their feelings in a personable and comforting experience. The robot will have interchangeable faces to convey dynamic emotions. It can help a user identify a person to talk to (such as a therapist or friend) and identify conversation goals. The robot can also help them workshop a conversation by incorporating data built throughout conversations, and also providing feedback based on the user’s goals.
Design Walkthrough
One main task a user can accomplish with the help of Prepare Bear is "Evaluating Emotional Needs," finding a path forward from intense emotions, which can be done by talking through an emotional event and defining goals with the system. A second task is "Preparing What You're Going to Say" in a conversation with another person, which is done by iteratively workshopping sentences with feedback from Prepare Bear. These are our focus tasks because we identified from our research two major barriers to communication being the inability to process what you're feeling and lack of concrete preparation. The way this system is designed behind the scenes is through multiple dialogue trees that guide the user through these conversation paths.
Dialogue Tree 1: a definition of the path the system takes to gather information, lead the user in processing their emotions, and help make a plan for if they need help
Dialogue Tree 2: A definition of the path the system takes to help the user prepare for a conversation, collecting information and goals, then looping through giving feedback on user sentences using the Feedback Generator
Feedback Generator: the logic for generating feedback in Dialogue Tree 2, giving feedback based on user defined goals, and allowing users to critique their feedback and edit their goals so feedback suits their needs
Evaluate Your Emotional Needs
User starts up PB with start phrase, and from here panda's face glows when listening to user, then responds
User tells PB about problem, PB offers sympathy, questions to understand, and confirms understanding
PB leads user into identifying if they need to speak to someone to improve the situation
PB answers user question with a suggestion, confirming they are satisfied with it before moving on
PB helps user identify their goals, usually iterating with them until they have no more goals left to add
PB confirms with user that they are satisfied with their goals and the user now has a concrete plan forward
Prepare What You Are Going to Say
User expresses need to prepare for conversation, PB is sympathetic and asks questions to define context
If not defined, panda and user define conversation goals together to be the foundation of feedback
PB prompts user for practice sentence and user gives sentence they plan to say in conversation
PB gives actionable feedback based on goals, and user can redo based on feedback in iteration until satisfied
Panda prompts for another sentence after last, loop sentences/feedback process until user feels secure
Once process done, user can say yes to a text summary of their process and read to recap their plan
Design Research and Key Insights
We lacked information about the reasons why someone might have trouble expressing their feelings in a conversation and what people want out of the before and after of a conversation. We knew that knowing the main barriers to communication and people's wants would be essential in building our product. To solve this issue, we conducted 3 research activities: two interviews with people who had experience with therapy and an online survey released to the online public receiving 36 responses, but with a large portion of responses coming from people with self-identified therapy experience. We targeted this demographic because we believed their experience with emotional communication would offer greater relevance and insight. To capture this, our questions for all activities were structured around the two research questions of "What barriers exist?" and "What makes communicating effective?". We conducted interviews to gain dynamic insights in the typical way, but put out a survey as well to reach more people in order to verify our understanding of this serious topic that we did not have expertise on.
Lack of Preparation Tools
Both interviews identified that the most common way for them to prepare for conversations was for them to run through the problem and the conversation in their head. Interviewee 1 stated: "I guess I would say, I go over the scenario in my head and, I don't know if that's a good or a bad thing sometimes." Interviewee 1 on multiple occasions expressed a real desire for some sort of tools that would help a person identify what they are feeling in preparation for communication. Many survey responses when asked for what they would want if they could have anything to prepare for a conversation, could not identify a specific thing, often writing "something" to help them feel better. One response wrote, "A guide that tells me about all the emotions and difficult feelings I'm feeling." These statements indicated to us that there was a need for a concrete tool to help prepare for emotional conversations compared to the nebulous existing methods.
This identified what motivated the concept of our project, and certain suggestions were implemented. Both the above statements influenced our "Identifying Needs" dialogue tree's process of walking through emotions. The interviews talking about running through conversations in their head motivated our "Prepare for a Conversation" dialogue tree as a practice and feedback session where the conversation partners' response is considered, streamlining this mentioned process of mental run-throughs into something more structured and with impartial help.
Initial section of DT1 influenced by the want of an emotional guidance tool
Loop structure of DT2 motivated by common mental walkthrough preparation
Concern about how others react deters individuals from opening up
Survey responses included a variety of reasons for why one can be afraid of communicating emotionally such as burdening the listener or not wanting to come off as vulnerable. Quote from one response: “fear of judgement and not wanting to worry them.” Interview 1 also identified it as one of three major barriers to communication in their life: Participant in full stated “Yeah. I think it’s those two, and then just the guilt of worrying or stressing people around me.” This shaped our design direction by motivating us to include features that could ease this worry by specially considering who the user wants to speak to and their possible reaction, and giving feedback based on this context during the "Preparation" phase. This in combination with generally easing worries through focusing on how to effectively communicate towards a wanted outcome attempts to make the user feel more confident in how they will converse with someone else.
Feedback system from DT2 giving feedback based on collected context about the conversation partner
People express a strong need for comfort/a way to ease anxiety before a difficult emotional conversation.
For the survey question asking what respondents would have to prepare if they could have anything, many answers stated similar things: “Something that makes me less anxious”, “A hug from a close friend or my partner”, “Something that makes me stop crying”, “A way to regulate overwhelming feelings”. Interviewee 1 identified worry about burdening others as a large barrier to communication, but stated that their therapist and parents reassuring them allowed them to overcome that fear and realize they wouldn’t be judged, making communication easier in the future. This shaped our design direction by heavily influencing the concept of our design as something that would appear comforting and sympathetic and also offer reassurance often. This manifests at many points in our dialogue tree processes, especially our entire "Sympathetic Responses" system, and one of our end-of-path messages directly being inspired by this Interview 1 insight which stated that people need to reassured that people in their life are willing to listen.
Sympathetic Response tree, added later but building off existing system of providing validation and comfort
Iterative Design and Key Insights
The physical paper prototype we used to test our dialogue trees during iteration
Similar to our peers, we started by creating a paper prototype of our system, which included our paper prototype bear and also our two main task dialogue trees we developed. Then we had inspection done by other groups, made edits based on the heuristic evaluation results we identified as most important, then we did usability testing with three participants by letting them speak to the physical bear and a tester speaking as the bear following the dialogue trees, then making small revisions to the prototype in between based on breakdowns encountered. Due to the nature of our project, what we were testing mainly was our dialogue trees and how people interacted with the process, which in usability testing really allowed us to identify problems with natural conversation flow, while heuristic evaluation helped identify features and best practice inclusions we overlooked because of our unique focus.
Allowing the User to Control Their Process
Many Usability Testing inspectors identified a violation of the "User Control" heuristic in that our dialogue trees did not allow for much user control on when to exit, redo, or navigate in a conversation, and seemed to be guide-railed on a certain path. We then added a stop tree to shutdown the bear at any point in the conversation, and a redoing statement function. We also experienced a breakdown during a usability test where a user stated they didn't like the feedback given and felt like the system was forcing them to follow it. It became clear to us that not only should users at any point be in full control of their position in the process, but also the system's underlying priorities. To do this, we redesigned the feedback system to allow users to edit their goals and feedback preferences if not satisfied with feedback received.
One revision from this insight, the revision tree handling user backtracking at any point
Making System Status and Processing Visible
What the system is doing and how it is interpreting the user's input should be made visible to the user in real-time or promptly after data collection. Also from Usability Testing, we received many reports identifying a lack of visible state and prompt feedback in violation of the "System Visibility" heuristic. Due to our system being predominantly a speech interface, we did not have many initial points where feedback was given to the user as we focused on developing a dialogue process, but we realized on the technical side and for clarity this was essential. Using our defined facial system that previously existed just to add a responsive human element, we added facial states that represents thinking and receiving input so that the user would be aware in real time that the system is pausing to process or that their audio is being picked up and listened to. Where this insight heavily factored into our iteration is our consideration of a change for exclusion: as we considered how to handle harder-to-parse accents or speech styles, we realized the whole system would benefit from more checkpoints that reported collected information back to users. We added many checkpoints along our dialogue trees in consideration of this, giving users that feedback soon after collection.
Addition to facial system displaying the level of audio input received to the user in real time
Addition to facial system displaying to user when system is loading/thinking for transparency and waiting
Supporting Natural Conversation Diversions
This insight came mostly from usability testing, where across tests many system breaks occurred when the user asked questions or made statements that were not anticipated by the dialogue tree. Participant 2 in particular asked many questions for advice and clarification. Participant 3 gave the feedback that the current process felt like a systematic interrogation in the way the system asked many questions without interruption. This identified to us that the rigid trees created needed to allow for natural side roads in conversation, and the user could even benefit from divergences in the form of questions.
So, we added a question-taking tree to our system, a sympathetic response system that made the system respond more naturally to user speech in between tree nodes, and an error tree handling all other diversions with a reminder of system state and ability to redirect if needed. We believe all of these changes improve user experience with improved natural dialogue, offering more user assistance, and handling error cases in defined ways.
Question taking tree added to the system, assisting users by handling a variety of useful questions throughout the whole process