Behind the game.
We study whether stories about life with multiple sclerosis can become structured event journeys and help us predict what comes next.
We study whether personal narratives about multiple sclerosis can be converted into structured event journeys and used to predict subsequent event categories. Multiple sclerosis is a condition affecting the brain and spinal cord; our project focuses on how people describe their experiences.
Compare predictions with recorded events
Turn predicted events into a possible story
We teach computer models to recognise event categories in text. We then compare different methods for finding patterns in the order of those events and predicting what comes next. We check those predictions against recorded events and explore turning them into possible stories.
What can an event be?
The study uses these 27 category labels to describe experiences recorded by people living with multiple sclerosis. They span medical experiences and everyday life. The category labels come from the project, with the condition’s name written in full.
Each game round offers just three choices to keep play simple. The full study works with the category set below.
- Change of Diet, Nutrition
- Desire for Children, Pregnancy, Childbirth, Miscarriage
- Disease Modifying Treatment & Side Effects
- Education, Apprenticeship, Studies
- Engaging with or becoming a member of (nonprofit) support organizations (e.g., the Swiss Multiple Sclerosis Society)
- Future, Future Anxiety, Future Outlook
- Hospital, Rehab, Hospitalization
- Housing Situation
- Impact and/or changes in daily life due to multiple sclerosis
- Interactions and Services Related to Disability Insurance (IV)
- Interpersonal Relationships (Family, Partner, Friends)
- Leisure, Hobbies, and Sports (Engaging, Giving Up, or Adapting)
- Loss of Loved Ones (Deaths)
- Multiple sclerosis diagnosis
- Mental Distress: Depression, Anxiety, Loneliness, Difficulty Coping
- Mental Health: Self-Confidence, Trust, Hope, Relief, Contentment
- Money, Financial Problems, Existential Anxiety
- Neurologist contacts, Diagnostic Procedures (MRI, Lumbar Puncture, etc.)
- OUTLIER
- Retirement (unrelated to Disability Insurance)
- Separation, Divorce, End of a Relationship
- Symptoms & Relapses
- Tiredness, Exhaustion, Fatigue
- Travel
- Wedding, Engagement, New Relationship, Joyful Moments in a Relationship
- Wheelchair
- Work, Work Adjustment, Termination/Resignation
OUTLIER covers text outside the event categories. The software also uses UNKNOWN when a category cannot be assigned; it is not an additional study category.
The project also explores how predicted events can be turned into short pieces of generated text. These describe possible continuations, rather than events known to have happened.
What the game demonstrates
The game follows three parts of our research: making sense of a story, predicting what might come next, and turning those predictions back into text.
1. Decode: from words to events
You match short excerpts with categories such as work, diagnosis, or relationships. The excerpts and their labels come from the study data. This is the kind of task we teach a text classifier to perform.
2. Predict: what comes next?
You see the first half of a recorded event sequence and guess the next event. Then you compare your answer with the recorded event and a prediction previously made by a computer model. Researchers use this approach to test predictions against evidence that was hidden from the model.
The game uses saved predictions, so everyone can explore the same examples without waiting for a model to run. The events stay in their recorded order, including repeated categories.
3. Match and continue: from events to text
You first match a text to the recorded event categories. Then the model’s predicted next category is revealed, and you choose a continuation that matches it. Both matches happen on the same screen. These text options were written for the game. They are synthetic examples, not patient quotations or text generated by the research model.
A story can sound convincing and follow its event categories faithfully while still describing a future that never happens. The game helps you explore that distinction.
How were the examples chosen?
We selected clear examples for teaching, including two correct model predictions and one mistake. They do not represent the model’s overall performance.
The game uses short excerpts without names, dates, locations, or patient identifiers. It explores patterns in recorded experiences and does not predict an individual’s medical future.