Abstract:
This post explores brainstorming ideas for an AI-powered system to transcribe, understand, and visualize family stories shared during gatherings. Key concepts include narrator detection to distinguish between speakers, mapping complex family relationships described in the narratives, identifying primary sources for stories, and constructing an interactive family history knowledge graph. A more speculative idea of guessing appearances from voice characteristics is also mentioned.
Estimated reading time: 2 minutes
Chinese New Year often means big family get-togethers, and with those gatherings come stories – wonderful, rambling tales that can weave back through several generations. It recently got me thinking: could Artificial Intelligence help us not just capture these precious oral histories, but also understand and visualize them in new ways?
I started brainstorming a few ideas for an AI-powered system to make sense of these family narratives:
-
Identifying the Storytellers (Narrator Detection): Often, a story isn’t told by just one person. You might have two aunties chiming in, sharing different parts of a memory about their parents, for example. The AI would need to distinguish between these different speakers. This is important for accurately attributing pieces of information and understanding different perspectives on the same event or person.
-
Mapping Roles and Complex Relationships: This is where it gets really interesting (and complicated!). For each person speaking, the AI would need to figure out their relationship to the people in the story.
- For instance, one auntie might be talking about her father. That father might have had two wives, and each wife had children. This auntie could be a child from the first wife.
- Simultaneously, another auntie contributing to the story might be, say, your mother-in-law. And these two aunties? They could be cousins. Perhaps your mother-in-law’s father had a younger sister, and that sister’s daughter is the other auntie telling the story. The AI would need to untangle these connections from the conversation.
-
Determining Primary Sources and Graphing the Family Tree: The system would need to identify who is narrating a particular story – these individuals would be marked as “primary sources” for that specific narrative. Then, it would need to detect all the other family members mentioned. Based on the content of the stories – who said what about whom – the AI could start to build out a family tree, or more accurately, a “knowledge graph” of relationships. This could go beyond simple parent-child links to include marriages, siblings, cousins, and other connections described in the anecdotes.
-
A Bit of Fun: Guessing Appearances from Voices? As a more speculative and fun idea, perhaps the system could even try to guess what the narrators (like the aunties) might look like based on characteristics in their voice. If you’ve already uploaded family photos, it might even propose some visual connections or similarities, though this is definitely more in the realm of “maybe someday” experimentation!
These are just some initial thoughts, of course, and each of these points comes with its own significant technical challenges. But the idea of using AI to preserve, explore, and deeply understand our family narratives in such a rich, interactive way is pretty exciting to me. It feels like a wonderful way to keep those multi-generational stories alive.
What other AI-powered features do you think could help bring family histories to life?
