Artificial intelligence for oral history and disaster experience: a systematic mapping study
This systematic mapping study reviews 35 primary studies to analyze how artificial intelligence is transforming oral history from static archives into dynamic, interactive fields, highlighting significant advances in transcription and analysis while emphasizing critical ethical challenges and the necessity of trauma-informed, human-centered approaches.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
For decades, the stories of human experience—survivors of war, witnesses to disasters, elders preserving forgotten traditions—have been captured on tape and disk, stored in archives as static recordings. These oral histories are vital because they hold the emotional truth and personal perspective that official documents often miss. However, a massive barrier has long stood between these voices and the people who need to hear them: the sheer difficulty of turning hours of spoken word into searchable text. For a researcher or a museum curator, listening to every minute of every recording to find a specific memory is a task that can take a lifetime. This bottleneck has left vast collections of human testimony locked away, accessible only to those with the time to listen. Now, a new field is emerging where artificial intelligence is being used to unlock these archives, not just to transcribe words, but to understand the feelings, patterns, and silences within them.
A team of researchers from universities in Canada recently conducted a comprehensive review to see exactly how this technology is being applied. They examined thirty-five studies published between 2004 and 2025, looking at every stage of how a spoken story becomes a digital record. Their work reveals that while computers have become remarkably good at the mechanical task of turning speech into text, the deeper challenge lies in preserving the human meaning behind those words. The researchers found that the field is moving rapidly from simple transcription toward complex analysis, yet significant ethical hurdles remain regarding how we handle sensitive memories and whose voices are truly heard.
The review began by mapping the journey of an oral history, which the authors divided into four distinct stages: creation, archiving, analysis, and exhibition. In the creation stage, where the story is first told, the use of artificial intelligence is still rare. Most studies in this area explore experimental ideas, such as using computer programs to interview people or generating simulated dialogues to fill gaps in the historical record. However, the vast majority of research focuses on the second stage: archiving. Here, the technology has made its most dramatic impact. By using advanced speech recognition systems, researchers can now automatically convert thousands of hours of audio into text. This has solved a critical problem that once made large collections unusable. The review notes that in 2024 and 2025 alone, nearly half of all the research in this field was published, signaling a shift from experimental prototypes to real-world tools that are actively processing massive backlogs of recordings.
The geography of this research tells a story of two different approaches. In the United States, where massive archives like the USC Shoah Foundation hold millions of testimonies from the Holocaust and other conflicts, the focus is on what the text can tell us about human psychology. Researchers there are using computer programs to analyze these stories for signs of trauma, such as predicting symptoms of post-traumatic stress in survivors of the 9/11 attacks. In contrast, a cluster of research in Central Europe, particularly Germany and the Czech Republic, is dedicated to the engineering itself. These teams are working to perfect the underlying technology, ensuring that computers can accurately hear and transcribe speech in different languages, dialects, and noisy recording conditions. While the American work asks "what does this story mean?", the European work asks "how do we make sure the computer hears it correctly?"
Beyond just writing down words, the third stage of the pipeline involves using artificial intelligence to find patterns and emotions within the stories. The review found that researchers are increasingly using these tools to group similar narratives together or to detect the emotional weight of a speaker's voice. Some studies go even further, analyzing non-verbal sounds like breathing, pauses, and silence. These sounds often carry as much meaning as the words themselves, indicating distress or hesitation that a simple transcript would miss. The fourth stage, visualization, is where these digital stories are brought back to the public. Here, artificial intelligence helps create immersive experiences, such as virtual reality exhibitions where visitors can walk through a digital space and interact with the testimonies, making the history feel immediate and personal.
Despite these technological leaps, the researchers identified serious concerns that cannot be ignored. The most pressing issue is the risk of losing the "human" element of the story. When a computer transcribes a recording, it often strips away the pauses, the breaths, and the emotional tremors in a voice, reducing a complex human experience to flat text. This is particularly dangerous when dealing with traumatic histories, where the way something is said is just as important as what is said. There is also the problem of bias. The artificial intelligence models used for these tasks are often trained on standard, high-resource languages and data. When these models encounter stories from Indigenous communities, speakers of rare dialects, or survivors of specific cultural conflicts, they often fail to understand the nuance, sometimes misinterpreting the sentiment or flattening the cultural context.
The review also highlights a growing divide between the historians who care for these stories and the engineers who build the tools. As archives rely more on complex, "black box" systems provided by commercial companies, it becomes harder for historians to understand how the technology is sorting and retrieving information. This lack of transparency raises questions about who controls the narrative. Furthermore, the researchers point out that the environmental cost of training these massive computer models is significant, adding a new layer of ethical consideration to the field. The authors emphasize that for artificial intelligence to be truly useful, it must be guided by a "human-in-the-loop" approach, where human experts remain in charge of the process, ensuring that the technology serves the story rather than distorting it.
Ultimately, this systematic review suggests that we are at a turning point. The technology to make oral history accessible to everyone is finally here, but the path forward requires more than just better software. It demands a careful balance between the efficiency of machines and the ethical responsibility of preserving human memory. The researchers conclude that the future of this field depends on developing new ways to measure success that go beyond simple accuracy. Instead of just counting how many words a computer got right, we must learn to evaluate whether the technology has preserved the emotional truth and cultural integrity of the stories it helps to tell. Only by addressing these challenges can we ensure that the voices of the past are not just heard, but truly understood.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.