When we read a text, we intuitively distinguish between a factual report, a set of instructions, and a personal story. We know the difference between someone saying "The medication is effective" and "I took the pill and felt better." For machines, however, this distinction is notoriously difficult. Traditional NLP often treats these sentences similarly because they share keywords like "medication" or "pill."
A growing body of research suggests that different modes of discourse—specifically personal experiences and narratives—leave behind unique "linguistic fingerprints." By identifying these structural and grammatical markers, researchers can move beyond simple keyword matching to understand the context of agency and perspective.
The Anatomy of a Story
The core intuition behind this research is that storytelling is not just about what is said, but how it is structured. Maria Antoniak and her colleagues, in their work on StorySeeker, demonstrate that narrative spans are characterized by a specific cluster of features: first-person singular pronouns ($I, me, my$), past-tense verbs, and concrete language.
This "fingerprint" acts as a discriminator. While an instructional post might rely on second-person pronouns ($you$) and future-tense verbs to give advice, a personal report anchors itself in the past and the self. Valentin Jijkoun and Maarten de Rijke refer to this as experience mining, noting that the presence of action-oriented past-tense verbs is a reliable signal that a user is reporting a direct interaction with a product or event rather than just sharing an opinion.
Identifying the Protagonist and Perspective
A major challenge in this field is distinguishing between a user sharing their own experience and a user reporting on someone else's. This is the problem of "agency." Arijit Chowdhury highlights this in the context of sexual harassment disclosures, where models often struggle to differentiate between a survivor's personal recollection and a news report about the same event.
To solve this, researchers look at the "narrative role" of the speaker. Janyce Wiebe and William J. Rapaport pioneered the tracking of subjective contexts, using psychological verbs and evaluative adjectives to identify when a narrative shifts from objective facts to a character's internal thoughts. Similarly, Jiwei Li and Alan Ritter found that extracting life events from social media requires filtering for factuality—distinguishing between a wish ("I want to get married") and an event ("I got married") by analyzing verb tense and pronoun alignment.
Factual vs. Emotional Argumentation
The linguistic fingerprint also extends to how we argue. Shereen Oraby and Ellen Riloff discovered that factual and emotional argumentation styles have distinct syntactic signatures. Factual discourse tends to use passive voice and complex prepositional structures (e.g., "The results of the study..."), whereas emotional or "feeling" discourse uses active voice and interpersonal critiques.
This distinction is crucial for argument mining. As Mihai Dusmanu notes, identifying factual arguments isn't necessarily about verifying the truth of a statement, but about identifying the structural presence of a claim or a citation. In contrast, Neele Falk and Gabriella Lapesa show that personal stories are often used as a form of "narrative evidence" that increases the perceived quality of an argument, even if it lacks formal logic.
Applications in Public Health and Beyond
The ability to detect personal experience has profound implications for pharmacovigilance and public health. Debanjan Mahata demonstrated that detecting personal intake of medicine on Twitter relies on identifying the context of agency—knowing if a user is actually taking a drug or just discussing it.
Furthermore, Anne Dirkson proposes using these experiential markers to mine patient forums for "experiential knowledge"—hidden medical insights like coping strategies or side effects that are often absent from clinical literature. By filtering for first-person, past-tense narratives, systems can isolate genuine patient experiences from general medical advice or hearsay.
Ultimately, this research direction moves NLP away from seeing text as a "bag of words" and toward seeing it as a structured expression of human perspective. By focusing on the grammatical and syntactic cues of storytelling, we can better understand the lived experiences embedded within our digital discourse.