TLDR: This research introduces a framework to formalize style in personal narratives by analyzing patterns in linguistic choices like processes, participants, and circumstances, automatically extracted using language models. A case study on dream narratives, particularly from a war veteran with PTSD, reveals distinct stylistic patterns (e.g., prevalence of verbal over mental processes), suggesting links between language use and psychological states, with potential applications in therapy and narrative generation.
The way we tell our personal stories, or narratives, is deeply influenced by our unique style – the distinctive linguistic choices we make to express ourselves. While this style is fundamental to how we convey subjective experiences, there hasn’t been a formal system to systematically analyze these choices. A new research paper introduces a novel framework that aims to fill this gap by formalizing style in personal narratives.
The researchers propose a new approach that defines style as patterns in the linguistic choices authors make when communicating their subjective experiences. This framework cleverly integrates insights from three distinct fields: functional linguistics, which views language as a system of meaningful choices; computer science, providing methods for automatically extracting and analyzing sequential patterns; and psychology, which helps in interpreting these patterns in a meaningful way.
At the heart of their methodology is the use of advanced language models. These models are employed to automatically extract linguistic features from narratives, such as ‘processes’ (actions, mental states, verbal communications), ‘participants’ (the entities involved in these processes), and ‘circumstances’ (details like time, place, or manner). Once these features are identified, each narrative is transformed into a symbolic sequence, essentially a string of linguistic choices.
The framework treats grammar not just as a set of rules, but as a network of meaningful choices. Authors select from available linguistic options to encode their experiences, and these selections form patterns. The research focuses on identifying recurring patterns within these sequences, specifically ‘substrings’ (contiguous blocks of choices) and ‘subsequences’ (ordered choices that are not necessarily contiguous). Drawing an analogy from computational biology, where DNA or protein sequences are analyzed, the researchers apply similar techniques to detect recurring linguistic patterns. To compare different narratives, they use measures like cosine similarity, which quantifies how similar two sequences of linguistic choices are. Unsupervised clustering methods are then used to group narratives that exhibit similar stylistic patterns, and ‘representative sequences’ are identified to summarize the characteristic features of each cluster.
A compelling demonstration of this framework’s applicability comes from a case study on hundreds of dream narratives from the DreamBank corpus. The researchers specifically highlight their analysis of a war veteran diagnosed with post-traumatic stress disorder (PTSD). Using a language model, Llama 3.1 8B Instruct, they automatically extracted linguistic features, focusing primarily on the types of processes used in these dreams.
The analysis of the war veteran’s dream narratives revealed distinctive stylistic patterns. For instance, the presence of verbal processes was found to be 40% more likely in his sequences compared to an average dreamer, while mental processes were 40% less likely. Furthermore, the veteran’s narratives showed a strong preference for sequences like ‘verbal.verbal’ and ‘verbal.verbal.verbal’, indicating a tendency to remain within verbal processes. Clustering of his narratives identified two main stylistic templates: one heavily dominated by action processes with minimal mental processes, and another more balanced between action and state processes, yet still featuring relatively few mental processes.
These findings offer intriguing psychological interpretations. The veteran’s inclination to frame experiences through action and verbal processes rather than mental ones could be linked to how combat trauma impacts cognitive and emotional processing. The researchers suggest that this formalization of style could have significant therapeutic implications. By identifying specific linguistic patterns associated with psychological states, therapists and patients could collaborate to restructure narratives, potentially aiding in trauma processing and healing.
Beyond therapeutic applications, the framework opens up several exciting avenues for future research. It could enhance ‘authorship profiling’ by identifying unique stylistic signatures that characterize individual authors. It also paves the way for ‘style-conditioned narrative generation,’ where language models could be guided to create stories based on specific stylistic attributes, offering new possibilities for both therapeutic interventions and creative writing. Future work may also delve into more complex linguistic features, explore non-contiguous subsequences, and apply complexity measures to quantify redundancies in narrative patterns, potentially revealing insights into cognitive flexibility or fixed mental representations.
Also Read:
- Enhancing Psychological Tests with AI: A New Method for Integrating Natural Language Data
- Modeling Human Consciousness in AI Through Psychoanalysis and Personality Theory
While the framework holds great promise, the authors acknowledge certain limitations. The automatic extraction of linguistic features by language models, though efficient, can introduce inaccuracies. Additionally, the connections drawn between linguistic patterns and psychological interpretations are currently correlational and descriptive, suggesting a need for validation against clinical assessments to strengthen claims about their therapeutic utility. Despite these limitations, this research, detailed further in the paper Formalizing Style in Personal Narratives, establishes a foundational approach for a sequence-based, linguistically informed analysis of personal narratives, offering valuable insights into the intricate ways we construct meaning from our experiences.
