Formalizing Style in Personal Narratives

1 Introduction

• Table 1 illustrates the pipeline from clause segmentation and feature identification to mapping a narrative into a symbolic sequence.Its example identifies features such as processes and participants for each clause.Source: Table 1 · p. 1
• Prior qualitative frameworks describe personal modes of thought but lack operational tools for capturing how style manifests linguistically.The authors also identify a need for formalization in therapeutic settings, where narrative reconstruction may facilitate healing.Source: §1 · p. 2
• The framework defines personal-narrative style as patterns in linguistic choices used to communicate subjective experience.It draws on systemic functional linguistics and represents narratives as sequences of choices.Source: §1 · pp. 1–2
• The approach combines functional linguistics, sequence analysis, and psychological interpretations to identify recurring patterns in how narratives encode experience.The paper presents automatic sequence-pattern identification as a methodological contribution.Source: §1 · p. 2
• A dream-narrative case study reports a war veteran’s distinctive process patterns, especially more verbal than mental processes, as suggestive of links to trauma processing.The authors describe therapeutic applications as a potential direction.Source: §1 · p. 2
• Proposed future directions include authorship profiling, style-conditioned narrative generation, and extensions using complexity science and formal language theory.These are presented as possible applications and theoretical extensions.Source: §1 · p. 2

2 Categorizing Linguistic Features

• Systemic functional linguistics treats language as a meaning-making resource shaped by contextual choices, and the framework focuses on its ideational metafunction.The ideational metafunction concerns how language represents experience through transitivity.Source: §2 · pp. 2–3

Processes, participants, and circumstances

• Transitivity analyzes experiential representation through processes, participants, and circumstances, with processes grouped into general types rather than finer categories.The paper limits finer process distinctions to save space and refers readers to appendix tables for participant definitions and circumstance examples.Source: Processes, participants, and… · p. 3

3.1 Formalizing Choice

• Authors’ linguistic choices are represented as symbols from finite feature sets, with combined alphabets enabling finer-grained process, tense, and aspect analyses.The dream case study uses process types alone, while larger alphabets could incorporate more features.Source: §3.1 · pp. 3–4

3.2 Sequence Analysis

• Sequence analysis distinguishes substrings, which are contiguous, from subsequences, which preserve symbol order but need not be contiguous.The study focuses on recurring substrings as potential stylistic patterns and notes that the framework can extend to subsequences.Source: §3.2 · p. 4
• Narrative similarity is measured by cosine similarity between vectors of substring frequencies, with higher scores indicating more similar distributions.The measure ranges from 0 for completely different to 1 for identical; one example scores 0.836 despite one sequence containing a verbal process absent from the other.Source: §3.2 · p. 4
• Hierarchical agglomerative clustering groups sequences into a dendrogram, letting analysts choose a level that reveals coarse or finer stylistic distinctions.The method does not require specifying the number of clusters in advance.Source: §3.2 · pp. 4–5
• A cluster’s representative sequence is the one with the smallest average distance to all other sequences in that cluster.Source: §3.2 · p. 5

4.1 Data and Methodology

• The study analyzes five DreamBank series, including 566 viet narratives, and compares them with a 720-narrative norm sampled across the series.The norm uses ten randomly sampled narratives from each series.Source: §4.1 · p. 5
• Language models extract processes, participants, and circumstances from dream narratives, replacing manual extraction by trained annotators; a 50-clause textbook sample matched the reference analyses.The workflow segments narratives into sentences and then clauses before feature extraction.Source: §4.1 · p. 5

4.2 Analysis

• Viet sequences contain verbal processes more often and mental processes less often than the norm; repeated verbal.verbal patterns have odds ratios of 2.00 and 1.75 for two- and three-process substrings.The reported single-process odds ratios are 1.4 for verbal and 0.6 for mental processes.Source: §4.2 · p. 6, Figure 1 · p. 6
• Clustering yields two viet sequence templates: one action-dominant and another more balanced between state and action processes, both with few mental processes.The templates cover 274 and 179 sequences, respectively; the dendrogram uses substrings of sizes one to three.Source: §4.2 · pp. 6–7, Figure 2 · p. 7
• The authors suggest viet’s action- and verbal-heavy framing may align with trauma-related cognitive and emotional processing, but say further research is needed to confirm that interpretation.They propose that identifying stylistic patterns could inform narrative restructuring in therapeutic settings.Source: §4.2 · p. 7
• The proof of concept uses only process types, and automatic labels may contain errors such as confusing mental and verbal processes.The authors identify adding participants, circumstances, tense, aspect, and participant concreteness as future extensions.Source: §4.1 · p. 6, §4.2 · p. 7

5 Related Work

• Earlier SFL parsers used brittle rules, dependency networks, or supervised verb classifiers that require expert-annotated data (Honnibal and Curran, 2007; O’Donnell, 2019; Costetchi, 2013; Yan and Webster, 2014).A survey characterizes SFL parsers as experimental, domain-sensitive, and labour-intensive to adapt (O’Donnell, 2019).Source: §5 · p. 7
• The paper uses few-shot language-model prompts to jointly label processes, participants, and circumstances, avoiding handcrafted grammars or verb lexicons and scaling to thousands of narratives.Source: §5 · p. 7
• Dream-analysis methods include manual coding, lexicons, semantic clustering, and classifier hybrids; emerging language-model approaches predict emotions or characters but mostly target content rather than linguistic construction (Bertolini et al., 2023).Source: §5 · pp. 7–8

• The framework combines functional linguistics, sequence analysis, and psychological interpretation to identify patterns in how narratives communicate subjective experience.The war-veteran case study found verbal processes more prevalent than mental ones, suggesting links between linguistic choices and trauma processing.Source: §6 · p. 8

Authorship profiling

• Distinctive substrings could characterize authors’ narrative construction and complement existing authorship-profiling approaches.The paper presents this as a possible application of its framework.Source: §6 · p. 8

Style-conditioned narrative generation

• Mapping sequences of linguistic choices back into narratives could support style-conditioned generation for therapeutic or literary uses.The inverse mapping is proposed as future work.Source: §6 · p. 8

Applying methods from complexity science and formal language theory

• Subsequence analysis and complexity measures could extend substring analysis to non-contiguous patterns and sequence redundancy.The paper suggests low complexity may indicate predictable patterns and recurring cognitive schemas, while higher complexity might suggest greater cognitive flexibility.Source: §6 · p. 8

• Language-model errors in segmenting clauses or classifying processes and participants can undermine the reliability of observed patterns.The authors say expert-annotated data could help validate or correct the automated annotations.Source: Limitations · p. 8
• Psychological interpretations remain speculative because the links between linguistic patterns and psychological states are correlational and descriptive.The authors say clinical-assessment validation is needed to strengthen claims about therapeutic utility.Source: Limitations · p. 8
• Future work could expand linguistic features, refine extraction, and validate psychological interpretations with clinical data.The current analysis relies solely on process types as a proof of concept.Source: Limitations · pp. 8–9

• DreamBank texts are publicly available, anonymized, and use pseudonyms supplied by the original collectors; dreamers consented to appear in the database.The study makes no attempt to infer individual clinical diagnoses beyond its descriptive case study.Source: Ethics Statement · p. 9
• The authors warn that open-weights language models can encode social bias and limit the framework’s intended use to exploratory research or therapist-in-the-loop settings.They release prompts, hyperparameters, and extracted sequences for third-party auditing and caution against fully autonomous clinical deployment.Source: Ethics Statement · p. 9

• The appendix reports per-series sequence-length distributions and additional sequence statistics, including distinct-substring counts and top substrings.Figures compare top substrings for blind, merri, izzy, and viet against norm.Source: A Appendix · p. 11, Figure 3 · p. 13, Figure 4 · p. 14
• Tables 3 and 4 present processes with their participants and types of circumstances with examples.Source: Table 3 · p. 12, Table 4 · p. 12
• Table 5 reports distinct-substring counts and counts of substrings with significant differences relative to norm by series and substring size.Source: Table 5 · p. 13