Files

No attached file found for this publication.

Details

Authors
Abstract
Forensic transcriptions, namely the process of converting spoken language into a written form for use in legal contexts (Lindsay and O’Connell, 1994), constitute a vital source of linguistic evidence in legal proceedings, playing a key role in the interpretation and evaluation of spoken testimony. The linguistic features embedded in these texts directly influence how utterances are understood and assessed in courtroom settings, with significant implications for evidentiary reliability (Eerland and van Charldorp, 2022). A detailed understanding of the linguistic features that characterize these forensic texts is therefore essential. Prior research by Venturi (2012) analysed Italian legal texts, identifying syntactic complexity and lexical density differences amongst various subgenres. Similarly, Dell’Orletta et al. (2013) and Montemagni (2013) focused on linguistic profiling to capture genre-specific characteristics across bureaucratic, literary, journalistic, educational and scientific texts. However, little attention has heretofore been given to the computational analysis of linguistic features in forensic transcriptions of police interviews. This study addresses this gap by conducting a computational and statistical analysis of two structurally distinct transcription types, namely narrative monologues and dialogic question-answer (Q-A) interviews, drawn from the Commissione Parlamentare Antimafia archives. These texts are also compared with a reference corpus composed of five distinct genre-based sub-corpora: parliamentary discourse, literary fiction, newspaper articles, legal texts, and oral discourse from semi-structured interviews. We applied automatic multi-level linguistic annotation to extract 143 lexical, morpho- syntactic, and syntactic features. Systematic comparisons between our two main genres and the five reference genres were carried out using either parametric or non-parametric statistical testing, supplemented by effect size analysis. A Random Forest classifier, trained via 10-fold stratified cross-validation, was used to assess the discriminative power of linguistic features across transcription types, and feature importance scores were leveraged to identify the most salient linguistic indicators, ensuring model robustness and interpretability in a forensic linguistic context. Model performance was measured using accuracy, precision, recall, and F1 scores to ensure reliable classification and generalizability. Our findings show that both transcription types share various stylistic features typical of oral and narrative discourses, such as long sentences, frequent subordination and many verbal heads per sentence. However, further analysis reveals substantial divergences. Narrative monologues exhibit frequent first-person singular verbs, consistent past-tense usage, dense syntactic embedding and a strong preference for canonical subject-verb order, indicating a form of structured, narrativized discourse closer to written legal-lay genres. In contrast, Q-A interviews exhibit features of spontaneous interaction, such as frequent pronouns, object fronting, syntactic dispersion, and longer distances between verbal elements and modifiers. They also contain more modal verbs and first-person plural forms, reflecting a cooperative and dialogic interaction style. 29 From a forensic perspective, the findings have direct implications for transcription practices and evidentiary evaluation. The high syntactic complexity and structural variability of Q-A formats may compromise transcription clarity and increase the risk of interpretative bias or misrepresentation in judicial contexts (Filipović, 2022). By contrast, narrative monologues, while linguistically dense, exhibit greater discursive cohesion and facilitate a clearer understanding of the overall story. However, they obscure the interactive dynamics inherent in the original police interviews, contributing to the perception of an autonomous and decontextualized discourse (Byrman and Byrman, 2018). These distinctions highlight the need for tailored transcription protocols and underscore the benefits of interdisciplinary approaches combining computational linguistics and forensic linguistics to better account for the variability of oral and legal discourse in judicial settings, ultimately aiming to support both investigation and trial processes.
Affiliations

Citations

  • APA
  • Chicago
  • FWB

Werner, R., Bitzer, S., & François, T. (2025). A Computational Forensic Linguistic Analysis of Narrative and Dialogic of Police Interview Transcripts. BKL Leuven 2025, Leuven. https://hdl.handle.net/2078.5/259477