Elamur Research⌕
☰
home / papers / 2026 / From Decision Tree to Detection Pipeline: Formalizing van Dijk's Socio-Cognitive Framework for Automated Anti-Language Identification in RICO Transcripts

From Decision Tree to Detection Pipeline: Formalizing van Dijk's Socio-Cognitive Framework for Automated Anti-Language Identification in RICO Transcripts

Paper · 2026 · Elena Morandini

Summary

Organized-crime groups hide criminal intent behind ordinary-sounding coded speech (anti-language), so threat detectors that rely on explicit lexical cues miss much of it. Morandini proposes a context-first pipeline that turns van Dijk's socio-cognitive CDA framework, previously a seven-step decision tree for human analysts, into ordered NLP subtasks. In sequence, these subtasks classify each speaker's role, detect ideological polarization, identify the genre, label moves and extract micro-level features. Each step narrows how the next one interprets the text. The approach was developed on a small corpus of authenticated American Mafia wiretap transcripts from RICO court records. The corpus was annotated by combining corpus-tool extraction with manual pragmatic coding. Validating a six-feature vector (F1–F6) showed patterns specific to anti-language, such as directives without mitigation and dense deontic language. Only part of the system works today: annotation, feature extraction, χ² validation and a relexicalization detector are implemented. The learned classifiers and the ALI score remain proposals. The author argues that the stepwise design makes outputs explainable enough to support expert testimony, which matters for intelligence triage and forensic linguistics.

Key points

  • Corpus: 14,072 words from 18 RICO court documents (2006–2023), obtained through PACER from New York and Pennsylvania districts. It covers Bonanno, Colombo, Gambino, Genovese and Lucchese members, plus Philadelphia families.
    • Text was digitized with ABBYY FineReader OCR and checked manually.
    • FBI transcription markers such as [U/I] and [ph] were kept, and dialect forms were left unnormalized.
  • Annotation:
    • Features were extracted with AntConc, Sketch Engine and spaCy, then coded by function in CATMA.
    • Deontic modals were split into Subjective Authority and Objective Necessity, and outgroup terms were tagged Neutral or Hostile.
    • A single annotator did two passes 12 months apart.
  • Validation result: The Negativity Amplification Ratio reached 2.35×, against a 1.5× threshold. Distinctiveness was tested with χ² against OANC and SBCSAE.
  • Limitations: Thresholds were set post hoc on the same corpus. Negative cases for false-positive calibration are not available through PACER.
  • Proposed next steps:
    • Build the planned components: fine-tuned RoBERTa, a CRF move labeler, domain-adapted NER and ALI scoring.
    • Develop three LEA uses: transcript flagging, context-sensitive decoding and hierarchy reconstruction.
    • Adapt the pipeline to encrypted messaging apps.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Fields, topics and concepts

📚 Shelves 2The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 works

Cite

Cite as Elena Morandini. 2026. From Decision Tree to Detection Pipeline: Formalizing van Dijk's Socio-Cognitive Framework for Automated Anti-Language Identification in RICO Transcripts. Proceedings of the Second International Conference on Natural Language Processing and Artificial Intelligence for Cyber Security.

The paper itself, not this page. BibTeX from the library's record of the work.

@inproceedings{morandini2026decision,
  title = {{From Decision Tree to Detection Pipeline: Formalizing van Dijk's Socio-Cognitive Framework for Automated Anti-Language Identification in RICO Transcripts}},
  author = {Elena Morandini},
  booktitle = {Proceedings of the Second International Conference on Natural Language Processing and Artificial Intelligence for Cyber Security},
  year = {2026},
  url = {https://aclanthology.org/2026.nlpaics-1.22/}
}

References

5 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. John Langshaw Austin. 1962. How to Do Things with Words. Oxford University Press, Oxford. find
  2. John W. Du Bois, Wallace L. Chafe, Charles Meyer, Sandra A. Thompson, Robert Englebretson, and Nii Martey. 2000–2005. Santa Barbara Corpus of Spoken American English, Parts 1–4. Linguistic Data Consortium, Philadelphia. https://doi.org/10.21415/T5VG6X — [doi:10.21415/t5vg6x](https://doi.org/10.21415/t5vg6x) ↗
  3. Helen Fraser. 2003. Issues in transcription: Factors affecting the reliability of transcripts as evidence in legal cases. Speech, Language and the Law, 10(2):203–226. https://doi.org/10.1558/sll.2003.10.2.203 — [doi:10.1558/sll.2003.10.2.203](https://doi.org/10.1558/sll.2003.10.2.203) ↗
  4. Helen Fraser. 2022. A Framework for Deciding How to Create and Evaluate Transcripts for Forensic and Other Purposes. Frontiers in Communication, 7:1–14. https://doi.org/10.3389/fcomm.2022.898410 — [doi:10.3389/fcomm.2022.898410](https://doi.org/10.3389/fcomm.2022.898410) ↗
  5. Evelyn Gius, Jan Christoph Meister, Marco Petris, Christian Bruck, Janina Jacke, Mareike Schumacher, Dominik Gerstorfer, Marie Flüh, and Jan Horstmann. 2023. CATMA 7.0. Zenodo. https://doi.org/10.5281/zenodo.1470118 \#\# 10 Conclusions Diego Gambetta. 1993. The Sicilian Mafia: The Business of Private… — [doi:10.5281/zenodo.1470118](https://doi.org/10.5281/zenodo.1470118) ↗

Updated

This page in Markdown → https://md.research.elamur.ai/free/en/papers/morandini-2026-decision-tree-detection-pipeline-formalizing/