pith. sign in

arxiv: 2605.15978 · v1 · pith:4ZN4IEYZnew · submitted 2026-05-15 · 💻 cs.CL · cs.AI· cs.LO

Ontology for Policing: Conceptual Knowledge Learning for Semantic Understanding and Reasoning in Law Enforcement Reports

Anita Srbinovska , Jansen Orfan , Adrian Martin , Ernest Fokou\'e This is my paper

Pith reviewed 2026-05-20 17:46 UTC · model grok-4.3

classification 💻 cs.CL cs.AIcs.LO
keywords law enforcement reportssemantic parsingcustom ontologypredicate mappingtemporal graphsproperty crimeevent extractionreasoning
0
0 comments X

The pith

Symbolic ontology mapping recovers core facts from police report narratives with strong agreement on initiation and stolen items.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The authors aim to show that narratives in law enforcement reports hold recoverable details about incidents that can be automatically turned into structured facts using symbolic techniques. They redact identifiers, parse the language, map verbs and predicates onto a policing ontology, and reason with time information and axioms to create temporal graphs. Testing this on hundreds of property crime reports reveals reliable mapping for most events and complete agreement among reviewers on basic facts such as the start of the incident and items taken. A sympathetic reader would care because this could let investigators and trainers access needed information without manually scanning every written report.

Core claim

The central claim is that a symbolic framework consisting of semantic parsing followed by predicate mapping through the PropBank-VerbNet-WordNet path to a custom ontology, combined with domain axioms for reasoning, can convert unstructured narratives into evidence-linked facts and temporal graphs. On a collection of 450 property crime reports this process produces events with over half at high confidence and maps almost all through the semantic path, while human review confirms full agreement on incident initiation, stolen items, and temporal cues.

What carries the argument

Predicate mapping to the custom ontology via the PropBank--VerbNet--WordNet semantic path, which serves to link natural language descriptions to formal facts and enable axiom-based reasoning over time cues.

If this is right

  • The extracted facts can be used to build searchable temporal graphs for incident review.
  • High agreement on core elements indicates the approach reliably captures essential details from text alone.
  • Domain axioms help interpret ambiguous parts of the narrative.
  • The method demonstrates the value of narratives beyond the structured fields in reports.
  • Lower agreement areas like forced entry highlight opportunities for refining the ontology.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Applying the same mapping to other types of reports could reveal patterns across different crime categories.
  • Integration with digital report systems might allow real-time fact checking during investigations.
  • This could support training programs by automatically generating scenarios from real reports.
  • Extending the axioms might improve handling of complex or ambiguous incident descriptions.

Load-bearing premise

The premise that mapping narrative predicates to the custom ontology through semantic paths and applying domain axioms will recover all necessary incident facts from the text without loss or distortion of details like the manner of forced entry.

What would settle it

A side-by-side comparison of system outputs with expert annotations on a new set of reports focusing on whether forced entry is correctly identified or not.

Figures

Figures reproduced from arXiv: 2605.15978 by Adrian Martin, Anita Srbinovska, Ernest Fokou\'e, Jansen Orfan.

Figure 1
Figure 1. Figure 1: Narrative redaction, extraction, AMR semantic normalization, and ontology outputs. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: AMR graph for the sentence “The suspect broke the rear passenger [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Partial class–level view of the ontology, showing event, entity, and [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Ranked frequency distribution of extracted AMR predicate senses [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Case–level symbolic graph from a redacted narrative. Solid edges show narrative sequencing and dashed edges show domain axioms. [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Ranked frequency distributions of extracted AMR predicate [PITH_FULL_IMAGE:figures/full_fig_p012_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Reviewer ambiguity in the five–case review. [PITH_FULL_IMAGE:figures/full_fig_p012_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Full temporal case graph from an example redacted report, showing [PITH_FULL_IMAGE:figures/full_fig_p013_8.png] view at source ↗
read the original abstract

Law enforcement reports contain structured fields and written narratives. However, many incident facts that are needed for review, police training, and investigations are in natural language and require manual reading. We propose a framework using symbolic methods for converting narratives into evidence-linked facts. Our objective is to measure the value of narratives to recover incident details only from the unstructured text and build temporal graphs with time cues and domain axioms. We achieve this by redacting personal identifiers, semantic parsing, predicate mapping to ontology, and reasoning. We evaluate the symbolic approach on 450 property crime reports and a short human review. Of the extracted events from the system, 54.1% had a confidence score of at least 0.80 and 93.7% were mapped through the PropBank--VerbNet--WordNet semantic path. 100% agreement was reached on incident initiation, stolen items, and temporal cues and lower agreement for forced entry interpretation.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The paper proposes a symbolic framework for extracting evidence-linked facts from law enforcement narrative reports. It redacts personal identifiers, applies semantic parsing, maps predicates to a custom policing ontology via PropBank–VerbNet–WordNet, and uses domain axioms to construct temporal graphs. On 450 property crime reports, it reports that 54.1% of extracted events achieve confidence ≥0.80, 93.7% map through the semantic path, with 100% human agreement on incident initiation, stolen items, and temporal cues, and lower agreement on forced entry.

Significance. If the central claim holds, the work demonstrates a practical integration of symbolic reasoning with lexical resources for domain-specific information extraction in policing, potentially reducing manual review effort. The use of standard resources (PropBank, VerbNet, WordNet) and reported human agreement on key elements are strengths, but the absence of baselines or error analysis limits assessment of added value over existing IE methods.

major comments (3)
  1. [Abstract / Evaluation] Abstract and Evaluation section: The reported 100% agreement on incident initiation, stolen items, and temporal cues is presented without details on the human review protocol (number of reviewers, annotation guidelines, or how disagreements on forced entry were resolved). This weakens the claim that the ontology plus axioms recover all necessary facts, as the paper itself notes lower agreement precisely for ambiguous predicates like forced entry.
  2. [Abstract] Abstract: No baseline comparison (e.g., to off-the-shelf SRL systems or rule-based extractors) or error analysis is provided for the 54.1% high-confidence events or the 93.7% mapping rate. Without these, it is difficult to determine whether the custom ontology and axioms provide a genuine improvement or simply reproduce standard semantic parsing outputs.
  3. [Evaluation / Ontology axioms] The central assumption that predicate mapping plus domain axioms recover incident facts without systematic loss for ambiguous elements (e.g., linguistic cues for forced entry such as “broke in” or implied damage) is load-bearing but unsupported by detailed case analysis. The lower agreement noted for forced entry directly challenges the completeness claim for temporal graphs.
minor comments (2)
  1. [Abstract] Clarify how confidence scores are computed (e.g., from mapping coverage, axiom satisfaction, or external validation) as this is referenced but not defined.
  2. Provide the size and composition of the custom policing ontology (number of concepts, axioms) to allow reproducibility assessment.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their detailed and constructive comments on our manuscript. We value the recognition of the potential for integrating symbolic reasoning with lexical resources in the policing domain. Below, we provide point-by-point responses to the major comments and indicate the revisions we will make to address them.

read point-by-point responses

  1. Referee: [Abstract / Evaluation] Abstract and Evaluation section: The reported 100% agreement on incident initiation, stolen items, and temporal cues is presented without details on the human review protocol (number of reviewers, annotation guidelines, or how disagreements on forced entry were resolved). This weakens the claim that the ontology plus axioms recover all necessary facts, as the paper itself notes lower agreement precisely for ambiguous predicates like forced entry.

    Authors: We agree that additional details on the human review protocol are necessary to substantiate the reported agreement levels. In the revised manuscript, we will expand the Evaluation section to describe the review process, including the number of reviewers, the specific annotation guidelines provided, and the method used to resolve any disagreements, with particular attention to the lower agreement observed for forced entry interpretations. This will strengthen the transparency of our evaluation. revision: yes

  2. Referee: [Abstract] Abstract: No baseline comparison (e.g., to off-the-shelf SRL systems or rule-based extractors) or error analysis is provided for the 54.1% high-confidence events or the 93.7% mapping rate. Without these, it is difficult to determine whether the custom ontology and axioms provide a genuine improvement or simply reproduce standard semantic parsing outputs.

    Authors: We acknowledge the value of baseline comparisons and error analysis for assessing the contribution of our custom ontology and axioms. The current work emphasizes the development of the policing-specific ontology and the use of domain axioms for temporal reasoning. In revision, we will incorporate a detailed error analysis of the extracted events and discuss how our approach differs from standard semantic role labeling outputs, although a full empirical baseline comparison would require additional experiments which we plan to outline. revision: partial

  3. Referee: [Evaluation / Ontology axioms] The central assumption that predicate mapping plus domain axioms recover incident facts without systematic loss for ambiguous elements (e.g., linguistic cues for forced entry such as “broke in” or implied damage) is load-bearing but unsupported by detailed case analysis. The lower agreement noted for forced entry directly challenges the completeness claim for temporal graphs.

    Authors: We recognize that the lower agreement on forced entry highlights potential limitations in handling ambiguous linguistic cues. The manuscript already acknowledges this lower agreement. To address the need for detailed case analysis, we will add specific examples in the Evaluation section illustrating how the predicate mappings and domain axioms handle (or fail to fully capture) forced entry scenarios, thereby providing a more robust justification for the completeness of the temporal graphs where possible. revision: yes

Circularity Check

0 steps flagged

No circularity: evaluation uses external resources and human agreement

full rationale

The paper presents a symbolic pipeline of redaction, semantic parsing, predicate mapping via PropBank-VerbNet-WordNet, and domain-axiom reasoning to build temporal graphs from narratives. Reported figures (54.1% events with confidence >=0.80, 93.7% mapped through the semantic path, 100% agreement on initiation/stolen-items/temporal cues) are direct outputs of applying this pipeline to 450 reports followed by independent human review; no equations, fitted parameters, or self-referential definitions are described that would make any result equivalent to its inputs by construction. The derivation therefore remains self-contained against external lexical resources and human evaluation.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The paper relies on standard semantic resources and introduces a domain-specific ontology plus axioms for temporal reasoning. No free parameters or invented physical entities are described.

axioms (1)
  • domain assumption Domain axioms for law-enforcement incidents and time cues
    Invoked to build temporal graphs from extracted predicates.
invented entities (1)
  • Policing ontology no independent evidence
    purpose: To map predicates to structured evidence-linked facts
    Custom ontology created for the law-enforcement domain.

pith-pipeline@v0.9.0 · 5702 in / 1285 out tokens · 64607 ms · 2026-05-20T17:46:42.704837+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

28 extracted references · 28 canonical work pages

  1. [1]

    Basic report writing,

    Metropolitan Police Department (Washington, DC), “Basic report writing,” 2025. [Online]. Available: https://mpdc.dc.gov/sites/default/fil es/dc/sites/mpdc/publication/attachments/3.1%20Basic%20Report%20 Writing%20-%20IA 071625.pdf

    work page 2025

  2. [2]

    Natural language, knowledge representation, and logical form,

    J. Allen, F., “Natural language, knowledge representation, and logical form,” inA Symposium on Future Directions in Natural Language Processing on Challenges in Natural Language Processing. USA: Cambridge University Press, 1993, pp. 146–175. [Online]. Available: https://apps.dtic.mil/sti/tr/pdf/ADA247389.pdf

    work page 1993

  3. [3]

    Allen,Natural language understanding (2nd ed.)

    J. Allen,Natural language understanding (2nd ed.). USA: Benjamin- Cummings Publishing Co., Inc., 1995. [Online]. Available: https: //dl.acm.org/doi/book/10.5555/199291

    work page doi:10.5555/199291 1995

  4. [4]

    Toward deep language understanding: Methods for learning conceptual knowledge from definitions,

    J. Orfan, “Toward deep language understanding: Methods for learning conceptual knowledge from definitions,” 2020, university of Rochester Institutional Publication Record. [Online]. Available: https://urresearch.rochester.edu/institutionalPublicationPublicView.acti on?institutionalItemId=35547

    work page 2020

  5. [5]

    Identifying underlying commonsense knowledge in definitions,

    J. Orfan and J. Allen, “Identifying underlying commonsense knowledge in definitions,” inProceedings of the Thirtieth International Florida Artificial Intelligence Research Society Conference (FLAIRS), 2017. [Online]. Available: https://aaai.org/papers/688-flairs-2017-15494/

    work page 2017

  6. [6]

    Computational Linguistics , volume =

    M. Palmer, D. Gildea, and P. Kingsbury, “The proposition bank: An annotated corpus of semantic roles,”Computational Linguistics, vol. 31, no. 1, pp. 71–106, Mar. 2005. [Online]. Available: https://doi.org/10.1162/0891201053630264

    work page doi:10.1162/0891201053630264 2005

  7. [7]

    Extending verbnet with novel verb classes,

    K. Kipper, A. Korhonen, N. Ryant, and M. Palmer, “Extending verbnet with novel verb classes,” inProceedings of the Fifth International Conference on Language Resources and Evaluation (LREC’06), Genoa, Italy, 2006, pp. 1027–1032. [Online]. Available: https://aclanthology.org/L06-1280/

    work page 2006

  8. [8]

    Wordnet: a lexical database for english,

    G. A. Miller, “Wordnet: a lexical database for english,”Commun. ACM, vol. 38, no. 11, pp. 39–41, nov 1995. [Online]. Available: https://doi.org/10.1145/219717.219748

    work page doi:10.1145/219717.219748 1995

  9. [9]

    Semlink 2.0: Chasing lexical resources,

    K. Stowe, J. Preciado, K. Conger, S. W. Brown, G. Kazeminejad, and M. Palmer, “Semlink 2.0: Chasing lexical resources,” inProceedings of the 14th International Conference on Computational Semantics (IWCS). Groningen, The Netherlands (online): Association for Computational Linguistics, 2021, pp. 222–227. [Online]. Available: https://aclanthology.org/2021.i...

    work page 2021

  10. [10]

    Survey of abstract meaning representation: Then, now, future,

    B. Mansouri, “Survey of abstract meaning representation: Then, now, future,” 2025. [Online]. Available: https://arxiv.org/abs/2505.03229

    work page arXiv 2025

  11. [11]

    Getting the most out of amr parsing,

    C. Wang and N. Xue, “Getting the most out of amr parsing,” in Proceedings of the 2017 Conference on Empirical Methods in Natural 10 Language Processing, Copenhagen, Denmark, 2017, pp. 1257–1268. [Online]. Available: https://aclanthology.org/D17-1129/

    work page 2017

  12. [12]

    Framenet: A knowledge base for natural language processing,

    C. F. Baker, “Framenet: A knowledge base for natural language processing,” inProceedings of Frame Semantics in NLP: A Workshop in Honor of Chuck Fillmore (1929–2014). Baltimore, MD, USA: Association for Computational Linguistics, jun 2014, pp. 1–5. [Online]. Available: https://aclanthology.org/W14-3001/

    work page 1929

  13. [13]

    Automatically deriving event ontologies for a commonsense knowledge base,

    J. Allen, W. Beaumont, L. Galescu, J. Orfan, M. Swift, and C. M. Teng, “Automatically deriving event ontologies for a commonsense knowledge base,” inProceedings of the 10th International Conference on Computational Semantics (IWCS 2013) – Long Papers, Potsdam, Germany, 2013, pp. 23–34. [Online]. Available: https://aclanthology.o rg/W13-0103/

    work page 2013

  14. [14]

    Ontology-driven automated reasoning about property crimes,

    F. Navarrete, L. A. Garrido, C. Bobed, M. Atencia, and A. Vallecillo, “Ontology-driven automated reasoning about property crimes,” pp. 687–710, 2024. [Online]. Available: https://link.springer.com/article/10 .1007/s12599-024-00886-3

    work page 2024

  15. [15]

    Actions and events in interval temporal logic,

    J. Allen and G. Ferguson, “Actions and events in interval temporal logic,” pp. 531–579, 1994. [Online]. Available: https://doi.org/10.1093/ logcom/4.5.531

    work page 1994

  16. [16]

    Acquiring commonsense knowledge for a cognitive agent,

    J. Allen, W. Beaumont, N. Blaylock, G. Ferguson, J. Orfan, and M. Swift, “Acquiring commonsense knowledge for a cognitive agent,” inAdvances in Cognitive Systems: Papers from the 2011 AAAI Fall Symposium (FS–11–01), 2011. [Online]. Available: https: //aaai.org/papers/04192-4192-acquiring-commonsense-knowledge-for -a-cognitive-agent/

    work page 2011

  17. [17]

    Problems with police reports as data sources: A researchers’ perspective,

    C. D. G ¨uss, M. T. Tuason, and A. Devine, “Problems with police reports as data sources: A researchers’ perspective,”Frontiers in Psychology, Volume 11, 2020. [Online]. Available: https://www.frontier sin.org/journals/psychology/articles/10.3389/fpsyg.2020.582428/full

    work page doi:10.3389/fpsyg.2020.582428/full 2020

  18. [18]

    Extracting meaningful entities from police narrative reports,

    M. Chau, J. J. Xu, and H. Chen, “Extracting meaningful entities from police narrative reports,” inProceedings of the 2002 Annual National Conference on Digital Government Research, ser. dg.o ’02. Digital Government Society of North America, 2002, pp. 1–5. [Online]. Available: https://dl.acm.org/doi/10.5555/1123098.1123138

    work page doi:10.5555/1123098.1123138 2002

  19. [19]

    Text mining and machine learning for crime classification: using unstructured narrative court documents in police academic,

    E. Bifari, A. Basbrain, R. Mirza, A. Bafail, S. Albaradei, and W. Alhalabi, “Text mining and machine learning for crime classification: using unstructured narrative court documents in police academic,” Cogent Social Sciences, 11(1), 2024. [Online]. Available: https: //www.tandfonline.com/doi/full/10.1080/23311916.2024.2359850

    work page doi:10.1080/23311916.2024.2359850 2024

  20. [20]

    Redacting sensitive information from the data,

    P. Rane, A. Rao, D. Verma, and A. Mhaisgawali, “Redacting sensitive information from the data,” in2021 International Conference on Smart Generation Computing, Communication and Networking (SMART GENCON), 2021, pp. 1–5. [Online]. Available: https: //ieeexplore.ieee.org/document/9645752

    work page arXiv 2021

  21. [21]

    Towards ai- driven policing: Interdisciplinary knowledge discovery from police body-worn camera footage,

    A. Srbinovska, A. Srbinovska, V . Senthil, A. Martin, J. McCluskey, J. Bateman, and E. Fokou ´e, “Towards ai- driven policing: Interdisciplinary knowledge discovery from police body-worn camera footage,” 2025, arXiv. [Online]. Available: https://arxiv.org/abs/2504.20007

    work page arXiv 2025

  22. [22]

    An overview of the tesseract ocr engine,

    R. Smith, “An overview of the tesseract ocr engine,” inProceedings of the Ninth International Conference on Document Analysis and Recognition (ICDAR 2007). IEEE Computer Society, 2007, pp. 629–633. [Online]. Available: https://research.google/pubs/an-overvie w-of-the-tesseract-ocr-engine/

    work page 2007

  23. [23]

    Research computing services,

    R. I. of Technology, “Research computing services,” 2026. [Online]. Available: https://www.rit.edu/researchcomputing/

    work page 2026

  24. [24]

    S. Bird, E. Klein, and E. Loper,Natural Language Processing with Python. O’Reilly Media, 2009. [Online]. Available: https://www.oreill y.com/library/view/natural-language-processing/9780596803346/

    work page arXiv 2009

  25. [25]

    Amr parsing as sequence- to-graph transduction,

    S. Zhang, X. Ma, K. Duh, and B. Van Durme, “Amr parsing as sequence- to-graph transduction,” inProceedings of the 57th Annual Meeting of the Association for Computational Linguistics, Florence, Italy, 2019, pp. 80–94. [Online]. Available: https://aclanthology.org/P19-1009/

    work page 2019

  26. [26]

    Penman: An open-source library and tool for amr graphs,

    M. W. Goodman, “Penman: An open-source library and tool for amr graphs,” inProceedings of the 58th Annual Meeting of the Association for Computational Linguistics: System Demonstrations. Online: Association for Computational Linguistics, jul 2020, pp. 312–319. [Online]. Available: https://aclanthology.org/2020.acl-demos.35/

    work page 2020

  27. [27]

    Hermit: An owl 2 reasoner,

    B. Glimm, I. Horrocks, B. Motik, G. Stoilos, and Z. Wang, “Hermit: An owl 2 reasoner,”Journal of Automated Reasoning, vol. 53, no. 3, pp. 245–269, Oct. 2014. [Online]. Available: https://doi.org/10.1007/s10817-014-9305-1

    work page doi:10.1007/s10817-014-9305-1 2014

  28. [28]

    Can we derive general world knowledge from texts?

    L. K. Schubert, “Can we derive general world knowledge from texts?” inProceedings of the Second International Conference on Human Language Technology Research, ser. HLT ’02. San Francisco, CA, USA: Morgan Kaufmann Publishers Inc., 2002, pp. 94–97. [Online]. Available: https://dl.acm.org/doi/10.5555/1289189.1289263 11 APPENDIX IX. SUPPLEMENTARYONTOLOGYSTAT...

    work page doi:10.5555/1289189.1289263 2002