Beyond Control-Flow: Integrating the Resource Perspective into Multi-Collaborative Process Modeling from Text

Alessandro Berti; Anton Antonov; Gyunam Park; Humam Kourani

arxiv: 2605.24546 · v1 · pith:FVC7FSFOnew · submitted 2026-05-23 · 💻 cs.AI · cs.IR

Beyond Control-Flow: Integrating the Resource Perspective into Multi-Collaborative Process Modeling from Text

Anton Antonov , Humam Kourani , Alessandro Berti , Gyunam Park This is my paper

Pith reviewed 2026-06-30 13:03 UTC · model grok-4.3

classification 💻 cs.AI cs.IR

keywords BPMN 2.0process modelinglarge language modelsresource perspectivecollaboration diagramstext-to-model generationintermediate languagebusiness process management

0 comments

The pith

An intermediate language with mandatory resource details enables LLMs to generate BPMN 2.0 collaboration diagrams from natural language text.

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

The paper aims to show that process modeling from text can go beyond control flow to include the resource perspective in collaborative settings. It proposes a pipeline that uses a compact intermediate language to capture organizations as pools and roles as lanes, then converts this to formal BPMN with message events for interactions. A layout routine handles spatial arrangement. Tests with nine LLMs on ten processes indicate good resource extraction while keeping control flow quality intact and with little extra computation time. This matters for automating the creation of models that reflect real multi-party business operations.

Core claim

The central claim is that rather than prompting LLMs directly for BPMN XML, using a compact executable intermediate language that requires resource details for pools and lanes allows generation of formal BPMN 2.0 collaboration diagrams from natural-language descriptions, with cross-organization dependencies represented by message events and automatic orthogonal layout.

What carries the argument

The compact, executable intermediate language with mandatory resource details defining both the organization (pool) and the role (lane), which structures the LLM output for conversion to BPMN.

If this is right

Resource discovery remains strong across multiple LLMs.
Control-flow quality from text-to-model is preserved.
Runtime overhead stays marginal.
Cross-organization interactions are materialized using message events in the diagrams.

Where Pith is reading between the lines

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

If the method scales, it could support automated generation of models for complex multi-organizational workflows.
This approach might connect to other areas like automated process simulation by providing richer models.
Testing on processes with ambiguous resource descriptions could reveal limits of the assumption.

Load-bearing premise

Natural language process descriptions contain sufficient, unambiguous information about organizations and roles that LLMs can reliably extract and that the intermediate language can represent without material loss of collaborative semantics.

What would settle it

A test set of ten processes where using the intermediate language results in significantly lower resource discovery accuracy or degraded control-flow quality compared to direct LLM prompting for BPMN XML.

Figures

Figures reproduced from arXiv: 2605.24546 by Alessandro Berti, Anton Antonov, Gyunam Park, Humam Kourani.

**Figure 1.** Figure 1: Motivating Example: Models generated by gemini-3-flash for the Complaint Handling process (p13). model therefore needs more than the control-flow perspective, i.e., the ordering of activities and events: it must also capture the resource perspective, namely who performs each activity and how participants communicate. This organizational view is essential for understanding responsibilities, handovers, and… view at source ↗

**Figure 2.** Figure 2: Overview of the proposed resource-aware generation pipeline: from [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** Figure 3: Boxplot analysis of execution time (seconds) per iteration. The consistent [PITH_FULL_IMAGE:figures/full_fig_p013_3.png] view at source ↗

**Figure 4.** Figure 4: Accuracy of resource assignment aggregated by LLMs. [PITH_FULL_IMAGE:figures/full_fig_p014_4.png] view at source ↗

read the original abstract

Process modeling is a sub-domain of Business Process Management (BPM) focused on the translation of process artifacts into formal models. This task traditionally requires extensive human input and domain expertise in both BPM notations and the specific business context. While Large Language Models (LLMs) can now automate much of this manual work, current text-to-model approaches focus predominantly on the control-flow perspective-ordering activities without considering the collaborative aspect of the processes. In this paper, we introduce a resource-aware generation pipeline that produces formal BPMN 2.0 collaboration diagrams from natural-language descriptions. Rather than solely prompting an LLM for raw XML, we describe a compact, executable intermediate language with mandatory resource details defining both the organization (pool) and the role (lane). Cross-organization dependencies are materialized using the standard formal notation for such interactions-message events-while an orthogonal layout routine automatically handles the spatial arrangement of elements within pools and lanes. Experiments on ten business processes with nine LLMs show strong resource discovery while preserving control-flow quality and adding only marginal runtime overhead. This approach moves generative modeling toward a more comprehensive, multi-collaborative representation of business operations.

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.

Desk Editor's Note private letter to a colleague

The paper gives a workable pipeline that adds mandatory resource details to LLM-generated BPMN collaboration diagrams via an intermediate language, with multi-LLM tests on ten processes showing usable results but narrow scope.

read the letter

The main point is a pipeline that uses a compact intermediate language to force resource information (pools for organizations, lanes for roles) into the output when turning natural language into BPMN 2.0 collaboration diagrams. It handles cross-organization links with message events and adds an automatic layout step instead of relying on raw LLM XML output.

What works is the empirical check across nine LLMs on ten processes: resource discovery comes through reliably while control-flow quality holds and extra runtime stays small. Testing multiple models gives more grounding than single-LLM demos, and the intermediate language is a straightforward engineering choice that keeps the output in standard BPMN.

The soft spot is the evaluation size. Ten processes is too small to establish how well this scales or handles ambiguous descriptions, and the paper does not show strong baselines against direct prompting or measure semantic loss in the intermediate step. The core assumption that natural-language texts contain clear, extractable org and role details is stated but not stress-tested on messier inputs.

This is for BPM researchers and tool builders who already work on automated process modeling and want to extend it to collaborative cases. A reader focused on LLM applications in business processes would find the pipeline and the multi-model results useful.

It deserves peer review because the method is concrete, the experiments are reported across independent models, and the contribution sits squarely in the applied BPM literature even if larger tests would strengthen it.

Referee Report

2 major / 1 minor

Summary. The paper introduces a resource-aware generation pipeline that produces formal BPMN 2.0 collaboration diagrams from natural-language descriptions. Rather than direct LLM prompting for XML, it defines a compact executable intermediate language that mandates resource details (organization as pool, role as lane), materializes cross-organization interactions via message events, and applies an orthogonal layout step. Experiments on ten business processes across nine LLMs report strong resource discovery, preserved control-flow quality, and marginal runtime overhead.

Significance. If the results hold, the work advances text-to-BPMN generation by systematically incorporating the resource perspective for multi-collaborative models, using standard BPMN notation and an intermediate representation that is executable. This pragmatic engineering contribution could support more complete automated modeling tools, though the narrow scope of ten processes limits claims of broad applicability.

major comments (2)

[Experiments section (abstract and implied evaluation)] The central empirical claim (strong resource discovery while preserving control-flow quality) is load-bearing but unsupported by reported metrics, baselines, quantitative scores, or process details. The abstract and evaluation description provide no precision/recall figures, control-flow similarity measures, or comparison to direct XML prompting, making verification impossible from the given information.
[Pipeline description (abstract and method overview)] The intermediate language is presented as a core contribution with 'mandatory resource details' and 'executable' properties, yet no syntax definition, semantics, or example mappings to BPMN pools/lanes/message events are supplied; this prevents assessment of whether collaborative semantics are preserved without material loss.

minor comments (1)

[Abstract] The abstract states results across 'nine LLMs' and 'ten business processes' but does not name the models or processes; adding this list would improve reproducibility without altering the core contribution.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments. The points raised correctly identify areas where the current manuscript lacks sufficient detail for independent verification, and we will revise accordingly.

read point-by-point responses

Referee: The central empirical claim (strong resource discovery while preserving control-flow quality) is load-bearing but unsupported by reported metrics, baselines, quantitative scores, or process details. The abstract and evaluation description provide no precision/recall figures, control-flow similarity measures, or comparison to direct XML prompting, making verification impossible from the given information.

Authors: We agree the referee's assessment is accurate: the provided abstract and evaluation overview contain no quantitative metrics, baselines, or process details to support the summarized claims. In the revised manuscript we will add precision/recall for resource discovery, control-flow similarity measures, a direct-XML-prompting baseline, and explicit details on the ten processes. revision: yes
Referee: The intermediate language is presented as a core contribution with 'mandatory resource details' and 'executable' properties, yet no syntax definition, semantics, or example mappings to BPMN pools/lanes/message events are supplied; this prevents assessment of whether collaborative semantics are preserved without material loss.

Authors: We agree that the current manuscript supplies neither a syntax definition nor semantics nor concrete mappings. The revised version will include a formal syntax (EBNF), semantics, and worked examples demonstrating translation to pools, lanes, and message events. revision: yes

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper describes an engineering pipeline that uses LLMs to populate a compact intermediate language (with mandatory pools/lanes) before emitting BPMN 2.0 collaboration diagrams; the central claims rest on empirical results across nine independent LLMs and ten processes rather than on any self-referential definition, fitted parameter renamed as prediction, or load-bearing self-citation. No equations, uniqueness theorems, or ansatzes are introduced that reduce to the authors' own prior outputs. The derivation chain is therefore self-contained against external benchmarks (standard BPMN notation and off-the-shelf LLMs).

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

Ledger constructed from abstract only; full paper may contain additional parameters or assumptions not visible here.

axioms (1)

domain assumption Large language models can reliably extract organizational resources and roles from natural-language process descriptions
This assumption underpins the entire resource-discovery step of the pipeline.

invented entities (1)

compact executable intermediate language no independent evidence
purpose: To enforce mandatory inclusion of pool and lane resource details before BPMN XML generation
New construct introduced by the authors to structure LLM output for resource awareness

pith-pipeline@v0.9.1-grok · 5740 in / 1219 out tokens · 49247 ms · 2026-06-30T13:03:44.294335+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

20 extracted references · 5 canonical work pages · 1 internal anchor

[1]

Chapman and Hall/CRC (2024) 16 A

Casella, G., Berger, R.: Statistical inference. Chapman and Hall/CRC (2024) 16 A. Antonov et al

2024
[2]

In: ITBPM@BPM

Corradini, F., Fornari, F., Polini, A., Re, B., Tiezzi, F.: Reprository: a repository platform for sharing business process models and logs. In: ITBPM@BPM. CEUR Workshop Proceedings, vol. 2952, pp. 13–18. CEUR-WS.org (2021)

2021
[3]

Ji, Z., Yu, T., Xu, Y., Lee, N., Ishii, E., Fung, P.: Towards mitigating hallucination inlargelanguagemodelsviaself-reflection.CoRRabs/2310.06271(2023).https: //doi.org/10.48550/ARXIV.2310.06271

work page doi:10.48550/arxiv.2310.06271 2023
[4]

CEUR Workshop Proceedings 4032, 304–311 (2025)

Klievtsova, N., Ehrendorfer, M., Mangler, J., Rinderle-Ma, S.: AutoBPMN.AI: Conversational Process Modeling and Automation. CEUR Workshop Proceedings 4032, 304–311 (2025)

2025
[5]

In: BPM (Demos / Resources Forum)

Köpke, J., Safan, A.: Introducing the BPMN-Chatbot for Efficient LLM-Based Process Modeling. In: BPM (Demos / Resources Forum). CEUR Workshop Pro- ceedings, vol. 3758, pp. 86–90. CEUR-WS.org (2024)

2024
[6]

Knowledge-Driven Hallucination in Large Language Models: An Empirical Study on Process Modeling

Kourani, H., Antonov, A., Berti, A., van der Aalst, W.M.P.: Knowledge-Driven Hallucination in Large Language Models: An Empirical Study on Process Model- ing. CoRRabs/2509.15336(2025)

work page internal anchor Pith review Pith/arXiv arXiv 2025
[7]

In: BPMDS/EMMSAD@CAiSE

Kourani, H., Berti, A., Schuster, D., van der Aalst, W.M.P.: Process Modeling with Large Language Models. In: BPMDS/EMMSAD@CAiSE. Lecture Notes in Business Information Processing, vol. 511, pp. 229–244. Springer (2024)

2024
[8]

In: IJCAI

Kourani, H., Berti, A., Schuster, D., van der Aalst, W.M.P.: ProMoAI: Process Modeling with Generative AI. In: IJCAI. pp. 8708–8712. ijcai.org (2024)

2024
[9]

Software and Systems Modeling (2025).https://doi.org/ 10.1007/s10270-025-01318-w

Kourani, H., Berti, A., Schuster, D., van der Aalst, W.M.P.: Evaluating Large Language Models on Business Process Modeling: Framework, Benchmark, and Self- Improvement Analysis. Software and Systems Modeling (2025).https://doi.org/ 10.1007/s10270-025-01318-w

work page doi:10.1007/s10270-025-01318-w 2025
[10]

Kourani, H., Park, G., van der Aalst, W.M.P.: Unlocking Non-Block-Structured Decisions: Inductive Mining with Choice Graphs. In: BPM. Lecture Notes in Com- puter Science, vol. 16044, pp. 144–161. Springer (2025)

2025
[11]

CoRRabs/2509.24592(2025)

Licardo, J.T., Tankovic, N., Etinger, D.: BPMN Assistant: An LLM-Based Ap- proach to Business Process Modeling. CoRRabs/2509.24592(2025)

work page arXiv 2025
[12]

CoRRabs/2408.01916 (2024)

Lin, L., Jin, Y., Zhou, Y., Chen, W., Qian, C.: MAO: A Framework for Pro- cess Model Generation with Multi-Agent Orchestration. CoRRabs/2408.01916 (2024)

work page arXiv 2024
[13]

In: ICSOC (1)

Nivon, Q., Salaün, G.: Automated Generation of BPMN Processes from Textual Requirements. In: ICSOC (1). Lecture Notes in Computer Science, vol. 15404, pp. 185–201. Springer (2024)

2024
[14]

OMG: Business Process Model and Notation (BPMN), Version 2.0. Tech. rep., Object Management Group (January 2011),http://www.omg.org/spec/BPMN/2.0

2011
[15]

Recker, J.: Opportunities and constraints: the current struggle with BPMN. Bus. Process. Manag. J.16(1), 181–201 (2010)

2010
[16]

Reimers, N., Gurevych, I.: all-minilm-l6-v2 sentence transformer.https:// huggingface.co/sentence-transformers/all-MiniLM-L6-v2(2021), hugging- Face model, accessed December 14, 2025

2021
[17]

Rosemann, M.: Potential pitfalls of process modeling: part B. Bus. Process. Manag. J.12(3), 377–384 (2006)

2006
[18]

In: Proceedings of the BPM (2025)

Safan, A., Köpke, J.: BPMN-Chatbot++: LLM-Based Modeling of Collaboration Diagrams with Data. In: Proceedings of the BPM (2025)

2025
[19]

Sander, G.: A Fast Heuristic for Hierarchical Manhattan Layout. In: GD. Lecture Notes in Computer Science, vol. 1027, pp. 447–458. Springer (1995)

1995
[20]

In: ICPM Workshops

Sola, D., Warmuth, C., Schäfer, B., Badakhshan, P., Rehse, J., Kampik, T.: SAP signavio academic models: A large process model dataset. In: ICPM Workshops. Resource-Aware Multi-Collaborative Process Modeling from Text 17 Lecture Notes in Business Information Processing, vol. 468, pp. 453–465. Springer (2022)

2022

[1] [1]

Chapman and Hall/CRC (2024) 16 A

Casella, G., Berger, R.: Statistical inference. Chapman and Hall/CRC (2024) 16 A. Antonov et al

2024

[2] [2]

In: ITBPM@BPM

Corradini, F., Fornari, F., Polini, A., Re, B., Tiezzi, F.: Reprository: a repository platform for sharing business process models and logs. In: ITBPM@BPM. CEUR Workshop Proceedings, vol. 2952, pp. 13–18. CEUR-WS.org (2021)

2021

[3] [3]

Ji, Z., Yu, T., Xu, Y., Lee, N., Ishii, E., Fung, P.: Towards mitigating hallucination inlargelanguagemodelsviaself-reflection.CoRRabs/2310.06271(2023).https: //doi.org/10.48550/ARXIV.2310.06271

work page doi:10.48550/arxiv.2310.06271 2023

[4] [4]

CEUR Workshop Proceedings 4032, 304–311 (2025)

Klievtsova, N., Ehrendorfer, M., Mangler, J., Rinderle-Ma, S.: AutoBPMN.AI: Conversational Process Modeling and Automation. CEUR Workshop Proceedings 4032, 304–311 (2025)

2025

[5] [5]

In: BPM (Demos / Resources Forum)

Köpke, J., Safan, A.: Introducing the BPMN-Chatbot for Efficient LLM-Based Process Modeling. In: BPM (Demos / Resources Forum). CEUR Workshop Pro- ceedings, vol. 3758, pp. 86–90. CEUR-WS.org (2024)

2024

[6] [6]

Knowledge-Driven Hallucination in Large Language Models: An Empirical Study on Process Modeling

Kourani, H., Antonov, A., Berti, A., van der Aalst, W.M.P.: Knowledge-Driven Hallucination in Large Language Models: An Empirical Study on Process Model- ing. CoRRabs/2509.15336(2025)

work page internal anchor Pith review Pith/arXiv arXiv 2025

[7] [7]

In: BPMDS/EMMSAD@CAiSE

Kourani, H., Berti, A., Schuster, D., van der Aalst, W.M.P.: Process Modeling with Large Language Models. In: BPMDS/EMMSAD@CAiSE. Lecture Notes in Business Information Processing, vol. 511, pp. 229–244. Springer (2024)

2024

[8] [8]

In: IJCAI

Kourani, H., Berti, A., Schuster, D., van der Aalst, W.M.P.: ProMoAI: Process Modeling with Generative AI. In: IJCAI. pp. 8708–8712. ijcai.org (2024)

2024

[9] [9]

Software and Systems Modeling (2025).https://doi.org/ 10.1007/s10270-025-01318-w

Kourani, H., Berti, A., Schuster, D., van der Aalst, W.M.P.: Evaluating Large Language Models on Business Process Modeling: Framework, Benchmark, and Self- Improvement Analysis. Software and Systems Modeling (2025).https://doi.org/ 10.1007/s10270-025-01318-w

work page doi:10.1007/s10270-025-01318-w 2025

[10] [10]

Kourani, H., Park, G., van der Aalst, W.M.P.: Unlocking Non-Block-Structured Decisions: Inductive Mining with Choice Graphs. In: BPM. Lecture Notes in Com- puter Science, vol. 16044, pp. 144–161. Springer (2025)

2025

[11] [11]

CoRRabs/2509.24592(2025)

Licardo, J.T., Tankovic, N., Etinger, D.: BPMN Assistant: An LLM-Based Ap- proach to Business Process Modeling. CoRRabs/2509.24592(2025)

work page arXiv 2025

[12] [12]

CoRRabs/2408.01916 (2024)

Lin, L., Jin, Y., Zhou, Y., Chen, W., Qian, C.: MAO: A Framework for Pro- cess Model Generation with Multi-Agent Orchestration. CoRRabs/2408.01916 (2024)

work page arXiv 2024

[13] [13]

In: ICSOC (1)

Nivon, Q., Salaün, G.: Automated Generation of BPMN Processes from Textual Requirements. In: ICSOC (1). Lecture Notes in Computer Science, vol. 15404, pp. 185–201. Springer (2024)

2024

[14] [14]

OMG: Business Process Model and Notation (BPMN), Version 2.0. Tech. rep., Object Management Group (January 2011),http://www.omg.org/spec/BPMN/2.0

2011

[15] [15]

Recker, J.: Opportunities and constraints: the current struggle with BPMN. Bus. Process. Manag. J.16(1), 181–201 (2010)

2010

[16] [16]

Reimers, N., Gurevych, I.: all-minilm-l6-v2 sentence transformer.https:// huggingface.co/sentence-transformers/all-MiniLM-L6-v2(2021), hugging- Face model, accessed December 14, 2025

2021

[17] [17]

Rosemann, M.: Potential pitfalls of process modeling: part B. Bus. Process. Manag. J.12(3), 377–384 (2006)

2006

[18] [18]

In: Proceedings of the BPM (2025)

Safan, A., Köpke, J.: BPMN-Chatbot++: LLM-Based Modeling of Collaboration Diagrams with Data. In: Proceedings of the BPM (2025)

2025

[19] [19]

Sander, G.: A Fast Heuristic for Hierarchical Manhattan Layout. In: GD. Lecture Notes in Computer Science, vol. 1027, pp. 447–458. Springer (1995)

1995

[20] [20]

In: ICPM Workshops

Sola, D., Warmuth, C., Schäfer, B., Badakhshan, P., Rehse, J., Kampik, T.: SAP signavio academic models: A large process model dataset. In: ICPM Workshops. Resource-Aware Multi-Collaborative Process Modeling from Text 17 Lecture Notes in Business Information Processing, vol. 468, pp. 453–465. Springer (2022)

2022