=Paper=
{{Paper
|id=Vol-2196/BPM_2018_paper_21
|storemode=property
|title=WoPeD goes NLP: Conversion between Workflow Nets and Natural Language
|pdfUrl=https://ceur-ws.org/Vol-2196/BPM_2018_paper_21.pdf
|volume=Vol-2196
|authors=Thomas Freytag,Philip Allgaier
|dblpUrl=https://dblp.org/rec/conf/bpm/FreytagA18
}}
==WoPeD goes NLP: Conversion between Workflow Nets and Natural Language ==
https://ceur-ws.org/Vol-2196/BPM_2018_paper_21.pdf
WoPeD goes NLP: Conversion between Workflow
Nets and Natural Language
Thomas Freytag, Cooperative State University (DHBW) Karlsruhe, Germany
thomas.freytag@dhbw-karlsruhe.de
Philip Allgaier, bpcompass GmbH, Karlsruhe and Hamburg, Germany
philip.allgaier@bpcompass.com
Abstract. WoPeD (Workflow Petrinet Designer) is an open-source
Java software for designing business processes in terms of workflow nets,
a common extension of Petri nets. This demo lays the focus on two re-
cently added features making use of Natural Language Processing (NLP)
algorithms in order support the conversion of a graphical process model
into a textual process description and vice versa.
Keywords: Petri net tool, workflow nets, NLP, Process2Text, Text2Process.
1 Introduction
Having started in 2003 as a simple Petri net editor and simulator, WoPeD has
been continuously extended over the last fifteen years towards a rich software
tool covering a variety of BPM topics. Based on the well-defined language of
workflow nets as introduced in [1], WoPeD has been frequently serving as an
educational tool for a hands-on experience to model, analyse and simulate busi-
ness process as well as to understand the underlying algorithms and verification
methods. For example, WoPeD can check a process model for soundness and
other semantic properties, can interactively construct the coverability graph, and
perform stochastic computations like capacity planning. Another feature allows
the transformation between WoPeD’s workflow net models and other common
process model languages like YAWL, BPMN, or EPC via an AProMoRe1 pro-
cess repository. Additionally, WoPeD has served serveral researchers in the BPM
community as an experimental platform to implement innovative project ideas
[4] [5]. Most of these features have been presented at past BPM conference demo
sessions (e.g. [2,3]). This demo paper focuses on the most recently added features
of WoPeD building the bridge between Natural Language Processing and BPM.
2 NLP and BPM
Graphical process modeling languages have become an industry standard, al-
lowing various stakeholders to collectively create visual process descriptions on
a conceptual level. However, graphical models frequently co-exist with informal
textual process descriptions, be it for lack of skills or tools or time to create
1
www.apromore.org
F. Casati et al. (Eds.): Proceedings of the Dissertation Award and Demonstration, Industrial Track
at BPM 2018, CEUR-WS.org, 2018. Copyright c 2018 for the individual papers by its authors.
Copying permitted for private and academic purposes. This volume is published and copyrighted
by its editors.
WoPeD goes NLP: Conversion between Workflow Nets and Natural Language
a graphical model or because a text-based description is required for a pro-
cess handbook as sometimes mandated by QA agencies. This has brought up
the question in BPM research transforming these two process representations
into each other, i.e. to automatically generate the textual equivalent of a given
graphical process model and to automatically generate an appealing visual pro-
cess model from a given textual representation. There are interesting practical
use cases for this functionality, e.g. to check the mutual compliance between a
textual and graphical version of a business process or to facilitate the inclusion of
visually impaired persons in process modelling activities by providing them with
a text-based interface for understanding and describing a business process. The
following two sections give an overview of how NLP algorithms have been inte-
grated into WoPeD: Firstly, to export the current process editor contents into
readable English text (“Process2Text”) and secondly to synthetize an workflow
net from a given verbal process specification (“Text2Process”).
3 Process2Text: Verbalization of a Workflow Net
In [6], an NLP-based approach has been introduced taking a BPMN diagram as
input and generating a naturual language text as output describing the modelled
behavior. Figure 1 shows the outline of the underlying algorithm. In the first step,
called “Text Planning”, the relevant linguistic information is extracted from the
workflow-net-based process model, mainly by analyzing labels of model elements,
control flow dependencies and task resource assignments.
Fig. 1. The Process2Text algorithm (adapted from [6])
The analysis is done by two freely available standard NLP tools, Stanford
Tagger2 and WordNet3 . The result is converted into a Refined Process Structure
Tree (RPST), a generic data structure to represent components of a process
2
http://nlp.stanford.edu/software/lex-parser.shtml
3
http://wordnet.princeton.edu
Thomas Freytag, Philip Allgaier
graph being introduced in more detail e.g. in [7]. After applying various heuristics
to restructure and fine-tune the RPST, the second major step of the algorithm
“Sentence Planning” is applied. Here, the leaves of the RPST are processed and
enriched with semantic information derived from the control flow in the graph.
After an additional refinement of the text fragments, a realizer tool (used here:
RealPro [8]) is applied in order to generate a naturally-sounding English text.
Fig. 2. Process2Text sidebar in WoPeD
The implementation in WoPeD takes major parts of the solution presented
in [6], however making the transformation process applicable to workflow nets
instead of BPMN diagrams and embedding the algorithm into the WoPeD pro-
cess editor by displaying the generated text together with the process graph.
Figure 2 shows a screenshot of the WoPeD editor with a graphical model and a
“Process2Text” sidebar containing the auto-generated text. The associated pro-
cess elements are highlighted once the user clicks on a specific text item line in
the sidebar (shown in red).
4 Text2Process: Generate a Workflow Net from Text
The opposite direction of the transformation is based on an algorithm well-
documented e.g. in [9] and [10]. The transformation uses the same freely-available
NLP tools as the Process2Text algorithm above. As a first step, the input string
is analyzed by the Stanford parser tool and split up into sentences and words
according to the underlying grammar rules. The subsequent steps make use of
two semantical word databases WordNet (see above) and FrameNet 4 .
4
http://framenet.icsi.berkeley.edu/fndrupal
WoPeD goes NLP: Conversion between Workflow Nets and Natural Language
Fig. 3. The Text2Process Algorithm (taken from [9])
This mainly identifies the semantically relevant elements of the textual pro-
cess model (e.g. actors, actions, business objects). The result is written into a
data structure called WorldModel where all contributing model elements and
their interrelations are stored. The WorldModel is converted into PNML, the
standard XML-based representation for workflow nets. The last step is to open
up a new WoPeD editor window and create coordinates for the automatically
generated model by calling WoPeD’s built-in layout beautifier. Figure 3 roughly
visualizes the algorithm. Figure 4 shows a screenshot of the dialog for entering
the textual process description (either by typing in or by uploading a text file)
and below the resulting auto-generated and auto-layouted workflow net.
Fig. 4. The Text2Process dialog in WoPeD
Thomas Freytag, Philip Allgaier
5 Conclusion and Outlook
The above features are still in an experimental stage suffering from the general
problem of semantical ambiguities in natural language processing in both di-
rections. However, the plan is to improve the transformation quality in future
releases of WoPeD. The software is open source and can be downloaded from
the homepage5 or directly from Sourceforge6 .
References
[1] Van der Aalst, W.M.P.; van Hee, K.M.: Workflow Management: Models, Methods,
and Systems MIT Press 2002.
[2] Freytag, T.; Sänger, M.: WoPeD - An Educational Tool for Workflow Nets. BPM
demo session 2014. http://ceur-ws.org/Vol-1295/paper3.pdf, accessed June 2018.
[3] Freytag, T.; Allgaier, P. Burratin A.; Danek-Bulius, A.: WoPeD - A “Proof-
of-Concept” Platform for Experimental BPM Research Projects. BPM demo
session 2017, http://ceur-ws.org/Vol-1920/BPM_2017_paper_190.pdf, accessed
June 2018.
[4] Reijers, H.A.; Freytag, T.; Mendling, J.; Eckleder, A.: Syntax highlighting in busi-
ness process models. Decision Support Systems 51(3): 339-349 (2011).
[5] Mendling, J.: Metrics for Process Models, Springer Verlag, Berlin, 2008.
[6] Leopold, H.; Mendling, J.; Polyvyanyy, A.: Generating Natural Language Texts
from Business Process Models. In: Ralyté, J., Franch, X., Brinkkemper, S., Wrycza,
S. (eds) Advanced Information Systems Engineering. CAiSE 2012. LNCS, vol 7328.
Springer, Berlin, Heidelberg 2012.
[7] Polyvyanyy A.; Vanhatalo J.; Völzer H.: Simplified Computation and General-
ization of the Refined Process Structure Tree. In: Bravetti M.; Bultan T. (eds)
Web Services and Formal Methods. Lecture Notes in Computer Science, vol 6551,
Springer, Berlin, Heidelberg, 2010.
[8] Lavoie, B.; Rambow, O.: A Fast and Portable Realizer for Text Genera-
tion Systems. 5th Conference on Applied Natural Language Processing, 1997,
www.aclweb.org/anthology/A97-1039, accessed June 2018.
[9] Friedrich, F.; Mendling, J.; Puhlmann, F.: Process Model Generation from Natural
Language Text. In: Mouratidis H., Rolland C. (eds) Advanced Information Systems
Engineering. CAiSE 2011. Lecture Notes in Computer Science, vol 6741, Springer,
Berlin, Heidelberg 2011.
[10] Riefer, M.; Ternis, S.; Thaler, T: Mining Process Models from Natural Language
Text: A State-of-the-Art Analysis. Processings of Multikonferenz Wirtschaftsinfor-
matik (MKWI-16), March 9-11, Illmenau, Germany, 2016.
5
www.woped.org
6
www.sourceforge.net/projects/woped/