Cognitive Concepts for Understanding Process Models
Basically, three different problem-solving “programs” or “processes” are known from cognitive psychology: search, recognition and inference . Search and recognition allow for the identification of information of rather low complexity, i.e., locating an object or the recognition of patterns. Most conceptual models, however, go well beyond complexity that can be handled by search and recognition. Here, the human brain as “truly generic problem solver”  comes into play. Thereby, cognitive psychology differentiates between working memory that contains information that is currently being processed, as well as long-term memory in which information can be stored for a long period of time . Most severe, and thus of high interest and relevance, are the limitations of the working memory. As reported in , the working memory can not hold more than 7+-2 items at the same time. In addition, information held in the working memory decays after 18–30 seconds if not rehearsed . To illustrate how severe these limitations are, consider the sequence A-G-K-O-M-L-J. The average human mind is just capable of keeping this sequence in working memory, and, after 18–30 seconds the sequence will be forgotten.
The importance of the working memory has been recognized and led to the development and establishment of Cognitive Load Theory (CLT), meanwhile widespread and empirically validated in numerous studies . Theories of CLT revolve around the limitations of the working memory and how these limitations can be overcome—especially the second question is of interest for the understandability of process models. Hence, subsequently, we will discuss three strategies for dealing with the working memory’s limits. First, we will discuss the chunking of information. Second, we will show how process models can support inference through computational offloading. Third, we will introduce external memory, which allows to free resources in the working memory.
Chunking and Schema Acquisition
Information is believed to be memorized in interconnected schemata rather than in isolation . Those schemata, which are stored in long-time memory, can be used to aggregate information and handle it as a “chunk” of information in the working memory. Each schemata requires only one slot in working memory when used, hence mental effort is effectively reduced, cf. [6, 7, 8]. The process of aggregating information to a chunk using a schema, in turn, is referred to as “chunking” .
To illustrate how chunking actually influences the understandability of business process models, an example is provided in Figure 1. An unexperienced reader may, as shown on the left hand side, use three chunks to store the process fragment: one for each XOR gateway and one for activity A. In contrast, an expert may, as shown on the right hand side, recognize this process fragment as a pattern for making activity A optional. In other words, in her long-time memory a schema for optional activities is present, thereby allowing her to store the entire process fragment in one slot in the working memory.
Figure 1: Chunking of an optional activity
In contrast to chunking, which is highly dependent on the internal representation of information, i.e., how the reader organizes information in his mind, computational offloading highly depends on the exact external presentation of the business process model, i.e., visualization of the process model. In particular, computational offloading “refers the extent to which differential external representations reduce the amount of cognitive effort required to solve information equivalent problems” . In other words, an external representation may provide features that help the reader to extract information. Instead of computing and inferencing respective information in the modeler’s mind, information can, as in the process of recognition, more or less be “read-off” .
To put computational offloading into the context of business process modeling, a simple example illustrating the described phenomenon is shown in Figure 2. The illustrated process models are information equivalent, i.e., the same execution traces can be produced based on both Model A and Model B. However, Model A is modeled in BPMN, whereas for Model B the declarative modeling language ConDec  was used (for a detailed explanation of declarative process models we refer to [10, 11]). Consider the task of listing all execution traces that can be inferred from the process model. A reader familiar with BPMN will probably infer within a few seconds that Model A supports execution traces <A, B, D> and <A, C, D>. Such information is easy to extract as BPMN provides an explicit concept for the representation of execution sequences, namely sequence flows. Thus, for identifying all possible execution traces, the reader simply follows the process model’s sequence flows—the computation of all execution traces is offloaded to the process model. In contrast, for Model B, no explicit representation of sequences is present. Rather, constraints define the interplay of actions and do not necessarily specify sequential behavior. Thus, the reader cannot directly read off execution traces, but has to interpret the constraints in her mind to infer execution traces. In other words, process model B, while information equivalent to Model A, does not provide computational offloading for extracting execution traces. Consequently, even for a reader experienced with ConDec, listing all supported traces is far from trivial.
Figure 2: Computational offloading
Finally, we would like to introduce another mechanism that is known for reducing mental effort, i.e., the amount of working memory slots in use. In particular, we will discuss the concept of external memory in the context of business process models. An external memory is referred to any information storage outside the human cognitive system, e.g., pencil and paper or a black board [1, 2, 9, 12]. Information that is taken from the working memory and stored in an external memory is then referred to as cognitive trace. In the context of a diagram, a cognitive trace would be to, e.g., mark, update and highlight information . Likewise, in the context of process modeling, the model itself may serve as external memory. When interpreting a process model, marking an activity as executed while checking whether an execution trace is supported, can be seen as leaving a cognitive trace.
For the illustration of external memory and cognitive traces, consider the process model shown in Figure 3. Assuming the reader wants to verify, whether execution trace <A, D, E, F, G, H> is supported by the process model. So far, as indicated by the bold path and the position of the token, she has “mentally executed” the activities A, D, E, F and G. Without the aid of external memory, she will have to keep in the working memory, which activities have been executed, i.e., sub-trace <A, D, E, F, G>, as well as the position of the token within the process instance. By writing down the activities executed so far, i.e., transferring this information from working memory to external memory (e.g., piece of paper), load on working memory is reduced. In addition, the process model even allows to store the “mental token”—either by simply putting a finger on the respective part of the process model or by marking the location of the token, as shown in Figure 3.
Figure 3: External Memory
For more details and background we refer to:
 Jill H. Larkin and Herbert A. Simon. Why a Diagram is (Sometimes) Worth Ten Thousand Words. Cognitive Science, 11(1):65–100, 1987.
 William J. Tracz. Computer programming and the human thought process. Software: Practice and Experience, 9(2):127–137, 1979.
 Fred Paas, Juhani E. Tuovinen, Huib Tabbers, and Pascal W. M. Van Gerven. Cognitive Load Measurement as a Means to Advance Cognitive Load Theory. Educational Psychologist, 38(1):63–71, 2003.
 George Miller. The Magical Number Seven, Plus or Minus Two: Some Limits on Our Capacity for Processing Information. The Psychological Review, 63(2):81–97, 1956.
 Maria Bannert. Managing cognitive load—recent trends in cognitive load theory. Learning and Instruction, 12(1):139–146, 2002.
 Peter Gray. Psychology. Worth Publishers, 2007.
 Allen Newell. Unified Theories of Cognition. Harvard University Press, 1990.
 AndrewW. Crapo, Laurie B.Waisel,William A.Wallace, and Thomas R.Willemain. Visualization and the process of modeling: a cognitive-theoretic view. In Proc. KDD ’00, pages 218–226, 2000.
 Mike Scaife and Yvonne Rogers. External cognition: how do graphical representations work? Int.J. Human-Computer Studies, 45(2):185–213, 1996.
 Maja Pesic. Constraint-Based Workflow Management Systems: Shifting Control to Users. PhD thesis, TU Eindhoven, 2008
 Marco Montali, Maja Pesic, W.M.P. van der Aalst, Federico Chesani, Paola Mello, and Sergio Storari. Declarative Specification and Verification of Service Choreographies. ACM Trans. Web, 4(1):1–62, 2010.
 John Sweller. Cognitive load during problem solving: Effects on learning. Cognitive Science, 12(2):257–285, 1988.