We present an initial framework resulting from our ongoing research concerning modeling strategies. Our approach is rooted in a subjectivist, communication-based view on modeling. Under this approach, models are viewed as the result of modeling dialogues, which are a specialized sub-type of the diverse conversations that constitute a system development conversation at large. By focussing on the process of modeling instead of properties of models or modeling languages, we expect, eventually, to be able to better understand and deal with some currently problematic aspects of modeling, in particular model validation in context. We sketch plans for an environment for studying modeling conversations and strategies. As the title suggests, the current status of our research efforts are at the level where we have an initial understanding of the fundamental workings of the playing field. In future research, the actual strategies will make their appearance.
what to do to get good models. In addition, some aspects of quality, it seems, can be better achieved through a good modeling process than by just imposing requirements on the end product. This holds in particular (though not exclusively) for matters of validation. In our view, valid models are only attainable by viewing a model in context, and therefore in relation to the actors who have created and agreed with some model. The process of modeling thus is a crucial link between the end product and its context.
Perhaps the most practical reason for studying which behavior leads to good models concerns human resources. Generally, few experts are available who are capable of, or willing to, perform high-quality modeling. In any case, this process takes much time and effort (especially formal modeling). However, an increasing number of high quality models (even “light formalizations”) are required in system development (e.g. formalized ontologies, business rules, requirements). Anything that helps guide modeling behavior and support the process would be helpful. 2
It makes good sense to consider (the requirements and functionality of) both
Ways of Modeling and Ways of Working on the basis of a fundamental Way of
Thinking [
The clearest aspect of knowledge transformations that can be observed is
the exchange of particular statements (in formal or informal language, verbal or
graphical) in order to achieve a knowledge goal. This is why we view “statements”
(with some function, stated in some language) as the most appropriate level at
which we can study what goes on in IS development and modeling. Two basic
sets of statements are to be distinguished here: the statements that are a record
of the modeling conversation, and the statements that constitute the knowledge
that is agreed upon by all participants. The former set is a recording of the
modeling conversation, the second essentially constitutes the model. Importantly,
the statements are grounded in their (social) context: the people that stated them
and agreed with them [
Modeling, then, to us is a process of gathering and refining statements: a sequence of questions and answers. Consequently, we reject the view of modeling describe translate
Model
as a linear process of first describing a (relevant aspect-of/abstraction-from a)
domain and then producing a model corresponding to this description. This is
illustrated in Figure 1. In this linear process, the resulting model essentially is
a translation from the (natural language) statements forming the description of
the domain to a to a (formal) model [
Domain
Questions Modeling process
Answers
Model
verbalize
Description
From a generic system development perspective, modeling is – at least in
theory – to be integrated with less typically “model-oriented stages” in the process,
e.g. explorative requirements gathering and negotiation, obtaining commitment
for models, and so on. We expect that some strategies that are useful for
modeling can also be fruitfully applied outside “pure modeling”. This perspective
reflects the wish to apply rational principles (communication-oriented or
otherwise) in system development where relevant and feasible [
As a starting point for our exploration of modeling behavior we take (aspects
of) the detailed ORM Conceptual Schema Design Procedure (CSDP; its Way of
Working) [
Our view of a modeling process is as follows. For particular contexts and
particular knowledge goals, Ways of Modeling are selected [
Modeling conversations are, of course, executed by participants. Each
participant shares knowledge statements with the others, trying to achieve agreement
on them. However, the knowledge goals of participants, as well as their
competencies [
The approach described is largely driven by what could be seen as the mother
of all conversation strategies: uncertainty reduction [
Ideally, the participants work towards mutually agreed set of statements in
a controlled language: natural language statements (or possibly, some agreed-on
graphical format) with a clear and simple syntax that reduces the danger of
ambiguity and confusion. This is in fact another high-level, generically applied
conversation strategy that is crucial in reducing linguistic uncertainty. In this
manner, the demands posed by the formalism underlying the way of modeling
can be fulfilled without discussing the syntax or semantics of the formalism
as such. In addition, interpretation of terms used in the conversation may be
explored in dedicated sub-conversations leading to more statements expressing
the meaning of the term, to the point of satisfaction of all participants [
The backbone of our framework is a “dialogue grammar”, which is used to
structure the sequence in which the actions can take place, and some further
restrictions imposed on it [
Withdraw(a, s) Actor a withdraws statement s. Withdraw is the opposite of a propose.
Accept(a, s) Actor a accepts statement s as a valid statement; it may eventually become part of its internal model Ma. A statement can only be accepted after it is proposed.
Reject(a, s) Actor a rejects statement s, because a finds s unacceptable even for further consideration. Reject is the counterpart of accept.
Ask(a, q) Actor a asks question q, to be answered by some actor. Queries can be withdrawn or answered.
Answer(a, q, s) Actor a answers question q with statement s; an answer functions as a special Propose.
As a mere illustration of our approach to modeling dialogues and strategies,
consider this very simple example, that reflects a possible strategy in
ORMlike modeling: we might call it the “delayed specificity strategy”. It comes in two
flavors: one taking the “gather more information” approach, the other the “make
assumptions” approach. A more elaborate, formalized analysis of this strategy
can be found in [
Course of action A Solve the non-specificity by asking the domain expert for missing information.
Course of action B Alternatively, solve the non-specificity by assuming the missing information. This requires the SA sub skill of conceiving and verbalizing plausible information.
Postcondition The non-specificity is resolved.
The strategy could lead to the following dialogue, which reflects a step-wise approach to the gathering of the information needed: propose DE ask SA
What kind of thing is John? ask/propose SA
propose DE ask/propose SA
Do we distinguish John from other persons by means of his name? accept DE,SA Yes [we distinguish John from other persons by means of his name].
Do you agree that John is a person with name John? accept DE,SA Yes [John is a person with name John].
For each “accept”, both DE and SA are registered as agreeing with the statement. In two cases, the action of asking and proposing is collapsed in one statement. As a result of this small sub-dialogue, the statement “John is a person with name John” is added to the separate set of “agreed statements”, on the basis of which a complete formalization can later be compiled (preferably automatically). In addition, by means of the “defoleating strategy” (i.e. removing instances from the statements, which is a typical step in “fact based modeling”), a type-level generalization can be derived (weeding out the instance “John”), suggested, and confirmed: ask/propose SA Do persons generally have names?
accept DE,SA Yes [persons generally have names]. 5
The core goal of our research-in-progress is to study, describe, validate, and
eventually improve strategies for (formal) modeling in a system development
context. As the title suggests, the current status of our research efforts are at the
level where we have an initial understanding of the fundamental workings of the
playing field. In future research, the actual strategies will make their appearance.
The only way of achieving satisfying results in an effort like ours is to prominently
include empirical data and validation in the study. Our initial ideas concerning
modeling and system development as a conversation (as sketched in this paper)
are grounded in literature, our own experiences, and interviews with experienced
modelers [
The research approach we intend to follow can be classified as “action
research” [
Therapeutic stage – This involves collaborative change experiments. In this
stage changes are introduced and the effects are studied [
Having passed the diagnostic stage at an exploratory level, we are currently preparing to start a more in-depth diagnostic study of actual modeling dialogues and strategies. However, we expect the study to move into the initial therapeutic stage before long: the moment the guiding of the dialogue is introduced. Various diagnostic and therapeutic issues will arise and be dealt with iteratively. Importantly, participants in the experiments will be “imported” from real life contexts as much as possible, so their knowledge and knowledge goals are familiar to them and optimally realistic, even if the modeling environment is, inevitably, not.
The environment for recording and guiding modeling dialogues will, at least initially, use verbal questions-and-answers in a workable controlled language. The design metaphor used is that of a “chatbox” (distantly related to the well known “telephone heuristic” in modeling), with an “automated modeling agent” (SA) as a participant in the chat. Creating an effective environment will be a study in itself. We intend to make opportunistic use of existing agent interaction and dialogue paradigms, and NLP techniques. We will start off by investigating information and domain modeling in ORM. Planned areas of application to be studied (i.e. contexts) are information modeling, pre-negotiation ontology construction, and requirements modeling.