What is a Concept? Perspectives from Design, Cognitive Science, and Social Theory

Introduction: The Problem of the Concept

In my previous post on conceptual modelling as design method, I articulated an approach to making visible the conceptual structures that underpin complex sociotechnical systems. What I did not examine was a more fundamental question: what is a concept?

This question has become pressing in my current work on the ADAPT/Pathway Generator project. When I attempt to model "vocational rehabilitation" or "machine learning" or "outcome prediction", I am treating these as concepts that can be extracted, structured, and visualised. But what exactly am I extracting? What kind of entity is a "concept", and what are the implications of different answers to this question for design practice?

In the early months of my doctoral work, orientating myself to the Swedish welfare context and trying to make sense of a project that seemed to promise impossible things, I found myself reading across disparate literatures - information systems, cognitive semantics, social theory - each of which offers a different account of what concepts are and how they function. This post is an attempt to synthesise that reading and articulate its implications for conceptual modelling in design.

The question matters because different accounts of "concept" imply different assumptions about what concept maps can represent, what they can reveal, and what they might fail to capture.

Three Traditions, Three Ontologies

I want to distinguish three broad traditions that offer different answers to the question "what is a concept?":

1. The Information Systems Tradition: Concepts as formal entities that can be modelled, structured, and computed upon
2. The Cognitive Science Tradition: Concepts as regions in multidimensional cognitive space, grounded in perception and prototype effects
3. The Social Theory Tradition: Concepts as socially constructed, performative, and emergent in situated interaction

These traditions are not hermetically sealed; there is cross-pollination and overlap. But they represent genuinely different ontological commitments about the nature of concepts, and these commitments have consequences for design practice.

1. The Information Systems Tradition: Concepts as Formal Entities

Mylopoulos and Conceptual Modelling

The information systems literature treats concepts as entities that can be formally represented in modelling languages. John Mylopoulos's foundational work on conceptual modelling (Mylopoulos, 1992) establishes this tradition. As Mylopoulos describes it, conceptual modelling is intended "to allow you to build a model of the situation in which the new system will operate" - a model that captures entities, activities, assertions, and their relationships.

Mylopoulos distinguishes three types of objects in his Requirements Modelling Language (RML): "entities, activities and assertions, all of which have attributes that relate them to other objects". This formal apparatus is intended to capture requirements in ways that can inform system development - making the implicit explicit, the tacit computable.

In his later work with Chung and Yu (Mylopoulos et al., 2002), this approach extends to goal-oriented requirements analysis. Here the conceptual apparatus includes not only domain entities but also "goals, agents, alternatives, events, actions, existence modalities and agent dependencies". The framework enables modelling of both functional and non-functional requirements, where "modeling a part of that world, the application domain, is a key activity in the software development process".

Guarino, Guizzardi, and Mylopoulos (2019) provide a more philosophically grounded account, arguing that "conceptual models are models of conceptual mental representations that cognitive agents use to construct their experience of the world". They emphasise that "a conceptual model has always a conceptual semantics, since the linguistic constructs they contain are always interpreted in terms of mental contents". This positions conceptual models as intermediaries between mental representations and formal specifications - capturing how domain experts think about their domains in a form that can inform system design.

The key claim is that conceptual models are "ontologically grounded" - they represent not merely arbitrary symbol structures but real aspects of how humans conceptualise domains.

Maass and Storey: Pairing Conceptual Modelling with Machine Learning

Particularly relevant to my current context is recent work by Maass and Storey (2021) on "Pairing Conceptual Modeling with Machine Learning". They argue that conceptual modelling and machine learning have developed largely independently but can complement each other. Referencing Mylopoulos (1992), conceptual modelling is described as "the activity for formally describing some aspect of the physical and social world around us for the purposes of understanding and communication".

Their framework proposes a progression from human mental models (tacit understandings held by domain experts) through conceptual models (explicit formal representations) to machine learning models (computational implementations). Crucially, they observe that "machine learning invariably relies on human mental models - representations of reality in the minds of data scientists or users of machine learning models, who either develop or interpret machine learning solutions, in light of their individual mental models".

This framework is directly pertinent to the ADAPT project, which promises to apply machine learning to vocational rehabilitation. Maass and Storey's insight is that making these mental models explicit through conceptual modelling can "help to structure requirements with the purpose of creating a shared understanding among various people during the design of a project".

Yet the framework also reveals a problem. If different stakeholders hold different mental models - if the caseworker's understanding of "rehabilitation" differs fundamentally from the data scientist's understanding of "predictive features" - then no amount of formal modelling can resolve that difference; it can only expose it. As they note, differences between stakeholders' mental models "inevitably lead to differences between the shared understanding of the information system design team and the behavior of the machine learning model".

Pfeiffer and Gehlert: Conflicts in Conceptual Model Comparison

Pfeiffer and Gehlert (2004) provide a taxonomy of conflicts that arise when comparing conceptual models. Their starting observation is that "conceptual models are a widely used means for documenting software systems as well as describing organisational structures" - yet comparing such models proves surprisingly difficult.

They identify three fundamental types of conflict:

• Type conflicts: "Type conflicts arise whenever the same fact of the application domain is represented by using different constructs of the modelling language". The same domain phenomenon might be represented as an entity in one model, an activity in another.

• Naming conflicts: Different terms used for the same concept (synonymy), or the same term for different concepts (homonymy). These conflicts reflect the fundamental polysemy of natural language.

• Structural conflicts: Equivalent domain content organised differently across models - different hierarchies, different decompositions, different granularities.

Their sobering conclusion is that "an automatic model comparison is theoretically not feasible since it cannot handle the real world semantics of the domain expressions involved". Fully automated comparison would require formalising all domain knowledge - an impossible task given the open-ended nature of real-world semantics.

This has profound implications for my practice. When I produce concept maps of different stakeholder perspectives, the conflicts I expose are not merely technical problems to be resolved through better modelling; they reflect genuine semantic heterogeneity that may be irreducible. A "purely syntactical model comparison process is accomplishable for a computer", but semantic comparison requires human interpretive judgment that cannot be fully automated.

An Example from Rehabilitation

All three conflict types are visible in my concept modelling of work rehabilitation. Consider how "depression" is represented across different models. In the JANUS Pathway Generator, I found two distinct depression variables: ht_dep (seemingly a general self-report measure) and dass_dep (derived from the DASS-21 standardised instrument). This is a type conflict - the same phenomenon represented through different constructs - compounded by a naming conflict, since neither variable name makes its assessment protocol explicit.

The structural conflicts are more profound. The BIP assessment model adopted by SCÖ captures five broad categories (health, opportunity, capability, collaboration, everyday functioning). The JANUS system uses over a hundred variables drawn from at least a dozen standardised instruments. Ludvigsson's Swedish model separates physical, cognitive, and social dimensions. The ICF Core Set for Rehabilitation organises functioning across body functions, activities, participation, and environmental factors. Each of these is a legitimate conceptual model of vocational rehabilitation. None is straightforwardly translatable into any other. And any attempt at machine learning across these models would need to navigate - not resolve - these incommensurabilities.

2. The Cognitive Science Tradition: Concepts as Geometric Regions

Gärdenfors and Conceptual Spaces

Peter Gärdenfors's (2000) theory of conceptual spaces offers a radically different account of what concepts are. Rather than treating concepts as symbolic entities defined by necessary and sufficient conditions, Gärdenfors proposes that concepts are regions in multidimensional cognitive space.

The theory "builds on a cognitivist view of semantics. This contrasts with both extensional and intensional realistic semantics" (Zenker & Gärdenfors, 2015). Where traditional semantics treats meaning as reference to objects in the world or to abstract intensions, the conceptual spaces approach treats meaning as location in cognitive geometry.

The dimensions of conceptual space correspond to quality dimensions - perceptual or cognitive dimensions along which stimuli vary. Colour, for instance, can be represented in a three-dimensional space (hue, saturation, brightness). More abstract domains can be represented in higher-dimensional spaces with dimensions corresponding to relevant properties. As Gärdenfors notes, "the dimensions form the framework used to assign properties to objects and to specify relations among them".

Crucially, Gärdenfors argues that "a natural property is a convex region of a domain in a conceptual space" (Criterion P). A convex region is one where any two points within it can be connected by a line that remains within the region. This geometric constraint captures the intuition that natural concepts have coherent, continuous structure - that if two things are both "red", then anything perceptually between them is also "red".

This account naturally accommodates prototype effects. "When natural properties are defined as convex regions of a conceptual space, prototype effects are indeed to be expected". The prototype is the central point of the region; membership grades off toward the boundaries. This explains why some exemplars seem more typical than others, why category boundaries are fuzzy, and why concepts resist definition by necessary and sufficient conditions.

Dessalles and the Operation of Contrast

Jean-Louis Dessalles, in his contribution to The Case for Geometric Knowledge Representation (Zenker & Gärdenfors, 2015), argues that in addition to the structure provided by conceptual spaces, "we also need the operation of contrast". A red face, for instance, is called "red" because it contrasts with other possible face colours, rather than being red in the prototypical sense.

Dessalles argues that "contrast, as it develops during a conversation, is an essential operation that converts perceptions into the predicates expressed in communication". Thus concepts, which belong to a conceptual space, "should be distinguished from predicates, which belong to language". The same conceptual region may give rise to different predicates depending on the contrast structure of the communicative context.

This has significant implications: "Relevance is an output, rather than an input, of the contrast operation". What makes a concept relevant in a given context is not intrinsic to the concept but emerges through contrast with contextual alternatives.

Dessalles further develops this by examining how "meanings are not only related to perception, but also to reasoning. Language is not used just to refer to things in the environment... It is used to convey propositional attitudes". These attitudes "express surprise, (dis)belief, and positive or negative emotions or desires". The concept is inseparable from the stance taken toward it.

Implications for Conceptual Modelling

Gärdenfors's and Dessalles's frameworks suggest that concepts are not discrete symbols but continuous regions in cognitive space, made salient through contrast operations. This has implications for conceptual modelling:

1. Concepts have fuzzy boundaries: There is no sharp line between "rehabilitation" and "support" or between "prediction" and "classification". Attempts to draw such lines impose artificial discreteness on continuous cognitive space.

2. Concepts are grounded in perception and experience: The dimensions of conceptual space are not arbitrary but rooted in how humans perceive and interact with the world. A concept of "vocational rehabilitation" will have dimensions related to experienced qualities of support, progression, barrier reduction.

3. Concepts are context-dependent: What counts as an instance of a concept depends on the contrast set in play. In one context, "AI" contrasts with "human judgment"; in another, it contrasts with "simple automation". The same term may invoke different conceptual regions.

4. Different stakeholders may inhabit different conceptual spaces: A caseworker and a data scientist may not merely disagree about where to draw boundaries within a shared space; they may operate in spaces with different dimensions entirely.

Work Ability as Conceptual Space

The concept of "work ability" (arbetsförmåga) is a useful test case. In my review of rehabilitation models, I found that work ability is not a single concept but a family of overlapping conceptual regions, each with different quality dimensions.

In Ludvigsson's model, the dimensions are physical, cognitive, and social - a three-dimensional space in which an individual's work ability can be located. Tengland adds environmental dimensions: the nature of the work, and the work environment itself. The ICF framework introduces "participation" as a distinct dimension, acknowledging that functioning depends not just on individual capacity but on engagement in life situations. Each model proposes a different geometry. To the caseworker meeting someone for the first time, "work ability" is grounded in perceptual qualities - how the person presents, what they say about their situation, what they seem capable of. The quality dimensions are experiential: energy, distress, motivation, coherence. To the data scientist reviewing the JANUS patient vector, "work ability" is a point in a feature space defined by DASS depression scores, Rosenberg self-efficacy scales, income categories, and housing status. The dimensions are quantitative and instrumentally derived.

Gärdenfors's framework helps explain why these aren't simply different views of the same thing. They are different conceptual spaces with different dimensional structures. Points that are "close together" in the caseworker's space (two people who present similarly in an interview) may be far apart in the data scientist's space (one scoring high on DASS anxiety, the other not). And Dessalles's contrast operation explains why "work ability" means something different depending on whether it contrasts with "disability" (a medical frame), "unemployment" (an economic frame), or "exclusion" (a social frame). The concept shifts as its contrast set shifts.

3. The Social Theory Tradition: Concepts as Social Performances

Goffman and Frame Analysis

Erving Goffman's (1974) concept of "frame" offers a sociological account of how meaning is constructed in interaction. Van Hulst and Yanow (2014) provide a useful synthesis of this tradition's implications for policy analysis.

One crucial feature of Goffman's interactionist approach is that the "definition of the situation... is not consciously created". Unlike the strategic framing emphasised by social movement theorists - who "typically focus on the strategic - conscious, intentional, cognitive - character of the different frame groups construct" - Goffman's frames emerge tacitly through ongoing practical accomplishment.

Van Hulst and Yanow argue that "an important part of the meanings of the acts, events, and things actors are confronted with resides not in those entities themselves or in the intentions of others, but in the interactional process through which they are conceived and expressed". Meaning is not retrieved from concepts but constructed in the framing process itself.

The framing process "enacts the sense-making work that enables what Rein and Schön (1977) called a normative leap from what is to what ought to be". Frames function as "implicit theories" of situations - not just descriptions but normative accounts that suggest what matters and what should be done.

This has immediate implications for the SCÖ context. The ADAPT project brings together stakeholders who frame what the project is in fundamentally different ways. For the UK data science group, the frame is research - testing whether the Pathway Generator algorithm can be generalised beyond Iceland. For SCÖ management, the frame is modernisation - demonstrating that the coordination association is adopting evidence-based, data-driven methods. For the caseworkers who would eventually use any resulting system, the frame might be threat or support, depending on whether they perceive ML as replacing their judgement or augmenting it. For the funder, the frame is innovation - international collaboration on a technically ambitious topic.

These are not merely different perspectives on the same project. They are different definitions of the situation, each carrying its own normative logic about what matters and what should be done. The concept of "federated learning for rehabilitation" is not a shared concept with different emphases; it is a different concept in each frame. And the frames are not consciously constructed or strategically deployed - they emerge from the institutional positions and professional identities of the actors involved. "Framing takes place - the Meadian/Goffmanian understandings or definitions of the situation at hand constructed by the multiple interpretive communities".

Giesen: Concepts as Sacred

If Goffman explains how concepts are framed differently by different actors, Bernhard Giesen's work on social performance explains why some concepts resist reframing - why certain shared understandings become so invested with collective meaning that questioning them feels transgressive.

Giesen, in his chapter "Performing the Sacred" (Alexander, Giesen & Mast, 2006), draws on the Durkheimian distinction between sacred and profane. The sacred represents that which transcends ordinary experience and carries collective significance - "the collective self" translated into shared symbols and representations. Not all concepts are sacred in this sense. "Depression" in a clinical handbook is a profane concept - useful, contested, revisable. But "innovation" in an organisation's strategic plan, or "evidence-based practice" in a professional community's self-understanding, may function as sacred representations: concepts that carry the weight of collective identity and aspiration.

What makes this relevant to the question of what a concept is is Giesen's account of constitutive rituals: "In their most elementary form rituals do not just describe or imitate an order of the external world that is also available by other representations". Ritual performances are iterations - they repeat events that have happened before - and through this repetition they constitute social reality rather than merely representing it. "No construction of social reality can entirely dispense with this constitutive poesis of the social - there has to be an ultimate horizon where we cannot step back anymore and assume the role of the observer".

This "constitutive poesis" - the way performance makes reality - suggests a fourth ontological possibility for concepts, beyond the three traditions outlined so far. Some concepts don't merely exist as formal entities, cognitive regions, or interactional accomplishments. They exist as sacred representations - concepts whose meaning is bound up with a community's collective identity, and whose questioning threatens not just intellectual disagreement but social dissolution. The concept holds the group together; to expose it to analytical scrutiny risks profaning the very thing that maintains organisational coherence.

Symbolic Interactionism

The symbolic interactionist tradition (Mead, Blumer) provides another social account of concepts. As contemporary summaries note, "Symbolic interactionism is a methodological framework derived primarily from the work of Mead and articulated by Herbert Blumer".

Blumer argued that the meaning of things "is derived from or arises out of the social interaction that one has with one's fellows" and "these meanings are handled in, and modified through, an interpretive process used by the person in dealing with the things he encounters". This approach "developed as a reaction to the positivistic sociological theories of the day that addressed society through structural models".

On this view, concepts are not representations of pre-existing entities but emerge through ongoing processes of social negotiation. The concept of "machine learning" in the ADAPT project is not a fixed technical definition but an ongoing accomplishment, shaped by who gets to speak authoritatively, whose definitions are institutionally enforced, and how the term functions rhetorically in funding applications and project reports.

Assessment as Ritual Performance

The social theory tradition offers a particularly illuminating lens on the assessment instruments used in vocational rehabilitation. When a caseworker administers the BIP questionnaire - asking a client to rate their health, everyday functioning, collaboration skills - this is not merely data collection. In Giesen's terms, it is a constitutive ritual: a performance that produces the client's "work readiness" as a social fact.

The BIP assessment follows a pattern: the caseworker asks standardised questions, the client responds, the responses are encoded in a scoring system, and a profile emerges. This profile then circulates through the institutional system - informing decisions about which interventions to offer, how long to fund them, what counts as "progress". The concept of the client's "work ability" is not something the assessment discovers; it is something the assessment constitutes through its performance.

Goffman's frame analysis deepens this. The same BIP conversation can be framed as support (we're helping you identify your strengths and needs) or as surveillance (we're checking whether you qualify for continued benefits). These frames aren't mutually exclusive; they coexist uncomfortably in the same interaction. And the client's responses - the "data" that might eventually feed a machine learning model - are shaped by which frame they perceive themselves to be in. A client who experiences the assessment as surveillance may present differently from one who experiences it as support. The "data" is not raw; it is a social accomplishment, saturated with the interactional dynamics of its production.

This has implications for my earlier questions about whether computational approaches can support relational care. If the data that feeds ML models is itself constituted through social performances - performances that are shaped by trust, power, framing, and the client's sense of who is asking and why - then the quality of the data is inseparable from the quality of the relationship. Automating or standardising the data collection may change the performance in ways that change what the data means.

Implications for Conceptual Modelling in Design

These three traditions offer genuinely different ontologies of concepts:

Tradition	What is a concept?	How does it exist?	What can modelling capture?
Information Systems	Formal entity with attributes and relationships	In models, languages, systems	Structure, relationships, requirements
Cognitive Science	Region in multidimensional cognitive space	In minds, grounded in perception	Dimensional structure, prototype effects, contrasts
Social Theory	Social accomplishment through performance	In practices, rituals, interactions	Nothing stable - only snapshots of ongoing construction

The Limits of Formal Modelling

The cognitive science and social theory traditions suggest fundamental limits to what formal conceptual modelling can achieve:

1. Concepts may not be discrete: If concepts are continuous regions in cognitive space (Gärdenfors), then the discrete nodes and edges of a concept map impose artificial structure. The crisp boundaries of a diagram belie the fuzzy gradients of actual conceptualisation.

2. Concepts may be context-dependent: If meaning is constructed dynamically through contrast (Dessalles) and framing (Goffman), then a concept map captures only one possible construction, not "the" concept. The map is always a map of this context, these contrasts.

3. Concepts may be performative: If concepts are socially accomplished through interaction (Goffman, Blumer), then modelling them may change them rather than simply represent them. The concept map is not a neutral mirror but an intervention that reconstitutes what it claims to represent.

4. Concepts may be sacred: If the binding force of shared concepts derives from their relationship to collective identity and transcendence (Giesen on Durkheim), then exposing them to analytical scrutiny may undermine the very meaning they carry. Making the sacred visible risks profaning it.

The Value of Modelling Nonetheless

Yet I would argue that conceptual modelling retains value precisely because it makes these problems visible. By attempting to formalise concepts, we discover where formalisation fails. By comparing models across stakeholders, we expose the conflicts and incompatibilities that Pfeiffer and Gehlert document. By creating explicit representations, we surface the tacit assumptions that different actors bring to shared terminology.

The concept map is not a neutral representation of pre-existing conceptual structure; it is an intervention that can reveal the heterogeneity, ambiguity, and contestation that informal discourse obscures.

But this revelation is not guaranteed to produce productive outcomes. What happens when the concept map exposes contradictions that stakeholders have invested in not seeing? What happens when making things visible threatens the "constitutive poesis" through which organisations maintain their coherence?

Connecting to the SCÖ Context

The three traditions illuminate the ADAPT/Pathway Generator project from different angles, and taken together they reveal why the conceptual landscape I'm navigating is so treacherous.

The information systems lens shows what I'm doing most visibly: producing formal conceptual models of the rehabilitation domain. My concept modelling work and data archaeology on the Pathway Generator are exercises in making implicit structures explicit. But Pfeiffer and Gehlert's taxonomy warns that the conflicts I'm surfacing - between BIP's broad categories and JANUS's granular variables, between Swedish rehabilitation paradigms and Icelandic ones, between clinical measures and self-report - may be irreducible. No synthesis will resolve them into a unified data model, because they reflect genuinely different ways of carving up the domain.

The cognitive science lens explains why these models resist integration. The caseworker, the data scientist, the client, and the policy maker don't just have different opinions about work ability; they inhabit different conceptual spaces with different quality dimensions. "Progress in rehabilitation" means different things depending on whether your dimensions are clinical outcomes, economic participation, or subjective wellbeing - and whether the contrast set is "continued sickness", "return to full-time work", or "a meaningful life". When someone on the project says "the algorithm will predict rehabilitation outcomes", the word "outcomes" is doing radically different conceptual work for different listeners.

The social theory lens explains why the gap between these conceptual spaces is so difficult to address. The concept of "AI for rehabilitation" doesn't just carry different meanings for different stakeholders; it performs different institutional functions. For the UK research group, it is another context in which to test or refine their algorithm. For SCÖ, it is a modernisation narrative - evidence that the organisation is forward-looking and evidence-based. For the funder, it represents innovation and reducing rates of long-term sickness or social isolation, or increasing the tax-base, or "enhancing" the social fabric. These functions are not incidental to the concept; they are constitutive of it. The "concept" of AI for rehabilitation is, in Giesen's terms, something closer to a sacred representation - invested with collective aspirations that transcend its technical content. This may explain why the project's continued existence does not seem to depend on its technical feasibility. It serves functions that operate at a different level entirely: signalling innovation, securing funding, maintaining organisational identity and the solidarity of the stakeholder network.

My earlier post on what design can contribute to ML argued that the design-ML literature operates at the product and interface levels, whereas the situation at SCÖ requires engagement at the level of discourse. The theoretical framework developed here suggests something more unsettling. At the discourse level, concepts are not stable entities that can be modelled and aligned. They are ongoing social accomplishments, shaped by power, performance, and the institutional functions they serve. My concept maps can expose the surface structure of these differences. What they cannot do is intervene in the social dynamics that produce and maintain them.

Conclusion: Concepts as Contested Terrain

The question "what is a concept?" has no single correct answer. Different intellectual traditions offer different accounts, each with implications for design practice.

What I take from this theoretical exploration is a heightened awareness of what my concept maps can and cannot do. They can formalise one possible structuring of a domain. They can reveal conflicts between stakeholders' implicit models. They can expose gaps between aspirational claims and material capabilities. In the rehabilitation context specifically, they can make visible the profound heterogeneity in how "work ability", "rehabilitation success", and "data-driven support" are conceptualised across different national traditions, professional communities, and institutional settings.

But what they cannot do is capture concepts as they actually exist: as continuous regions in the cognitive spaces of caseworkers who know their clients through years of interaction; as rituals performed in BIP assessments where "work readiness" is constituted rather than measured; as sacred representations of institutional identity where "AI for rehabilitation" carries meaning far beyond its technical content. The map is not the territory; the concept map is not the concept.

Whether this limitation undermines the value of conceptual modelling or simply defines its proper scope is a question I continue to grapple with. What I can say is that in my current context - where grand claims about AI collide with the absence of data infrastructure, where Icelandic depression variables resist translation into Swedish categories, where different stakeholders seem to inhabit different conceptual universes - some method of making these differences visible seems necessary, even if the method's theoretical foundations remain contested.

The question that these theories cannot answer - and that my doctoral work may need to address - is what happens when visibility itself becomes the problem. When making things visible triggers not recognition but resistance. When the concept map reveals contradictions that the organisation has reason to protect rather than resolve. When the formal model of vocational rehabilitation exposes that the stakeholders who agreed to collaborate on "data science" do not share a concept of what rehabilitation data is, let alone what science should be done with it.

---

References

Alexander, J.C., Giesen, B. & Mast, J.L. (Eds.) (2006). Social Performance: Symbolic Action, Cultural Pragmatics, and Ritual. Cambridge University Press.

Blumer, H. (1969). Symbolic Interactionism: Perspective and Method. University of California Press.

Dessalles, J.L. (2015). From Conceptual Spaces to Predicates. In Zenker, F. & Gärdenfors, P. (Eds.), The Case for Geometric Knowledge Representation (pp. 57-76). Springer.

Gärdenfors, P. (2000). Conceptual Spaces: The Geometry of Thought. MIT Press.

Giesen, B. (2006). Performing the Sacred: A Durkheimian Perspective on the Performative Turn in the Social Sciences. In Alexander, J.C., Giesen, B. & Mast, J.L. (Eds.), Social Performance (pp. 325-367). Cambridge University Press.

Goffman, E. (1974). Frame Analysis: An Essay on the Organization of Experience. Harvard University Press.

Guarino, N., Guizzardi, G. & Mylopoulos, J. (2019). On the Philosophical Foundations of Conceptual Models. Information Modelling and Knowledge Bases XXX. IOS Press.

Maass, W. & Storey, V.C. (2021). Pairing Conceptual Modeling with Machine Learning. Data & Knowledge Engineering, 134, 101909.

Mylopoulos, J. (1992). Conceptual Modelling and Telos. In Loucopoulos, P. & Zicari, R. (Eds.), Conceptual Modeling, Databases, and CASE. Wiley.

Mylopoulos, J., Chung, L. & Yu, E. (2002). From Object-Oriented to Goal-Oriented Requirements Analysis. Communications of the ACM, 42(1), 31-37.

Pfeiffer, D. & Gehlert, A. (2004). A Framework for Comparing Conceptual Models. In Proceedings of the Workshop on Evaluating Modeling Methods for Systems Analysis and Design (EMMSAD 2004).

Van Hulst, M. & Yanow, D. (2014). From Policy "Frames" to "Framing": Theorizing a More Dynamic, Political Approach. American Review of Public Administration, 46(1), 92-112.

Zenker, F. & Gärdenfors, P. (Eds.) (2015). The Case for Geometric Knowledge Representation. Springer.

---

Revision Note: This post reflects reading and reflection from the early months of doctoral work (2022), as I attempted to orientate myself theoretically to the challenges of the SCÖ/ADAPT context. The synthesis presented represents my engagement with these literatures at that time.

The question raised at the end - what happens when visibility triggers resistance rather than recognition - would become central to the doctoral research. The concept modelling work described here and in the previous post produced exactly the dynamics hinted at: the maps exposed contradictions that the project could not absorb, and the response was not alignment but marginalisation of the analysis.

Whether this constitutes a limitation of the method or a demonstration of its power remains an open question. The theories surveyed here - particularly Giesen's account of the sacred and the constitutive function of performance - may offer resources for understanding why certain kinds of visibility become unbearable for organisations invested in their own imaginaries.