Interdisciplinary Introduction to Categorization: Interview with Amie Thomasson (philosophy), Dvora Yanov (political sciences), and Thomas Serre (artificial intelligence)

Par Stéphane Baele
Université de Namur (Belgique)

Versions : HTML | PDF Papier

[Fr] Dans cet entretien, nous avons posé le même jeu de sept questions à trois chercheurs de premier rang dans les études de la catégorisation. Dans un esprit de multidisciplinarité, nous avons formulé des questions « naïves » – c’est-à-dire les plus intuitives et les moins disciplinairement marquées – et les avons posées à chacun de manière cloisonnée. Amie Thomasson est professeure de philosophie à l’Université de Miami, Dvora Yanow est professeure visiteur en science politique à l’Université d’Amsterdam – après avoir assuré la « strategic chair » de sciences sociales à la Vrije Universiteit, et Thomas Serre est chercheur en intelligence artificielle à l’Université Brown. Une fois de plus, nous les remercions. Mots-clefs : catégorisation, interview, enjeux, disciplines scientifiques.

[En] This interview provides the answers of three prominent categorisation scholars to the same set of seven questions they were sent. These questions should be seen as «naïve », in the sense that they were aimed to be as intuitive and disciplinary-independent as possible. This has been done in order to allow contrasting answers to emerge. Amie Thomasson is professor in philosophy at the University of Miami, Dvora Yanow has hold a « strategic chair » at the social and political sciences faculty at the Vrije Universiteit and is now visiting professor in political science at the University of Amsterdam, and Thomas Serre is researcher in artificial intelligence at Brown University. Once again, we warmly express our gratitude for their participation. Key words: categorisation, interview, issues, scientific disciplines.


1. Notre première question est aussi la plus naïve : qu’est-ce que la catégorisation ? Et quel est le meilleur exemple qui l’illustre ? Our first question is also the most naïve one: what is categorisation? And what could be its most illustrative example?

Amie Thomasson (A.T.): Categorization in philosophy is classically thought of as a way of dividing entities into highest kinds or genera. The earliest philosophical example of categorization comes from Aristotle’s Categories, in which he lists ten highest categories, including: Substance (e.g., man, horse), Quantity (e.g., four-foot, five-foot); Quality (e.g., white, grammatical); Relation (e.g., double, half) ; Place (e.g., in the Lyceum, in the market-place); Date (e.g., yesterday, last year); Posture (e.g., is lying, is sitting); State (e.g., has shoes on, has armor on); Action (e.g., cutting, burning) ; Passion (e.g., being cut, being burned). But while many past and contemporary philosophers see the study of categories as the study of the most basic kinds of entity there are (this is known as the ‘realist’ approach), others see the task instead as a matter of working out what the most basic categories are in terms of which we conceptualize, experience, or speak of things in the world (this is the ‘conceptualist’ approach, illustrated by Kant and Strawson among others).

Dvora Yanow (D.Y.): I’ll give a “naïve” answer: “categorization” means “making categories.” We all do it, every day, in everyday ways. Go into a supermarket: its produce and other goods are categorized. Have breakfast in the morning, lunch midday, dinner at night – you are engaging in categorization. Identify people you pass in the street as “infants,” “children,” “youth” or “teenagers,” “adults,” “elderly,” and you have categorized them. And so forth, from my perspective as an ethnographer. Because it is so common-place, there is no “most illustrative” example. The more important point to note is that because of its commonplace-ness, we do not attend to our category-making. That is the source both of its power and of its problematic character: category structures (“taxonomies”) entail tacit knowledge which is made explicit at times only with difficulty; and that knowledge is created from a point of view that also goes unattended – unless we make it a point to focus on this everyday, common sense, unspoken, unwritten knowledge.

Thomas Serre (T.S.): In cognitive science, categorization refers to the process that allows objects and people to be recognized, differentiated and understood. For instance, finding a friend’s face in a crowd or finding keys on a messy desk are two categorization tasks that people solve routinely.

2. Doit-on différencier catégorisation et catégorisation sociale ? Should we differentiate between categorisation and social categorisation?

A.T.: Yes, we probably should, depending on what is meant by each term. One might take the latter to be a matter of trying to determine what the categories of entities in the social world (e.g. student, teacher, citizen, immigrant...) are, as opposed to categories of entities in the natural world (e.g. biological or chemical categories; or basic metaphysical categories like substance, property, event...). Or one might take ‘categorization’ to be a matter of trying to discern what the (most basic) categories of things are in the world (in the realist spirit), and ‘social categorization’ to be a matter of trying to discern what categories members of different societies use to conceptualize the world around them (a version of the conceptualist approach mentioned above). The categories used to conceptualize the world around us certainly vary culturally—at least for high-level categories (though perhaps not for a few ‘basic’ categories). Some defend the view that basic categories are not culturally relative, holding that the most basic categories of things in the world are simply ‘out there’ in the world to be discovered independently of anyone’s concepts (as some philosophers, arguably including Aristotle, suppose). Or alternatively, one may take the conceptualist transcendentalist view that the basic categories are those that must be employed in any human understanding (e.g. object, property, event, person...) or language, even if the higher-level categories, including social categories (e.g. race and religion categories, gender categories, food categories, art categories) vary from culture to culture (we see something along these lines in the work of Kant and Strawson).

D.Y.: I don’t think there is a general answer to that question, at least from my perspective as a policy/political/organizational ethnographer. Categories – all of them – do not exist in nature; they are human creations, made collectively (by a smaller group – e.g., an organization – or a larger entity – a society). From the perspective of practices, whether you separate these or not depends on what is of interest to you.

T.S.: The branch of computer science that deals with the problem of categorization is called ‘machine learning’. From the machine learning perspective, there is no real distinction between categorization tasks. These are all similar in that, in principle, the same machine learning algorithms can be used to solve them. The key difference between different categorization tasks lie in the nature of the representation used for a specific problem rather than the categorization process itself.

3. L’étude des catégories et des processus de catégorisation est aujourd’hui fragmentée entre plusieurs disciplines, elles-mêmes divisées en différents courants théoriques parfois difficilement compatibles. Face à cette multiplicité, quelle serait aujourd’hui l’approche – ou la combinaison d’approches – la plus adéquate pour conduire une telle étude ? Current studies of categories and categorisation processes are still fragmented between various disciplines, which are themselves places of (sometimes) strong theoretical oppositions. What would be the approach – or combination of approaches – that could transcend these antagonisms and put forward a satisfactory understanding of categorisation?

A.T.: I think the first step involves understanding what the practical purposes of different systems of categories are. To the extent that different disciplines are pursuing different goals in their divisions of things into categories, the disagreements among their resulting systems may be merely apparent. Another important move is to try to become clear about what methods are to be employed in distinguishing categories (in the relevant discipline, for the relevant purpose). Again, if we can agree about the methods for distinguishing categories and determining category differences, we can have a better chance of agreeing in our conclusions (or if we don’t, we can perhaps see the differences in results as resulting from different uses of the word ‘category’ tied to different methods, and thus being merely verbal disagreements).

D.Y.: From where I sit, the fact that category-making has been studied in various disciplines is not a difficulty. On the contrary, it means that we can understand, potentially, different aspects of categorizing practices. The distinctions drawn in theories – between a “slotting” approach and a “prototype” approach, for instance – are not so much incompatible as they are calling attention to different features. I think both can be used in a single study; but for me, the more important point is to start with the character of the categories being studied and the ways in which they are being used (the practices), and to let those determine what sort of theoretical and/or methodological approach is more useful.

T.S.: The short answer is: computational models and theories. Computational models and theories force ideas to be made precise (in order to be implemented and simulated on a computer) and can help integrate knowledge across several levels of analysis. Computational models can provide a much-needed framework for summarizing and integrating existing data and for planning, coordinating and interpreting new experiments.

4. Face à cette multiplicité d’approches et de théories que nous avons invoquée, quelles sont les compatibilités et incompatibilités qui vous apparaissent cruciales, sur lesquelles le doute n’est que peu possible ? In the light of the aforementioned antagonisms, which theories appear to connect rather well, and which theoretical combinations are more doubtful?

A.T.: I don’t know enough about the use of categories in other disciplines to comment on this well, though I know there have been fruitful interactions between philosophers, cognitive scientists, psychologists, and computer scientists about categories.

D.Y.: Given my view of the previous question, this one makes no sense to me. I have no answer for it.

T.S.: I am not sure how to answer that…

5. Si l’on considère uniquement votre discipline, pensez-vous qu’il est exagéré de dire que la catégorie et la catégorisation constituent son nœud central ? La question de la catégorie n’est-elle pas en effet à chaque fois, dans chaque courant, déterminante ? Considering your discipline alone, would you agree that categories and categorisation constitute the central issue? Isn’t categorisation a core problem within each strand of the discipline?

A.T.: In philosophy, categorization is certainly a central issue, though I would not claim that it is the central issue; the field is too diverse for that (subdisciplines such as ethics, epistemology, philosophy of science, etc. don’t directly concern categorization at all). In the sub-discipline of metaphysics, categories play a much more central role: sometimes directly, as philosophers dispute which categories there are, or which are most basic; more often indirectly, e.g. in discussions of which sorts of entity to accept into one’s ontology (which presupposes a division into sorts or categories), in arguments about whether or not various assertions involve a category mistake, and so on.

D.Y.: I work in several fields – public policy analysis, organizational studies, race-ethnicity, and so on. Although categorization is a central practice within each of them – what discipline does not categorize, beginning with its “units of analysis”? – I cannot say that category-making is the “core problem” within them. Policy analysis and organizational learning, or practice studies, each has set up other kinds of questions at its core. I do think that for race-ethnic studies, category-making is central – but that’s almost definitional, for that area of enquiry, and not something surprising.

T.S.: Categorization is today a central topic for many researchers interested in creating intelligent machines at the intersection between mathematics, neuroscience and computer science.

6. Sommes-nous condamnés à ne rien comprendre à la catégorisation, puisque cette dernière apparaît comme le mécanisme le plus essentiel de notre perception et de notre esprit, mécanisme nécessaire à sa propre compréhension ? Faut-il abandonner toute étude des catégories et les accepter comme données, en tant que « tout ce que nous pouvons connaître », dans une perspective néo-positiviste ? Since categorisation is such a fundamental process in perception, mind and action, are we therefore condemned to misunderstand it? In other words, can we build theoretical accounts of categorisation through thinking – that is, by using categorisation? Or should we abandon such enquiries and accept positivistic worldviews?

A.T.: I do not think it is hopeless to attempt to understand categorization. There are actually two enterprises here: one, trying to determine the fundamental categories into which things in the world fall; the other, trying to determine the fundamental categories we use in conceptualizing or experiencing the world (some of course would argue that these go together). You seem to have the latter, conceptualist, approach in mind here. Working out what categories we use to speak of or experience the world will indeed require different techniques than direct empirical investigation of the world itself—perhaps some form of transcendental argument, and/or a form of semantic analysis (analysing the differences in the sorts of fundamental rules governing our terms or concepts) is needed, rather than empirical analysis made using our categories, but that is not to say that it is impossible. (Another attempt at developing a relevant methodology can be found in Husserl’s methods of phenomenological reduction—arguably designed to reveal the ways in which, and meanings in terms of which, we experience the world rather than revealing any facts about the world itself.)

D.Y.: I can’t make much sense of this question. What understanding positions “positivism” against the study of categorization? Many realist-objectivist studies rest on categories; but so do constructivist-interpretive ones. This question seems to be mixing apples and pears, so to speak. I can understand it a bit more with the French version; and then I think that it equates “categorization” with “sense-making” in general. At this very moment, trying to make sense of this question, I am thinking in ways that are not only “categorical.” Metaphoric reasoning, for instance, while moving meaning from one domain to another, is not involved in determining “A”-ness and “not-A”-ness – not with the same discriminatory sense-making as one engages in taxonomy-building (whether slotting or prototyping). And in any event, since when have scholars given up on trying to understand something because it’s complex and to accept it as given?

T.S.: In recent years, progress in machine learning has been very significant. At a theoretical level, we now have a formal mathematical theory, called statistical learning theory, to describe the categorization problem. This theory has permitted the rapid development of modern computer vision systems capable of interpreting and recognizing complex visual scenes. I think a good measure of our understanding of the categorization process is the widespread availability of automated face recognition algorithms in consumer electronics.

7. Finalement, croyez-vous que deux manières différentes de penser la catégorisation entraînent deux éthiques différentes au sujet des problèmes réels – parfois graves – dus à la catégorisation sociale ? The last question relates to ethics: do you agree with the idea that two diverging ways of considering categorisation involve two different moral stances on social problems that emerge from categorisation?

A.T.: That’s an interesting question. Certainly many categories (e.g. substance, object, property) don’t obviously have anything to do with moral stances, and differing views about which of these basic metaphysical categories are fulfilled, or which fundamental, certainly need not go with different moral stances. But perhaps where social categories are concerned, e.g. race, gender, national identity categories, or other institutional categorizations (e.g. student, convict, citizen, etc.) one could make the case (following a sort of Heideggerian idea) that what is in part constitutive of membership in that category is being subjected to certain norms of treatment. (I have argued this point regarding categories of artefact.) If that case can be made out, then different ways of distinguishing these social categories may come with different norms for treating the individuals (or objects) that fall under them, and so may in that sense involve different moral stances.

D.Y.: Here, too, I do not understand the question. Which two do you have in mind? If we take the classic split between seeing categorical divisions as clear and bounded (the slotting of a set of mailboxes, for instance), versus seeing them as entailing prototypes under a “normal curve,” I do not see that these invariably translate into different moral stances. For me, this does not entail a generalizable, universalizable answer but, instead, rests on the context in which they are being applied. And even then, I’m not sure that category-making itself entails a morality. Where morality enters for me is in their use, especially when it is the state that is creating categories and imposing them on persons without giving the latter the opportunity of self-identification or self-definition (as is currently the situation in many places with respect to designating membership in “race-ethnic” groups). This also has consequences for how those persons may be conceived of and/or treated by others. Perhaps if I understood the intent of the question, this would make more sense to me.

T.S.: I have no opinion on this.

Article publié en version papier aux Presses universitaires de Louvain dans le n° 8 de la revueÉmulations de décembre 2010 : 
Catégories Politiques