DynaPsych Table of Contents

Understanding the Evolution of Categorization :

An Interdisciplinary Approach

I. Saillot, M. Patou-Mathis

Institut de Paléontologie Humaine, Muséum National d’Histoire Naturelle de Paris, 1 rue René Panhard, F-75013 Paris, France.

Corresponding author : I. Saillot, isaillot@mnhn.fr, tel : +33 1 45 47 51 69, fax : +33 1 45 47 06 62


Recent         Advances in the psychology of goal oriented actions show how it is possible to construct the semantic networks of procedural knowledge, directly giving insight on the categorization associated to a non verbal task. According to this approach, the elements of the analysis are the objects being manipulated, and the functional properties they’ve being given by the subject. Such conditions perfectly match a large amount of archaeological data, because reconstructing the past is always reconstructing behavior, namely associating each artefact to the actions it has undergone. It is hence possible to link into binary relationships the archaeological artefacts with the procedures they have been applied. These relationships can be processed by calculating a Galois lattice. The final graph is the Hasse diagram of the lattice. The hierarchy obtained is a visual representation of the categorization associated to the task, how anciant it might be. Such an interdisciplinary approach, involving close collaboration between prehistorians and psychologists, could allow psychologists to easily investigate the evolution of categorization, especially if comparisons between data sets of different periods were analysed this way.



                   Categorization, semantic network, Galois lattice, ancient behavior.


A rigorous approach to ancient categorization


                   Giving a rigorous description of the categorization capacity of past men is one of the main goals of the semantic analysis applied to anciant human cognition (Saillot, 2001). These concepts are based, first, on the use of recent theoretical advances from goal oriented actions psychology, and second, on the evidence for compatibility between some archaeological data and psychological models regarding categorization. We suggest that gathering these two fields could help providing evolutionary cognition with common discussion bases among specialists (Saillot, 2000-2001).


This modelling not only allows of description but also interpretation. The cognitive data can give evolutionary or functional insights about man evolution and allows of comparisons between sites from different periods or locations. This article also proposes a preliminary modelling of the anciant categorization associated with remains from a paleolithic cave (Tournal, France). 


Archaeological data are often goal oriented actions


Goal or        Goal oriented actions are actions performed by a subject during a task at the laboratory. Let’s consider man subsistance behaviour during paleolithic : it is typically a goal oriented action (Richard J.-F. & al., 1988) because a task is being performed (acquisition and processing of animals) and goals are pursuied (feeding, habitat…). The processed data comes after the perception level which consists for instance in identifying animals, places, contexts, ethologic or climatory meanings (Patou-Mathis M., 1993). We will recall here how dividing subsistance behaviour into elementary units allows of a description and an interpretation of paleolithic man cognitive activity regarding the act of categorization, according to psychology of actions (I.Saillot & al., 2002)


The computational modelling


Procedures applied to objects can be seen as properties of these objects exactly the same way than color or shape for instance : objects have functional properties as well as surface ones. This fact doesn’t seem to have been considered so far nor by archaeologists, nor by psychologits. It is nevertheless extremely interesting for cognitive archaeology because procedures are often known by deduction along with the artefacts directly found on the sites. In fact, according to T. Wynn, this is even the very scope of archaeology as a science, namely reconstructing human behaviour according to the artefacts (Wynn, 2002). This is particularly the case about subsistance behaviour during paleolithic (Patou-Mathis, 1996 ; 1997). Of course it is even clearer in the case of more recent periods, where subsistance activities become a lot more diversified, as systematic fishing proves it for instance, a behaviour too often neglected by archaeologists (Maingeot, 1999).


What we know about actions is structured in our memory like what we know about objects. It is hence possible to define a semantic action network as it is possible to define an object semantic network (Poitrenaud, 2001). The first step of the analysis is hence analysing men activities step by step, defining precisely the objects and the procedures they underwent by the men on the site. Regarding paleolithic periods, what past men did whith the objects found on the site is the field of palethnology, and the palethnologic results are the starting point for the anciant categorization investigation.

Firstly the relevant palethnologic data must be isolated from the other results in the litterature or directly on the site. Then the objetcs and their properties are gathered in a double entry table objects – properties, which establishes the binary relationships between them. Each time an object displays a property, the intersection cell is checked.


The computer eventually gives a graphic representation of the semantic network linked to the categorization of the task on the site. The description and the interpretation about anciant categorization on the site is obtained from this graph according to cognitive psychology concepts.


The computer program mathematically gives the simplified Hasse diagram of the Galois network generated by the combinations objects – properties of the initial table. Galois networks have been used to describe memory structure since Wille (1992) showed that nodes could be viewed as concepts. Preliminary studies have been done with STONE (Semantic Tree Oriented Navigator and Editor), a software designed by S. Poitrenaud from University Paris 8 (Poitrenaud S., 1998). These results have demonstrated the validity of a semantic analysis applied to anciant cognition. A new program was then devised at the IPH, SIMBOL, following the same process. SIMBOL (Système Informatique de Modélisation par Buts et Objets Liés) developped with Excel VBA draws, like STONE the hierarchy enclosed within the data table. The hierarchy given is displayed along a general – specific axis, which is the aim of the Galois lattice.


The graph is a tree which displays oriented arcs with arrows and commented nodes. Several arcs following each other are called a line. Each node displays a set of objects and properties : from the cognitive point of view, it’s a category (sometimes even called a concept). The arcs are the relationships between the categories, sorted according to a hierarchy from the most specific ones to the less. The graph directly displays the categories activated by the past men, but also the relationships between them : we attend the past categorization process itself.


Preliminary semantic analysis of a prehistoric behavior : the case of the Tournal cave


We chose this site because it’s very rich regarding the remains of men activities at this place and the marks and artefacts are very well preserved from taphonomic alterations. The archaeological data about this layer is hence very rich too  (Tavoso, 1987 ; Patou-Mathis, 1994). The numerous anthropic marks gave way to an accurate knowing of the human gestures that occured on the site as far as during this middle paleolithic stay. The human type is Neanderthal man.


Our analysis deals with the bones displaying marks which anthropic origin can’t be doubted. Selected items are : phalanxes, metapodiale, humerus, femur, tibia, radius et mandible. Taxonomic level is not relevant in this study. The procedures are selected according to the goals that oriented the actions, namely the extraction of skin, tendons, marrow and also the fact of burning something.

The marks found on the surface of the bones allow to identify the gestures that generated them and it is then possible to ascribe goals to these elementary actions. These goals are one of the sets of data of our semantic study, the other one being the objects.


This last step of the analysis is often not completely sure and a lot of data must be put aside from the archaeozoologic results. For instance numerous anthropic marks are very clear but the goal is not with certainty. In fact it is often possible to know  which step of the functionnal processing chain let the marks, but these steps are not goals in themselves and the artefacts can’t be selected if no other relevant information is available.


Only two of these marks have been chosen for our palethnologic study : the ones on the deer phalanx because it’s nearly certain that they’ve been generated due to the extraction of the tendons, and the marks on the hyena mandible which clearly prove the fur acquisition. When long bones fracturation occurs, we consider that it proves the intention of getting the marrow inside.

Some bones have been intentionaly burned (Patou-Mathis, 1994). As it’s been intentional, burning has been placed within the goals series and not the elementary gestures ones.


Table I presents the palethnological results selected from the litterature about the mousterian layer of the cave. Objects, actions and goals are presented in a synthetic way. The selection criteria is the following : the object found on the layer must be understood into its action context, in other words, the psychologist must know what process the object has undergone, either elementary gestures or final goal.






humerus (horse, bovines, ibex)



radius (horse, bovines)



tibias (horses, bovines)



tibia (ibex)

fracturation striation


metapodiale (bovines)



mandible (hyena)



phalanx (deer)



metatarsus (bovines)



phalanx (bovines)



humerus (equidae)



radius (equidae)



femur (equidae)



metacarpus (equidae)



phalanxes (indéterminées)




TABLE I : Paleo-ethnologic data selected for the semantic study



In table II, the data is presented so as to be directly processed by SIMBOL. The entry line shows the objects, whereas the entry column shows the properties ascribed to the objects, by paleolithic man. The software needs binary relationships between objetcs and their properties, namely here the goals that lead the actions on the objects. When an object has a property, the intersection cell is checked. This table is simply a different way to display the data enclosed in table I.











extract skin








extract tendons








extract marrow

















Table II

Hasse diagram obtained with SIMBOL

( Binary relationships between objects and their properties)


The following graph has been generated by SIMBOL. It gives the logical structure enclosed in the initial table. The blue boxes, nodes of the graph, are sets of objects and their linked properties, in other words each blue box is a category, according to cognitive psychology. The organization of the categories is presented along a hierarchy from the most specific (source of the hierarchy) to the most general categories (top of the lines). The arrow on the lines show the heritage direction, hence the hierarchy direction. Such a graph is a semantic action network and gives way to interpretation from cognitive psychology.


Figure 1 : Hasse diagram of the prehistoric layer



The description allowed by SIMBOL is a usual description from goal oriented actions cognitive psychology. It opens the way to knowledge tranfer between psychologists and archaeologists and even between archaeologists themselves thanks to new common discussion bases.


The interpretation suggested here, especialy evolutionary ones, can’t be directly validated by cognitive psycvhology because it overwhelms its field. These suggestions are hence open.


Description and interpretation of the graph

Number of categories : 5

Number of relationships : 3

Number of abstraction levels : 2


Superordonated categories are those of extraction of marrow and burning. They are the most general and deal with the maximum of objects. On the contrary, categories regarding tendons or skin seem to be more specific, their extension is lower. But due to the low number of abstraction levels, the hierarchy may not be very relevant to describe categories. So it’s impossible to go further than a description and give an interpretation. One of the questions to ask in the future is to know wether the particular topographic location of the category about marrow (here at the top of the hierarchy) could be related to a important goal for prehistoric men, either during the butchery task, and during human evolution.


The categories at the top of the lines (marrow and burning) heritate from the properties of their lower neighbors. Burning and extracting skin are close together on the graph : could the process of skin have occured before the voluntary burning of bones during paleolithic ?


No category has more than two neighbors on the graph, and one is even isolated. This rather low quantification could suggest that for the selected procedures in the initial table, the cognitive fluidity remains relatively low. The number of close neighbors on the graph lets few possibility to cognitive mechanisms like sense sliding and analogy, as defined by D. Hofstadter (1995).

These mental properties suggest a rather low cognitive complexity for the mousterian men of Tournal because the ability to generate analogies is the core of complex cognitive tasks like problem solving or elaborating strategies (hunting, habitat, etc.). Nevertheless it’s impossible to conclude anything from this particular result about mousterian men generaly speaking and more data must first be obtained.


The extraction of tendons is related to a category close to the hierarchy source and we could interpret this fact suggesting that extracting tendons is a frequent task. Could we infer that the use of ligated or sewed objects was frequent too, even as far as in middle paleolithic ? The topographic position of categories linked to tendons and marrow are close to each other and we know that extracting marrow has been a typical human activity since a long time. Are tendons in the same position ? We suggest that our semantic analysis could allow the psychologists to learn something about actions that do not let visible tracks on the archaeological remains.


Summary and perspectives


The semantic action analysis of the Tournal Cave mousterian level gave a first view on a categorization process during middle paleolithic. According to archeozoological and palethnological results, the cognitive categories of a neandertal man or population, regarding elementary tasks of the subsistance behaviour are displayed by SIMBOL. Cognitive complexity and fluidity seem in a medium range and will necessitate further investigations to deliver relevant information. But the structure of the graph already shows that the categories linked to the use of tendons and marrow are close to each other, maybe suggesting the antiquity of tendons use, which lets no archaeological remains. Directly describing human cognition, semantic analysis could certainly yield information about man activities beyond the scope of traditional archaeology.


The semantic analysis should be applied on each prehistoric site, systematically associated to the digging process, like any other caracterization regarding datation, climatic, faunal, geologic or lithic data. This new paradigm will deliver all its power when there are enough results to allow comparisons between sites of different periods (giving insights about evolution) or different places of the same period (giving insights about prehistoric culture).


References Cited











·        Richard J.-F. & Poitrenaud S. (1988). Problématique de l’analyse des protocoles individuels d’observations comportementales. In Caverni J.-P., Bastien C., Mendelsohn P. & Tiberghien G. (Eds.), Psychologie cognitive : concepts et méthodes, P. U. G., Grenoble.






·        Wille R. (1992). Concepts lattices and conceptual knowledge systems. In Computers Math. Application, volume 23, pp493-515.