Archaeologists work by destroying their object of study. An archaeological excavation is a process of deliberate destruction of the site being dug, during which relevant information is recorded. Since the original site is destroyed in the process, information must be recorded with special care, because archaeologists cannot revisit the site to check dubious information or complete missing data.
During an excavation, information is gathered about multiple kinds of things: the place itself, structures that may be discovered (such as walls or pavements), objects found (such as pieces of pottery, stone or bone), etc. When recording information about something, archaeologists usually employ the following approach: first, the thing being described is categorized. Is it an arrow head, a fragment of a pottery beaker, the remains of a wooden door, an ancient pavement? Once the category is clear, this gives them a number of properties that need to be documented. For example, if a pottery fragment is being described, they will probably note down things such as its dimensions, what part of the vessel it comes from (such as the rim, body or base), or whether it has some decoration. Each of these properties is described on paper or computer, and thus an information object is generated that represents the pottery fragment.
According to 1, this approach is very powerful, but may lead to the kind of category bias that Stephen Jay Gould used to call “Walcott’s shoehorn” 2: when we are asked to categorize something according to a collection of predefined categories, we tend to choose one even when none of them is suitable. In other words, we “push” the thing being described in the least bad category, even when this category is quite bad. Gould described the negative consequences that a famous case of the Walcott’s shoehorn had in palaeontology; in archaeology, the risk is the same, because we are constantly finding things that we know very little about. If something is ascribed to an ill-suited category, we are likely to focus on irrelevant details and miss crucial information.
To avoid or, at least, mitigate this, we propose in González-Pérez (2012) an approach called typeless information modelling. This means that things are not categorized before they are described. Instead, they are described straightaway by choosing individual properties that seem relevant. Imagine an archaeologist using this approach. When they find a pottery fragment during an excavation, they would describe properties such as its size, colour, weight, shape, material and texture, because these are what our senses allows us to perceive. Also, the archaeologist would try very strongly not to decide whether this thing is a beaker fragment, a loom weight or just a lump of workshop waste. This is difficult, because our mind tends to categorize everything in order to organize the world. If the archaeologist manages to do it, then an information object would be generated like before, but instead of being determined by a category, it would consist of a cluster of freely-chosen values. Information objects like these are called valsters, which is short for “value clusters”.
The fact that things are not categorized when they are described does not mean that we cannot use categories at all. On the contrary, the authors explain that a posteriori categorization can be helpful. After having described many things in a category-less fashion and having a sizeable collection of information objects, archaeologists would be able to enter declarative rules in a computer system to define emergent categories. For example, they could say something like “an Artefact is anything with values for at least two coordinates, plus Material and Cultural Ascription properties”, and immediately observe what things are selected as part of the Artefact category. They could tweak the definition and experiment with different options, receiving immediate feedback on what things get in and out of focus as the definition varies. Membership of the category would not be fixed, as in the conventional approach, but emerge as they explore possibilities.
To make this possible, the authors created a computer-based valster language, and developed some software that allowed them to experiment with the approach, including the definition of a posteriori categories.
The authors also realized that this approach has some issues. First of all, and as we said above, working against the natural tendency to categorize is difficult. In addition, the cognitive processes that people employ to construct and select properties are subject to bias as much as the processes that they use to construct and select categories. However, they argue that the bias is smaller with the proposed approach, because selecting the wrong property may add incorrect details to an information object, but selecting the wrong category would create an incorrect structure for the information, which would be a more serious problem.
To test whether the valster language and the overall approach could work in real-life settings, the authors rolled out some experiments with people 3. They selected ten archaeologists of varied profiles, arranged them in pairs, and asked them to describe a replica of a large protohistoric ceramic pot [Figure 1] by using valsters.
Also, participants were asked to fill in some questionnaires before and after the modelling exercise, in order to assess three variables as defined by 4: their perceived ease of use, the perceived usefulness, and the intention to use in the future. Most of the participants agreed that the approach was easy to use and useful, and all but one said that they would try to use it in the future. Participants expressed multiple advantages of the valster approach as compared to others, including the ability to minimize ambiguities without classifying the information a priori, or the possibility to describe information in a disorderly manner, relying on lists of properties that allow specialists not to start from scratch and not to forget properties, and for non-specialists to consider properties that they would not have thought of. They also mentioned that some degree of subjectivity is involved in selecting properties. The models produced were of acceptable quality, although some were uncappable of avoiding categorization, having incorporated properties that were, in fact, describing what kind of thing the ceramic pot was.
The authors conclude that the valster approach may be useful as a first step in archaeological information modelling: specialists would describe things without a priori categorization, and then they would be able to refine the resulting models by using more structured techniques such as conceptual modelling languages. Since the time when these results were published, further advances have been made in ConML, a modelling language for the humanities and social sciences, which aims to address the second step that would come after valster modelling. Whether this works or not is still to be determined.
- Gonzalez-Perez, C., 2012. Typeless Information Modelling to Avoid Category Bias in Archaeological Descriptions, in Thinking Beyond the Tool: Archaeological Computing & the Interpretive Process, A. Chrysanthi, P. Murrieta Flores, and C. Papadopoulos (eds.). BAR International Series, 2344. Archaeopress. Oxford (UK) ↩
- Gould, S.J., 2000. Wonderful Life: The Burgess Shale and the Nature of History. New Edition. Vintage. ↩
- Hug, C. and C. Gonzalez-Perez, 2012. Qualitative Evaluation of Cultural Heritage Information Modelling Techniques. ACM Journal on Computing and Cultural Heritage. 5(2). doi: 10.1145/2307723.2307727 ↩
- Moody, D.L., 2003. The method evaluation model: A theoretical model for validating information systems design methods, in Proceedings of the 11th European Conference on Information Systems (ECIS 2003). AIS Electronic Library. ↩