Preface to Supplement 6, 2014.

The study of complex archaeological systems with the support of the philosophy of Artificial Intelligence is a research project that evaluates the historical meaning of the relationships between archaeological documents, intended as an essentially human construction, reaffirming, in this way, the importance of Analytical Archaeology, and updating it on the basis of the progress made by Cognitive Science, Neuroscience and Cybernetics through the simulation of the principles regulating memory, orientation, classification and interpretation of reality. It is important to highlight that these models, unlike others, require a precise encoding of the documents and acquire an important role in the research only when the results they produce become the hyper-surface to continue, update, refine or open the analysis itself. In the time of techniques it is still too predictable that the last perceptible limit is still that of the relationship (metaphorical, nuanced or allusive) between 'mind and machine'. Besides, in this age, it is almost instinctive to replicate the function of knowledge, to retrieve its origin and to postulate a backstory for it. On the other hand, the models seeking a place in this discipline, by drawing their inspiration both from other dissimilar disciplines and from the theories that try to explain the cognitive function, would be absorbed by the recreation, even though minimal or impossible, of intelligences, first the Cybernetic and then the Artificial Intelligence. The other model they would be inspired by is reason as a tool and this becomes, today, the condition for interpreting and communicating the historical, archaeological and anthropological complexity of the human being.

This paper describes the philosophy of Artificial Adaptive Systems and compares it with natural language, revealing some striking parallels. Artificial sciences create models of reality, but their ability to approximate the 'real world' determines their effectiveness and usefulness. This paper provides a clear understanding of the expectations created by the use of this technology, an evaluation of the complexities involved, and expresses the necessity of continuing with an open mind to unexpected and still unknown potentials. Supervised and unsupervised networks are described here.

This contribution represents a further attempt to synthesise and to introduce the research activities of the Analytical Archaeology and Artificial Adaptive Systems Laboratory (LAA and AAS) recently instituted at La Sapienza University of Rome thanks to the award of the project ARCHEOSEMA and to the institutional collaboration of the Department of Antiquities and the Department of Intercultural and European Studies and Physic Department. The main didactic and empirical activities of the Laboratory are related to the applicative simulations of Artificial Adaptive Systems to the analysis of complex natural and cultural phenomena through the lens of Analytical Archaeology. These complex phenomena are essentially understood to be the product of cognitive behaviour, in other words models and ideal-types which represent it and can be analysed on a formal logical level. This introductory exploration leads to a significant syntactic diversification of logical inferences and a progressive human attempt to trace them back to the simulation of cognitive complexity. Artificial Adaptive Systems, as Natural Computation mathematical tools which express these emulative properties, are historiographically involved in the connectionist reaction to behaviourism and therefore they effectively form the social sciences’ attempts to ascribe the complexities developed by our brains to advanced, non-linear and dynamic computational models. The LAA and AAS results will be examined in a historical perspective, but it is of great importance to consider the epistemological implications of this new approach since it is moved by the idea that every kind of language can be studied after being transferred into a non-linear sequence of variables.

Understanding the world around us is usually a hard task. All dynamically evolving phenomena in the natural world are produced by a strong interaction among a great number of causes and, often, only a few amounts of them are visible or measurable. Moreover, the phenomena may be so widely distributed over space and time, like the weather evolution, that only a small number of measurements can be taken, making the understanding of the overall system difficult and approximated. Some characteristics of systems can produce a very strange behaviour, even when the elements constituting the system are a small number. All these elements and their mutual interaction can produce the so-called complexity. Artificial neural networks (ANNs) form an interesting class of dynamic systems, as a paradigm of natural and spontaneous computation. ANNs are founded on bases inspired by the neurophysiological nature of neurons and their mutual connectivity. In this paper the historical reasons that led to the former mathematical models of neuron and connectionist topologies will be detailed. Over time, they have evolved through the feed-forward systems, Self-Organizing Maps, the associative memories up to the latest models in artificial cerebral cortex.

The Artificial Adaptive Systems (AAS) have had several applications in different technical and scientific fields, in medical research, life sciences, and financial and insurance studies. These systems have had, so far, poor implementation in social sciences. Among the latter, the main examples can be found in research about urban models in which AAS are usually used together with GISystems. By their nature, neural networks are suitable for interpreting complex phenomena like the social ones. Their limited use is, therefore, surprising. It is just to explain a complex phenomenon that AAS have been used in the SECOA project. The project deals with the study of environmental conflicts in coastal areas. Environmental conflicts are, by nature, complex phenomena, multidimensional and multiscalar. In SECOA 26 conflicts in 17 regions were analysed. The AAS were used to generate an explanatory model that would allow to describe, through its essential elements, the relationship between conflicts and territories. AAS are not only an ordinary complement to the spatial analysis toolbox but a new paradigm for spatial analysis and mining. In particular Geo-SOM (Geo-Self-Organizing Map) is a tool to identify homogenous regions for which predictive analysis can be done using tools that make the visualisation of positive and negative correlation possible. Increased use of AAS and GIS, and the good results this method produced, contributed to a more precise identification of a GIScience in general and its research agenda in particular.

In this paper the theoretical and methodological aspects of some of the tools applied to the archaeological, geographical and linguistic problems posed by ARCHEOSEMA project will be analysed. In particular, the single steps of the process of generation of outputs, from the initial analysis of the dataset, the subsequent procedures of pre-processing and encoding of the data to the characteristics of the processing algorithms will be described. For this purpose we will use a so-called toy dataset known in the literature. Using the same dataset, we will illustrate the main output produced, Minimum Spanning Tree maps. Along with the use of classical literature measurements, such as the Pearson linear correlation and Prior Probability, both used as metrics for the generation of these outputs, we have tried to show the innovative contribution of a new artificial neural network, the Auto-Contractive Map, designed by P.M. Buscema at the Semeion Research Center.

For several years now archaeology has made use of methodologies based on Artificial Intelligence (AI) and Artificial Adaptive Systems (AAS). However, there are still only a few experiments that involve the spatial aspect, and in particular spatial analyses of the territory. Moreover, we are often faced with theoretical approaches, procedures that cannot be used or repeated by the scientific community because they are based on proprietary or undivulged algorithms. The first part of the paper is focused on a short historical retrospective of the applicative experiences of AI and GIS, from the New Archaeology pioneers to the latest experiments in predictive approaches. Subsequently, we present an open source application, both from the software as well as the procedural point of view, oriented to the creation of predictive maps and focused in particular on the study of ancient settlements.

Methods borrowed from Information Theory are applied to the traditional text criticism. A critique of the raw cladistic methods and an interpretation of the dichotomy-phenomenon are offered. The same methods are applied to 13th century Italian poetry to determine authorship attributions and to verify commonly accepted literary taxonomy. Philology is a human science primarily applied to literary texts and traditionally divided into lower and higher criticism. Lower criticism tries to reconstruct the author’s original text and higher criticism is the study of the authorship, style, and provenance of texts. The use of methods borrowed from information theory makes it possible to bring together methodologically some of the sectors of the two fields. The outcome of the experiments in both text criticism and text attribution has been encouraging. In the former, the tests performed on three different traditions have provided results very similar to those obtained by traditional methods requiring a great amount of time. The experiments carried out both on 13th century Italian poets and schools have shown that it is possible to draw texts closer to one another. Furthermore, the method we have used makes it possible to attribute anonymous writings.

Fernando Pessoa represents an extreme case in the context of contemporary author’s philology. The breadth of his legacy, the large number of unpublished works at his death, the disorganisation and incompleteness of his materials and the entropy caused by the early processes of inventory produced an archive, now largely in the possession of the Portuguese National Library, partially refractory to the application of traditional text-criticism methods. This paper will demonstrate, through some application examples, that a careful study of material aspects concerning the originals of the Pessoa archive, made through the use of Artificial Adaptive Systems, will shed new light on the complex and multi-layered writing system created by Pessoa and identify new genetic relationships among his works, useful for the construction of an overall mapping of his literary output.

A sample of administrative texts from the Early Syrian state archives of Ebla were coded and processed through the model known as Auto-Contractive Map (Auto-CM). The results of this study led us to focus on some basic issues related to the structure of the Eblaite administrative records which deal with transactions of textiles. This first step is oriented toward the development of a methodology which would allow us to outline some concrete proposals for reconstructing the content of badly preserved tablets.

The paper illustrates how the application of a specific version of Artificial Neural Networks, Self-Organizing Maps (SOMs), enabled a more accurate analysis of human dental morphology. SOMs enable the processing of individual samples (dentitions) because they can cope with missing data. In fact, in archaeological samples of human remains, teeth are often broken or missing making a complete set of morphological traits often impossible to achieve. Other classification methods like Principal Component Analysis, Multidimensional Scaling, Mean Measure of Divergence, Multiple Correspondence Analysis do not handle missing descriptors and incomplete data matrices have to be filled in, thus leading to a certain approximation in the outcome with a lack of geographical or temporal resolution, as many incomplete samples have to be merged into a virtual one that does not present missing descriptors. Our discussion about the proficiency of SOMs, and ANNs in general, in the exploration and classification of anthropological databases concerning morphology is based on a specific case study, that is the classification of a Neanderthal sample. Through this example we would like to attract the attention of anthropologists and archaeologists to a very flexible methodology that is seldom applied, despite being widely used in many other disciplines.

The analysis of a corpus of seals belonging to the period of Isin and Larsa, carried out through the use of the Artificial Neural Network Auto-Contractive Maps, allows us to understand the complexity of the relationship of the different elements of the visual domain and its variety. The point of view adopted here is that of reading the iconology and iconography of the so-called presentation scene by offering an interpretation that goes beyond the concept of standardised and homogeneous production without any special innovative connotations.

This study aims at exploring both the identity of the maenads and their multiple interactions with the satyrs on Athenian red-figure vases by presenting the preliminary results of an ANN-based analysis applied to a dataset of 114 vases representing 478 figures (maenads and satyrs). The encouraging results seem to confirm the highly significant role of ANN-based methodologies as innovative tools for the organisation, visualisation and analysis of complex data in History of Art and Archaeology. Further explorations of these methodologies, associated with higher levels of data formalisation, should open new perspectives for the research on Athenian iconography and iconology.