Maritime Networks in the Ancient Mediterranean World 1108429947, 9781108429948

Citation preview

PREFACE

This book owes its initial impetus to a year of collaboration between the coeditors at the University of Toronto in 2012–2013, which allowed for fruitful discussions of maritime networks spanning three millennia between the Bronze Age and late antiquity. It was this interest in comparing and contextualizing network behavior across different Mediterranean worlds that gave rise to the November 2013 Toronto workshop on which this volume is based, Networks of Maritime Connectivity in the Ancient Mediterranean: Structure, Continuity and Change over the Longue Durée. In the midst of that venture came Cyprian Broodbank’s provocative proposal that the group author a collective “manifesto” capturing some of the key ideas and insights of our focused discussions (Leidwanger et al., “A manifesto for the study of Mediterranean maritime networks,” Antiquity+ 342 (2014), at http://journal.antiquity.ac.uk/projgall/lei dwanger342d), a stimulating process that prompted further reflection and informed the present contributions. We thank the Social Sciences and Humanities Research Council of Canada for the Postdoctoral Fellowship and the Connections Program grant that made this collaboration and workshop possible. Support for the event was also provided by the Department of the History of Art, the Aegean Material Culture Laboratory, and the Archaeology Centre at the University of Toronto. The Royal Ontario Museum kindly hosted a public lecture in association with the workshop. Assistance in the running of the event came from graduate students Paula Gheorghiade, Rachel Kulick, and Elana Steingart. We wish to thank Beatrice Rehl, the editorial team, and the reviewers for Cambridge University Press, who have helped shepherd the volume along. Most importantly, though, we wish to acknowledge the participants who made the workshop such a stimulating success, some of whom were not able to contribute to the present volume.

xiii Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:36:21, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.001

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:36:21, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.001

CHAPTER ONE

MARITIME NETWORKS, CONNECTIVITY, AND MOBILITY IN THE ANCIENT MEDITERRANEAN Justin Leidwanger and Carl Knappett

CREATING CONNECTIONS

In an exponentially hyper-connected modern world it is tempting to imagine that the past was a different place, one of sedentary villages in which most people barely ventured beyond familiar confines. Indeed, for Mediterranean prehistory, it is farms and hamlets that dominate the settlement record (Whitelaw 2017, 118). One might, then, easily assume that in such societies most interactions were with family and neighbors, and of a frequency and regularity that made for an almost intuitive communication. In the study of antiquity this perspective is perhaps best encapsulated in Finley’s assertion that ancient societies must have operated primarily on a face-to-face basis (Finley 1973). With this notion of the face-to-face, it is all too easy to portray society as static (Moatti 2006; see also Osborne 2011, 217). Mobility becomes an optional add-on, something that might well have happened, but certainly not an inherent societal condition (Clifford 1997). A strong response to this sedentarist bias emerged in the form of a so-called “mobility turn” that put movement center stage (Clifford 1997; Moatti 2004; Cresswell 2011). What are the implications of a perspective privileging mobility for the study of antiquity? That there was considerable movement in the ancient Mediterranean is hardly in doubt; it is quite clear from written sources and artifact distributions (de Ligt and Tacoma 2016). Furthermore, the sense of it being a precondition for Mediterranean life emerges once one takes into 1 Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:30:55, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.002

2

JUSTIN LEIDWANGER AND CARL KNAPPETT

account the region’s heterogeneous environment and unpredictable ecology; it would have been highly beneficial in many cases not to limit oneself to any one ecozone (Woolf 2016; Broodbank 2013). This is a just a general overlay, and over time there would have been diverse motives behind mobility, motives we should no doubt continue to explore. But the more basic question of how mobility was made possible has certainly received insufficient attention. This is a question not only of transportation technology—more on this below—but also of the fundamental conditions enabling communication beyond the faceto-face. For example, Moatti (2006) has identified “translation” as a key process in movement between cultures, and not just in a textual sense. She also describes translation in relation to art, and specifically the “translations” of Greek art in the Roman world. When it comes to the movement of people, the Roman world had various means for assigning identity to migrants, though it was far from categorical; ancient sources tell us that migrants may have carried recommendation letters, or been asked to narrate biographical details, while particular insignias or objects may also have helped establish identity (Moatti 2006; see also Moatti 2004). For prehistory, we may not have access to the documents that helped establish the identity of a migrant, which in turn could open the door to trust and communication. But the insignias and objects of identity ought not to be completely lost to us. And we need not limit ourselves to such obviously symbolic artifacts. Perhaps it was not only artifacts associated with personal identity that helped establish the conditions for interaction beyond the face-toface. Might not more prosaic and less personalized artifacts also have provided some of the means for regulating and establishing the basis for interaction? We generally think of artifacts like transport amphoras as impersonal commodities, and as such mere symptoms of movement: “material diasporas, the scattering of objects left behind by human vectors,” according to Woolf (2016, 442). But might we not also argue for the infrastructural support provided by things and technologies, themselves variably mobile (Knapp and van Dommelen 2010; Knappett and Kiriatzi 2016)? This tension between seeing artifacts as simply a reflection of human movement and exchange relations, on the one hand, and as actively constitutive of social relations, on the other, is played out in the history of archaeological approaches to exchange. The processual archaeology of the 1970s saw attempts to systematize the relationship between the distributions of circulating materials and their underlying social mechanisms (e.g., Renfrew 1975; Sabloff and Lamberg-Karlovsky 1975; Oka and Kusimba 2008), though since then archaeologists have become less certain that any kind of predictable link exists. The idea slowly began to emerge that rather than just being the material outcome of social processes, circulating artifacts may themselves contribute to the formation of social ties across regions, and in turn to the creation of social place.

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:30:55, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.002

MARITIME NETWORKS, CONNECTIVITY, AND MOBILITY

3

Ancient historians and archaeologists have introduced considerable theoretical sophistication into studies of space and place in the ancient world, from the domestic space of the household to sacred realms of sanctuaries and vast landscapes of power; in these approaches, the active role of artifacts is pivotal (e.g., Smith 2003; Khatchadourian 2016). Such work is mostly concerned with terrestrial landscapes, however; approaches to analyzing the human geography of maritime space and place remain comparatively underdeveloped (though see Knapp and van Dommelen 2010). In general, the sea represents either a flat and featureless plane free to be crossed or a deterministic mix of environmental constraints (winds, currents, visibility, etc.) that essentially predefine a few major vectors of movement and communication. WE ARE SAILING (OR CANOEING)

Whether we consider maritime connectivity as uniquely enabling or constraining, as offering unparalleled benefit or prohibitive cost, we surely must recognize its uniqueness in circumventing proximity, in collapsing space—and to some extent time—in contexts like the Mediterranean. If the face-to-face basis of interaction is undermined by connectivity, then perhaps maritime mobility offers a particularly dramatic challenge to that principle. Travel overland largely involves a graded movement, such that one culture gives way to another gradually; or an abrupt transition will be marked by some kind of frontier. At sea, such frontiers—to the extent that they actually existed in concept or practice (Rougé 1966, 41–44; Lytle 2012)—cannot be marked, and the unpredictability of maritime movement might throw one upon unexpected shores. These circumstances create more acute challenges for establishing interaction and communication. While maritime research has often focused on the obvious physical constraints enforced by water transport, the sea also influences the development of social bonds centered firmly on maritime rather than terrestrial space. Engagement with seafaring should force us to grapple at once with both the physical and the social factors of mobility. When we think of the unique benefits of maritime movement that fundamentally distinguish it from terrestrial movement, we might focus on its capacity for fast and reliable longer-distance voyages (noted above) and hence easier and often direct access to exotics of low bulk but high value. Alternatively, its greater transportation capacity for bulky commodities—especially mineral resources, building supplies, and agricultural staples—might serve as the primary driver behind its development. If both motivations are relevant in different conditions, and forms of seaborne exchange are carried out variously over short, medium, and long distances, then what infrastructural and technological considerations come into play for these different kinds of maritime movement? How might persistent patterns of maritime interaction play a role in structuring other

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:30:55, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.002

4

JUSTIN LEIDWANGER AND CARL KNAPPETT

political, social, and economic relations? Does the higher opportunity cost—in technical skill and resource investment—of boats and ships render maritime transport more or less relevant for different individuals, commodities, and mechanisms of exchange? Is there a fundamental distinction between formal connections and “routes” prescribed by those administering exchange and less official geographies derived simply from repeated opportunistic movements and shared experiences among seafarers? And might such seaborne routes persist, exhibiting a form of institutional memory, on the basis of embedded social structures and knowledge, ongoing needs or desires for resources and goods, or simply the continuity of environmental parameters and coastal topography? These and many other broad questions quickly emerge when investigating how maritime interaction shapes past societies, and any such modeling undeniably requires consideration of both environmental and social variables. When interrogating the interrelated environmental factors and social structures behind Mediterranean maritime activity, a distinction must be made— and here we borrow from Woolf (2016), who draws in turn on Horden and Purcell (2000)—between connectivity as potential or precondition, and mobility as the instantiation and realization of that potential. To reframe some of our questions above, we might ask the degree to which mobility was shaped by the distinctly connective landscape of the maritime Mediterranean world, and thus persistent over time regardless of political and social change. Or were mobilities completely reconfigured in light of changing social conditions? The need to consider connectivity and mobility, environment and society, together is thus obvious, yet there often remains a polarization in approaches to ancient maritime interaction. Archaeologists studying this sphere can become too narrowly focused on particular parameters like winds and currents, harbor and ship technologies. On the other hand, more social approaches to human mobility across the sea can at times be essentially unfettered by such key constraints. It is no accident that this conceptual separation runs parallel to a specialist division between, to put it bluntly, historians and prehistorians respectively. With comparatively few clues offering direct insights into maritime technologies like ships and harbors (Wachsmann 1998), the prehistorian has a limited set of parameters for understanding connectivity. Considerable emphasis is therefore placed on broad environmental conditions and constraints (Morton 2001), which have allowed prehistorians to be among the most active in constructing models focused on connectivity. Yet this lacuna allows, or perhaps forces, more freedom and flexibility for prehistorians in discussions of place. On the other hand, an abundance of technical information on the ships, harbors, and even specialized sails and transport jars can leave a Romanist feeling less compelled to engage with social worlds that inhabited these spaces even though the period was ripe with vivid testimony of individual voyages and patterns of mobility.

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:30:55, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.002

MARITIME NETWORKS, CONNECTIVITY, AND MOBILITY

5

Such a picture of separation is, of course, a caricature, particularly for the Mediterranean that forms the focus of this volume.1 In this region a number of pivotal studies have had a global impact on the scholarly treatment of maritime space. Fernand Braudel’s La Méditerranée et le monde méditerranéen à l’époque de Philippe II (1949) was transformative in promoting the sea as an integral factor in structuring the awareness and experience of past Mediterranean populations, followed by Peregrine Horden and Nicholas Purcell’s The Corrupting Sea (2000), and recently joined by Cyprian Broodbank’s The Making of the Middle Sea (2013). Along with other important contributions (e.g., Sherratt and Sherratt 1993; Harris 2005; Abulafia 2011; Tartaron 2013), these serve to create a more nuanced perspective on how ancient communities viewed, experienced, and exploited maritime space for different social, economic, and political goals. In the eastern Mediterranean, for example, scholars have recognized that some large islands like Crete and Cyprus function effectively as “miniature continents” (Brun 1996; Rackham and Moody 1996; Cadogan et al. 2012), while some continental landmasses like the Peloponnese are almost archipelagic. We should not minimize the interwoven environmental and socioeconomic constraints—including a remarkably heterogeneous resource landscape and fragmented “micro-ecologies” on the one hand, and diverse communities with varied consumer needs, interests, and institutions on the other—that framed maritime connectivity and promoted seaborne mobility for communication and exchange. Both island-studded and with an “inside-out geography” (Horden and Purcell 2000)—water surrounded by land rather than vice versa—the Mediterranean nudged its coastal populations toward the sea as an obvious topography of interaction and recourse for livelihoods. Yet patterns of mobility in either direction from the coast, both across the sea and inland, contributed vitally to the development of community identities in shared social space. That the sea was not only a source of immense potential but one laden with uncertainty and even great risk is evident in coastal raiding and the resulting fear of seaborne visitors, the fortunes and lives lost to a sea capable of sudden transformation into a tempest; the tragic refugee plight reminds us of the ongoing precariousness of maritime mobility even into the modern Mediterranean. Despite important advances in how to approach such a heterogeneous space, scholarship still struggles to create the sorts of meaningful dialogue between specialists of different periods that are so essential. We certainly do not wish to project the notion of a “Great Divide” discussed nearly four decades ago (Renfrew 1980), but significant gaps do remain between different scholarly traditions. Some of these differences, as noted above, are more the product of the contrasting forms of evidence at our disposal than of any profound epistemological split. For the broadly historic context of the Greco-Roman period onward, scholars are able to consult a number of sources that offer quite direct

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:30:55, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.002

6

JUSTIN LEIDWANGER AND CARL KNAPPETT

testimony to patterns and structures of maritime activity. Texts encompassing a wide range of topics and written for a host of purposes—periploi and geographies, poetic narratives, historical sources, epigraphic inventories—can offer indirect, and occasionally direct, evidence (e.g., Arnaud 2005). Of course, direct archaeological evidence for actual sea routes, ephemeral by their nature, in the densely interconnected Mediterranean is hard to find, posing particular challenges for the study of human interaction (e.g., Rice 2016). To the extent that prescribed routes ever existed, they are largely invisible, possibly arising and persisting over generations only from tradition, marine knowledge, technology, opportunities, and hazards (Pomey 1997; Andreau and Virlouvet 2002; Arnaud 2014). To some extent there may be fewer significant “sunk costs” for sea travel than in overland transport and communication, where infrastructure aggregates over time as roads and passes remain in use well beyond an initial investment (Laurence 1999). Communication by sea can be undertaken from widely varying maritime installations, some as simple as unadorned and only seasonally protected sandy beaches, yet certain forms of interaction may only be practical with built all-season harbors that (if well attended) can represent longlived installations and landscape features (Marriner and Morhange 2007; Oleson and Hohlfelder 2011). Shipwrecks provide another critical and increasingly abundant source of data in the form of artifact movements, particularly when cargos of transport amphoras or other durable goods are sufficiently diagnostic to ascertain origins (e.g., Parker 1992; 2008). Even so, identifying the origin of a cargo object is hardly the same as understanding the origin of a cargo, and the ultimate destinations of such wrecks can only rarely be discerned with any precision or confidence. Ceramics, anchors, and other objects that gradually accumulate at architecturally invisible anchorages, opportunistic ports, or other points of casual maritime activity among individuals or groups may likewise provide underutilized evidence for evaluating patterns of maritime connectivity and the inscription of distinctive social words on coastal and sea space (Ilves 2009; 2011; Leidwanger 2013; 2018). The adoption of maritime cultural landscape studies, originally a feature of Scandinavian archaeology but now more widely incorporated in both historic and prehistoric contexts in the Mediterranean, has helped to remedy some of this dichotomy (Ford 2011). The landscape approach to long-term patterns of maritime and coastal activity has broadened the traditional focus of such studies, in effect embracing the full material and nonmaterial record for connectivity and mobility: rock carvings, mooring devices, portages, canals, shipyards, ship types, landing sites, beliefs, ritual, mythology, folklore, symbolism, and the like (Westerdahl 1992; 2010; 2011). Yet this relative abundance of Mediterranean material and historical evidence has perhaps encouraged descriptive approaches of connectivity tied to physical geography that have sometimes inhibited the development of more

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:30:55, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.002

MARITIME NETWORKS, CONNECTIVITY, AND MOBILITY

7

social explorations of maritime mobility and the formation of place. After all, when it comes to prehistory, what can we say directly and securely about Neolithic or Bronze Age seafaring in the Mediterranean? From the Bronze Age Aegean, for example, we have only a tiny number of shipwrecks, with at best a couple that may be slightly earlier than the Late Bronze Age and certainly none that can comfortably inform pre-Bronze Age models. To this we might add a handful of contexts where boats are depicted; mostly stylized renderings serving purposes generally unknown but certainly other than our own, these need not be especially accurate or representative (Basch 1987; Wachsmann 1998; Strasser 2010). The contrast is therefore quite pronounced with the Roman period, where far more direct evidence is available: hundreds of shipwrecks surveyed or excavated (Parker 1992; McCormick et al. 2013; Strauss 2013), scores of larger and smaller harbors, and numerous literary and iconographic portrayals of ships from a variety of contexts (Rougé 1966; Casson 1995). Arnaud’s (2005) comprehensive study of the seaborne routes documented in the surviving sources for the Roman Mediterranean is a particularly strong case study in what can be done to promote a holistic view of the structure of maritime activity for one crucial period from just the historical, literary, and epigraphic record. This is not to say that prehistorians have not problematized maritime interaction in their period; it is simply a reflection that the theme has, by necessity, been tackled primarily indirectly, by focusing more on connectivity than on mobility. One powerful example is provided by Broodbank’s (2000) landmark work on the Early Bronze Age (EBA) Cyclades. Armed with a very small number of boat depictions and a generally limited range and resolution of archaeological data (i.e., site size and location, artifact imports), he used basic network analysis to model some likely interaction patterns between islands and to explore how the location and centrality of certain sites may have arisen from these dynamics. This is a rather uncommon example not only of recognizing in principle and acknowledging the paramount importance of specifically maritime connectivity, but also of structuring a formal inquiry around such marine dynamics despite the limited evidence available. For other prehistoric periods where we have significant maritime mobility, like the Middle and Late Bronze Ages, similar investigations have been thin on the ground (see early work on Delos and centrality in Davis 1982, and recent analysis of Mycenaean interaction in Tartaron 2013). This observation suggests that the fundamental dilemma in addressing questions of maritime connectivity and mobility across the Mediterranean is primarily one of method. The brilliant but rare syntheses of huge datasets into convincing narratives can inspire the field but can also leave the wider swath of scholarship in their wake. It is one thing to appreciate the success of wellconstructed and well-analyzed case studies of maritime mobility, but quite another to derive detailed methods and implement them more broadly against a

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:30:55, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.002

8

JUSTIN LEIDWANGER AND CARL KNAPPETT

long-term backdrop of connectivity across periods and increasingly complex bodies of material and other evidence. How can scholars effectively pursue similar fundamental maritime questions for other periods, regions, and datasets? Broodbank’s seminal Aegean study was inspired by similar use of proximalpoint analysis in another archipelagic setting, Oceania (e.g., Terrell 1977; Hage and Harary 1991; 1996). The success of such network methods inspired one of us (CK) to employ a similar approach to later periods of the Bronze Age in the Aegean, stimulating a long-term collaboration far beyond the traditional disciplinary boundaries to engage with particle physicists (e.g., Knappett, Evans, and Rivers 2008; 2011; Evans, Knappett, and Rivers 2009; Rivers, Knappett, and Evans 2013; Rivers, Evans, and Knappett 2016). Notwithstanding certain earlier applications in archaeology more broadly (for reviews, see Knappett 2011; Brughmans 2013), Broodbank’s was a pioneering example of the successful application of formal network analysis to a maritime problem, particularly for the Mediterranean. Broad network metaphors had long been employed in discussions of early trade and interaction (“trade routes,” “hubs,” etc.), but rapid advances from the social and physical sciences regarding formal network analysis over the past ten to fifteen years (e.g., Newman, Barabási, and Watts 2006) have offered new opportunities for engaging systematically with the breadth and dynamism of structures of socioeconomic interaction within complex societies. The practical impact of network thinking is evident in the boom that began in the mid- to late 2000s and continues apace. Classical archaeology may at times appear behind the vanguard of innovative methodological approaches compared to other branches of the discipline, but this field too has been quite active in the uptake of network approaches drawing from complexity science as well as social network analysis (SNA) (e.g., Graham 2006; Isaksen 2008; Larson 2013). One of the central aims of the present volume is therefore to advance this trend in the maritime realm, to promote network thinking broadly across the distinctive problems and potential of maritime themes within the Mediterranean. Several important reasons lead us to believe that networks can provide a strong methodological common ground where both prehistorians and historians can productively tread. First, networks allow for a conceptual starting point in either physical or social space. For example, one can begin with a spatial distribution of artifacts (as one tends to find archaeologically) or a set of attested social interactions (as might be described in texts). Though network analysis has not always prioritized combining these two facets, they can indeed help us bridge the persistent gap noted above between the physical and the social. Second, networks encompass a wide range of approaches from which one might choose. This in turn creates its own new challenges that must be addressed explicitly, but it also underscores how the generally flexible network framework can accommodate both data-poor and data-rich scenarios as described above. For example,

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:30:55, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.002

MARITIME NETWORKS, CONNECTIVITY, AND MOBILITY

9

if a prehistorian aims to model some most likely interactions across a certain space, it is entirely feasible to do so with only minimal inclusion of data beyond basic details of site location. Similarly, if a Roman archaeologist has masses of data on quantities, types, and co-occurrences of amphoras, then this too can be addressed through the same basic network approach, albeit with certain modifications to account for numbers and variability in the dataset. This flexibility within a single overall method is likely to be a significant factor in the long-term success or otherwise of network approaches in archaeology and history; these are evidently already seeing significant use and adaptation across a broad range of global contexts, from the American Southwest to the North Sea,2 a phenomenon underscored by the final commentary by Mills, which aims to contextualize further the central contributions to the present volume. As a methodological lynchpin, networks can accommodate a broad range of epistemological positions, from the humanistic network metaphors of Constantakopoulou (2007) or Malkin (2011) to the more formal scientific approaches of Evans or Rivers (this volume; see also Knappett 2016). To rephrase these strengths in terms of our prior discussion, network approaches allow us to bridge connectivity (as network potential) and specific patterns of human mobility. A DEEP HISTORY OF MEDITERRANEAN MARITIME INTERACTION?

By adopting a specifically network approach to the archaeological and historical evidence for seaborne communication and exchange in the Mediterranean world, this volume examines the predominant model of maritime connectivity with analytical tools that can shed light on continuity and discontinuity of mobilities across periods and areas. What long-term and interregional trajectories can we identify in the networks that guided movement, communication, and exchange? The Mediterranean offers an unparalleled diachronic case study for maritime network structures across millennia from before the Neolithic up to the early modern era; here our focus is squarely on the pivotal period extending from the Bronze Age into the early medieval world, though many of the themes and perspectives have much broader temporal and spatial relevance. The region has attracted much large-scale multi-period research, focused, however, predominantly on environmental angles (e.g., Vita-Finzi 1969; van der Leeuw 1998; Leveau et al. 1999; Grove and Rackham 2001; Butzer 2011). What is sorely needed now is the fuller integration of different social variables as active agents, which in turn bring their own challenges as scholars attempt to bridge multiple disciplines, principally archaeology, Classics, and history. Drawing together a range of experience among researchers in these allied fields, the contributions collected here advance network approaches to maritime connectivity and mobility in the ancient world. In particular, we aim to

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:30:55, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.002

10

JUSTIN LEIDWANGER AND CARL KNAPPETT

promote applications of diverse network thinking as well as methodologies that investigate the motives, behaviors, and experiences of seaborne movement and exchange by proposing and testing specific models of the Mediterranean archaeological and historical record (see also Leidwanger et al. 2014). The rapid growth in the size and availability of complex datasets in recent years— including databases of primary maritime material evidence for ancient shipwrecks and ports3—challenges us to employ new management and analysis tools that will allow us to capitalize on these earlier investments in data collection. Network methodologies offer the opportunity to maximize the utility of the multifaceted and often uneven archaeological and historical evidence in a systematic and measurable way (e.g., Preiser-Kapeller 2015). Moreover, this volume aims, where possible, to bridge what are traditionally viewed as transitional junctures between periods, regions, cultures, and disciplines: for example, between the end of the Bronze Age and the Iron Age, from the late classical era to the international Hellenistic world and the rise of the Roman Empire, and across the dissolution of the Roman state in the early medieval west and its resurgence in the late antique east. Under what conditions do maritime networks manifest a form of “memory,” continuing to inform movement through the physical and social landscape despite significant political and cultural change? As needs of exchange and interaction shift over time, to what extent should we expect to see resilience and continuity in the patterns of maritime mobility? When do networks, by contrast, change on their own or to fit new sociopolitical realities? Do significant changes in maritime technology, such as the innovation and widespread adoption of the sail between the late 4th and the 3rd millennium, correspond to new networks? When basic seafaring technologies remain essentially unchanged—as seems to be the case from the Hellenistic into the Roman world—should we expect resilience and robustness in the nodes and links of maritime networks, even in the face of shifting supply and demand as well as evolving political and other institutions? A necessary concern therefore centers on the notion of institutional “memory” in Mediterranean maritime networks. Are there entanglements and locked-in trajectories of such a kind that, once connections are firmly rooted, it takes an extraordinary event or rupture to destabilize or even de-establish them? To understand late Roman networks, then, would one need to understand how their sea routes were inherited from earlier Roman and Hellenistic traditions? Would evaluating this in turn require projection back to the classical and Archaic periods, and perhaps even beyond? Early Iron Age exchange networks themselves may not have represented entire reinventions, but rather piggybacked on certain residues of pre-existing maritime structures from the end of the Bronze Age. One could obviously continue back eventually to the Neolithic and perhaps even earlier. This is no doubt an extreme example, but

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:30:55, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.002

MARITIME NETWORKS, CONNECTIVITY, AND MOBILITY

11

given the persistence of many of the essential environmental and ecological parameters of connectivity, might significant continuities have crossed some of these critical transitions? Or were networks constantly reinvented time and again according to changing socially constructed variables from technologies to products and markets? This question brings to the fore a critical second issue related to how we balance our models of interaction between “routes” on the one hand and “Brownian motion” on the other. The former we can imagine as prescribed and perhaps institutionalized, but hardly the latter. It is worth comparing this situation and looking for possible parallels with that between regular and random networks, a dichotomy that was only mathematically overcome two decades ago with a landmark paper defining “small-world” networks (Watts and Strogatz 1998). Should we not, then, be able to conceive of some synthesis or intermediary model between these two forms of interaction? In a Mediterranean world with such diverse mechanisms of exchange as attested for the Roman era, can we model some elements through more formal routes while imagining other seaborne activity more closely paralleling the random background noise of Brownian motion? A modern historical example might suffice to demonstrate the intriguing dilemma, for which we draw on research by John Leonard (2005, 716–740). With the arrival of British rule on Cyprus came new administration of ports and an attempt, particularly in the early decades of the 20th century, to centralize maritime distribution around the several improved harbors at larger coastal centers. The reaction of the local sailors—including those involved in the long-standing carob trade—was swift and indignant. Many continued to use their traditional makeshift ports to transport goods not only among the island’s smaller and larger harbors, but also between the island and the neighboring mainland. Many elements of this traditional maritime network persisted, much to the dismay of the British authorities. Documented in the official records, their efforts continually emphasize the suppression of such activity, which now fell formally under the rubric of smuggling, and the tightening of their grip on customs duties and flow of goods to wider Mediterranean markets. If one network model cannot reasonably fit all situations—something that should hardly come as a surprise given the discussion above about diversity within the material evidence— should the particular pattern depend on the nature of the objects or individuals moving (elite or everyday) or perhaps the geographical distances being traversed? The notion of multiple maritime operational scales, whether complementary or contrasting, presents a third area of necessary concern. Should we envision a series of structurally similar but nested networks, from the local to the global? Is it critical, or even beneficial, to work simultaneously across all scales in an effort to understand a complex and geographically expansive

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:30:55, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.002

12

JUSTIN LEIDWANGER AND CARL KNAPPETT

phenomenon like Greek colonization or Roman commerce? If one works only at a macro scale, does the sea just become an undifferentiated expanse that fails to reveal the remarkable variation and nuance evident in the ways humans experienced and exploited maritime space? In our efforts to address the complexity of “social” maritime space, we cannot lose sight of the complexity of its physical space, its undeniable topographical and environmental variability as well as its heterogeneity in the spatial allotment of resources. We must recall the differential geographical behavior cited above, wherein large islands have the capacity to function conceptually or practically like “miniature continents” while peninsulas or other continental landmasses assume an insular identity. Rather than rigidly framing maritime connectivity, environmental realities can themselves become secondary, subsumed under socioeconomic considerations. The island-studded and distinctly “inside-out” geographies noted previously point to the possible advantages afforded by such opportunistic maritime landscapes and underscore the need to address each at its own regional scale, yet any coherent synthesis of Mediterranean maritime networks must at once look up and down and across these scales. Alongside these issues of analytical scale we should raise a fourth obvious question: where should an individual shipwreck cargo fit? From a spatialnetwork perspective of maritime exchange, these sites take on an absolutely fundamental importance as direct evidence for mobility, for connectivity in action, but they also present pronounced methodological difficulties (e.g., Greene this volume; Leidwanger 2017). Approaching shipwrecks simply as “mobile nodes” would seem to present a certain dilemma in that they hardly represent the logical equivalent of settlements or other traditional analytical units within network studies. A shipwreck ostensibly strikes right at the heart of our inquiry: people and goods in motion, the distribution stage of economic movement, the purposeful assembly into one journey of different items that often contain traces of earlier journeys and other network dynamics. Yet the geographical position of a wrecked vessel and cargo may hold disappointingly little meaning from a basic spatial perspective. After all, the amphora pile that most often marks such sites did not complete the journey; these jars were not meant to reside on the seabed. Just comparing like with like, a ship carrying Cypriot pottery that wrecks off the coast of Spain while nearing its destination will look instinctually quite different by virtue of its location than it will if it sinks off the coast of Cyprus on the first day of its journey. Such examples should give us pause, and this is only the most straightforward of scenarios. In the case of complex cargos, how might we distinguish effectively between those picked up through cabotage and those assembled at one or more large warehouse ports? In the latter case, how should we model the multiple interlocked networks so tantalizingly implicit in this copresence of materials within a single ship’s hold? How can we synthesize the cargo, the galley wares or

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:30:55, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.002

MARITIME NETWORKS, CONNECTIVITY, AND MOBILITY

13

personal possessions of a crew, and the material remains of the vessel itself—let alone such accidental travelers as the stowaway mouse from the Late Bronze Age Uluburun shipwreck (Cucchi 2008)—into a holistic network story that speaks at once to the individual journey as well as to the broader mechanisms and socioeconomic world behind such an assemblage? Distilling networks from bulk artifact distributions raises a glaring fifth issue concerning how to integrate the spatial patterns—or, more properly, the quantified records of consumption at different sites—for the selective objects we happen to be capable of tracing archaeologically. That is, we can offer network representations based on individual classes of artifacts or products (most often through the surviving containers for wine, oil, etc.), however broadly we define these classes. Brughmans and Poblome (2015; 2016) have endeavored to test network methods through the diachronic distribution of particularly widespread classes of Roman finewares of known origin across the Mediterranean. Leaving aside the obvious caveat that such patterns will be strongest for only the most diagnostic artifacts, certainly a network built from the distribution of another object (e.g., transport jars) of the very same region could appear quite different, reflecting a range of alternative mechanisms at play (Autret et al. 2014). Given the many interconnected relationships underpinning maritime interaction, we should aim to embrace not just many complicated overlapping network diagrams, but also a holistic reflection and integrated approximation of real complexity. Our final aim is perhaps straightforward and implicit from the previous discussion, but it should by no means be left underemphasized: we hope to push forward the study of Mediterranean maritime interaction through explicit and accessible network approaches, which may be more or less mathematical, but are all “models” in a basic sense and hence intellectually helpful if inevitably reductive. The range of reactions to network analysis underscores this point quite nicely, as does the constantly expanding repertoire of ways that scholars have usefully appropriated the term “network” and network concepts. Understanding how researchers focusing on different periods and traditions —for example, prehistoric archaeologists and ancient historians—value and employ (or eschew) some of these network ideas can offer new insights into how such tools might be bent to the questions and research concerns of multiple increasingly interdependent fields. To this end, the contribution by Tim Evans (Chapter 2) provides a succinct overview and comparative evaluation of formal network methods for modeling cultural interaction and exchange. These range from the traditional proximal point analysis to stochastic models (including his own collaborative creation, ariadne) that offer a more natural approximation of real-world social systems. The next chapter (Chapter 3), by Ray Rivers, takes up these formal approaches to test the diachronic behavior of networks surrounding the

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:30:55, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.002

14

JUSTIN LEIDWANGER AND CARL KNAPPETT

important eruption of Thera and the transition between the Middle and Late Bronze Age. He shows how the ariadne model effectively captures the functioning of the maritime interaction network before and after the removal of an evidently crucial node at Akrotiri on Thera. Working from a qualitative rather than quantitative approach, Thomas F. Tartaron (Chapter 4) pulls the discussion back into the practical lives of mariners in this Bronze Age Aegean setting, offering thoughtful critique of SNA and other formal network models commonly applied to date in studies of this early period. Augmenting the multi-scalar model from his previous work (Tartaron 2013) with ethnoarchaeological observation, he emphasizes the generally overlooked role of “local” as crucial to the formation and maintenance of connectivity and short-distance links that formed a powerful and persistent core of much seaborne activity. Barbara Kowalzig (Chapter 5) offers a richly textured case study of the overlapping and mutually reinforcing economic and religious networks that extended across the Greek world. Organizing many facets of life amid Horden and Purcell’s (2000) Mediterranean maritime environment of opportunity and unpredictability was a regional “cultic cabotage” that, she finds, follows many of the behavioral properties of social networks. These strong links structured both economic integration and shared religious practice, and served as the foundation for preferential attachment of longer-distance “weak links” that could create “small-world” phenomena. The maritime material record of this same multi-scalar interaction forms the basis of Elizabeth S. Greene’s contribution (Chapter 6), which attempts to situate shipwreck assemblages conceptually and formally within broad network thinking. She adopts SNA visualization to posit a framework for Archaic Mediterranean maritime interaction based on cargo composition, while also exploring formally the internal connections within a full assemblage—not only cargo, but shipboard materials and even the ship itself—that attest to the individuals and broader cultural circuits in which objects and ideas were traveling. The amphoras that serve as our most robust testimony to Greek maritime commercial networks are the core data informing the simulation by Mark L. Lawall and Shawn Graham (Chapter 7), which explores how different network structures might create varied patterns in the introduction and adoption or failure of shapes of transport jars. The spatial results of their iterated economic simulations can be productively compared with the archaeological record, although contrasting network configurations can create indistinguishably similar patterns; a prime example of equifinality at work, this analytical reality suggests that while similar structures may be hypothesized for historical situations, proof of one particular structure or another is often elusive. The appropriate systematic use of network models in making archaeological hypotheses forms the theme to Tom Brughmans’ contribution (Chapter 8) exploring, through statistical approaches and visualization, similarities and

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:30:55, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.002

MARITIME NETWORKS, CONNECTIVITY, AND MOBILITY

15

differences in the distribution of eastern Roman tablewares. He emphasizes the crucial role of specific dependence assumptions in creating models that can be formally tested alongside other analytical methods and deployed when evaluating potential mechanisms behind artifact distribution patterns. Ceramic distributions form a core dataset also for Paul Arthur, Marco Leo Imperiale, and Giuseppe Muci (Chapter 9), who construct a Byzantine network based on shared material culture spanning the 8th-century Mediterranean and beyond. The particular properties of this connective structure are used to understand the maritime context—evidently a shrinking economy and perhaps top-down administrative directives—that allowed the multipurpose and now standardized globular amphora form to purge, at least temporarily, the range of competing jars. In her commentary on the preceding chapters, Barbara J. Mills (Chapter 10) weaves together themes common among the contributions, addressing some of the guiding questions behind the volume as well as underdeveloped but productive paths for future work. Drawing on contemporary archaeological perspectives, including from her own network-based studies in the American Southwest, she situates maritime connectivity and mobility within the scholarly dialog on network theory and analysis of cultural interaction. For coast-hugging ancient populations, much of whose communication and travel was necessarily seaborne, the measurement of maritime networks takes on paramount importance for understanding cultural interaction, but at the same time these networks demand new and focused interdisciplinary approaches to the sorts of complex datasets that characterize connectivity and mobility. This volume brings together scholars of Mediterranean archaeology, ancient history, and complexity science to integrate theoretical approaches and analytical tools into models of maritime interaction within its social and spatial context. By bringing both theoretical approaches and analytical methods from network science to patterns of maritime communication, resource procurement, and exchange, we seek to understand the evolving structure and nature of socioeconomic connectivity that guided Mediterranean interaction manifested in the material and historical records. The contributions gathered here build connections not only among subfields of ancient studies spanning four millennia of human activity, but also across the traditional boundaries of humanities, social sciences, and physical sciences. In gathering such studies together for this volume, we hope to make the most of what the study of the Mediterranean can offer: unrivaled case studies for long-term perspectives on maritime network structures across varied geography, institutions, and millennia. Networks provide a flexible analytical framework geared specifically toward addressing large and complex spatial and relational information, and as such hold considerable potential not only as general conceptual tools but also as formal methods for modeling

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:30:55, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.002

16

JUSTIN LEIDWANGER AND CARL KNAPPETT

archaeological and historical data across the Mediterranean. Yet systematic applications of network methodologies to explore the overarching themes of maritime connectivity remain surprisingly few, in part no doubt because the challenges of such an approach outlined above are acute: balancing complexity and reduction, working across scales, and requiring research to bridge multiple disciplines and particular evidence sets, principally archaeology, Classics, and history. Despite such cautions, network perspectives provide a promising way forward. This general outlook has recently attracted attention in studies of ancient exchange and cultural interaction in diverse contexts that span the globe, offering the advantage of including here an additional perspective from outside the period and region. The recent emergence of a community of archaeologists and historians engaging with explicit network approaches to the ancient world provides occasion to build on earlier studies of Mediterranean maritime connectivity, and to integrate new formal concepts and tools from the social and physical sciences within this general framework of comparative historical inquiry. NOTES 1.

2.

3.

There are other arenas with interesting work that might form a focus of such a study, like the Caribbean, the Indian Ocean, or the Baltic. We have tried to leaven our Mediterranean maritime bias through inclusion of commentary from a specialist working in the completely landlocked environment of the American Southwest. E.g., Knappett 2013; Mills et al. 2013; see also the Connected Past initiative at http://con nectedpast.soton.ac.uk, with a follow-up publication (Brughmans, Collar, and Coward 2016), as well as a guest-edited volume of the Journal of Archaeological Method and Theory (Collar et al. 2015). For shipwrecks, see the important catalogs provided and expanded in Parker 1992; 2008; Strauss 2013; http://oxrep.classics.ox.ac.uk/databases/; http://darmc.harvard.edu/icb/icb .do. For growing comparative studies of ports, see www.ancientportsantiques.com; http:// awmc.unc.edu; www.portusproject.org; portuslimen.edu.

REFERENCES

Abulafia, D. 2011. The Great Sea: A Human History of the Mediterranean. Oxford: Oxford University Press. Andreau, J., and Virlouvet, C. 2002. L’information et la mer dans le monde antique. Rome: École Française de Rome. Arnaud P. 2005. Les routes de la navigation antique: Itinéraires en Méditerranée. Paris: Editions Errance. Arnaud, P. 2014. Ancient mariners between experience and common sense geography. In K. Geus and M. Thiering (eds), Features of Common Sense Geography: Implicit Knowledge Structures in Ancient Geographical Texts, 39–68. Zurich: LIT Verlag. Autret, C., Dillon, M., Rauh, N.K., and Zoroğ lu, L. 2014. The trading networks of ancient rough Cilicia. In S. Ladstätter, F. Pison and T. Schmidts (eds), Hä fen und Hafenstä dte im ö stlichen Mittelmeerraum von der Antike bis in byzantinische Zeit: Neue Entdeckungen und

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:30:55, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.002

MARITIME NETWORKS, CONNECTIVITY, AND MOBILITY

17

aktuelle Forschungsansä tze (Harbors and Harbor Cities in the Eastern Mediterranean from Antiquity to the Byzantine Period: Recent Discoveries and Current Approaches), 593–618. Byzas 19. Istanbul: Ege Yayınları. Basch, L. 1987. Le musée imaginaire de la marine antique. Athens: Hellenic Institute for the Preservation of Nautical Tradition. Braudel, F. 1949. La Méditerranée et le monde méditerranéen à l’époque de Philippe II. Paris: Colin. Broodbank, C. 2000. An Island Archaeology of the Early Cyclades. Cambridge: Cambridge University Press. Broodbank, C. 2013. The Making of the Middle Sea: A History of the Mediterranean from the Beginning to the Emergence of the Classical World. London: Thames and Hudson. Brughmans, T. 2013. Thinking through networks: a review of formal network methods in archaeology. Journal of Archaeological Method and Theory 20, 623–662. Brughmans, T., Collar, A., and Coward, F. (eds) 2016. The Connected Past: Challenges to Network Studies in Archaeology and History. Oxford: Oxford University Press. Brughmans, T., and Poblome, J. 2015. Pots in space: an exploratory and geographical network analysis of Roman pottery distribution. In E. Barker, S. Bouzarovski, C. Pelling, and L. Isaksen (eds), New Worlds from Old Texts: Revisiting Ancient Space and Place, 255–80. Oxford: Oxford University Press. Brughmans, T., and Poblome, J. 2016. Roman bazaar or market economy? Explaining tableware distributions through computational modelling. Antiquity 90, 393–408. Brun, P. 1996. Les archipels égéens dans l’antiquité grecque. Besançon: Les Belles Lettres. Butzer, K.W. 2011. Geoarchaeology, climate change, sustainability: a Mediterranean perspective. Geological Society of America Special Papers 476, 1–14. Cadogan, G., Iacovou, M., Kopaka, K., and Whitley, J. (eds) 2012. Parallel Lives: Ancient Island Societies in Crete and Cyprus. Proceedings of the Conference Held at Nicosia 1–3 December 2006. London: British School at Athens. Casson, L. 1995. Ships and Seamanship in the Ancient World. 2nd edn. Baltimore: Johns Hopkins University Press. Clifford, J. 1997. Routes: Travel and Translation in the Late Twentieth Century. Cambridge, MA: Harvard University Press. Collar, A., Coward, F., Brughmans, T., and Mills, B., 2015. Networks in archaeology: phenomena, abstraction, representation. Journal of Archaeological Method and Theory 22, 1–32. Constantakopoulou, C. 2007. The Dance of the Islands: Insularity, Networks, the Athenian Empire, and the Aegean World. Oxford: Oxford University Press. Cresswell, T. 2011. Mobilities I: catching up. Progress in Human Geography 35(4), 550–558. Cucchi, T. 2008. Uluburun shipwreck stowaway house mouse: molar shape analysis and indirect clues about the vessel’s last journey. Journal of Archaeological Science 35, 2953–2959. Davis, J.L. 1982. Thoughts on prehistoric and Archaic Delos. Temple University Aegean Symposium 7, 23–33. De Ligt, L., and Tacoma, L.E. 2016. Approaching migration in the early Roman Empire. In L. de Ligt and L.E. Tacoma (eds), Migration and Mobility in the Early Roman Empire, 1–22. Leiden: Brill. Evans, T., Knappett, C., and Rivers, R. 2009. Using statistical physics to understand relational space: a case study from Mediterranean prehistory. In D. Lane, D. Pumain, S. van der Leeuw and G. West (eds), Complexity Perspectives on Innovation and Social Change, 451–480. Berlin: Springer. Finley, M.I. 1973. The Ancient Economy. Berkeley: University of California Press.

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:30:55, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.002

18

JUSTIN LEIDWANGER AND CARL KNAPPETT

Ford, B. (ed.) 2011. The Archaeology of Maritime Landscapes. New York: Springer. Graham, S. 2006. Networks, agent-based models and the Antonine itineraries: implications for Roman archaeology. Journal of Mediterranean Archaeology 19(1), 45–64. Grove, A.T., and Rackham, O. 2001. The Nature of Mediterranean Europe: An Ecological History. London and New Haven: Yale University Press. Hage, P., and Harary, F. 1991. Exchange in Oceania: A Graph Theoretic Analysis. Oxford: Clarendon Press. Hage, P., and Harary, F. 1996. Island Networks: Communication, Kinship and Classification Structures in Oceania. Cambridge: Cambridge University Press. Harris, W.V. 2005. The Mediterranean and ancient history. In W.V. Harris (ed.), Rethinking the Mediterranean, 1–42. Oxford: Oxford University Press. Horden, P., and Purcell, N. 2000. The Corrupting Sea: A Study of Mediterranean History. Oxford and Malden: Blackwell. Ilves, K. 2009. Discovering harbours? Reflection on the state and development of landing site studies in the Baltic Sea region. Journal of Maritime Archaeology 4, 149–63. Ilves, K. 2011. Is there an archaeological potential for a sociology of landing sites? Journal of Archaeology and Ancient History 2, 1–31. Isaksen, L. 2008. The application of network analysis to ancient transport geography: a case study of Roman Baetica. Digital Medievalist 4, at https://journal.digitalmedievalist.org/ articles/10.16995/dm.20. Khatchadourian, L. 2016. Imperial Matter: Ancient Persia and the Archaeology of Empires. Oakland: University of California Press. Knapp, A.B., and van Dommelen, P. 2010. Material connections: mobility, materiality and Mediterranean identities. In P. van Dommelen and A.B. Knapp (eds), Material Connections in the Ancient Mediterranean: Mobility, Materiality and Identity, 1–18. London: Routledge. Knappett, C. 2011. An Archaeology of Interaction: Network Perspectives on Material Culture and Society. Oxford: Oxford University Press. Knappett, C. (ed.) 2013. Network Analysis in Archaeology: New Approaches to Regional Interaction. Oxford: Oxford University Press. Knappett, C. 2016. Networks in archaeology: between scientific method and humanistic metaphor. In T. Brughmans, A. Collar, and F. Coward (eds), The Connected Past: Challenges to Network Studies in Archaeology and History, 21–33. Oxford: Oxford University Press. Knappett, C., Evans, T., and Rivers, R. 2008. Modelling maritime interaction in the Aegean Bronze Age, Antiquity 82, 1009–1024. Knappett, C., Evans, T., and Rivers, R. 2011. The Theran eruption and Minoan palatial collapse: new interpretations gained from modelling the maritime network. Antiquity 85, 1008–1023. Knappett, C., and Kiriatzi, E. 2016. Introduction. In E. Kiriatzi and C. Knappett (eds), Human Mobility and Technological Transfer in the Prehistoric Mediterranean, 1–17. British School at Athens Studies in Greek Antiquity 1. Cambridge: Cambridge University Press. Larson, K.A. 2013. A network approach to Hellenistic sculptural production. Journal of Mediterranean Archaeology 26(2), 235–260. Laurence, R. 1999. The Roads of Roman Italy: Mobility and Cultural Change. London: Routledge.

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:30:55, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.002

MARITIME NETWORKS, CONNECTIVITY, AND MOBILITY

19

Leidwanger, J. 2013. Modeling distance with time in ancient Mediterranean seafaring: a GIS application for the interpretation of maritime connectivity. Journal of Archaeological Science 40, 3302–3308. Leidwanger, J. 2017. From time capsules to networks: new light on Roman shipwrecks in the maritime economy. American Journal of Archaeology 121(4), 595–619. Leidwanger, J. 2018. The power of coastal resources: assessing maritime economic opportunity in the Roman Mediterranean. In E. Holt (ed.), Water and Power in Past Societies, 217-240. Albany: State University of New York Press. Leidwanger, J., Knappett, C., Arnaud, P. Arthur, P., Blake, E., Broodbank, C., Brughmans, T., Evans, T., Graham, S., Greene, E.S., Kowalzig, B., Mills, B., Rivers, R., Tartaron, T., and van de Noort, R. 2014. A manifesto for the study of Mediterranean maritime networks. Antiquity+ 342. http://journal.antiquity.ac.uk/projgall/leidwanger342. Leonard, J.R. 2005. Roman Cyprus: harbors, hinterlands, and hidden powers. Unpublished PhD dissertation, State University of New York at Buffalo. Leveau, P., Trément, F., Walsh, K., and G. Barker (eds) 1999. Environmental Reconstruction in Mediterranean Landscape Archaeology. Oxford: Oxbow Books. Lytle, E. 2012. ‘H θάλασσα ϰoινή: fishermen, the sea, and the limits of ancient Greek regulatory reach. Classical Antiquity 31(1), 1–55. McCormick, M., Huang, G., Gibson, K. et al. 2013. Summary geodatabase of shipwrecks AD 1–1500, status 2008. The Digital Atlas of Roman and Medieval Civilizations (DARMC) Scholarly Data Series, Data Contribution Series #2013-1. Cambridge: Center for Geographic Analysis, Harvard University. https://darmc.harvard.edu. Malkin, I. 2011. A Small Greek World: Networks in the Ancient Mediterranean. Oxford: Oxford University Press. Marriner, N., and Morhange, C. 2007. Geoscience of ancient Mediterranean harbours. Earth-Science Reviews 80, 137–94. Mills, B.J., Clark, J.J., Peeples, M.A., Haas, W.R., Jr., Roberts, J.M., Jr., Hill, J.B., Huntley, D.L., Borck, L., Breiger, R.L., Clauset, A., and Shackley, M.S. 2013. Transformation of social networks in the late pre-Hispanic Southwest. Proceedings of the National Academy of Sciences 110(15), 5785–5790. Moatti, C. (ed.) 2004. La mobilité des personnes en Méditerranée, de l’antiquité à l’époque moderne: Procédures de contrôle et documents d’identification. Rome: Ecole française de Rome. Moatti, C. 2006. Translation, migration, and communication in the Roman Empire: three aspects of movement in history. Classical Antiquity 25(1), 109–40. Morton, J. 2001. The Role of the Physical Environment in Ancient Greek Seafaring. Leiden: Brill. Newman, M.E.J., Barabási, A.-L., and Watts, D.J. (eds) 2006. The Structure and Dynamics of Networks. Princeton: Princeton University Press. Oka, R., and Kusimba, C.H. 2008. The archaeology of trading systems, Part 1: towards a new trade synthesis. Journal of Archaeological Research 16, 339–395. Oleson, J.P., and Hohlfelder, R.L. 2011. Ancient harbors in the Mediterranean. In A. Catsambis, B. Ford, and D.L. Hamilton (eds), The Oxford Handbook of Maritime Archaeology, 809–833. Oxford: Oxford University Press. Osborne, R. 2011. The History Written on the Classical Greek Body. Cambridge: Cambridge University Press. Parker, A.J. 1992. Ancient Shipwrecks of the Mediterranean and the Roman Provinces. BAR International Series 580. Oxford: Tempus Reparatum.

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:30:55, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.002

20

JUSTIN LEIDWANGER AND CARL KNAPPETT

Parker, A.J. 2008. Artifact distributions and wreck locations: the archaeology of Roman commerce. In R.L. Hohlfelder (ed.), The Maritime World of Ancient Rome, 177–196. Ann Arbor: University of Michigan Press. Pomey, P. (ed.) 1997. La navigation dans l’antiquité. Aix-en-Provence: Edisud. Preiser-Kapeller, J. 2015. Harbours and maritime mobility: networks and entanglements. In J. Preiser-Kapeller and F. Daim (eds), Harbors and Maritime Networks as Complex Adaptive Systems, 119–139. RGZM Monograph 23. Mainz: Rö misch-Germanischen Zentralmuseum, 2015. Rackham, O., and Moody, J. 1996. The Making of the Cretan Landscape. Manchester: Manchester University Press. Renfrew, C. 1975. Trade as action at a distance: questions of integration and communication. In J.A. Sabloff and C.C. Lamberg-Karlovsky (eds), Ancient Civilization and Trade, 3– 59. Albequerque: University of New Mexico Press. Renfrew, C. 1980. The great tradition versus the great divide: archaeology as anthropology? American Journal of Archaeology 84(3), 287–298. Rice, C. 2016. Shipwreck cargoes in the Western Mediterranean and the organization of Roman maritime trade. Journal of Roman Archaeology 29, 165–192. Rivers, R., Knappett, C., and Evans, T. 2013. Network models and archaeological spaces. In A. Bevan and M. Lake (eds), Computational Approaches to Archaeological Spaces, 99–126. London: UCL Press. Rivers, R., Evans, T., and Knappett, C. 2016. From oar to sail. In C. Ducruet (ed.), Maritime Networks: Spatial Structures and Time Dynamics, 63–76. Routledge Studies in Transport Analysis. London: Routledge. Rougé, J. 1966. Recherches sur l’organisation du commerce maritime en Mé diterrané e sous l’Empire romain. Paris: S.E.V.P.E.N. Sabloff, J., and Lamberg-Karlovsky, C.C. (eds) 1975. Ancient Civilization and Trade. Albuquerque: University of New Mexico Press. Sherratt, A., and Sherratt, S. 1993. The growth of the Mediterranean economy in the early first millennium BC. World Archaeology 24, 361–78. Smith, A.T. 2003. The Political Landscape: Constellations of Authority in Early Complex Polities. Berkeley: University of California Press. Strasser, T.F. 2010. Location and perspective in the Theran Flotilla Fresco. Journal of Mediterranean Archaeology 23(1), 3–26. Strauss, J. 2013. Shipwrecks Database, Version 1.0. Oxford Roman Economy Project. oxrep. classics.ox.ac.uk/databases/shipwrecks_database. Tartaron, T.F. 2013. Maritime Networks in the Mycenaean World. Cambridge: Cambridge University Press. Terrell, J. 1977. Human Biogeography in the Solomon Islands. Chicago: Field Museum of Natural History. van der Leeuw, S.E. 1998. The Archaeomedes Project: Understanding the Natural and Anthropogenic Causes of Land Degradation and Desertification in the Mediterranean Basin. Luxembourg: Office for Official Publications of the European Communities. Vita-Finzi, C. 1969. The Mediterranean Valleys: Geological Changes in Historical Times. Cambridge: Cambridge University Press. Wachsmann, S. 1998. Seagoing Ships and Seamanship in the Bronze Age Levant. College Station: Texas A & M University Press. Watts, D. J. and Strogatz, S.H. 1998. Collective dynamics of “small-world” networks. Nature 393, 440–442.

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:30:55, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.002

MARITIME NETWORKS, CONNECTIVITY, AND MOBILITY

21

Westerdahl C. 1992. The maritime cultural landscape. International Journal of Nautical Archaeology 21(1), 5–14. Westerdahl, C. 2010. “Horses are strong at sea”: the liminal aspect of the maritime cultural landscape. In A. Anderson, J.H. Barrett, and K.V. Boyle (eds), The Global Origins and Development of Seafaring, 275–287. Cambridge: McDonald Institute for Archaeological Research. Westerdahl C. 2011. The maritime cultural landscape. In A. Catsambis, B. Ford, and D.L. Hamilton (eds), The Oxford Handbook of Maritime Archaeology, 733–762. Oxford: Oxford University Press. Whitelaw, T. 2017. The development and character of urban communities in prehistoric Crete in their regional context. In Q. Letesson and C. Knappett (eds), Minoan Architecture and Urbanism: New Perspectives on an Ancient Built Environment, 114–180. Oxford: Oxford University Press. Woolf, G. 2016. Movers and stayers. In L. de Ligt and L.E. Tacoma (eds), Migration and Mobility in the Early Roman Empire, 438–461. Leiden: Brill.

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:30:55, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.002

CHAPTER TWO

ROBUST SPATIAL NETWORK ANALYSIS* Tim Evans

INTRODUCTION: GENERAL APPROACH TO MODELING IN ARCHAEOLOGY

Modeling Scales There are many roles for modeling in social sciences. A key decision to be made is the modeling scale: micro, meso, or macro. Working at a microscopic level, be it in terms of geography, social, or political levels, means that we are trying to capture the interactions of entities at the smallest level we can imagine. Agentbased models (ABMs) lend themselves to such modeling (e.g., Wilkinson et al. 2007; Graham 2006) and I would place geographic information systems (GIS) packages in this category as they often encode information on small spatial scales (e.g., Bevan 2010). Mesoscopic approaches work with aggregations of the smallest-scale entities (a “coarse graining”). Most network models and Monte Carlo approaches lie in this category. Finally, one can imagine a macroscopic scale where the behavior of whole systems is considered. Typically, a few variables, such as the populations and the resources they require, are tracked via nonlinear mathematical equations. While such a classification is useful to frame our thinking, today’s computing power and * I would like to thank Carl Knappett and Ray Rivers for their long collaboration and many discussions about these issues. I have also worked on different aspects at various stages with Giovanni Brandani, Edmund Hunt, and Eric Beales.

22 Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:37:28, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.003

ROBUST SPATIAL N ETWORK ANALYSIS

23

increasing number of large datasets mean that such divisions have become blurred. It is possible to solve mathematical equations for thousands of parameters using GIS information as a key input. One may run simplified ABMs on large networks. In this chapter, we will focus on network-based examples but many of the same issues and solutions are relevant to all of these approaches.

Sites and Vertices Traditional social network analysis uses known people as the vertices and has been employed in cases particularly for historical research (e.g., Padgett and Ansell 1993). However, archaeological information is often “site-centric,” with information from a single excavation concentrated in lengthy reports. The challenge is, then, to see how such information fits into a wider pattern of regional and global interactions. The network models considered here emphasize such mesoscopic interactions and are ideal vehicles to overcome the inherent site-centricity. At the same time, networks function with a discrete set of vertices (i.e., nodes or actors), so the site-centric aspect makes archaeological sites the natural candidates for the vertices of a network. Island archipelagos are a particularly good example. Even today, the sea can isolate groups, suggesting that we represent each small island community as a single network vertex. In the context of our work on the Bronze Age Aegean we find that coastal sites on Crete or mainland Greece are often separated by the mountainous geography and the difficulty of overland travel (Knappett, Evans, and Rivers 2008; 2011; Evans, Knappett, and Rivers 2009; 2012b; Rivers, Knappett, and Evans 2013).

Interactions, Distances, and Edges The edges (i.e., ties or bonds) of a network can be used to represent the interactions between sites in a variety of ways. In modern situations, we may have direct information such as mobile phone call volumes (e.g., Expert et al. 2011) or explicit lists of maritime movements (e.g., Fournier 2016). The most obvious example of this approach in a historical context would use explicit descriptions of links between sites found in texts to define edges. For example, the Antonine Itineraries give insights into Roman Baetica (Isaksen 2007), while the descriptions of the travels of Saint Ansgar contained in the Vita Ansgarii (Sindbæk 2007b) create a network of travel routes across the Viking world. More generally, the appearance of sites in the same text can be used to suggest interactions. However, in most cases, even for relatively modern times, it is rare to have sufficient detail on explicit interactions. One of the main goals of spatial modeling is therefore to create a model of possible interactions. In all cases we need some sort of proxy for the interaction strength, so the key input to such models will be some measure of the distance between sites. There are, however, many types of distance. Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:37:28, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.003

24

TI M EV AN S

Artifact counts provide one such distance measure. For every pair of sites, we compare the numbers of each artifact found at each site and use this to produce a measure of the similarity of these sites (e.g., Terrell 2010; Sindbæk 2007a). The sites most similar to each other are then considered to be the closest and so forth. In this context, we are placing our sites in some sort of artifact space and exploiting some sense of distance within this space. Mathematics has a rich set of tools for this job. However, our vertices carry a geographical location, so deriving the interactions between vertices from their geography is a natural approach and, not surprisingly, a common one in the literature (e.g., Terrell 1977; Irwin 1983; Hage and Harary 1991; Broodbank 2000; Collar 2007; Bevan 2010). The remaining discussion will be in this geographical context, but much of what follows has a natural generalization to the higher-dimensional artifact spaces mentioned above. Even if we restrict ourselves to interactions controlled by geographical distances, there are still many ways of quantifying what we mean by distance. There are many “physical” distances, including direct “as the crow flies” in kilometers, shortest route in kilometers, quickest time, or lowest cost. Once we have a measure of distance in some unit, we have the option to convert this into a ranked distance scale. Instead of working in terms of the absolute value of the distance, when considering the interactions between one source site and its neighbors (the target vertices) we consider the nearest neighbor to be at distance (rank) 1, the second-nearest neighbor to be at distance (rank) 2, and so forth. Such ranked distances are used in proximal point analysis (PPA), one of the earliest network models used in archaeological contexts (e.g., Terrell 1977; Irwin 1983; Hage and Harary 1991). In fact, PPA is the simplest realization of the much earlier intervening-opportunities model (Stouffer 1940), which is based on ranked distances. The idea that the connection to potential targets depends only on the order of proximity irrespective of physical distance might seem odd, but in some situations it is natural. For instance, you would usually prefer to visit the most accessible hospital in an emergency irrespective of the actual physical distance to the hospital. We should note that the distances used in spatial models need not have the usual properties we associate with normal distance measurements or the precise properties of generalized distance concepts found in mathematics. For instance, the distances in these models need not be symmetric; that is, the distance from one site to another need not be considered to be the same as the effective distance for a trip in the opposite direction. Prevailing winds and currents make the time of travel across the sea asymmetric. MODELING INTERACTIONS

Once we have decided how to represent the distances between our sites, then we still need a model to turn these distance measures into the interactions, the

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:37:28, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.003

ROBUST SPATIAL N ETWORK ANALYSIS

25

actual edges, of our network. The most appropriate model will reflect the nature of the problem and the questions being asked. For island archipelagos, where the vertices are the major population or resource centers, the edges are often taken to represent some sort of exchange between sites. This could be physical trade of goods or transmission of culture. The contacts can be direct or the links might represent island-hopping trips. In situations of these types, the interactions are controlled by physical limitations of ancient sea travel; technology which enables daytime travel only gives a range of about 100 kilometers. There are two main types of model. What we term “zones-of-control” models give geographical zones acting in some sense as a single collective entity. Sometimes the models specify a single site as the dominant controlling entity for each region, a “regional capital.” The simplest of such models would be the Thiessen polygons (Voronoi diagrams) that assign each site its own region of influence assuming equal site sizes (see Aurenhammer, Klein, and Lee 2013, for example). The Xtent model (Renfrew and Level 1979) is a generalization of Thiessen polygons that uses information about the different site sizes to assign regions to one site. The region centered on a large site may include several other smaller sites that are then in the sphere of influence of the dominant site of their region. I would also put the Rihll and Wilson (1987; 1991) gravity model (RWGM) for some parameter ranges into this category since it can produce networks that are dominated by star-like features. That is, sites on the periphery of the stars have only one interaction, namely with the site at the center of their star. The networks in these models tend to be relatively simple, dominated by a single level of hierarchy. Different levels of hierarchy can be described in such models by changing model parameters. The alternative type of model we refer to as “interaction models.” These use geography to produce a map of the likely interactions, perhaps with some measure of the strength of each interaction. Questions such as likely zones of control may be derived as part of secondary analysis of the interaction pattern. Models of interaction (rather than control) can be classified by the way they treat different aspects: distance, costs of travel, and input/output constraints. We confine our discussion to such models of interaction. Before turning to look at some of the key examples of interaction network models, let us outline our notation. NETWORK DESCRIPTION

We will work with the following notation (Figure 2.1). We will assume a fixed measure of the distance from site i to site j, denoted dij. Each site will have a fixed site size or capacity, Si. This can represent the resources—physical, social, or political—of both the site and its hinterland. In this sense, we assume

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:37:28, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.003

26

TI M EV AN S

i

Si

dij Sj

Fij

j

2.1. Illustration of the network parameters used in this paper.

we have already performed some sort of coarse graining, using single sites to represent the influence of small regions. Most of the microscopic details of the geography are therefore lost, but the hypothesis is that the site locations and distances between sites encode the most important geographical aspects, and that the remaining details are largely irrelevant at the scales represented by our choice of sites. The interaction or flow from site i to site j will be represented by Fij. This reflects the weight of the corresponding edge in the network, and the matrix of values F is simply the adjacency matrix of the network. In general edges are weighted and directed, but in some models the edges are simply given values of either one or zero (Fij = 0 or 1) and may carry no directional information (Fij = Fji). EXAMPLE: GRAVITY MODELS

Gravity models are one of the oldest spatial models, having been used, for instance, to model modern transport systems (see Ortúzar and Willumsen 1994 for a good overview). Such models can be seen as implementing a cost–benefit viewpoint; that is, all trips are deemed equally likely but are subject to constraints on the total “cost” of the whole network, which can be expressed in the language of a maximum-entropy principle (Wilson 1967). This constraint means that routes that are “cheaper” are used more often. This cost is rarely actual financial cost. Rather it is the geographical distance, however defined, that sets the cost. Typical examples are a linear cost model, costs cij = dij, or a logarithmic cost model cij = log(dij) in which long-distance trips are only slightly more expensive than short-distance trips. Gravity models are not interpreted in terms of this cost, which is rarely observable, but by the way it alters the relative likelihood of finding interactions as a function of the distance. It is the actual flows of their direct effects that are observed. In gravity models, we find that the flows are proportional to a deterrence function, Fij ∝ f(dij), which contains all of the information about the cost function. In the gravity models the deterrence function is of the form Fij ∝ f(dij) ∝ exp(–b cij). Therefore, in a linear cost model

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:37:28, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.003

ROBUST SPATIAL N ETWORK ANALYSIS

27

100

linear

80

Cost

60 log 40

20

0

0

20

40

60

80

100

Distance

2.2. Sketch showing the forms of the typical cost functions used in gravity models: the linear cost as a solid line to the logarithmic cost as a dotted line, which is higher (lower) at short (long) distances.

the frequency of making a trip will fall off exponentially with distance, producing very few long-distance trips, while the unlikely sounding logarithmic trip cost model produces a power-law deterrence function which allows more longdistance trips (Figure 2.2). The sizes of sites in this model merely serve to scale the resulting flows and we find a simple gravity model (SGM) F ij ¼ S i S j ðd ij Þγ if cij ¼ log d ij while F ij ¼ S i S j expðd ij Þγ if c ij ¼ d ij

ð1Þ

In most cases we have some information about constraints on the inputs and/or the outputs at each individual site. Perhaps we could use the size of a port to estimate total maritime flows at each site. Such constraints are implemented in the doubly constrained gravity model (DCGM), where the flow Fij from site i size Si to site j size Sj is Fij ¼ ai bj Si Sj f ðdij Þ where ðai Þ1 ¼

X j

bj Sj f ðdij Þ and ðbj Þ1 ¼

ð2Þ X i ai Si f

ðdij Þ

ð3Þ

for deterrence function f(dij). Although f is formally just the negative exponential of the cost cij associated with each link, it is more normal to specify the deterrence function directly, usually taking the exponential or power-law

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:37:28, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.003

28

TI M EV AN S

forms in equation (1). Formally, the DCGM is a minimization of a cost–benefit function, specifically ! ! X   X X X H ¼ κ Fij lnðF Þ  1  β C  F c þ α S  F ij

i

þ

X j

α0j Sj 

ij ij

X

!

ij

Fij

i

i

i

ij

j

ð4Þ

i

The first term, weighted by parameter κ, allocates the flows to each pair of sites equally if no other terms are present. The second term, proportional to β, fixes the total cost of all trips to be C. This term sets the deterrence function f(dij) equal to the negative exponential of the costs cij. The last two terms fix the input and output flows of each site i to be equal to Si. In general, the input and output values at each site can be different, but here we fix them as equal for simplicity. However, for almost all purposes it is easier to look at the solution to this minimization problem that is the form given in equation (2). EXAMPLE: PROXIMAL POINT ANALYSIS

In PPA, sites are treated equally (Figure 2.3). Each site is connected to the k nearest neighbors regardless of the absolute value of these distances, making it the simplest example of a model using ranked distances. Generally the direction is ignored. With only one discrete valued parameter, it is a remarkably simple model yet it has proved useful in several situations (e.g., Broodbank 2000). Idealized versions of PPA are known as random k-nearest-neighbor graphs in the mathematics literature (e.g., Balister et al. 2005). EXAMPLE: MAXIMUM-DISTANCE NETWORK

The maximum-distance network (MDN) is the network formed when we connect two sites if they are within distance D of each other (Figure 2.4). This is the same method used to connect randomly placed sites to create what are known as random geometric graphs in mathematics, widely used in relation to ad hoc wireless networks. The only difference between MDN and PPA is that the former uses physical distances and the latter uses ranked distances. EXAMPLE: STOCHASTIC MODEL, ARIADNE

In a stochastic model, there is intrinsic volatility so that for the same input parameters, for the same sites in the same positions, a different result will be obtained every time the model is run. This makes interpretation of the model

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:37:28, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.003

ROBUST SPATIAL N ETWORK ANALYSIS

29

2.3. PPA for k = 2. Note the middle sites have more than two connections as the site at the top is not one of their closest two neighbors, but the middle sites are the closest two sites for this site at the top.

D

Connected Core

2.4. A simple MDN example. The circles centered on each site indicate the area within distance D from that site. Connections are made to any other site falling within this circle, as indicated by the links. This visualization uses measurements for direct (“as-the-crow-flies”) distance.

harder. Typically, we make several runs and then look at averages of any quantity of interest. Gathering statistical information on a varying quantity is a well-established procedure so this aspect is straightforward. In any case such volatility is a better reflection of the intrinsic variation of real social systems. In this sense, it represents a step forward from the single definitive answer put forward by the models considered until now. Another characteristic of these models is that they can be very flexible. Since they are invariably implemented numerically through the specification of some cost function to be minimized, this can be made as complicated as the programming skills allow. Of course, it is tricky to achieve a balance between having very few parameters giving predictive power and clarity, and including

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:37:28, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.003

30

TI M EV AN S

2.5. A screenshot showing an ariadne network. The software used to produce and investigate such networks and the other models described here is available online from the author (see also Evans, Knappett, and Rivers 2012a, 2012c).

a large number of parameters to match some actual data. ABMs face a similar choice. Most stochastic models can be conceptualized as special examples of ABMs in which the update rules are derived from very precise mathematics behind the optimization of some cost or entropy, such as the detailed balance criteria of Monte Carlo methods often used to implement such stochastic models. The exponential random graph models (a.k.a. p* models) are wellknown examples of stochastic models used in the context of generic network modeling (see Robins et al. 2007 for a review). However, I will illustrate this class of model with what I will call ariadne (Knappett, Evans, and Rivers 2008; 2011; Evans, Knappett, and Rivers 2009; 2012b; Rivers, Knappett, and Evans 2013). In the ariadne model the input sites and the output connection strengths vary in size (Figure 2.5). The site size can be thought of as representing the

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:37:28, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.003

ROBUST SPATIAL N ETWORK ANALYSIS

31

population while the outgoing edge strengths represent the frequency of trips from a site. The network is chosen such that it gives a low value for some “cost” function but will not generally be the lowest possible value, reflecting our expectation that systems are rarely perfectly optimized. X X H ¼ κ 4Si vi ð1  v Þ  λ ðS v Þe V ðd =DÞðS v Þ i i i ij ij j j i

þj

X

X Si vi þ μ Si vi eij

i

i;j

ð5Þ

i;j

The first term proportional to κ makes the optimal site size equal to the fixed input values Si, but the output variable νi represents the fraction of this optimal value that the site actually achieves (and this can be greater than 1.0). The second term represents the benefits of an interaction between two sites i and j which falls off as a function of distance as described by a deterrence function V (see Rivers, this volume). In my work, I have used a form which falls off slowly with distance and then falls steeply around the scale D (another input parameter of the model), but then has a long tail of weak interaction over larger distances. This λ term controls the strength of interaction between two sites, with the output variables eij representing interaction strength from site i to site j as a fraction of total size of site i, namely (Siνi). The third term controls the total site size and the last term represents a distance-independent cost of maintaining any interaction. COMPARING NETWORKS

We have discussed just a few of the many spatial network models available. Fundamentally, they all take the same arrangement of sites as their input but each model gives networks of very different appearances. At the most basic level, sites may be given new values, edges may or may not have values assigned, and values may change in different directions between pairs of sites. However, the fundamental input for all of these models is the same, the site locations, so surely some of these networks are representing essentially the same information. The real question is, how can we tell? If there are differences, do models fall into a few broad classes, where all models in one class give essentially the same type of network? We have identified zone-of-control models and interaction models but perhaps there are more types. How can we classify spatial network models? These questions all require us to be able to make comparisons between different networks. If we can create broad categories of network model types, then we can pick a model from the most appropriate category: models of zones of control, models of mutual interaction, etc. Once we have identified several models of the same type, we can use them to

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:37:28, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.003

32

TI M EV AN S

table 2.1. Table of the basic features of several spatial network models

Model

Distance

Input–output constraints Site size

Deterence function

Network type

Maximum-distance network (MDN) Simple gravity model (SGM) Doubly constrained gravity model (DCGM) Rihll and Wilson gravity model (RWGM) Alonso

Physical

No

Equal

Threshold

Simple

Physical

No

Fixed

Any

Physical

Both

Fixed

Any

Weighted, directed, dense Weighted, directed, dense

Physical

Both

Variable

Any

Weighted, directed, dense

Physical

Both

Variable

Any

Ranked

Output

Equal

Threshold

Weighted, directed, dense Simple

Ranked

Output

Fixed

Power Law

Ranked

Output

Fixed

Any

Physical

Output

Variable

Any

Proximal point analysis (PPA) Radiation Interveningopportunities model ariadne

Weighted, directed, dense Weighted, directed, dense Weighted, directed, dense

examine whether any conclusions we draw are robust. The qualitative comparison in Table 2.1 demonstrates the differences among models, serving to reinforce the need for a quantitative approach. A RECIPE FOR COMPARING NETWORKS QUANTITATIVELY

The networks produced by different models often appear very different even if they are all derived from the same data and are meant to represent the same phenomena. However, visualizations of networks, while sometimes useful, can be misleading. Instead, we should focus our attention on each network reflecting how the system functions and we should try to compare the behavior of processes on the network. We are looking for networks which lead the system to function in similar ways regardless of how they look. My approach for each network is as follows. First, we measure several quantities associated with vertices. Quantities defined only for simple networks, such as those based on path lengths, are not particularly useful. Measures which probe the structure of the whole network are more likely to tell us something about the regional and global properties of our network. Some of the obvious choices are PageRank or current betweenness measures

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:37:28, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.003

ROBUST SPATIAL N ETWORK ANALYSIS

33

table 2.2. A simple example of the types of vector used to compare different networks. Each column is the vector of of values produced for one network. In this case only a single measurement is made, that of PageRank for the arrangement of sites (all assumed to be equal) on the left. The network shown is that for PPA with k = 2. For each model a single exemplary choice of parameters has been made. Rather than use the absolute values of PageRank, we have ranked these values for each network. For MDN three vertices (1, 2, 3) networks have the same large value, so these are given the average of the first, second, and third rank. The models are PPA, directed PPA (DPPA), MDN, doubly constrained gravity model (DCGM), RWGM, and Monte Carlo (MC, i.e. the ariadne model).

4 1

2

Vertex

PPA

DPPA

MDN

DCGM

RWGM

MC

1 2 3 4

1.5 1.5 3 4

1.5 1.5 3 4

2 2 2 4

2.5 2.5 2.5 2.5

1.5 1.5 3.5 3.5

1.5 1.5 4 3

3

table 2.3. Correlation matrix for the Kendall’s tau method using the PageRank values of Table 2. A value of 1.0 indicates that two network vectors are identical. For instance, every network is identical with itself, but we also see that PPA and DPPA rank the vertices by PageRank in exactly the same way. On the other hand, values close to zero indicate low similarity, so MDN and the MC (ariadne) examples are quite different.

PPA DPPA MDN RWGM MC

PPA

DPPA

MDN

RWGM

MC

1 1 0.82 0.94 0.78

1 1 0.82 0.94 0.78

0.82 0.82 1 0.58 0.27

0.94 0.94 0.58 1 0.94

0.78 0.78 0.27 0.94 1

since these are based on theoretical diffusion processes, or broadcast-type processes such as Eigenvalue centrality. This gives us a list (vector) of values, one value per vertex and per quantity. For each different network, we have a vector representing its functional properties. The next step is to use these vectors to produce a measure of the similarity of each pair of networks. Standard approaches would involve using Pearson correlation coefficients, Kendall’s tau, or Spearman’s rank correlation coefficient if we have outliers. Finally, we analyze the resulting correlation matrices using standard multivariate analysis methods. Principal-component analysis and hierarchical clustering methods are typical approaches that can produce helpful visualizations. Tables 2.2 and 2.3 offer a simple example of such vectors. This exercise creates a problem. How do we choose values for different networks when we want to make a comparison? Should we be comparing a k = 2 PPA or a k = 4 PPA to a D = 100 km MDN? Every model has a few parameters and can produce a vast range of networks. If we have the computing power we can produce comparisons for thousands of different values but we will be left with a vast amount of data to sift through.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:37:28, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.003

34

TI M EV AN S

(a) 1400

Distance measured

1200 1000

〈D〉

800 600 400 Δ

200 0

1

2

3

4

5

6

7

8

9

Number of outgoing edges

(b) 600

〈D〉

Distance measured

500 400 300 200

Δ

100 0 40

60

80

100

120

140

160

Distance Scale

2.6. On the left, a comparison of the number of edges per site in PPA against distance measures. Diamonds are used for D, the distance averaged over all edges in the network, while the squares give the average weighted distance Δ. On the right, a similar comparison for the MDN model. The line shows the MDN distance parameter D, which is also the horizontal axis. The average weight parameter clearly scales with D but is not equal, or even simply proportional, to D.

When making comparisons between models of simple graphs in normal network theory, one would normally compare models with the same numbers of edges and vertices. However, in the spatial context an edge always carries some information (its distance in whatever units we are using), so not all edges are equal. In many of our models the edge has a derived weight, the flow along that edge. We must look for models with the same “physical” characteristic. We could study the parameters of models that give a “just” fully connected

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:37:28, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.003

ROBUST SPATIAL N ETWORK ANALYSIS

35

(a)

Principal Component 2 (6%)

2 0.6

4

0.4 5

6

0.2

1

0.0

3 –0.4

–0.2

0.0

0.2

0.4

Principal Component 1 (93%) (b)

Distance

0.6

0.4

0.2

0.0

4

5

2

1

3

6

2.7. Example of the use of hierarchical clustering and principal-component analysis. The spatial data used are from aegean39S1L3a, while PageRank and weighted betweenness values are found for each site and then a Pearson correlation matrix is found. The networks are: (1) PPA k = 2, (2) DPPA k = 3, (3) DCGM D = 40 kilometers, (4) RWGM D = 60 α = 1.03, (5) RWGM D = 55 α = 1.18, (6) SGM D = 75. These networks all had an average weighted distance Δ of around 70 kilometers, chosen because this is where we tended to have a fully connected network. In this simple example, it is clear that RWGMs (4) and (5) are very different from the other network models here. The DPPA model (directed PPA, i.e., PPA where directions are retained on edges) (2) also seems to be reasonably distinct from other models. On the other hand, PPA, DCGM, and SGM seem to be giving similar results.

network (Evans 2016). But this is a special type of network close to a critical point in the language of physics, and we might be interested in other regions of parameter space. Given the spatial context, some measure of the typical distance scale in the network might be better. After all, in simple graphs, the average shortest path length is a classic measure of graph connectivity, but we

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:37:28, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.003

36

TI M EV AN S

need a generalization of this that reflects the real distances between sites as well as the strengths of interactions. A natural answer is to use the average weighted distance, which we shall call Δ. We take the average value of the strength of each edge, the flow Fij, multiplied by the distance of that edge dij. X Fij dij ð6Þ Δ¼ X Fij Figure 2.6 shows how this average weight parameter Δ varies along with model parameters and other distance measures. The way it rises gradually and relatively smoothly as we add more edges or strengthen those edges illustrates that this parameter works well. Alternatives, such as those shown in Figure 2.6, are not so useful for this purpose. Our proposal is therefore that we compare networks with the same average weighted distance. Once we have selected networks with the same average weighted distance, we can proceed as before. We generate a vector of values for the vertices and different measurements in each network. From this we create a similarity matrix. The final step is to use multivariate analysis methods such as hierarchical clustering and principal-component analysis. A small and simple example is shown in Figure 2.7. Even in this limited illustration we can see that the RWGMs appear to be quite distinct from other models. Perhaps more surprisingly, it appears that the simple PPA model is not so distinct from the gravity models used here. CONCLUSIONS

We have set out a recipe that allows us to compare different spatial network models. In principle, this could help us classify different models into broad families. Such a classification would allow practitioners to select one model from the family most suited to their needs. By comparing results with other models from the same family, they could check that any conclusions drawn are robust (Evans and Rivers 2017). To reach this point and any definite conclusions, we would need to apply this methodology on an extensive set of data. REFERENCES

Aurenhammer, F., Klein, R., and Lee D.-T. 2013. Voronoi Diagrams and Delaunay Triangulations. Hackensack, NJ: World Scientific. Balister, P., Bollobás, B., Sarkar, A., and Walters, M. 2005. Connectivity of random k-nearest-neighbour graphs. Advances in Applied Probability 37, 1–24. Bevan, A. 2010. Political geography and palatial Crete. Journal of Mediterranean Archaeology 23, 27–54.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:37:28, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.003

ROBUST SPATIAL N ETWORK ANALYSIS

37

Broodbank, C. 2000. An Island Archaeology of the Early Cyclades. Cambridge: Cambridge University Press. Collar, A. 2007. Network theory and religious innovation. Mediterranean Historical Review 22, 149–162. Evans, T.S. 2016. Which network model should I use? Towards a quantitative comparison of spatial network models in archaeology. In T. Brughmans, A. Collar, and F. Coward (eds), The Connected Past: Challenging Networks in Archaeology and History, 149–173. Oxford: Oxford University Press. Evans, T.S., Knappett, C., and Rivers, R. 2009. Using statistical physics to understand relational space: a case study from Mediterranean prehistory. In D. Lane, S. van der Leeuw, D. Pumain, and G. West (eds), Complexity Perspectives in Innovation and Social Change, 451–479. Berlin: Springer Methodos Series. Evans, T.S., Knappett, C., and Rivers, R., 2012a. Ariadne, Figshare http://doi.org/10 .6084/m9.figshare.97746. Evans, T.S., Knappett, C., and Rivers, R. 2012b. Interactions in space for archaeological models. Advances in Complex Systems 15, 1150009 http://doi.org/10.1142 /S021952591100327X. Evans, T.S., Knappett, C., and Rivers, R., 2012c. 39 Minoan sites, Figshare http://doi.org /10.6084/m9.figshare.97395. Evans, T.S., and Rivers, R. 2017. Was Thebes necessary? Contingency in spatial modelling. Frontiers in Digital Humanities 4(8) http://doi.org/10.3389/fdigh.2017.00008. Expert, P., Evans, T.S., Blondel, V.D., and Lambiotte, R. 2011. Uncovering space-independent communities in spatial networks. PNAS 108(19), 7663–7668. Fournier, M. 2016. Venetian maritime supremacy through time: a visualisation experiment. In C. Ducruet (ed.), Maritime Networks: Spatial Structures and Time Dynamics, 77–91. London: Routledge. Graham, S. 2006. Networks, agent-based modeling, and the Antonine Itineraries. Journal of Mediterranean Archaeology 19, 45–64. Hage, P., and Harary, F. 1991. Exchange in Oceania: A Graph Theoretic Analysis. Oxford: Clarendon Press. Irwin, G. 1983. Chieftainship, kula and trade in Massim prehistory. In J. Leach and E. Leach (eds), The Kula: New Perspectives on Massim Exchange, 29–72. Cambridge: Cambridge University Press. Isaksen, L. 2007. Network analysis of transport vectors in Roman Baetica. In J.T. Clark and E.M. Hagemeister (eds), Digital Discovery: Exploring New Frontiers in Human Heritage, CAA2006. Computer Applications and Quantitative Methods in Archaeology. Proceedings of the 34th Conference, Fargo, United States, April 2006, CD-ROM, 64–76. Budapest: Archaeolingua. Knappett, C., Evans, T.S., and Rivers, R. 2008. Modelling maritime interaction in the Aegean Bronze Age. Antiquity 82, 1009–1024. Knappett, C., Evans, T.S., and Rivers, R. 2011. The Theran eruption and Minoan palatial collapse: new interpretations gained from modelling the maritime network. Antiquity 85, 1008–1023. Ortúzar, J. d. D. and Willumsen, L. 1994. Modelling Transport. New York: Wiley. Padgett, J. F. and Ansell, C.K. 1993. Robust action and the rise of the Medici, 1400–1434. American Journal of Sociology 98, 1259–1319.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:37:28, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.003

38

TI M EV AN S

Renfrew, A.C., and Level, E.V. 1979. Exploring dominance: predicting polities from centres. In A.C. Renfrew and K. Cooke (eds) Transformations: Mathematical Approaches to Culture Change, 145–167. London: Academic Press. Rihll, T.E., and Wilson, A.G. 1987. Spatial interaction and structural models in historical analysis: some possibilities and an example. Histoire & mesure 2, 5–32. Rihll, T.E., and Wilson, A.G. 1991. Modelling settlement structures in ancient Greece: new approaches to the polis. In J. Rich and A. Wallace-Hadrill (eds), City and Country in the Ancient World, 59–95. London: Routledge. Rivers, R.J., Knappett, C., and Evans, T.S. 2013. Network models and archaeological spaces. In A. Bevan and M. Lake (eds), Computational Approaches to Archaeological Spaces, 99–126. London: UCL Press. Robins, G., Snijders, T., Wang, P., Handcock, M., and Pattison, P. 2007. Recent developments in exponential random graph (p*) models for social networks. Social Networks 29, 192–215. Simandiraki, A. 2005. Minoan archaeology in the Athens 2004 Olympic Games. European Journal of Archaeology 8, 157–181. Sindbæk, S.M. 2007a. Networks and nodal points: the emergence of towns in early Viking Age Scandinavia. Antiquity 81, 119–132. Sindbæk, S.M. 2007b. The small world of the Vikings: networks in early medieval communication and exchange. Norwegian Archaeological Review 40, 59–74. Stouffer, S.A. 1940. Intervening opportunities: a theory relating to mobility and distance. American Sociological Review 5, 845–867. Terrell, J. 1977. Human Biogeography in the Solomon Islands. Chicago: Field Museum of Natural History. Terrell, J. 2010. Language and material culture on the Sepik Coast of Papua New Guinea: using social network analysis to simulate, graph, identify, and analyze social and cultural boundaries between communities. Journal of Island and Coastal Archaeology 5(1), 3–32. Wilkinson, T. J., Gibson, M., Christiansen, J. H., Widell, M., Schloen, D., Kouchoukos, N., Woods, C., Sanders, J., Simunich, K.-L., Altaweel, M., Ur, J. A., Hritz, C., Lauinger, J., Paulette, T., and Tenney, 2007. Modeling settlement systems in a dynamic environment: case studies from Mesopotamia. In T. Kohler and S.E. van der Leeuw (eds), Model-Based Archaeology, 175–208. Santa Fe, NM: School for Advanced Research Press. Wilson, A.G. 1967. A statistical theory of spatial distribution models. Transport Research 1, 253–269

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:37:28, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.003

CHAPTER THREE

NEW APPROACHES TO THE THERAN ERUPTION Ray Rivers

OVERVIEW

In his chapter for this volume, Tim Evans presents an overview of theoretical models of greater and less sophistication that can help in our understanding of historic exchange networks. They only lightly disguise their antecedents in econophysics, anthropology, transport, migration, and urban planning. Nonetheless, despite these contemporary origins, they have already been applied with some success to historical contexts as diverse as Late Geometric Greece (Rihll and Wilson 1987; 1991; Wilson 2012), Bronze and Iron Age Mesopotamia (Davis et al. 2014), Latenian Northern Europe (Filet 2017), the Bronze Age Eastern Mediterranean (Rivers, Evans, and Knappett 2016), and, more specifically, the Middle (MBA) and Late Bronze Age (LBA) south Aegean (Evans, Knappett, and Rivers 2009; Knappett, Evans, and Rivers 2008; Knappett, Rivers, and Evans 2011). Further applications are in progress. There is a major problem in navigating such a wide set of models. Evans has made a strong beginning in finding measures that enable us to identify their similarities and differences, but it is still often difficult to know how to proceed. There is a framework of model discrepancy (see Kennedy and O’Hagen 2001) and topological sensitivity (see Babtie, Kirk, and Stumpf 2014) which, in principle, can help address these issues, but it is devised for systems for which the data are quantitatively good, such as the economic and other systems for which these models were devised. Archaeologists can justifiably be proud of their ability to 39 Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.004

40

RAY RIV ERS

extract data from unpromising circumstances but in general it is not of a quality to permit the level of statistical analysis required. This suggests that, in the absence of robust statistical analysis, we adopt a pragmatic approach, based upon a judicious choice of models that make as few assumptions about the prehistoric societies of the Mediterranean as possible. In fact, such success as they have can be construed as disappointing, in that it indicates that Mediterranean society is doing the generically “obvious,” without the need to invoke the specific social narratives of status, dynasty, strife, and calamity which play such a part in other explanations of the record. This fits with the network viewpoint that what makes a site important is its socially neutral attributes within the network centrality, betweenness, etc., as measured by simple indices of exchange and resources (Newman 2010). Even then we anticipate a certain muddiness in our ability to distinguish between models, particularly if we attempt to retrodict behavior for a specific time interval in the past, since the data may be so poor as to make competing models for a given time-slice effectively indistinguishable. A much better test for discriminating between models is to see to what extent they can describe the evolution of networks in time. In general this is difficult to assess, but in this chapter I shall take advantage of one of the most significant “natural experiments in history” (Diamond and Robinson 2010), the early LBA volcanic eruption of the island of Thera, to test models both before and after the event. We began this exercise in an earlier contribution (Knappett, Rivers, and Evans 2011), but our approach there was limited to our own cost–benefit model. In this chapter I continue this modeling with the wider palette of models previously discussed by Evans. At the height of Minoan influence, the eruption of Thera, a single journey from the northern coast of Crete for contemporary sail-driven vessels, removed what might have been expected to be a crucial gateway between north Crete, the center of Minoan society, and the Cyclades, and hence to the Peloponnese and Dodecanese (see Figure 3.2 below). Surprisingly, after the eruption, exchange throughout the south Aegean arguably intensified rather than diminished until the “burning of the palaces” some years later (Knappett, Rivers, and Evans 2011, with references therein). This sets a high bar for our models since, superficially, removing a key site would be expected to lead to a weaker network with diminished activity. Subsequently, I shall examine the effect of removing Thera for a variety of models taken from Evans’s chapter. As we might anticipate, they place different emphasis on the relative importance of geographical and social factors to give distinguishably contrasting exchange patterns of greater or lesser plausibility. However, because the interplay of maritime technology and geography is very special in this Minoan Aegean, these patterns do not fit easily into Evans’s analysis.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.004

NEW APPROA CHES TO THE THERAN ERUPTION

41

AGENCY

Although not always stated explicitly, the models from Evans’s chapter encode different assumptions as to how and why social networks arise, the nature of the underlying agency, or even the lack of it. Roughly, they lead to networks that fall into three groups: i. the “most likely” networks: maximum-entropy models, gravity models, “retail” archaeology; ii. the “easiest” networks: proximal point analysis (PPA) and interveningopportunity models; iii. the “best” networks: stochastic cost–benefit models.

For the Bronze Age maritime networks of interest here, the nodes of our networks are primarily coastal sites with variable resources. Even when there are overland links, this travel is usually more difficult than by water, and the links between them largely represent maritime “exchange” (Evans, Knappett, and Rivers 2009). MOST LIKELY NETWORKS

The “most likely” networks are the most straightforward to understand in that they attempt to quantify the following question: “all other things being equal, what would I expect to have occurred?” That is, based on what we know about the system, they provide the most likely outcome commensurate with our limited knowledge. This is an old approach, usually linked by name with Laplace as the principle of insufficient reason (Laplace 1825), reframed by Keynes as the principle of indifference (Keynes 1921) and by Jaynes as the principle of maximum ignorance (Jaynes 1973). In the present context the principle states that if we can list the ways a network can occur that comply with the totality of our knowledge—and therefore we have no reason to believe that one way will occur preferentially compared to another, since that would have meant withholding information—the network will be equally likely to have occurred in any of these ways. Essentially we make a snapshot of each potential network, put these snapshots in a pile, and then determine which of them represents the “most likely” network. By definition, since we have not used any information that is not given in the statement of the problem, this network does not have to be understood through a historical or social narrative beyond our initial assumptions about social behavior. Computationally, the prior listing of acceptable networks looks forbidding but it is eased by the realization (Jaynes 1957) that the “most likely” networks are those with maximum entropy. Entropy is colloquially understood in terms of the disorder of a system, but it has a parallel meaning in terms of the

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.004

42

RAY RIV ERS

information we possess about a system. The entropy is directly related to the number of questions we need to ask about a system to get complete knowledge of it, whence the equivalence. This approach is used extensively for transport systems (Ortúzar and Willumsen 1994) and urban planning (Wilson 1970; Wilson and Bennett 1986; Wilson 2010). At its simplest, consider a transport system in which all we know is that a certain number of passenger miles will be traversed at a certain total cost within a given fare scheme. Maximizing entropy gives a passenger flow Tij from a site i to a site j which falls off exponentially with cost. Of course, historic travel is not that of a contemporary transport system. There is no “fare” structure and human mobility is only part of the more general social and economic exchange that connects sites. With regard to the former, when we restrict ourselves to sea travel, the time of travel, which can be rephrased as “effective distance,” becomes a reasonable proxy for “cost.” For the latter, we do not distinguish between different types of vessel—canoe, multi-oared vessel, sailing ship—nor between their cargos. For the MBA at the time of the eruption of Thera, we know that sailing vessels were capable of traveling the distances necessary to enable the Minoan exchange network to thrive. With their large carrying capability, we think of them as the main vehicles for exchange which, between sites i and j, we flatten to the single variable Tij. Large Tij denotes strong exchange, while low Tij indicates weak exchange. More generally, when we take site sizes into account (where the size Si of site i is usually interpreted as its “population”), we recover the simple gravity model (Jensen-Butler 1972) Tij α Si Sj V(dij /D),

(1)

(α denotes “proportional to”) where dij is the effective distance between sites i and j, and D, with some caveats (Evans and Rivers 2017), is most simply taken as the effective distance that the traveler (or the artifact) can expect to make in a single journey, at this period largely a journée. The function V(x), which encodes the relationship between “cost” and “distance,” is the “deterrence function” for network travel and exchange. If it is large, exchange is easy; if small, exchange is difficult. Maximizing entropy does not fix V beyond it being exponential in “cost.” This is one place where we have to make assumptions about social behavior, although only to a limited extent. Figure 3.1 shows two simple choices that bracket the more obvious possibilities. The dashed line is the exponential falloff with distance that we get if we take “cost” or effort as proportional to distance, perhaps understood as “down-the-line” exchange (Renfrew 1977). For sea travel this does not seem realistic. We do not expect a doubling of effort or time on doubling short journeys because of the effort of embarking and

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.004

NEW APPROA CHES TO THE THERAN ERUPTION

43

1.0 0.9 0.8 0.7 V (x)

0.6 0.5 0.4 0.3 0.2 0.1 0

1

2

3

4

x = distance/D

3.1. The deterrence function V(x) for costs proportional to distance (dashed line) and with a short-distance “shoulder” (continuous line).

disembarking or loading and unloading cargo. In what follows we have adopted a functional form (continuous line) which gives a “shoulder” for short distances before falling off rapidly. There are circumstances in which the difference between the two can be important (Rivers and Evans 2014; Evans and Rivers 2017), but we do not anticipate that here. We take the simple gravity model of (1) as our first model to apply to the eruption of Thera. For “geographic” inputs Si and dij (taken as given) and chosen V(x), the remaining input parameter is the journey distance D of the prevailing maritime technology of sail. We note that the structure of the network plays no role, with Tij only dependent on sites i and j, a primitive first guess. Links are reciprocal, reflecting another major simplification. Without wishing to specify social behavior in more than the simplest generalities, for the next step we can impose constraints on its entropy as a trade-off for further knowledge (see Evans this volume; Wilson 1970). Most simply, on restricting outflows of sites, Tij now depends on the network as a whole. Again, distance D is the variable parameter. It is one step further to derive the retail gravity model as used by Rihll and Wilson (1987; 1991) in their analysis of Greek city states. In addition to limiting outflows, the new ingredient involves constraining inflows by means of a quality termed “attractiveness” to characterize the benefits (or not) of concentrating the resources of small separate sites into one extended site, understood here as a restriction on the entropy of inflows. It was developed in urban planning to understand the benefits of shopping malls, with their concentration of retail outlets. Without having to ask in detail how such benefits might accrue, it follows that the resulting networks have a tendency to move from relatively homogeneous to inhomogeneous site importance as the benefits

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.004

44

RAY RIV ERS

become apparent. It provides a natural language with which to describe settlement formation, most recently in Latenian Gaul (Filet 2017). Retail gravity provides our second application. EASIEST NETWORKS

Rather than rely on the passive act of making the best of our limited knowledge, alternative models impute different active agency to the society they describe. For the case of “easiest” networks we assume that exchange begins close to home in that, in the first instance, we choose to interact predominantly with our near neighbors as stepping stones from which a network evolves. The simplest model is PPA, in which, without invoking entropy, we simply restrict outflows. The assumption is that any site only has the resources to interact with a few (typically three or four) of its nearby neighboring sites, ranked as nearest (#1), second-nearest (#2), and so forth. Irrespective of separation, exchange is taken to be uniform for these few sites and zero for the rest, a gross simplification. Making the links in the model directed (directed PPA) avoids the problem of reciprocal exchange, but does not modify the unweighted on/off nature of interactions. Despite the deficiencies of PPA models, in particular the absence of weak links that are so important for network robustness (Granovetter 1973; Csermely 2004), the models have been used extensively because of their simplicity (e.g., Terrell 1977; 1986; Broodbank 2000; Groenhuijzen and Verhagen 2017). We take directed PPA analysis as our third model for the eruption. Given distance site ranking, the only input is the number k of nearby sites to which each site connects. In fact, although the characteristic trip distance D does not appear as a parameter in PPA through an explicit deterrence function, it is present implicitly. Because the networks are built from neighboring sites, there is a local distance scale di (the distance from the site to its furthest interacting (kth) nearest neighbor) at each site i such that, unless D > di some of the links will be very difficult to implement and the model is inappropriate. This sets a minimum scale for D for the model to be useful. To remedy these deficiencies we need to go to more general interveningopportunity models which, like maximum-entropy/gravity models, originated in part for describing mobility. First developed by Stouffer (1940), in the later formulation by Schneider (1959) for describing passenger flows, it is assumed that the number of trips from one site to a destination site is proportional to the number of “opportunities” at the destination and inversely proportional to the number of intervening “opportunities.” Nonreciprocal exchange of varying strength follows. The translation of these ideas into archaeological exchange is fairly evident. For simple networks (e.g., linear coastal sites), exchange between

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.004

NEW APPROA CHES TO THE THERAN ERUPTION

45

sites now falls off exponentially with respect to the number of intermediate sites, a smoothed version of PPA. A particular intervening-opportunity model is the radiation model of Simini et al. (2012), which, while relaxing the conditions under which a journey stops, surprisingly leads to more restricted outcomes. Particularly, whatever levels of willingness we consider to be uniformly acceptable for transactions to occur, there are no free parameters beyond the given site sizes and distance rankings. However, there is again an implicit minimum scale D for the model to be sensible. The radiation model provides our fourth application. BEST NETWORKS

“Best” networks start from a different position, that of cost–benefit analysis. It is assumed that there are identifiable “benefits” and “costs” in sustaining a network, and that those networks that have survived were those that maximized the difference between them. This, again, is an old idea. Laplace’s principle of insufficient reason noted earlier was a response to the earlier principle of sufficient reason by Leibniz, from which he (Leibniz) evolved to a belief that “we live in the best of all possible worlds” (e.g., Murray and Greenburg 2013). Quantified by Maupertuis (1744) as the principle of least action, the basis for much contemporary physics, it leads naturally to cost–benefit analysis for suitable actions. Many different cost–benefit models can be constructed, but they have important features in common. The cost–benefit model that I shall discuss here, termed ariadne (created by Tim Evans and developed with Carl Knappett and myself), encodes the obvious costs of sustaining the populace and maintaining the network, and the benefits of local resources and exchange. The given inputs are now the deterrence function V(x), the Si (now understood as the carrying capacities of the sites), and the separations dij. The variable inputs are the relative magnitudes of the costs and benefits. The outputs are the link strengths Tij and the populations (or relative use of resources) Pi. The action here is a “social potential” (Butts 2007) describing the differences between costs and benefits, or a social “Hamiltonian,” in our language (Evans, Knappett, and Rivers 2009). However, there is an ingredient not present in the earlier models, that of stochastic behavior. The idea is straightforward. Because it requires global knowledge a network does not arrive fully formed in the configuration best able to minimize costs and maximize benefits. This would be particularly the case after an event like an eruption that severely dislocated exchange. We can instead imagine a series of local adjustments which tend to make the network “better” step by step. These adjustments can be thought of as taking place within a landscape of networks, in which each point of this landscape is

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.004

46

RAY RIV ERS

a network, the lowest point of this landscape (the least action) corresponding to the “best” network. These sequential local adjustments do not guarantee to achieve the “best” result, which depends on all sites and all links, since local communities do not have access to this information. Nonetheless, if we think of this landscape as a multidimensional space full of hills and valleys (whose elevation is the Hamiltonian or social potential), the effect of these adjustments is, in general, to drive us “downhill.” Whether we are driven to the lowest of all valleys is a moot point, but it is enough to be sufficiently downhill. That is, the resulting network may not be the “best,” but hopefully it is “good enough.” This “satisficing” approach of “bounded rationality” (Simon 1957) is familiar to us in our everyday lives, but it does have the consequence that, beginning from identical initial conditions, we can still have different outcomes for the same model parameters, each of which is “good enough” without being the best. This is our stochastic behavior, which we might wish to think of in terms of contingency. It is important that this contingency be small since, otherwise, slightly different decisions made by communities in the earlier stages of adjusting to changing circumstances will lead to very different and unpredictable outcomes. This is unlike approaches such as PPA and gravity models, which are effectively deterministic in their formulation. Finally, the epistemic approach of entropy maximizing looks at odds with the ontic approach of the easiest and best modeling. This is often a false dichotomy, although a useful tactic, and our division of models into the three categories is somewhat simplistic. Wilson has shown how interveningopportunity models can often be recast in the language of entropy maximization, although sometimes this requires contortions (Wilson 1970). At the same time the retail model has a direct interpretation in terms of retail footfall which is anything but “the obvious.” This blurring of the difference between epistemic and ontic approaches can be taken further if we couch all models in the language of “social thermodynamics.” From this viewpoint, maximumentropy models can be expressed in terms of microcanonical ensembles and cost–benefit models in terms of macrocanonical ensembles, formally equivalent for large enough networks. In fact, the Aegean network that we shall consider is very small—the thirty-nine nodes of Figure 3.2—and we shall treat all models as conceptually separate. To be satisfactory, a network model must be able to address the periods both immediately before and after the eruption, where we understand “immediately” in the sense of archaeological haste, perhaps a generation. However, because of the relative uncertainty of the events leading to the “burning of the palaces,” it is more difficult to constrain model behavior well after the eruption (perhaps two or three generations), but we shall see that the network collapse is arguably a further discriminant. As indicated, I shall restrict myself to

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.004

NEW APPROA CHES TO THE THERAN ERUPTION

47

33 39 26

29

36 27 38 37

20

14

18 23

35

31

13 28

24 11

34 19

32

17

25 10

12

9

16

15 22 1 4

21

2 8

3 5

30 6 7

100 km

3.2. Important sites for the MBA Aegean, including Knossos (1) and Thera (10). The sea journey from the northern Cretan coast to Thera is a little over 100 kilometers, but Knossos is a little inland.

• • • • •

the simple gravity model the retail gravity model directed PPA the radiation model the ariadne stochastic model

As representative models, the first two are maximum-entropy models, the next two intervening-opportunity models, and the last a cost–benefit model. MBA MINOAN CULTURE AND THE ERUPTION OF THERA

The Aegean Bronze Age is characterized by “dark ages” between which the MBA Minoan culture has a fairly well-defined span from about 2000 BC to 1400 BC. Our understanding of the links between sites relies largely on ceramic evidence. Although finds are limited, we would argue that the south Aegean forms a coherent whole and can be treated as an approximately autonomous region, even though some external links are important: for example, the importation of tin from Anatolia. In Figure 3.2 we give some idea of its reach, in which we have taken thirty-nine of the important sites of that period,

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.004

48

RAY RIV ERS

table 3.1. The sites enumerated in Figure 3.2, including the size of their local resource base: small (S), medium (M), and large (L). Where relevant (e.g., ariadne, but not PPA) we take typical resource ratios to be 1:2:4 or 2:3:6. Qualitatively the differences between these choices are unimportant. 1. Knossos (L) 2. Malia (L) 3. Phaistos (L) 4. Kommos (M) 5. Ayia Triadha (L) 6. Palaikastro (L) 7. Zakros (M) 8. Gournia (L) 9. Chania (L) 10. Thera (M) 11. Phylakopi (M) 12. Kastri (M) 13. Naxos (L)

14. Kea (M) 15. Karpathos (S) 16. Rhodes (L) 17. Kos (M) 18. Miletus (L) 19. Iasos (M) 20. Samos (M) 21. Petras (L) 22. Rethymnon (L) 23. Paroikia (M) 24. Amorgos (S) 25. Ios (S) 26. Aegina (M)

27. Mycenae (L) 28. Ayios Stephanos (L) 29. Lavrion (M) 30. Kasos (S) 31. Kalymnos (S) 32. Myndus (M) 33. Cesme (M) 34. Akbuk (M) 35. Menelaion (S) 36. Argos (M) 37. Lerna (M) 38. Asine (S) 39. Eleusis (M)

from which Knossos (site 1) and Akrotiri (site 10), the main settlement on Thera, are key sites in our subsequent discussion. We have discussed the eruption of Thera elsewhere and the next few paragraphs provide a brief reprise. The reader is referred to Knappett, Rivers, and Evans (2011) for more detail on the sites. It is sufficient for our purposes just to name sites when we need, but a complete list is given in Table 3.1. The eruption of Thera, one of the largest of the last 10,000 years, buried Akrotiri beneath meters of tephra. Happening sometime between 1627 BC (Manning et al. 2006) and 1525 BC (Wiener 2009), what matters is that it occurs at the end of the Late Minoan IA (LM IA) period and is followed by Late Minoan IB (LM IB). The evidence from the site shows that Thera was a significant trade gateway between north Crete and the rest of the Aegean. As such, the island’s destruction jeopardized an important exchange route since its distance from north Crete (approximately 100 kilometers) is about the limit of one day’s sailing. However, our current knowledge suggests that, despite this disruption of the exchange network, the LM IB period immediately following the destruction sees no reduction in overall exchange across the south Aegean, to the extent that LM IB shows a late flowering of Minoan culture (Hood 1971; Warren 1975; Cadogan 1976; Mountjoy 2008). Crete appears to obtain the resources it needs without difficulty, and evidence for continuing interregional exchange is apparent at many sites (Wiener 1991, 341). PRE-ERUPTION NETWORKS

Back-tracking a step, prior to the Theran eruption the south Aegean was the scene of a thriving maritime exchange network. There is evidence for intensive

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.004

NEW APPROA CHES TO THE THERAN ERUPTION

49

Distance 0

20

40

60

80

100

120

Karpathos Kasos Chania Rethymno A.Triadha Phaistos Kommos Knossos Malia Gournia Zakros P−kastro Petras Phylakopi Amorgos Akrotiri Ios Naxos Paroikia Kea Lavrion Asine Lerna Mycenae Argos Aegina Eleusis Kastri A.Stephanos Menelaion Cesme Rhodes Miletus Samos Kalymnos Kos Myndus Iasos Akbuk

3.3. Dendrogram for the sites of Figure 3.2. The distance is the shortest geographical distance between sites (kilometers) on negotiating headlands. A horizontal line cutting the vertical axis at distance scale D cuts the dendrogram into a number of disconnected branches below the horizontal line. Each of these separate branches defines a cluster of sites within which any pair of sites is separated by a distance of D or less. It is not possible to move between sites in different branches without following a link of length greater than D. (Produced using the R statistics package: R Development Core Team 2010.)

trade within regional clusters: e.g., within Crete itself, and within the area of the Dodecanese and coastal Anatolia. However, these clusters do not make a network of any reach. What permits them to create active links is the relatively new technology of sail, which, in favorable circumstances, permits single journeys with substantial cargos of up to around 100 kilometers, the distance from north Crete to Thera, but not just Thera. By coincidence, we see from the dendrogram of separations in Figure 3.3 that this distance is the tipping

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.004

50

RAY RIV ERS

point for connectivity in the south Aegean. Although we have taken headlands into account, we have ignored winds and tides to be as simple as possible. For D < 100 kilometers the network breaks up into regions, for D > 100 kilometers it is a unified whole with, crucially, the Cyclades roughly equidistant from Crete and the two mainlands. We note that D is not the absolute maximum distance traveled, but the one for which longer journeys are increasingly difficult. This confluence of maritime technology (i.e., the ability to travel a distance D) with geography (the need to travel a distance dij) enables the exchange to happen for the larger distances (dij ≈ D ≈ 100 kilometers) that we need for a vigorous network and is the key ingredient in the subsequent analysis. While it is true that, prior to the widespread use of sail, it was possible to travel large distances by oar or to make more circuitous routes, this requires more resources and less cargo. It needs the relative democratization of sail for exchange volumes to be sufficiently large for the thriving network that we see to come into existence. As well as the connections to Akrotiri, the outcome is exchange between Crete and the Dodecanese and coastal Anatolia, between the mainland and the Cyclades, and between the Cyclades and Dodecanese and coastal Anatolia. For references for all of these connections and more, see Knappett, Evans, and Rivers (2008) and Knappett, Rivers, and Evans (2011). It follows that only models which encode this confluence seriously can characterize the pre-eruption network. I have discussed this in part already (Rivers, Knappett, and Evans 2013a) so I shall be brief here. It needs to be stressed that these models have few variable parameters. At best they help our understanding of how the “real world” works rather than demonstrate what happens in detailed reality. There is a general issue, already touched upon, as to how our models incorporate the trip distance D. We have seen that generalized gravity models and cost–benefit models incorporate D explicitly, whereas interveningopportunity models incorporate it implicitly. We know that the network would not exist if D were significantly smaller than 100 kilometers. Should we therefore reject models that do not show this on varying D, on the ground that they cannot explain how the network could have come into existence? Alternatively, given that D ≈ 100 kilometers for sail, do we need to vary D at all? If we assume the first, then the retail gravity model can be discarded because of its insensitivity to D, still giving significant networks for, say, D = 60 kilometers. Further, directed PPA and the radiation model would also look to be inapplicable. On the other hand, with D implicit in their formulation, it follows that a minimum value D ≈ 100 kilometers, appropriate for the MBA, is sufficient for our models to make sense. This is a happy coincidence that we accept and, if we

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.004

NEW APPROA CHES TO THE THERAN ERUPTION

51

are to reject any models, it will be for other reasons. This reflects a change of position on my part (Rivers, Knappett, and Evans 2013a). For example, the record suggests that, despite the extent of the network, the most intense activity still lies within the regional clusters rather than between them. To realize this requires models with variable link strengths, prohibiting simple on/off models like directed PPA. More subtly, we require north Crete to play a significant role in the network, in terms of centrality rank, population “weight,” and “betweenness” to Akrotiri. We should not attach too much importance to the details. Beyond that, what is important is that exemplary networks show regional clusters that are strongly linked within themselves, but linked with each other by weak links and a few strong links. We have no concerns about sensitivity to D for the simple gravity model which, should we wish, shows disconnection for small D. Moreover, it gives strong interregional connectivity and highlights the importance of north Crete and Akrotiri. Although it provides reciprocal links, its other virtues are enough for its retention. The retail gravity model is more subtle in that, on varying “attractiveness,” connectivity undergoes a transition from a fairly homogeneous distribution of sites to a heterogeneous distribution with a scattering of dominant sites. However, in neither case does north Crete play a significant role, with low betweenness to Akrotiri, and the only way that strong links arise is for the regional “star” network with nontrivial attractiveness that is at variance with the data (Rivers, Knappett, and Evans 2013a; 2013b). We have only included directed PPA because of its common use by archaeologists. For the MBA Aegean it is a total failure. Not only is the network largely a ring around the mainland coastlines and Crete, but the absence of weak links also makes it useless. The radiation model is more subtle, with similarities to the retail gravity model for low attractiveness. Like the gravity model, it has strong regional links but only weak links between regional clusters. Yet there is no particular role for Thera. We give the unique preeruption network in Figure 3.4a. The application of our cost–benefit model has been discussed in detail by Knappett, Rivers, and Evans (2011), to which we refer the reader. We highlight here two points. The first is that, in comparison to the simple gravity model, ariadne has one or two more parameters than the other models, which by itself suggests that it should inevitably provide a better fit to the data. The reality is more subtle. Cost–benefit models tread a delicate balance between “boom” and “bust.” A small change in the parameters can flip the network either toward disconnectivity, with very weak links between the clusters—as happens when benefits of interaction between sites are small in comparison to the benefits from exploiting local resources—or toward hyperconnectivity, with an abundance of strong links and large sites expanding their

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.004

52

RAY RIV ERS

a

100 km

b

100 km

3.4. Upper figures: the radiation model for Aegean networks before (a) and after the eruption of Thera (b). The networks have sites labeled (in size) by rank and links (in thickness) by exchange flow. The absence of links between Crete and the Cyclades post-eruption (right) is enough to eliminate the model. Lower figures: ariadne networks for D = 100 kilometers before (c) and after (d) the eruption of Thera. The shift in the pattern of exchange after the eruption is very striking, with an emphasis on the northwest of the network, for which there is archaeological evidence for post-eruption activity. In all cases necessary land travel has been penalized by a friction factor 3.0.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.004

NEW APPROA CHES TO THE THERAN ERUPTION

53

c

100 km

d

100 km

3.4. (cont.)

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.004

54

RAY RIV ERS

populations and trade preferentially. It happens that a sensible scenario for LM IA can be generated as an output from the model, but only with a fairly narrow range of inputs. An example is given in Figure 3.4c. In the familiar contemporary context of financial instability this is a “Goldilocks” scenario, a network that, as far as the intensity of exchange is concerned, is neither “too hot” nor “too cold.” This is commensurate with the following observation (Broodbank, Kiriatzi, and Rutter 2005, 90): For the south Aegean islands in the late Second and Third Palace periods, an age of intensifying trans-Mediterranean linkage and expanding political units, there may often have been precariously little middle ground to hold between the two poles of (i) high profile connectivity, wealth and population, or (ii) an obscurity and relative poverty in terms of population and access to wealth that did not carry with it even the compensation of safety from external groups.

The second point concerns the stochastic nature of the output networks that we discussed earlier. Identical inputs will lead to different contingent networks. Yet again the balance between technology and geography is crucial. For these differences to be small, it seems empirically that D is less than or equal to the larger dij that we need for the network to thrive. This is what happens here. If, on the other hand, D > dij, it is too easy to find acceptable alternative choices and contingency is high (Rivers and Evans 2014). In deciding which models to take through to the next round of posteruption analysis, we should not be quick to dismiss since we need models that work well enough over both LM IA and LM IB. From the previous discussion, ariadne and the simple gravity model go through, despite the reciprocal exchange of the latter. For comparison I shall also include the radiation model. It downplays Thera but it could be argued that Akrotiri’s importance is overstated just because it looks so central. Although it plays a significant role because of the dramatic pottery finds, Cretan imports even here only constitute around 10–15% of the total pottery consumed. IMMEDIATE POST-ERUPTION NETWORKS

Although the site of Akrotiri is completely destroyed, ending occupation on the island for generations (Doumas 1983), it is clear from the archaeological evidence that Cretan centers continued to participate in exchange networks in LM IB (Wiener 1991; Warren 2001). For example, Warren (2001, 116) notes that the palace building at Zakros (7) makes “little sense on its own in its relatively isolated and circumscribed landward position . . . Its wealth makes better sense as part of a larger structure, plausibly as a, or even the, eastern port of Knossos in LM IB.” However, the network will be obliged to redistribute its

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.004

NEW APPROA CHES TO THE THERAN ERUPTION

55

exchange links to accommodate the hole in its center that was Thera. How this happens is less clear. Thus it has been argued that Cretan influence in the Cyclades post-eruption (LM IB) is less than it was pre-eruption (LM IA) (Mountjoy 2004, 399–404). In particular, chemical analysis of LM IB/Late Helladic (LH) IIA marine-style pottery from Melos (11) and Kea (14) seems to show that it is imported from the Greek mainland rather than from Crete (Mountjoy and Ponting 2000). On the other hand, “the evidence points to, if anything, an increase in Minoan trading activity in LM IB, particularly in our excavations at Ayia Irini, Keos [14] where we literally had thousands of LM IB vases imported from outside” (Davis 1980, 336; Cummer and Schofield 1984). Whatever the case, the fact that the removal of Thera causes a reorganization of exchange flows with comparable overall activity puts strong constraints on acceptable network models. We mimic the aftermath of the eruption by removing Akrotiri from the list of sites and repeating the analysis with otherwise the same parameters. For our surviving models the effect of removing Thera is as follows: • For the simple gravity model we erase the links between Akrotiri and the rest of the south Aegean, leaving the others unchanged. The model fails because this is not a rearrangement of the network in the sense above, just its diminution, contrary to the data. • For the radiation model there is a rearrangement, although not of any particular note (Figure 3.4b). This model fails in turn because the absence of any links between Crete and the Cyclades post-eruption does not match the data. • For ariadne the removal of Thera causes an instantaneous remodeling of the social potential landscape with different low-lying valleys. The effect is to cause a major rearrangement of flows (Figure 3.4d). Phylakopi [11] takes over as the link between the Cyclades and other regions, with these islands now strongly connected to the mainland. Stochastic effects mean that the details vary but, in general, there is a shift to the northwestern Aegean, along the lines of Davis (1980), quoted above, with Kastri maintaining a major presence.

There is no doubt that ariadne—and only ariadne among the models considered here—is capable of describing the main characteristics of the pre- and posteruption south Aegean. LATER POST-ERUPTION: NETWORK COLLAPSE

We have little to say about later events here beyond the fact that, even though “civilization” carries on much as before, we should not discount changes in Minoan society as a result of the disruption. Sometime later, LM IB comes to an end through violent destructions by fire. Rather than the eruption, this is the catastrophe that ends Minoan culture. There are competing explanations, most

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.004

56

RAY RIV ERS

simply that it was a consequence of internal strife, perhaps an amplification of pre-eruption vulnerability (Riede 2014). This may have facilitated the invasion of the Mycenaean mainlanders whose culture now takes predominance (Preston 2008), or there may have been a direct invasion. We also cannot preclude earthquakes. In the post-eruption scenario outlined above, the network is quite stable, no less so than pre-eruption. Other than the removal of Thera, we did not touch the models. This is unrealistic, because it is likely that the removal from the network of a trading hub would have changed the overall “costs” of trade. The suggestion from the immediate post-eruption efflorescence is that the costs did not increase immediately, but took a while to be implemented. These costs concern the greater distance needed to travel from Crete to trade with the Cyclades, and perhaps a concomitant need for more ships, and more crews, as well as perhaps new storage and distribution facilities. This is where, even had we retained it, the radiation model fails totally on general grounds. There are no variable parameters. In this context its uniqueness makes it uniquely wrong. The simple gravity model is equally at fault. Of our original models only ariadne is capable of spanning from LM IA to the end of LM IB. We have explored this possibility in Knappett, Rivers, and Evans (2011), to which we refer the reader. An increase in exchange costs can easily lead to fewer but stronger links as the major sites respond by putting what eggs they have in fewer baskets. In one sense this may seem to be a valid response, in that there can be more total trade passing through the limited number of links. However, with a diminishing number of stabilizing weaker links and contingent strong links, the system breaks down easily, in accord with Renfrew’s (1980, 337) suggestion of a centralised economy which may be working under some adversity which might be increased population . . . people coming in from Thera . . . What I think you would expect to see is not a gradual decline, but an increasing intensity in the various sub-systems of the culture system, including an increasing level of trade, until the system breaks down altogether . . . There is a parallel here. . . with a stock exchange collapse.

It is not surprising that ariadne, with its econophysics antecedents (Jackson 2008) primed to describe “boom” and “bust,” can easily accommodate cultural collapse. Of course, this is not the only possible outcome. Less dramatic choices of parameter change can be made that enable the network to continue to thrive, or go into gentle decline, perhaps requiring the deus ex machina of an earthquake or external invasion to bring it to an end.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.004

NEW APPROA CHES TO THE THERAN ERUPTION

57

OUTLOOK

The MBA and early LBA Aegean offers a very special place and time for the development of exchange networks. For the first time in the Bronze Age, improvements in maritime technology, namely the introduction of sail, permit travel over the distances necessary for a network to thrive across the south Aegean. Concretely, if D (technology) is the distance scale for single journeys that sailing vessels allow and d (geography) is the required journey distance for networks to be easily effected, then we have D ≈ d (≈ 100 kilometers) and the network models must reflect this. As such, the networks that support our understanding of the archaeological record are not the gravity models that maximize entropy, nor those which hope to thrive on the basis of local interactions, the “easy” option. Rather, the relevant networks, particularly after the eruption, are plausibly the consequence of a bounded rationality in which the network adjusts to being as effective as possible on an Aegean-wide scale, permitting substantial rearrangement. It is this matching of D and d and the concomitant sensitivity of the network to the value of D that singles out ariadne so cleanly. The effect is that, for this specific case, the analysis of Evans in this volume on model similarity and difference is largely unnecessary. For different times and places we cannot be so dismissive. Empirically, a key determinant is the ratio of D to d. We have already said that the Bronze Age Aegean is known for the punctuations of its “dark ages” which demarcate cultural change. In the Early Bronze Age, for which the Cyclades in particular seem to possess a largely self-contained coherence (Broodbank 2000), we have D < d. For the Late Bronze Age, with improved rigging and vessels capable of extended tramping, we have the effective D > d. We have argued elsewhere (Rivers, Evans, and Knappett 2016) that this, of itself, conditions our choice of models, but that is a different story. REFERENCES

Babtie, A.C., Kirk, P., and Stumpf, M.P.H. 2014. Topological sensitivity analysis for systems biology. Proceedings of the National Academy of Sciences 111(52), 18507–18512. Broodbank, C. 2000. An Island Archaeology of the Early Cyclades. Cambridge: Cambridge University Press. Broodbank, C., Kiriatzi, E., and Rutter, J.B. 2005. From Pharaoh’s feet to the slave-women of Pylos? The history and cultural dynamics of Kythera in the Third Palace Period. In A. Dakouri-Hild and E.S. Sherratt (eds), Autochthon: Studies Presented to Oliver Dickinson on the Occasion of His Retirement, 70–96. BAR International Series 1432. Oxford: Archaeopress. Butts, C. 2007. Statistical mechanical models for social systems. In A. Bejan and G. Merkx (eds), Constructal Theory of Social Dynamics, 197–223. New York: Springer. Cadogan, G. 1976. The Palaces of Minoan Crete. London: Routledge.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.004

58

RAY RIV ERS

Csermely, P. 2004. Strong links are important, but weak links stabilize them. Trends in Biochemical Science 29, 331–334. Cummer, W.W., and Schofield, E. 1984. Keos III. Ayia Irini: House A. Mainz: Philipp von Zabern. Davis, J.L. 1980. Comment in discussion of session 7A, in C. Doumas (ed.) Thera and the Aegean World. Vol. 2, 336. London: Thera Foundation. Davis, T., Fry, H., Wilson, A., Palmisano, A., Altaweel, M., and Radner, K. 2014. Application of an energy maximising and dynamics model for understanding settlement structure: the Khabur triangle in the Middle Bronze and Iron Ages. Journal of Archaeological Science 43, 143–154. De Maupertuis, P.L.M. 1744. Accord de différentes lois de la nature qui avaient jusqu’ici paru incompatibles. Mémoires de l’Académie des Sciences de Paris, 417. Diamond, J.M., and Robinson, J.A. 2010. Natural Experiments of History. Cambridge, MA: Belknap Press. Doumas, C. 1983. Thera: Pompeii of the Ancient Aegean. London: Thames and Hudson. Evans, T., Knappett, C., and Rivers, R. 2009. Using statistical physics to understand relational space: a case study from Mediterranean prehistory. In D. Lane, D. Pumain, S.E. van der Leeuw, and G. West (eds), Complexity Perspectives on Innovation and Social Change, 451–479. Berlin: Springer. Evans T.S. and Rivers, R.J. 2017. Was Thebes necessary? Contingency in spatial modeling, Frontiers in Digital Humanities 4. https://doi.org/10.3389/fdigh.2017.00008. Filet, C. 2017. An attempt to estimate the impact of the spread of economic flows on Latenian urbanization. Frontiers in Digital Humanities 3. https://doi.org/10.3389/fdigh .2016.00010. Granovetter, M.S. 1973. The strength of weak ties. American Journal of Sociology 78(6), 1360–1380. Groenhuijzen, M.R., and Verhagen, P. 2017. Comparing network construction techniques in the context of local transport networks in the Dutch part of the Roman limes. Journal of Archaeological Science Reports 15, 235–251. Hood, S. 1971. The Minoans. London: Thames and Hudson. Jackson, M.O. 2008. Social and Economic Networks. Princeton: Princeton University Press. Jaynes, E.T. 1957. Information theory and statistical mechanics. Physical Review 106(4), 620–630. Jaynes, E.T. 1973. The well-posed problem. Foundations of Physics 3, 477–493. Jensen-Butler, C. 1972. Gravity models as planning tools: a review of theoretical and operational problems. Geografiska Annaler. Series B, Human Geography 54, 68–78. Kennedy, M.C., and O’Hagan, A. 2001. Bayesian calibration of computer models. Journal of the Royal Statistical Society, Series B 63, 425–464. Keynes, J.M. 1979. A Treatise on Probability. Reprint. New York: AMS (first published London: Macmillan, 1921). Knappett, C., Evans, T., and Rivers, R. 2008. Modelling maritime interaction in the Aegean Bronze Age. Antiquity 82, 1009–1024. Knappett, C., Rivers, R., and Evans, T. 2011. Modelling maritime interaction in the Aegean Bronze Age, II: the eruption of Thera and the burning of the palaces. Antiquity 85, 1008–1023. Laplace, P.-S. 1995. Philosophical Essay on Probabilities. Trans. Andrew I. Dale. New York: Springer-Verlag (originally published as Essai philosophique sur les probabilités. Paris: Bachelier, 1825).

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.004

NEW APPROA CHES TO THE THERAN ERUPTION

59

Manning, S.W., Bronk Ramsey, C., Kutschera, W., Higham, T., Kromer, B., Steier, P., and Wild, E. 2006. Chronology for the Aegean Late Bronze Age. Science 312, 565–569. Mountjoy, P. 2004. Knossos and the Cyclades in Late Minoan IB. In G. Cadogan, E. Hatzaki, and A. Vasilakis (eds), Knossos: Palace, City, State, 399–404. BSA Studies 12. London: British School at Athens. Mountjoy, P. 2008. The Cyclades during the Mycenaean period. In N. Brodie, J. Doole, G. Gavalas, and C. Renfrew (eds), Horizon, Όρίζων: A Colloquium on the Prehistory of the Cyclades, 467–477. Cambridge: McDonald Institute for Archaeological Research. Mountjoy, P., and Ponting, M. 2000. The Minoan thalassocracy reconsidered: provenance studies of LH IIA/LM IB pottery from Phylakopi, Ayia Irini and Athens. Annual of the British School at Athens 95, 141–184. Murray, M., and Greenberg, S. 2013. Leibniz on the Problem of Evil. In Edward N. Zalta (ed.), The Stanford Encyclopedia of Philosophy (Spring 2013 edn). http://plato.stanford.edu/arc hives/spr2013/entries/leibniz-evil. Newman, M. 2010. Networks: An Introduction. Oxford: Oxford University Press. Ortúzar, J. de D. and Willumsen, L.G. 1994. Modelling Transport. New York. Wiley. Preston, L. 2008. Late Minoan II to IIIB Crete. In C. Shelmerdine (ed.), The Cambridge Companion to the Aegean Bronze Age, 310–326. Cambridge: Cambridge University Press. Renfrew, C. 1977. Alternative models for exchange and spatial distribution. In T.K. Earle and J. Ericson (eds), Exchange Systems in Prehistory, 71–90. New York: Academic Press. Renfrew, C. 1980. Comment in discussion of session 7A, in C. Doumas (ed.), Thera and the Aegean World, Vol. 2., 337. London: Thera Foundation. Riede, F. 2014. Towards a science of past disasters. Natural Hazards 71, 335–362. Rihll, T.E., and Wilson, A.G. 1987. Spatial interaction and structural models in historical analysis: some possibilities and an example. Histoire & Mesure 2, 5–32. Rihll, T.E., and Wilson, A.G. 1991. Modelling settlement structures in ancient Greece: new approaches to the polis. In J. Rich and A. Wallace-Hadrill (eds), City and Country in the Ancient World, 59–95. London: Routledge. Rivers, R., and Evans, T. 2014. New approaches to Archaic Greek settlement structure. Les nouvelles de l’archéologie 135, 21–27. Rivers, R., Evans, T., and Knappett, C. 2016. From oar to sail. In C. Ducruet (ed.), Maritime Networks: Spatial Structures and Time Dynamics, 63–76. Routledge Studies in Transport Analysis. London: Routledge. Rivers, R., Knappett, C., and Evans, T. 2013a. Network models and archaeological spaces. In A. Bevan and M. Lake (eds), Computational Approaches to Archaeological Spaces, 99–126. London: University College London Press. Rivers, R., Knappett, C., and Evans, T. 2013b. What makes a site important? Centrality, gateways and gravity. In C. Knappett (ed.), Network Analysis in Archaeology, 125–150. Oxford: Oxford University Press. Schneider, M. 1959. Gravity models and trip distribution theory. Papers of the Regional Science Association 5, 51–58. Simini, F., González, M.C., Maritan, A., and Barabási, A.-L. 2012. A universal model for mobility and migration patterns. Nature 484, 96–100. Simon, H. 1957. Models of Man: Social and Rational. New York: John Wiley and Sons. Stouffer, S.A. 1940. Intervening opportunities: a theory relating mobility and distance. American Sociology Review 5, 845–867.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.004

60

RAY RIV ERS

Terrell, J. 1977. Language and material culture on the Sepik Coast of Papua New Guinea: using social network analysis to simulate, graph, identify, and analyze social and cultural boundaries between communities. Journal of Island and Coastal Archaeology 5, 3–32. Terrell, J. 1986. Prehistory in the Pacific Islands. Cambridge: Cambridge University Press. Warren, P. 1975. The Aegean Civilisations. London: Elsevier-Phaidon. Warren, P. 2001. Book review of “The Troubled Island,” by J. Driessen and C. Macdonald. American Journal of Archaeology 105, 115–118. Wiener, M. 1991. The nature and control of Minoan foreign trade. In N.H. Gale (ed.), Bronze Age Trade in the Mediterranean, 325–350. Jonsered: P. Aströ ms Fö rlag. Wiener, M. 2009. The state of the debate about the date of the Theran eruption. In D.A. Warburton (ed.), Time’s Up: Dating the Minoan Eruption of Santorini, Acts of the Minoan Eruption Chronology Workshop, Sandbjerg (November 2007), 197–206. Monographs of the Danish Institute at Athens 10. Athens: Danish Institute at Athens. Wilson, A.G. 1970. Entropy in Urban and Regional Modelling. London: Pion. Wilson, A.G. 2010. Entropy in urban and regional modelling: retrospect and prospect. Geographic Analysis 42, 364–394. Wilson, A.G. 2012. Geographical modelling for archaeology and history: two case studies. Advances in Complex Systems 15, 1–14. Wilson, A.G., and Bennett, R.J. 1986. Mathematical Methods in Human Geography and Planning, New York: Wiley-Blackwell.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.004

CHAPTER FOUR

GEOGRAPHY MATTERS Defining Maritime Small Worlds of the Aegean Bronze Age Thomas F. Tartaron

INTRODUCTION

Archaeologists of the ancient Mediterranean concerned with maritime interactions of all kinds have been encouraged in recent years to explore networks from both qualitative and quantitative perspectives. In this chapter I wish to assess the appropriateness of recent social network analysis (SNA) language and models for capturing maritime networks of the Greek (Aegean) Bronze Age, c.3100–1050 BC, in their entirety; that is, at all scales and embracing as many actors as possible. The main point I wish to make is that the realities of the Bronze Age placed constraints on communication that are not relevant to the traditional concerns and subjects of SNA as they emerged from sociology (or of network theory in physics). Indeed, these studies demonstrate that the formation and intensity of modern social networks are often not driven or constrained by physical proximity. I contend that for the vast majority of Bronze Age coastal dwellers, distance was a decisive factor in the maritime networks in which they participated, which were as much social as economic. Thus terminology and models drawn from SNA may not fit particularly well with the conditions of prehistory. This is especially the case for local-scale maritime networks, which have also received insufficient attention from archaeologists. I summarize a multi-scalar, diachronic model of Bronze Age maritime networks that is based on nested geographical scales, from local to international. This model emphasizes local and microregional scales, but also reveals how the 61 Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:14, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.005

62

THOMAS F. TA RTARON

interplay of interactions and events at larger scales shapes life locally. Further, the model is a qualitative one not born of any explicit engagement with SNA, but should be amenable to modification and development as SNA becomes a more useful tool in archaeology. GEOGRAPHY, DISTANCE, AND THE BRONZE AGE

Carl Knappett (2013, 7–10) and others have pointed out that current network theory is in some ways a poor fit for networks of the ancient past. Network analysis in the field of sociology treats location and distance as metaphorical rather than literal; networks transcend real-world geography. In our era, one can form a network remotely without physical contact, and an acquaintance 2,000 miles away can be closer in network terms than a neighbor down the street. Obviously, technology has made this possible. Technology obliterates geographical distance and very nearly time. It is now an ordinary experience for me to stand in a cotton field in rural northern Greece and speak on the phone with my wife in Philadelphia, often with a better connection than we get locally at home. Distance and geography did matter for premodern sea travel, however, unlike the World Wide Web, the organization of air traffic, or other models that network theory typically invokes. The factors that facilitated or hindered maritime travel in the Aegean Bronze Age were both environmental and social, including distance, weather conditions, navigational hazards, boat technology, skill in navigation, and esoteric knowledge of distant people and places. The capabilities of Bronze Age boats— seaworthiness, propulsion—and prevailing navigational skills placed practical limits not only on the range and frequency of long voyages, but also on the group of specialized sailors capable of mounting them. These factors, some quasi-quantifiable and some not, have not yet been captured well in SNA or other network approaches. For example, the sea is typically represented as a flat, untextured plain that takes no account of winds, currents, hazards, or stochastic events such as the rapid-onset storms characteristic of the Aegean Sea. Daily distance ranges are mentioned for specific boats, such as forty kilometers per day for a paddled Early Bronze Age Cycladic longboat (Broodbank 2000, 287–289) or 100 kilometers per day for a Bronze Age sailing ship (Knappett et al. 2008, 1014), but they typically do not incorporate specific sea characteristics. A few attempts have been made to arrive at more realistic maritime travel times and parameters—notably in dissertations by David Conlin (1999) and, within a more explicit network analysis framework, by Justin Leidwanger (2011). Leidwanger’s research assembles wind and current data for much of the Mediterranean and shows how a voyage out can be very different from the voyage back. The need for this kind of refinement was made abundantly clear in experiments in the Aegean with a reconstructed “prehistoric” reed canoe

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:14, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.005

GEOGRAPHY MATTERS

63

(Tzalas 1995). The voyage in early October from Lavrion to Melos (about 120 kilometers as the crow flies) encountered unseasonable but hardly extraordinary rough weather, with heavy rain, high winds, and choppy waves. The trip required seven days of paddling, but also eight days spent at anchor at Seriphos island, during which time winds of 7–8 Beaufort made conditions too dangerous on the water. The return voyage was not made, but it would have faced headwinds and opposing currents. Thus a voyage calculated to take six days based on a daily range estimate for canoes of twenty kilometers (Broodbank 2000, Table 3) took fifteen, and the round trip could easily have taken a month or more. A rowed galley or a sailing ship may have cut this time considerably, but all Bronze Age journeys were prone to delays and dangers imposed by environmental conditions. SCALAR ISSUES IN LATE BRONZE AGE MARITIME NETWORKS

I was motivated to write a book about Late Bronze Age Aegean (i.e., Mycenaean, c.1500–1050 BC) maritime networks by more than twenty years of coastal archaeology in Greece, during which time I observed some disturbing gaps in research that made it difficult to reconcile the local archaeological record with the prevailing discourse on maritime networks (Tartaron 2013). Specifically, I perceived: (1) a lack of interest in, and serious analysis of, localscale social and economic networks; (2) virtually no knowledge about where Mycenaean harbors actually were, or about the communities that inhabited them; and (3) little account taken of coastal change or paleocoastal reconstruction when discussing networks or network nodes. The Mycenaeans were contemporaries of the Hittite New Kingdom, the Egyptian New Kingdom, and the Canaanite cities of the Levant. It is often assumed that the Mycenaean Greeks were great seafarers, interpreting optimistically the empirical evidence of the thousands of Mycenaean artifacts found at coastal sites throughout the eastern and central Mediterranean, from Egypt and the Levant to Italy and Sicily (Figure 4.1). There are also a few iconographic representations of ships in the Aegean. The most famous of these is the so-called Flotilla Fresco, preserved by the volcanic eruption on the island of Thera in the late 17th or 16th century BC (Morgan 1988; Warren 1979). Although not a Mycenaean site or a Mycenaean iconographic tradition, the interest for us is that apart from large, festooned ships in ceremonial procession, there are also a number of small boats: three canoe-like craft pulled up onto a sandy beach, two larger boats lying at anchor in a larger harbor, and another small boat rowing out to meet the fleet. This is virtually the only record we have of such small, local craft in the Greek Bronze Age. Representations of boats found on fresco fragments from Pylos (Brecoulaki et al. 2015) and Iklaina (Cosmopoulos 2011) in Messenia reflect a Minoan iconographic tradition, with differences

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:14, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.005

4.1. Map of the Mediterranean (this page) and Aegean Sea region (following page), with important sites mentioned in the text.

64 Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:14, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.005

4.1. (cont.)

65 Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:14, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.005

66

THOMAS F. TA RTARON

that may indicate changing boatbuilding practices or simply interpretation by Mycenaean artists. Finally, the Linear B administrative tablets from the handful of Mycenaean palaces mention nothing directly about trade at any scale, or about interactions abroad, which can only be inferred from circumstantial internal evidence, such as mentions of Cypriot products and people, or slave women from Asia Minor at Pylos. Some Hittite tablets concern the Ahhiyawa, probably Mycenaean Greeks who colonized Miletus and plied the seas perhaps as far as northern Syria (Cline 1991). In the end, however, there are no actual ship remains from a Mycenaean vessel; the famous Late Bronze Age wrecks at Cape Gelidonya and Uluburun are almost certainly Levantine in origin (Pulak 1998), and we do not know if the distribution of Mycenaean artifacts means that they actually traveled to all those places, for example voyaging all the way to Egypt rather than doing business through Cypriot or Levantine middlemen. All of these vestiges of cross-cultural maritime interaction, and the interpretive problems they present, are well documented. What has been missing is any systematic consideration of maritime networks at the local scale. The missing local scale exposes a serious imbalance, since I feel confident in asserting that in the Late Bronze Age, the vast majority of coastal dwellers rarely, if ever, ventured more than a few tens of kilometers from their communities. Long-distance travel would have been rare by contrast, dwarfed by the density of nodes and connections active at the local scale. These local networks composed vibrant worlds buzzing with activity and connectivity. The lifeline for these communities lay in small-scale networks for subsistence and trade, intermarriage and other social ties. Because of these strong ties, local networks are hypothetically more stable and enduring than very large maritime structures, such as empires and thalassocracies, which tend to be artificially configured and susceptible to rupture with changing political fortunes. By contrast, local networks are easier to maintain from a practical point of view, since distances and environmental obstacles are less inhibitive, and they are often founded on long-established and deeply embedded social ties. On this point it is possible to speak of a locally embedded economy in which economic and social transactions are closely intertwined. It is no surprise that these links can persist even during periods of external domination, and revert to familiar patterns once released from external control (see case studies in Horden and Purcell 2000; Kramer 2016). It is at the local scale that we must look for the true fabric of Mycenaean life, and in view of the coastal topography of the Aegean, with its extraordinarily long coastline and innumerable islands, we particularly need a better understanding of maritime connectivity at this scale. A worthy objective is to write diachronic maritime histories in which the local context is central, while larger political entities with their larger-scale maritime networks move into and out of the picture. Accordingly, my aim was to create and

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:14, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.005

GEOGRAPHY MATTERS

67

systematize a set of concepts to theorize these networks and methods to recover them archaeologically. A MULTI-SCALAR FRAMEWORK

The multi-scalar framework I propose shifts emphasis to the local and microregional scale of “definite places” (Horden and Purcell 2000), which I call “Mycenaean coastal worlds.” I begin with the discovery and investigation of coastal archaeological sites, draw partly upon network theory to model webs of maritime interaction at multiple scales, and seek textual and ethnographic evidence to shed light on the people and their lives. Because the data are so fragmentary, it is an exercise not so much in network analysis as in network reconstruction or “synthesis” (Sindbæk 2013). The outlines of the problem can be seen as methodological—how can we use archaeology, geomorphology, and other research tools to reconstruct parts and characteristics of networks?— and conceptual—how were networks configured and how did they work?

Methodological Problem: Where Were the Mycenaean Anchorages and Harbors? We have little secure knowledge about where the Mycenaean harbors and other landing sites were, partly due to lack of attention to the local scale. Equally important, however, is that the harbors of the Mycenaean period have been rendered virtually invisible on the Greek coastal landscape of today, erased by millennia of geomorphological change and doomed to obscurity by the practices of the Mycenaeans themselves. There is little evidence that the Mycenaeans built durable harbor infrastructure, like quays, jetties, or breakwaters, which seem to be a post-Bronze Age phenomenon in the Aegean. Instead, like the Homeric heroes of the Odyssey, they relied on natural anchorages where smallish boats with minimal draft could be pulled up onto sandy shores, or anchored or moored just offshore in locations protected naturally from winds and waves. Many would have been used only episodically or opportunistically as safe havens, leaving few or no material traces. Altogether, these places have very low archaeological visibility. This problem is exacerbated by long-term coastal change. Coastal zones are among the most dynamic settings on Earth, constantly reworked by long-term, natural processes of erosion, deposition, and tectonics. Over time, coastal features change, appear, and disappear, and these changes can affect the relationship of humans to the sea in profound ways. Although global sea level change since the Bronze Age has not been a major factor in the Aegean, progradational (advancing seaward) or recessive (eroding landward) shorelines can alter coastal configuration dramatically. A prominent example is the massive sedimentation of the great rivers of the Aegean’s eastern coast, silting

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:14, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.005

68

THOMAS F. TA RTARON

in the once great natural harbors at Troy, Ephesus, and Miletus. The processes of plate tectonics are even more insidious because their effects can be so variable and localized. The Aegean sits directly over a subduction zone where the African tectonic plate is moving northward and grinding underneath the Eurasian plate. This results in volcanoes, frequent earthquakes, mountain building, and a complex set of faults underlying the Greek landmass and seabed, creating variable tectonic effects on regional and local scales. A good example at the regional scale is the Corinthia, which is generally tilting downward from west to east: the Corinthian Gulf coast being uplifted while the Saronic Gulf coast is subsiding. Hence, in antiquity the Corinthian Gulf port of Lechaion was apparently put out of commission by co-seismic uplift, while the Saronic port at Kenchreai subsided in a series of earthquakes and was submerged (Noller et al. 1997; Wells 2001). But tectonic effects are also quite localized, because coastal configuration is often controlled by local fault systems. As a result, regional or pan-Mediterranean models of coastal evolution may be invalid for any particular local setting, and experience has shown that locations even a few kilometers apart may have different tectonic histories (Nixon, Reinhardt, and Rothaus 2009). In the search for potential Mycenaean harbors, there is no getting around a proper geomorphological analysis, which would include some or all of the following (Marriner and Morhange 2007): examination of coastal landforms for features like fossil barrier reefs, lakes, lagoons, sandy coastal plains, dunes, or tombolos; geophysical survey to profile the marine basin and detect anomalies potentially associated with harbor activity; underwater dive surveys to investigate anomalies and discover submerged features; and programs of coring for samples of sediment across modern wetlands or alluvial plains. These cores contain microfauna that are sensitive to salinity and temperature, allowing experts to track changing coastal environments, such as marine embayment, lagoon, marsh, or freshwater lake. Sediment grain size and sorting can also give clues to the depositional environment and can indicate human interference in the form of artificial harbor works. Organic material suitable for radiocarbon dating is usually present, from which chronologies for the changes seen in sediment and microfaunal species can be derived. These studies should be closely co-ordinated with archaeological survey and excavation. Survey offers the opportunity to explore coastlines on a large scale, and may lead to the discovery of coastal sites and activity areas to populate coastal worlds. Modern and known historical harbors can be investigated for their histories, and settings of certain kinds, such as natural embayments, deltas, and coastal wetlands, should be targeted.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:14, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.005

GEOGRAPHY MATTERS

69

Conceptual Problem: Building Networks The process of finding evidence for coastal activity and working back to paleocoastal environments is only the first step in reconstructing maritime networks for the Mycenaean period. SNA can provide useful tools for building network models, but as Knappett (2013, 7–8) has pointed out, social network models originated in conditions where all of the actors were known and the problem was simply to analyze nodes and links that could be recovered empirically, and, by representing them in mathematical or graphical form, to come to a better understanding of the structure and operation of the networks. The problem for archaeology, of course, is that the actors are long dead and the material evidence that we use as proxy for their interactions is fragmentary even in the best scenarios, and can be highly ambiguous since their specific actions, motivations, and intentions are mostly lost to us. Modeling maritime networks for prehistoric periods, in the absence or virtual absence of texts, is especially challenging. Nonetheless, the attempts by Broodbank (2000) and Knappett and colleagues (Knappett, Evans, and Rivers 2008; 2011) to do just that for the Bronze Age Aegean have been significant advances in network modeling. Broodbank’s pioneering network model applied a simple proximal point analysis (PPA) to simulate interaction networks in the Early Bronze Age (EBA) Cycladic islands given certain assumptions about the number and location of interacting nodes (in this case, settlements) and certain rules about how they connect. PPA predicts patterns of connections between points distributed in space, conventionally by connecting each point with the three closest to it. The webs formed by these connections generate network clusters, as some points accumulate more links by virtue of their proximity to a larger number of other points. The denser clusters hypothetically mark out interaction “centers” where communication ought to flow most easily. Broodbank addressed the problem of fragmentary site date by placing known sites on the map and then adding points to simulate the growth of population over time. He created four different network models (PPA 1–4) by adding a point for every 150, 100, 75, and 50 square kilometers, and matched the results with the apparent settlement patterns of the Neolithic to EBA Cyclades. The limitations of this PPA were recognized by Broodbank and have been well characterized elsewhere (e.g., Knappett, Evans, and Rivers 2008; 2011). In the model, communities form links with their nearest neighbors because longer voyages are risky and time-consuming with the available propulsion technologies of paddling and rowing. Thus geographic proximity is the structuring principle of network formation. Sites are taken to be of roughly equal size and distributed evenly in space among the islands. The links between them are similarly undifferentiated and non-preferential: one node can connect with any other directly or through a series of short hops.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:14, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.005

70

THOMAS F. TA RTARON

Sea travel is uniform in all directions. While this set of rules and assumptions obviously oversimplifies and distorts the reality of these networks, Broodbank’s PPA was successful because it was designed for the limited geographical world of the Cyclades at a time when boats were propelled by human power alone. Although the fit between the model and the archaeological record is not perfect, Broodbank’s analysis did demonstrate that location and network centrality can be closely correlated under conditions of relatively limited mobility. PPA is unlikely, however, to simulate well eras with large travel ranges, or to translate easily to greater geographical scales. Knappett, Evans, and Rivers (2008) sought to devise a more sophisticated network model with wider geographic and historical applicability. Their model of “imperfect optimization” uses a complex mathematical equation to express the notion that participants in a network tend to strike a reasonable, though never perfectly optimal, balance between the costs and benefits of maintaining maritime connections. To assess the likelihood of connection between two sites, or the connective potential of any single site in a network, each site is coded with several variables, including an estimate of importance based on site size, population, and available resources. These values lead to a set of equations to calculate in quantitative terms the energy balance between the costs of supporting the local population versus maintaining distant links, and the benefits of exploiting local resources versus acquiring distant resources. The connectivity between any two sites is measured by the energy required to maintain contacts, derived as a combination of the physical distance between them and the fraction of effort each devotes to the interaction. To each variable a constant can be attached to assign its relative weight in decisions about connectivity; these constants can be varied to test the implications of different strategies. This “imperfect optimization” is more flexible because it incorporates more of the variables that influence connectivity and allows the weight of each variable to be modified, either experimentally or to reflect current understandings of the archaeological record. Thus the model admirably serves as a tool to explore alternative interpretations of the archaeological data. Some aspects of the model articulate powerfully with emerging concepts in network theory. A central assumption is the network centrality of large sites, like Knossos, which are better connected than smaller sites and attract new connections preferentially because of their greater ability to acquire and control the resources needed to sustain and benefit from overseas contacts. Network theorists Albert-László Barabási and Réka Albert (1999) describe two common properties of networks: continuous growth by the addition of new nodes, and preferential attachment by which new nodes attach disproportionately to sites that are already well connected. A node that acquires more connections than others will accumulate them at an increasing rate, causing the difference in connectivity between the two to multiply as the network grows (Barabási and

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:14, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.005

GEOGRAPHY MATTERS

71

Albert 1999, 511). Further, large communities tend to target each other, creating longer-distance connections and network hierarchies. This “gravitational pull” can aid in linking distant settlements and holding large-scale networks together. The constant addition of new nodes and the creation of shortcuts between well-connected nodes links local clusters into “small worlds” (Watts and Strogatz 1998) and further into large-scale networks in which powerful centers can connect directly over long distances; and certain sites such as emporia that are well positioned in network terms may attract links from the entire sailing universe. These dynamics may help explain the meteoric emergence of Mycenae as a central place during the Shaft Grave era, or the rise of Knossos to an unparalleled position on Crete. (It would also be interesting to synthesize the enormous network connectivity of a true emporium, such as Bronze Age Ugarit in northern Syria.) With advances in seafaring technology and the emergence of large centers in Protopalatial Crete, conditions were set for Aegean-scale networks to grow, requiring a model of greater scope and variability than Broodbank’s PPA. Knappett and colleagues applied the model of “imperfect optimization” to the Middle Bronze Age (MBA) Aegean by adjusting the constants to simulate an incremental increase in the benefits of trade (Knappett, Evans, and Rivers 2008, 1015–1016, Figure 4). At each increment, the links between geographically distant areas of the Aegean— the mainland, Cycladic islands, Crete, the Dodecanese, and Asia Minor— strengthened, and particularly well-positioned sites such as Akrotiri on Thera became crucial “intermediate” nodes in holding the larger network together, in spite of their modest size. Removing these nodes, as when Thera was destroyed in a volcanic cataclysm in the middle of the 2nd millennium BC, can (and did) cause major disruptions in the broader network (Knappett, Evans, and Rivers 2011). Knappett’s model of “imperfect optimization” can be manipulated to simulate admirably enough the kinds of network that plausibly existed in the MBA Aegean, but it carries its own assumptions and simplifications. Most problematic is the challenge of quantifying human behavior and representing it by means of mathematical formulas and graphical output, in view of the fragmentary archaeological record and our limited knowledge of human motivations and actions in the distant past. The model accommodates flexibility in its mathematical variables and constants, but what is the procedure for establishing numerical values for abstract concepts? For example, what is the basis for quantifying the “fraction of effort” that one site puts into its relationship with another? The values of the constants and variables can be changed to simulate different allocations of resources, but is this based on a clear rationale grounded in behavior or are they merely being tweaked until they seem to fit a known historical scenario? If the mathematical outcome of a test run looks rather like what we see in the archaeological record, does that mean that we have

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:14, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.005

72

THOMAS F. TA RTARON

explained something about the past? In other words, can we validate the choices we make in each step of model building, or does it begin to look like a house of cards, piling assumption on assumption? Ambiguities arise and confidence varies as an inescapable result of the fragmentary nature of the archaeological record. One need only read the “technical appendix” (Knappett, Evans, and Rivers 2011, 1022–1023) to appreciate the difficult range of variables for which values must be derived to run the “imperfect optimization” model: population size and density, site size, carrying capacity, “costs” and “benefits,” etc. Often these are simply not available for most settlements in a region under study, at least not without wide margins of error. The problem is acute for Mycenaean archaeology, because the settlement and cemetery data tend to be inferior to those available for the Cyclades and Crete, the areas analyzed by Broodbank and Knappett. Even geographic distance by sea, fundamental to the calculation and among the more quantifiable variables, uses normative figures for daily travel ranges and features an untextured sea. In the absence of reliable quantitative data for these and other categories, calculations of site importance or cost–benefit for local and long-distance interaction are open to challenge. The double quandary of acquiring robust numerical values for structural features and then translating them through mathematical equations into social behavior has led many historians and archaeologists to adopt a cautious attitude (e.g., Malkin, Constantakoploulou, and Panagopoulou 2007, 6). It is not only refining the models that we use, but also addressing critically the quality of our empirical data, that must draw our attention. Concerted effort on both fronts will help to move network analysis into the mainstream of archaeological practice. A MULTI-SCALAR NETWORK MODEL

The network model I devised to address Mycenaean maritime networks is multi-scalar and qualitative. I envisioned networks forming at multiple, nested geographical scales from local to international. Although it is possible heuristically to consider each scale independently, it is crucial to bear in mind the following characteristics: (1) the boundaries are fuzzy, never hard and fast; (2) the shapes and frontiers of the scales are not static, but dynamic, susceptible to change over time; (3) the larger scales intrude upon the smaller, not only as long-distance travelers penetrate local worlds, but also because life at the local and microregional scales responds to, and often is transformed by, events and currents unfolding at larger geographical scales. This dynamism, and the inseparability of local history from larger processes, are well documented in Broodbank’s Cycladic island networks, and by the many case studies in Horden and Purcell’s The Corrupting Sea (2000). The nested geographical scales, from small to large, are the coastscape, the maritime small world, the regional/intra-cultural

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:14, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.005

GEOGRAPHY MATTERS

73

maritime interaction sphere, and the interregional/inter-cultural maritime interaction sphere (Figure 4.2). These are summarized briefly here; for detailed discussion, see Tartaron 2013, 188–203.

Coastscape The local scale is represented by the coastscape, the coastal zone defined by habitation, interaction, practice, and perception (Figure 4.2a). The coastscape is both liminal and central. Where land meets sea, coastal dwellers occupy a liminal transition between contrasting ecological zones and their productive resources. But it is also central and connective, the meeting place for land and sea dwellers. The coastscape includes the following components: (1) the shoreline, the settlement, and the adjacent coastal lowland inhabited and exploited by a maritime community; (2) the connective routes and openings into the interior, following natural paths connecting coast and hinterland—the landward limit is often defined by ridges or mountains, but can also be a cultural barrier; (3) the inshore waters utilized on a daily basis for economic and social purposes; and (4) the visual seascape, the everyday field of view that defines the cognitive horizon in the seaward direction. Ideally, coastscapes are defined by a combination of topography, archaeological survey, and phenomenology of place.

Maritime Small World Maritime small worlds are microregional interaction spheres that form as aggregates of many neighboring coastscapes (Figure 4.2b). They are constituted by habitual face-to-face interaction and cohesion based on shared origin, cultural traditions, language, economic ties, social networks, mutual protection arrangements, and so forth. The relationships among these communities may be hierarchical, orbiting around a powerful polity, but will often be nonhierarchical or heterarchical. Proximity, intervisibility, and ease of travel enhance the cohesion of small worlds. The small world is the scale that dominates maritime interaction. The Saronic Gulf, as described below, is an ideal Bronze Age maritime small world of intensely interacting coastscapes orbiting around Kolonna on the island of Aigina. But if we trace the long-term history of this small world, we see it oscillating between cohesion and fragmentation, affected by internal as well as external forces. The term “small world” is now used regularly to describe networks of interaction, but it currently indicates two distinct and contradictory streams of meaning, so there is a need to clarify what it means and why. In social network theory, small worlds address not geographic distance, but “interaction distance,” measured by the ease and frequency of interactions.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:14, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.005

4.2. Maps representing four nested geographical scales of maritime networks in the Aegean and eastern Mediterranean.

74 Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:14, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.005

GEOGRAPHY MATTERS

75

Small worlds form when the addition of a few key nodes links smaller networks and causes them to grow into larger networks, thus decreasing the barriers to interaction among nodes that were previously “distant” in interaction terms. As the number of intermediate nodes it takes to link two nodes together decreases, interaction distance decreases, creating wellconnected “small worlds” that may be spatially expansive. Some archaeologists have followed this notion to construct small worlds of impressive geographic size. In Andrew and Susan Sherratt’s article entitled “Small worlds: interaction and identity in the ancient Mediterranean” (Sherratt and Sherratt 1998), the term “small worlds” does not appear in the article itself, but the authors are clearly concerned with long-distance trade networks stretching across the entire Mediterranean. Similarly, in a recent book, Irad Malkin (2011, 5) envisions Greek colonization of the 8th to 6th centuries BC as “turning the vast Mediterranean and the Black Sea into a ‘small world’.” On the other hand, “small worlds” grounded in real-world geography and a more literal interpretation of “small” are also established in Aegean Bronze Age archaeology. Cyprian Broodbank (2000) used the term to describe closely spaced, intensely interacting island communities in the Early Bronze Age Cyclades. The fact that he also referred to them as “local worlds” and “local interaction networks” confirms his commitment to geographical scale. For my purposes, the term fits logically in a nested geographic scheme, and I was directly inspired by Broodbank’s scale of analysis, which I find entirely appropriate for the Aegean Bronze Age. I emphasize geographical scale because it matters in the Bronze Age Mediterranean, so it ought to be a real-world measure rather than an abstraction.

Regional/Intra-cultural Maritime Interaction Sphere Voyaging beyond the small world, a crucial transition occurred. Moving beyond “the safe and familiar,” maritime travel was relatively infrequent and was in the hands of specialist sailors and merchants plying the seas in seagoing vessels. They possessed knowledge of sea routes, navigation in a range of conditions, open-sea and coastwise sailing, winds, currents, storms, landing sites en route and at the final destination, and personal relationships with people along the way. This transition finds support in ethnographic examples of recent seafaring in the South Pacific: most young men learn to navigate in local waters for fishing and visiting, but only a few achieve the navigational skill required for long-distance, open-sea voyaging (Feinberg 1988, 88–91). We also see this difference in the locally seagoing farmers of Hesiod’s Works and Days versus the hardened sea captains in the Odyssey, two roughly contemporary written works. As part of Hesiod’s rant against his lazy brother Perses

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:14, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.005

76

THOMAS F. TA RTARON

(Works and Days 645–665), he offers advice about taking to the sea, engaging properly in maritime trade, and storing a boat over the winter. Ultimately, however, he admits that he has actually been on the sea only once when he crossed the narrow gulf to Euboea to compete in funeral games. Hesiod’s knowledge of the sea is no more than conventional folk wisdom and he is familiar only with local-scale maritime activity. In Homer’s Odyssey, on the other hand, we meet captains, helmsmen, sailors, and rowers possessing intimate knowledge of seafaring. Odysseus was a hardened sea captain who voyaged far and wide over the sea. He understood stellar navigation, as we learn when he departs from Calypso’s island and, with her instructions, navigates by the Pleiades, Arctophylax, Ursa Major, and Orion to reach the island of the Phaeacians in eighteen days (Odyssey 5, 270–281). We can mark out a rough “Mycenaean maritime culture region” of the 13th century BC (Figure 4.2c). It was crisscrossed by innumerable sea-lanes, but not by fixed boundaries. At different times, a maritime voyage from Mycenae to Dimini or to Knossos might be an intra-cultural or a cross-cultural journey.

Interregional/Inter-cultural Maritime Interaction Sphere The interregional/inter-cultural maritime interaction sphere involves interactions and networks that extend beyond the Mycenaean maritime culture area (Figure 4.2d). Sporadic visits of Mycenaeans to far-flung lands outside the Aegean seem assured for Cyprus and the northern Levantine coast in the East, as well as for the shores of southern Italy and Sicily in the West. Activity in this sphere is best represented by the non-Mycenaean Uluburun and Cape Gelidonya shipwrecks. FROM THEORY TO PRACTICE: A SARONIC GULF MARITIME SMALL WORLD

With many coastal and island settlements, the Saronic Gulf is an ideal maritime small world because it is well bounded by the enclosing landmasses of the Argolid, Corinthia, and Attica (Figure 4.3). Sea voyaging in the relatively calm gulf waters is considered easy, and a high level of intervisibility promotes intensive interactions in local-scale social and economic networks. The analysis that follows focuses on two Bronze Age sites: Kolonna on the island of Aigina, the largest and most prominent settlement of the Bronze Age Saronic region, and Kalamianos, a small, peripheral coastal settlement located on a rugged segment of the gulf’s western shore. Kolonna dominated this small world for a millennium from about 2500 to 1400 BC, but during that time the small world oscillated between cohesion and fragmentation, primarily because small worlds are

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:14, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.005

GEOGRAPHY MATTERS

77

4.3. Map of the Saronic Gulf and surrounding land masses; the black circles represent known sites of the Mycenaean palatial period of the 14th and 13th centuries BC.

enmeshed in, and respond to, larger networks and historical processes unfolding at larger geographical scales. Beginning in the Early Bronze Age, relations of small coastal settlements such as Kalamianos with Aigina waxed and waned as the attention of Kolonna’s inhabitants shifted into and away from the gulf.

Kolonna Kolonna is a highly complex fortified site, with nine separate urban phases in the Bronze Age, and a center without peer in the mainland region until the political expansion of Mycenae incorporated the gulf into its own larger sphere of influence (Felten 2007). During the Early Bronze Age phases Early Helladic (EH) II (c.2700–2200 BC) and EH III (c.2200–2000 BC), Kolonna grew from a modest settlement of mud-brick houses to one of the most significant urban centers of the Aegean: a densely populated, heavily fortified town with monumental stone buildings and sophisticated town planning (Figure 4.4). Evidence of economic specialization includes the production of pottery and textiles, storage of agricultural surplus, and smelting of copper. The so-called Weißes Haus was a monumental building of the “corridor house” type, like those found at contemporary mainland sites, which possibly played a central administrative role in the community.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:14, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.005

78

THOMAS F. TA RTARON

4.4. Plan of a portion of Bronze Age architecture at Kolonna on Aigina, showing different phases of the fortification walls as well as other features. (After Gauß and Smetana 2007, 58, Figure A)

During EH II, Kolonna was one of many peer sites participating in a time of increasing complexity around the Aegean and an “international spirit” characterized by high maritime connectivity (Renfrew 1972, 451–455). Exotic items with presumably high social value, including bronze daggers and tools, metal jewelry, fine drinking and pouring vessels of metal and ceramic, and marble vessels and figurines, circulated among the coasts and islands of the Aegean Sea. Competition and some level of maritime threat are implied in the appearance of fortifications at many coastal sites. EH III witnessed dramatic changes. Settlements dispersed or disappeared all over the southern Greek mainland and islands. In some areas settlement did not recover until late in the Middle Helladic (MH) period, the so-called Middle Helladic hiatus of up to 500 years. By contrast, Kolonna emerged as the singular, dominant power in the Saronic in the late 3rd millennium. Beginning in EH III, the Aiginetans imported pottery from the Peloponnese, central Greece, and the Cycladic islands. By the beginning of the Middle Bronze Age, these same areas had begun to import Aiginetan table ware, storage vessels, and cooking pots. Ties with Minoan Crete were also strong: alongside Minoan imports a local industry of imitation Minoan ceramics emerged, perhaps operated by resident Cretan craftsmen. This evidence suggests that the Aiginetans shifted their focus to more distant trading partners partially in response to the demographic crash on the Greek mainland.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:14, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.005

GEOGRAPHY MATTERS

79

The Aiginetans maintained their focus on this extra-Saronic network until developments of the Shaft Grave period, most prominently the rise of Mycenae and the recolonization of the interior of the Greek mainland, revived intensive interaction with the Saronic and northeastern Peloponnese starting in MH III and peaking in Late Helladic (LH) I–II. This was the time of greatest cohesion of the Kolonna-centered Saronic world, as indicated by the abundance of imported Aiginetan pottery at most sites in this orbit. There are also signs of emerging competition between Kolonna and Mycenae. The rarity of early Mycenaean painted pottery at Kolonna and in the circum-Saronic region despite easy trade routes may indicate a deliberate exclusionary strategy on the part of the Aiginetans. In the 15th century (LH II in pottery terms), Mycenaean-style pottery spread for the first time into the Saronic region. Still, Kolonna’s pottery export industry declined only after 1400 BC, coinciding with the construction of the first palace at Mycenae itself in LH IIIA. By this time, it appears that Mycenae had begun to expand politically as well as economically, poised to replace Kolonna as the dominant power in the Saronic Gulf. The construction of the palace at Mycenae ushered in the palatial period, and the number of sites in the Saronic almost doubled. These new foundations show strong influence from Mycenae. By LH IIIB1, c.1300 BC, the Saronic region, including Aigina, was fully incorporated politically and culturally into the Mycenaean state. Mycenae had broken apart the old Saronic world and incorporated the region into its own sphere of land and sea connections.

Kalamianos Let us now insert Kalamianos into this narrative. Kalamianos was discovered in 2001 and is the focus of the Saronic Harbors Archaeological Research Project (SHARP). Strikingly, the Bronze Age harbor was situated at the currently exposed location at Kalamianos, and not at the well-sheltered modern harbor of Korphos, an excellent illustration of dramatic change in coastal configuration over millennia, in this case caused by local tectonics (Figure 4.5). A program of paleocoastal reconstruction established the likelihood of a harbor basin with sheltered anchorages to the east and west of the promontory on which the site was built (Figure 4.6; Dao 2011; Tartaron et al. 2011, 570–575). The settlement itself is preserved as a large architectural complex of the 13th century BC (i.e., the later palatial period), with more than fifty buildings, many of them monumental, exposed on the surface as foundations and walls spread over eight hectares. A plan of the architectural complexes reveals two main foci of construction and two phases of enclosure wall. Coastal subsidence has submerged part of the site (Figure 4.7). Whereas Kalamianos reached its acme in the 13th century, the Korphos area has a longer history as a Saronic coastscape. In EH II, Kalamianos was a small

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:14, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.005

80

THOMAS F. TA RTARON

4.5. The modern coastline at Korphos, showing the location of the unlikely harbor at Kalamianos. (Satellite image © 2010 Google Earth.)

4.6. The reconstructed Late Bronze Age harbor basin at Kalamianos. (Courtesy Joseph I. Boyce and the archives of the Saronic Harbors Archaeological Research Project.)

but significant harbor tied into a nascent Saronic “small world” centered at Kolonna. Two major settlements, one at Kalamianos and an even larger one at Stiri high on a coastal cliff, were founded. Obsidian from the island of Melos, 170 kilometers away, was imported as raw nodules and processed at Kalamianos, as was andesite from Aigina, found both as raw nodules and as finished ground-stone implements. In the hinterland, surface survey discovered

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:14, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.005

GEOGRAPHY MATTERS

81

4.7. Plan of architecture exposed on the surface of the Mycenaean site at Kalamianos.

Early Bronze Age stone cairns and enclosures, evidence of a highly humanized and exploited landscape. Beginning in EH III and lasting for more than 500 years, Kalamianos becomes almost invisible archaeologically, like so many other small settlements in southern Greece. Only a few sherds with standard Aiginetan potmarks give evidence of sparse human presence during a time that corresponds to the period in which Kolonna’s attention lay outside the Saronic. Subsequently, in the transitional time (LH II–IIIA) when Kolonna and Mycenae vied for hegemony in the Saronic, Kalamianos was part of a contested periphery, set almost exactly halfway between them. But it was only at the end of that period that we see the first signs, in architecture and pottery later in LH IIIA (later 14th century), that Kalamianos was re-established and interacting with the outside world. By that time, it seems that Mycenae’s economic and political influence had extended to envelop the Saronic Gulf. Sometime around 1300 BC, during the mature stages of the Mycenaean palace period, the urban port was founded and built at Kalamianos, most likely by Mycenae. The port may have served two objectives: as a foothold for maritime economic and military activity in the Saronic, and as a definitive statement of Mycenae’s hegemonic position in the Gulf. This meaning is encoded in the monumentality of the architecture, marking Kalamianos as a second-order center and probably Mycenae’s principal Saronic harbor.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:14, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.005

82

THOMAS F. TA RTARON

During the 13th century BC, the Mycenaeans developed the economic potential of both lowland and upland zones by building a second settlement at Stiri, with expansive views overlooking the Saronic Gulf and fertile basins and hills that could have been used for growing grain and tree crops, and for grazing livestock, just as they are today. From there, a natural pass through the mountains leads to the west, to the interior of the Corinthia and the Argolid, and to Mycenae itself. On the slopes, the Mycenaeans erected an extensive system of agricultural terrace walls to maximize productive capacity (Kvapil 2012). The role of Kalamianos as a harbor can be established by the evidence of imported materials, as well as recent underwater work, mentioned above, which clarified the evolution of the Bronze Age harbor basin by identifying several episodes of tectonic subsidence and the changing configuration of the shoreline. Kalamianos was not a long-lived harbor town, however. We have not found even a sherd of LH IIIC, meaning that, shortly after 1200 BC, the settlement and indeed the region were abandoned. The fate of the harbor seems closely tied to the demise of the Mycenaean palatial system early in the 12th century, as it was to the vibrant maritime life of that system. Korphos–Kalamianos exhibits the hallmarks of a coastscape, with the development of the local zone for habitation, exploitation of the sea, connecting routes to the interior, and a visual seascape opening to the Saronic maritime small world. With mountains inhibiting views to the interior, the daily frame of visual reference for Kalamianotes was the Saronic Gulf. This view incorporates the inshore waters where people of Kalamianos fished and traveled, as well as the visual seascape: not of boundless sea, but of many islands and coastlines, each with their own coastscapes. Kolonna looms in the center of the Gulf. The visual connection and the relative ease of maritime travel to these nearby places bound these communities in a maritime small world.

Conclusion This brief example shows that the Saronic was susceptible to the emergence of a “maritime small world” because visual contact, ease of movement by sea, and moderate distances facilitated connectivity and the experiential sense of a coherent world. Interweaving the stories of Kolonna and Kalamianos over time allows us to move beyond static maps to access the dynamism of a small world oscillating between cohesion and fragmentation over time, responding to internal forces as well as shifting centers of power and demographic trends played out beyond the Saronic. This scale of analysis is important because most Mycenaeans lived and died within these small-scale settings, yet Kalamianos became prominent only in periods of strong interregional connectivity: EH II with its nucleation of population and strong maritime orientation, and LH III

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:14, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.005

GEOGRAPHY MATTERS

83

with the incorporation of large territories by the Mycenaean palaces. In each case, the harbor at Kalamianos and its hinterland were developed to articulate with economic and political systems of greater scope than the Saronic. Adding in the stories of other Bronze Age coastscapes in the Saronic, such as Kanakia on Salamis, or Megali Magoula on the mainland across from Poros, would allow for an increasingly nuanced narrative of the diachronic network patterns in a maritime small world from multiple points of reference. ETHNOARCHAEOLOGY OF THE MARITIME COASTAL COMMUNITY

I turn finally to the last element of the framework, using anthropological techniques to draw out the coastal community and its people, about whom Linear B and even archaeology are almost silent. Ethnoarchaeology studies living traditional societies and technologies as a way to provide analogies and insights on societies of the distant past, including insight into patterns of the archaeological record. Analogy is fundamental to all archaeological research, but a cross-cultural study that attempts to build a bridge between the present and the past carries certain explicit assumptions. We must demonstrate that patterns of material culture and behavior observed in a contemporary society have some analogues in past societies, and we must clarify both similarities and differences: what coastal communities share across the world and what makes each one unique. The ethnoarchaeological component is based on an ongoing program of oral history interviews in Greece and India. Between 2007 and 2009, my colleague Lita Tzortzopoulou-Gregory conducted oral history interviews with elder fishermen and -women from Korphos village as part of SHARP.1 In 2014, I collected similar interviews with Greek colleagues on the Aegean coast of Thrace in the villages of Porto Lagos, Maroneia, and Imeros,2 and with Indian colleagues in the southwestern Indian state of Kerala.3 In each case, we sought out older fishermen and -women who lived and worked in the years before mechanization, federal government intervention, and globalization; in practice before the end of World War II. What follows is merely a sketch of some preliminary results.

Assumptions and Hypotheses The theoretical foundation for my use of cross-cultural analogy is based on three key assumptions: (1) there are certain aspects of engagement with the sea and life in a maritime community that are universal or at least widespread across the world and through time, a kind of “structuring logic” to coastal life; (2) in many places, the lives of people in maritime communities in the period before World War II were more similar to those of the ancient past than to 21stcentury maritime life; (3) oral histories and other observations of behavior and

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:14, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.005

84

THOMAS F. TA RTARON

material culture can provide enlightening ways of thinking about and interpreting the archaeological record of coastal communities, particularly where textual sources are lacking. Tempering these assumptions must be a careful consideration of the inevitable differences among examples drawn from disparate times and places. These differences may include geographical or environmental setting; social, political, and economic organization; and so forth. In some cases the differences may be sufficiently decisive that comparison is rendered difficult or impossible. Based on the world ethnographic literature and the oral interviews in Greece and India, I have developed a series of hypothetical characteristics that seek to bridge ancient and modern “traditional” coastal life in the context of a nested geographic network model. Here I mention and discuss just three of these: (1) there exists an esoteric body of maritime knowledge transmitted from generation to generation in the form of practical instruction and maritime “lore”—that is, a maritime habitus; (2) the coastscape is often a physical space of segregation from the broader society where the attributes of liminality and centrality play out; and (3) coastal dwellers often bear a peripheral status relative to inland centers whose power is based on agriculture, herding, and the acquisition of exotic goods in long-distance trade, with the result that the coastscape becomes a locus for the formation of a distinct identity.

Maritime Habitus In both Greece and India, the systematic transmission of knowledge across generations has played a fundamental role in the survival of the community, its conservatism, and its identity, an excellent example of habitus in Bourdieu’s terms (Bourdieu 1977; 1990). The many aspects of fishing—knowing where and how to fish, navigational skills, maintenance of equipment, marketing the catch—are not easy to learn and master. Prior to World War II, around 90% of the male population of Korphos was engaged in fishing or merchant activities on the Saronic Gulf. Young boys learned by doing, accompanying their fathers and grandfathers on the sea at an early age. The fishermen worked in local waters and preferred the rich fishing ground near Kalamianos. Fishing communities in Kerala (India) exhibit a particularly strong maritime habitus. As in Korphos, they learn their trade from their fathers or other male relatives, sometimes with formal instruction but mainly by observation and participation. As one fisherman observed, They learned exactly the same way from their parents. He has learned from his father. It was handed down exactly from father to son. What your father has done you will learn and will teach your next generation. During

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:14, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.005

GEOGRAPHY MATTERS

85

4.8. Sherd of the 12th century BC from Kynos, showing net fishing strikingly similar to modern net fishing techniques in Greece (gripos) and India (karamadi). (After Dakaronia 2002, 100, Plate 6.)

my time you will learn from me what I have shown you; thus it goes on . . .

Because fishermen depended on the catch for their meager subsistence, deviation from taught practice typically brought correction or even a strong rebuke. In Kerala a typical response went like this: “If something goes wrong the father or someone who is an expert will correct them and instruct them how to do it.” Virtually all respondents spoke of learning basic stellar navigation since they routinely went to fish at night, and at the time there were few, if any, lights on the shore to guide them. They spoke of specific stars. One fisherman from Thrace commented, “You had to learn them even if you didn’t want to.” These factors explain why the fishing life tends to be conservative, with limited scope for experimentation and innovation. So it should not surprise us too much to encounter fascinating parallels such as net fishing, in Kerala called karamadi and in Greece gripos, in which a net secured to extremely long ropes is taken offshore by a small boat, and then slowly dragged back in to shore, represented also in depictions of communal net fishing on painted pottery of the 12th century BC in Greece (Figure 4.8). The division of labor is not uniform, even across Greece or Kerala, for example, but there is a high occurrence of women marketing the fish while the men focus on the equipment and their work at sea. At Korphos and in Kerala, women transported the fish to market by foot, often long distances. In Kerala, women often walked a dozen kilometers or more to market with a heavy basket of fish on their head. At Korphos, one very old woman remembered walking over rugged terrain to bring fish to the inland village of

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:14, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.005

86

THOMAS F. TA RTARON

Sophiko, a few hours by foot. In all cases, women did not normally go to sea, but helped with cleaning and mending the nets and other shore-based tasks.

Liminality and Centrality The coastscape as a place both liminal and central is perhaps best illustrated by Korphos. The smaller group of sea-traders in the village lived a more varied life with better economic prospects than the fishermen. In the early 20th century, Korphos was a major port in a vibrant Saronic maritime small world with nodes on coasts and islands and innumerable links connecting them. The sea-traders purchased fish and local agricultural and forest products and exported them to Saronic markets. There was not a single dominant port in the Saronic, but instead a handful of large, bustling nodes of maritime connectivity. Several interviewees recalled bringing wood, charcoal, resin, and manure to markets at Piraeus, Eleusis, Salamis, Aigina, Poros, Nea Epidauros, and elsewhere. In exchange, the Korphiotes sought food and staples. From Aigina they imported flour and water jugs (still in modern times tempered with the volcanic inclusions that enhanced their performance and made them desirable in the Bronze Age), fruits and vegetables from Nea Epidauros, and foodstuffs and water from Piraeus, among many other items. Upon returning to Korphos, the merchants brought their wares inland, where local buyers acquired them and distributed them further on. The people of Korphos had strong ties of kinship and intermarriage with the inland village of Sophiko, but when prompted about the orientation of the community, the elders were unanimous that the Korphiotes have always thought of themselves as an island people: they looked to the sea for their livelihood, wore island dress, listened to island music and danced island dances, and created networks of interaction with coastal and island people in the Saronic Gulf. They found spouses on Aigina and Salamis, and many emigrated to those islands after marriage. They contrasted their outlook with that of the Sophikites, whom they considered inland, “mountain” people. That they nevertheless maintained close social and economic ties with Sophiko indicates the dual orientation of a maritime coastal community, and exemplifies the inland/coastal symbiosis that is an important feature of the dynamism of coastal life. The Sophiko–Korphos–Saronic system in the early modern period bears the stamp of a microregion in Horden and Purcell’s terms, and Korphos emerges as a coastscape and a maritime coastal community. The people of Korphos forged the link between the terrestrial and maritime worlds. The small-world scale of the Korphiotes is echoed in Thrace and in Kerala. The best fishing near Korphos was only a couple of kilometers from the harbor. At Thracian Porto Lagos, one fisherman mentioned occasionally sailing along the coast as far as Molyvoti, about twenty kilometers distant, to reach good

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:14, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.005

GEOGRAPHY MATTERS

87

fishing grounds. In Kerala, fishing is measured in distances out to the open ocean from shore, and the traditional range was up to about twenty-five kilometers. When queried about storied seafarers of the past, informants emphasized not those who could navigate best or sail farthest, but those who could fish in all weather conditions and bring in the largest catches. These are the people of Hesiod’s world, not Homer’s.

Peripheral Status and Identity Formation The maritime orientation of the Korphiotes marks out a distinct identity in contrast to the inland people of Sophiko. In Kerala, this distinction is even more pronounced and it strongly affects the social networks the communities form. In recent centuries, the fishing folk of Kerala have been marginalized by inland centers of power because of their low status in the caste system and their predominantly Christian religion. They were excluded from power, and ignored in historical and archival texts, save for narrow interests such as taxation or conscription. When we examine the record of long-distance traders who came to Kerala, including Romans, Jews, Arabs, Dutch, British, and Indians, we see them operating in an entirely separate, parallel maritime system, feeding the demands of foreign or inland centers and bypassing fishing villages, in part because their landing sites were too shallow for large cargo ships. This history helps us to understand the different ways that village coastscapes cohere to form maritime small worlds. The early 20th-century Saronic Gulf, with its calm waters, short distances, extreme intervisibility, and lack of a strong political hierarchy, was an ideal incubator for a heterarchical maritime small world with dense nodes and omnidirectional links. The Bronze Age Saronic small world had similar natural properties, but experienced different network configurations due to the strong gravitational effects first of Kolonna and later of Mycenae. The southern coast of Kerala presents a very different configuration, similarly circumscribed but arrayed in a long, linear series of communities sharing common caste, occupation, and religion. Interaction was particularly intense because the villages are cheek by jowl. People visited one another along the coast and intermarried, often finding spouses many villages up or down the coast. Thus the coastal villages were tied inextricably through kinship; their small world was shaped by social networks that were more binding even than economic ones. In spite of their distinct identities and orientations, coastal and inland people came together regularly to do business. Any “traditional society” with a reasonable level of resource differentiation and transportation by foot or animal-drawn cart should have a proliferation of local marketplaces, both formal and informal. Kerala offers an opportunity to observe market dynamics of the recent past. In Kerala there were many markets. Usually a small market

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:14, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.005

88

THOMAS F. TA RTARON

within a kilometer or two of the beach stocked basic essentials. But the women typically had to walk to larger roadside or village markets in the range of five to twelve kilometers to sell fish and obtain necessities such as rice, meat, fruit, and vegetables from inland producers. Hindus, Muslims, and Christians came together peacefully at these markets. We can understand these as neutral, liminal spaces where people of different social groups interacted and exchanged complementary resources. Part of the current research in India is to plot the locations of markets of all descriptions in southern Kerala as a way to identify the spatial attributes of a partly self-organizing system that developed organically to serve the needs of diverse producers and consumers. Bearing in mind the obvious differences, Kerala does at least provide one model that can be tested against the cultural and environmental landscapes of Bronze Age Greece; for example, the geography of Messenia derived from the Linear B tablets combined with intensive surveys already accomplished there. MARITIME COMMUNITIES IN THE AEGEAN BRONZE AGE

We might imagine a similar situation in the Mycenaean palatial period, with small coastal villages and palace centers engaged in networks at quite distinct, but geographically overlapping and sometimes interacting, scales. While we should not overstate the idea of discrete and independent spheres in the Mycenaean economy, note that unlike sheep, goats, cattle, wheat, or flax, fish and marine products were both highly perishable and too widely available to be easily monopolized or converted to profit by a palace. There is scant testimony of coastal activity in the Linear B tablets at Pylos except for references to shipbuilding and conscription of crews presumably for naval ships. Like the agents from the inland cities of Xanthi and Komotini who came to the Thracian coast to buy fish, the palaces may have sent representatives to procure products directly. This may have been part of their responsibilities to collect taxes, recruit rowers, and monitor the movement of exotic goods from palace to shore. After all, we have never explained how exactly the palaces managed to control the safe passage of these goods and to restrict them from wide dissemination. Though apparently not recorded in the Linear B archives, salt may have been a key commodity harvested by coastal people and coveted by the palaces. The coastal regions of western India are locations where salt flats yield prodigious amounts of the resource. Salt is an essential part of the Indian diet, and was of such value that the British monopolized the harvesting and trade of salt, and forbade Indians to engage in them. Salt as a symbol of oppression was so powerful that Gandhi chose disobedience of this law as the focus of nonviolent protest against British rule in 1930, the famous Salt March. The western coast of Greece’s Peloponnese, with many coastal wetlands demonstrated for the Bronze Age, was a potential salt producer. Despite the apparent absence of

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:14, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.005

GEOGRAPHY MATTERS

89

salt as a commodity in the Linear B archives, it is worth investigating in light of the proximity of the palace at Pylos to those coastal wetlands. The coastal–inland symbiosis characteristic of Korphos and Sophiko in modern times may be analogous to the relationship between coastal Kalamianos and upland Stiri in the Mycenaean period. Kalamianos was integrated into a Saronic maritime small world, but at the same time the intervisible site at Stiri provided a link to the interior, including paths west to Mycenae. As the research continues, intriguing similarities and differences are emerging among coastal communities separated by time and space. With these few preliminary observations I hope to have suggested what might be learned from a cross-cultural ethnoarchaeological approach. I would like to stress that this is just one complementary component alongside archaeological and geoarchaeological fieldwork, and study of texts and artifacts, one that aims to address different questions for which our usual approaches fall short. It helps me to think about how coastal people in any setting negotiate their status as simultaneously peripheral, liminal, and central. They form the link between land and sea and the people, products, and ideas that pass between. Observing life among so-called traditional people, who can tell us in their own words about their experiences, opens up new ways of interpreting what we find in the coastal archaeological record, or projecting where we might find features like marketplaces. I return to the basic assumption that these elders lived in worlds more like antiquity than the 21st century, and this is a valuable, but of course not infallible, link. A final point is that we are almost out of time to do this kind of research. These people, along with the memory of their ways of life, will very soon vanish. CONCLUSIONS

Network analysis offers new and enlightening ways to explore the components of ancient maritime networks and the variables that condition their diachronic trajectories. The models and the various iterations run on them may lead to convincing reconstructions of network inception, growth, decline, collapse, and so forth, but some caution is warranted regarding this work in progress. The case of the Aegean Bronze Age, where visionary work has been done, illustrates some of the challenges that lie ahead. I shall make just a few summary points. First, the quality of the empirical data is the limiting factor to the robustness of the inputs to a model, and ultimately therefore to the model’s outputs. Assessment of the data inputs should be an area of the greatest concern and effort. Second, there has not been sufficient attention to multi-scalar approaches that synthesize maritime networks holistically from local to international scales. In particular, the local scale, represented in the model presented here by the coastscape and the maritime small

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:14, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.005

90

THOMAS F. TA RTARON

world, has not been adequately explored. Understanding Bronze Age maritime networks in terms of nested geographical scales is one way to isolate patterns of interaction that are different while also characterizing how the different scales interpenetrate and influence one another diachronically. Third, for prehistory and other cases where texts are unavailable, or wherever there are large lacunae in basic behavioral information, ethnographic and ethnoarchaeological approaches can help to fill the gaps with plausible proxy characteristics. These must be used judiciously with careful attention to environmental and social context, but, as I have suggested, it may be possible to identify certain structuring principles of coastal life that transcend time and space to offer testable scenarios such as market location or fishing catchments. The model I have proposed is a qualitative one, because I am not confident that it is possible to derive sufficiently robust variables and constants for a quantitative model with the Mycenaean data available to me. My hope is that further refinement of the data, coupled with the oral history interviews, might one day make that a possibility. Our colleagues in this volume have done the hard and necessary work of modifying off-the-shelf approaches from sociology and physics for the very different questions and problems of archaeology, and we can expect this work to yield fascinating new insights into the ways that ancient maritime networks operated. NOTES 1. 2.

3.

Many thanks to Dr. Tzortzopoulou-Gregory for allowing me to mine her much more extensive and wide-ranging interviews for this information. I was assisted by Dimitra Adamantidou, Giorgos Makris, and Demetris Brellas. This is a subproject of the Molyvoti, Thrace Archaeological Project (MTAP), a collaboration between the American School of Classical Studies at Athens and the 19th Ephoreia of Prehistoric and Classical Antiquities (Komotini), codirected by Professor Nathan Arrington (Princeton University) and Domna Terzopoulou and Marina Tasaklaki, representing the Ephoreia. The Kerala Maritime Communities Project is a collaboration of the author with Professor Sanal Mohan of Mahatma Gandhi University in Kottayam, Kerala, and Professor V. Selvakumar of Tamil University, Thanjavur, Tamil Nadu.

REFERENCES

Barabási, A.-L. and Albert, R. 1999. Emergence of scaling in random networks. Science 286, 509–512. Bourdieu, P. 1977. Outline of a Theory of Practice (trans. R. Nice). Cambridge: Cambridge University Press. Bourdieu, P. 1990. The Logic of Practice. Palo Alto: Stanford University Press. Brecoulaki, H., Stocker, S.R., Davis, J.L., and Egan, E.C. 2015. An unprecedented naval scene from Pylos: first considerations. In H. Brecoulaki, J. L. Davis, and S. R. Stocker (eds), Mycenaean Wall Paintings in Context: New Discoveries, Old Finds Reconsidered,

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:14, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.005

GEOGRAPHY MATTERS

91

260–291. ΜΕΛΕΤΗΜΑΤΑ 72. Athens: Research Center for Greek and Roman Antiquity. Broodbank, C. 2000. An Island Archaeology of the Early Cyclades. Cambridge: Cambridge University Press. Cline, E.H. 1991. A possible Hittite embargo against the Mycenaeans. Historia 40, 1–9. Conlin, D. 1999. The wind-made world: the nautical geography of the Mycenaean Peloponnese. Unpublished PhD dissertation, Brown University. Cosmopoulos, M. 2011. A group of new Mycenaean frescoes from Iklaina, Pylos. Paper presented at the Mycenaean Wall Paintings in Context: New Discoveries and Old Finds Reconsidered conference, February 11–13, 2011. Athens: American School of Classical Studies at Athens. Dakoronia, F. 2002. Anatoliki Lokrida: i istoria tis mesa apo ta mnimeia kai tis archaiologikes erevnes. In F. Dakoronia, D. Kotoulas, E. Balta, V. Sythiakaki, and G. Tolias (eds), Lokrida: Istoria kai Politismos, 19–100. Athens: Ekdoseis Ktimatos Hatzimichali. Dao, P. 2011. Marine geophysical and geomorphic survey of submerged Bronze Age shorelines and anchorage sites at Kalamianos (Korphos, Greece). Unpublished MA thesis, McMaster University. Feinberg, R. 1988. Polynesian Seafaring and Navigation: Ocean Travel in Anutan Culture and Society. Kent, OH: Kent State University Press. Felten, F. 2007. Aegina-Kolonna: the history of a Greek acropolis. In F. Felten, W. Gauß, and R. Smetana (eds), Middle Helladic Pottery and Synchronisms, 11–34. Ägina-Kolonna Forschungen und Ergebnisse I. Vienna: Verlag der Österreichischen Akademie der Wissenschaften. Gauß, W., and Smetana, R., 2007. Aegina Kolonna, the ceramic sequence of the SCIEM 2000 Project. In F. Felten, W. Gauß, and R. Smetana (eds), Middle Helladic Pottery and Synchronisms: Proceedings of the International Workshop held at Salzburg, October 31st–November 2nd, 2004, 57–80. Vienna: Verlag der Österreichischen Akademie der Wissenschaften. Horden, P., and Purcell, N. 2000. The Corrupting Sea: A Study of Mediterranean History. Oxford: Blackwell. Knappett, C. 2013. Introduction: why networks? In C. Knappett (ed.), Network Analysis in Archaeology: New Approaches to Regional Interaction, 3–15. Oxford: Oxford University Press. Knappett, C., Evans, T., and Rivers, R. 2008. Modelling maritime interaction in the Aegean Bronze Age. Antiquity 82, 1009–1024. Knappett, C., Evans, T., and Rivers, R. 2011. The Theran eruption and Minoan palatial collapse: new interpretations gained from modelling the maritime network. Antiquity 85, 1008–1023. Kramer, M. 2016. Mycenaean Greece and the Aegean: Palace and Province in the Late Bronze Age. Cambridge: Cambridge University Press. Kvapil, L. 2012. The cultivation terraces of Korphos-Kalamianos: a case study of the dynamic relationship between land use and socio-economic organization in prehistoric Greece. Unpublished PhD dissertation, University of Cincinnati. Leidwanger, J. 2011. Maritime archaeology as economic history: long-term trends of Roman commerce in the northeast Mediterranean. Unpublished PhD dissertation, University of Pennsylvania. Malkin, I. 2011. A Small Greek World: Networks in the Ancient Mediterranean. Oxford: Oxford University Press.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:14, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.005

92

THOMAS F. TA RTARON

Malkin, I., Constantakoploulou, C., and Panagopoulou, K. 2007. Preface: networks in the ancient Mediterranean. Mediterranean Historical Review 22, 1–9. Marriner, N., and Morhange, C. 2007. Geoscience of ancient Mediterranean harbours. Earth-Science Reviews 80, 137–194. Morgan, L. 1988. The Miniature Wall Paintings of Thera: A Study in Aegean Culture and Iconography. Cambridge: Cambridge University Press. Nixon, F.C., Reinhardt, E.G., and Rothaus, R. 2009. Foraminifera and tidal notches: dating neotectonic events at Korphos, Greece. Marine Geology 257, 41–53. Noller, J.S., Wells, L.E., Reinhardt, E., and Rothaus, R.M. 1997. Subsidence of the harbor at Kenchreai, Saronic Gulf, Greece, during the earthquakes of A.D. 400 and A.D. 1928. EOS 78, 636. Pulak, C. 1998. The Uluburun shipwreck: an overview. International Journal of Nautical Archaeology 27, 188–224. Renfrew, C. 1972. The Emergence of Civilisation: The Cyclades and the Aegean in the Third Millennium B.C. London: Methuen. Sherratt, A., and Sherratt, S. 1998. Small worlds: interaction and identity in the ancient Mediterranean. In E. Cline and D. Harris-Cline (eds), The Aegean and the Orient in the Second Millennium, 329–342. Aegaeum 18. Liège: Université de Liège. Sindbæk, S. 2013. Broken links and black boxes: material affiliations and contextual network synthesis in the Viking world. In C. Knappett (ed.), Network Analysis in Archaeology: New Approaches to Regional Interaction, 71–94. Oxford: Oxford University Press. Tartaron, T. 2013. Maritime Networks in the Mycenaean World. Cambridge: Cambridge University Press. Tartaron, T.F., Pullen, D.J., Dunn, R.K., Tzortzopoulou-Gregory, L., and Dill, A. 2011. The Saronic Harbors Archaeological Research Project: investigations at Mycenaean Kalamianos, 2007–2009. Hesperia 80, 559–634. Tzalas, H. 1995. On the obsidian trail with a papyrus craft in the Cyclades. In H. Tzalas (ed.), Tropis III: Third International Symposium on Ship Construction in Antiquity, Athens 1989, 441–469. Athens: Hellenic Institute for the Preservation of Nautical Tradition. Warren, P. 1979. The miniature fresco from the West House at Akrotiri, Thera, and its Aegean setting. Journal of Hellenic Studies 99, 115–129. Watts, D., and Strogatz, S. 1998. Collective dynamics of “small world” networks. Nature 393, 440–442. Wells, L. 2001. Archaeological sediments in coastal environments. In J.K. Stein and W.R. Farrand (eds), Sediments in Archaeological Context, 149–182. Salt Lake City: University of Utah Press.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:33:14, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.005

CHAPTER FIVE

CULTS, CABOTAGE, AND CONNECTIVITY Experimenting with Religious and Economic Networks in the Greco-Roman Mediterranean* Barbara Kowalzig

INTRODUCTION

In the ancient Mediterranean, networks of maritime connectivity were constituted in both economic and religious terms. The etiological legend for the cult of Aphaia on Aigina, an island known for its role in trade, offers an insight into how these two aspects intersected, interacted, and often coincided across time and space. The story of Aphaia’s cult begins at Kydonia, Aigina’s only “colony,” in western Crete. Diktynna, a virgin huntress often thought similar to Artemis, fleeing the pursuit of the thalassocrat Minos, throws herself into the nets of local fishermen and travels inside them to Aigina, where she “appears”—or makes herself “invisible”—and turns into the famous goddess Aphaia (Aphaia < phainesthai or < aphanes). On the way, “Netty” (Diktynna < diktyon, “net”) casts her net widely to capture a whole set of localities central to maritime economic mobility, such as the important harbor of Las in Lakonia, Athens, and the islands—as if she had stopped off there on the way. In other * I am grateful to Alex Knodell, Corinna Riva, Ian Rutherford, Alexandra Villing, the editors of this volume, and the anonymous referees for their suggestions and comments; to Kevin Daly and Donald Haggis for leading a trip to the Euboian Gulf in spring 2016 together with a student group from the American School of Classical Studies at Athens, which helped to refine my argument and broaden my vision; to Erika Milburn and Stephan Neitmann for assisting with the editing, and to Kate Morton for creating maps 5.2 and 5.3. Above all I would like to thank participants of the original workshop for introducing me into methods of quantitative network analysis, which made me rethink some of the traditional models developed in the humanities.

93 Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:31:07, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.006

94

BARBAR A KOWALZIG

versions, she takes the epic route instead, starting off from Phoenicia and traveling to Argos and Kephalonia, perhaps tracing the maritime nodes of the Odyssey, before moving on to Crete and then to Aigina.1 This is just one among many religious traditions by which the island of Aigina positioned itself at the center of connectivity by sea through myth and cult, a reflection not least of its stint as a thalassocracy in the late 6th and early 5th centuries (Kowalzig 2010, 2011). Aphaia’s sanctuary itself bears all the signs of a gateway for, and close ally of, those traveling and controlling longstanding and attractive sea routes. Finds on the site include ship graffiti, boats’ eyes, clay ships, and lavish dedications by likely traders. Inscribed Chian pottery sherds are part of their feasting debris, just as is the case for their peers at Naukratis (Sinn 1988, 151–153; Williams 1983, esp. 184–186; Furtwängler 1906, 368 no. 10, Plate 25.2). The combined evidence points to a sanctuary as a node of seaborne connections supported by always evolving and adapting sets of myth and ritual, moving alongside actual goods and people. Though the material and mythical webs do not match precisely, this case nonetheless suggests that maritime, religious, and economic connectivity are profoundly intertwined. It also illustrates how long-distance links originate from a local milieu, here that of the fishermen in western Crete, where a group of Diktynna’s cults embraced a small-scale maritime world.2 This local world is continuously stretched and expanded, often by the addition of merely one far-flung link—something that, as I will argue, reveals how the Greeks may be seen to conceptualize the Mediterranean in terms of networks carried by the sea, and according to patterns identified by network theory. It is the behavior of such maritime “networks,” and particularly the relationship between religious and economic connectivity, that will interest me in this chapter. Before I zoom in on a series of quite specific issues, let me provide some background to how thinking about connectivity, and about “networks” arising from such connectivity, has informed recent work on the Mediterranean, and what religious data may contribute to assessing the dynamics of maritime networks and to network analysis. Studies such as Horden and Purcell’s The Corrupting Sea (2000), and more recently Broodbank’s The Making of the Middle Sea (2013), have significantly changed our vision of the Mediterranean. The Mediterranean with which we work today is a network of interconnected routes of travel without centers and peripheries; it is a web of incessantly interacting but highly fragmented major and minor seaborne ecologies, a world exposed to volatile climates, uneven resource division, irregular productivity, necessitating frequently changing patterns of redistribution and exchange. These conditions make it a thoroughly transcultural space where maritime mobility was as much a strategy of survival and risk mitigation as a source of unlikely opportunities.

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:31:07, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.006

CULTS, CABOTAGE, AND CONNECTIVITY

95

Indeed, in recent years, interest in maritime connectivity has seen a surge (e.g., Tartaron 2013, Leidwanger 2013a; 2013b; 2013c; 2014; Kowalski 2012). However, none of these studies has fully exploited what I consider key to holding all this connectivity together as a lived space: the creative role of religion in carrying, structuring, and managing these dynamic economic interfaces. My broader proposition is that religious practice and imagination, for long periods of Mediterranean history, played a vital role in counteracting, while simultaneously benefiting from, the unpredictability of this maritime world, organizing the sea cognitively and psychologically, socially and economically. Maritime religious networks in particular emerge as strategic social —and thus institutional—responses to economic conditions, offsetting nature’s volatility by regularizing, and stabilizing, mobility by sea. In doing so, they reduce what economists call transaction costs—that is, the cost of acquiring information and building trust in environments of uncertainty. Religious networks enable the aggregation and alignment of information over time, turning this knowledge into “economic knowledge.” The dynamics of maritime networks, in formal and quantitative terms, has been most systematically studied within prehistoric archaeology, beginning with Cyprian Broodbank’s (2000) work on the early Cyclades, the models developed by Knappett, Evans, and Rivers (2008; 2011) for the Mid- to Late Bronze Age, and most recently Tom Tartaron’s lively study on the Mycenaean world (2013). Network science is increasingly used elsewhere to address questions of maritime connectivity and its economic implications in archaeology (Collar et al. 2014; Collar et al. 2015; Brughmans, Collar, and Coward 2016; Brughmans 2010; 2013; Evans, Knappett, and Rivers 2009; Knappett 2011; 2013; 2014; Graham 2006a; 2006b; 2009; 2014; Graham and Weingart 2015). Here too, however, religious practice and imagination—cult, myth, ritual— have largely been left out of current models, in the case of the Bronze Age perhaps due to an insufficient understanding of religion and the working of sanctuaries within a wider connected world.3 In the Iron Age Mediterranean, as the example of Aigina’s Aphaia shows, sanctuaries feature more visibly as centers of gravity and should be considered as a major factor in representing and quantifying maritime mobility. In fact, the nature of Greek cult itself underscores the potential for network thinking. Religious sites do not exist in isolation, but are part of a larger, interlocking system of cults often appearing in clusters, with myths and rituals tying them to one another on a local, regional, or pan-Mediterranean level. More broadly, Greek, and to some extent Mediterranean, polytheism might itself be understood as a network. The basic insight and legacy of French structuralism is that Greek polytheism is profoundly relational, that gods (and cults and myths) do not operate in isolation; their powers emerge and are effective only in relation to one another (Vernant 1983; 1990; Detienne and

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:31:07, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.006

96

BARBAR A KOWALZIG

Vernant 1991). Conversely, the constant forging and reforging of relations— between gods, people, and places—is not just a feature of, but a dynamic intrinsic to, Greek polytheism. The resulting flexibility, malleability, and drive toward innovation of the polytheistic divine develop their own momentum, but also their own agency, their own self-regulation. Cults, myths, and rituals may hence offer new variables to add to the existing parameters in building quantitative network models of maritime connectivity in a historical perspective. Sanctuaries could be thought of as “nodes,” while the myths relating them to one another in a shared story-world function as “ties” or “edges”; similar and shared rituals link shrines into clusters of religious practices, and, in cases of pilgrimage, materialize religious ties between larger cultic hubs and smaller religious nodes. It would be easy enough, if entirely experimental, to give numerical values to the strength of the religious nodes, measuring the expansion and contraction of their catchment areas over time, as manifested in epigraphy and the votive record. Similarly, values could be attributed to the degree of attention paid by the ancient sources to sets of myths associated with individual sanctuaries, which present great variation over time; shrines emerge and vanish from the evidentiary surface. Indeed, Greek myth, so rooted in geography, is a strong indicator of the density of religious connectivity over time; if, for example, 5th-century Athenian drama turns a series of local heroes of the Saronic Gulf into Athenians, adopting them into the city, this surely suggests a certain tightness of religious connectivity, arguably part of Athenian imperial strategy (Kowalzig 2006). Any model based on religious evidence would, of course, be laden with assumptions underlying the evaluation of the data used: are we really justified in claiming that Euripides is a better source than a late antique antiquarian? Should one small votive from a far-flung place be given more weight than a cluster of local monumental dedications? Nonetheless, religious practice and imagination offer a rich and privileged set of data that might ultimately lead to a greater balance between quantitative and qualitative analysis in network theory. This sort of formal network analysis for religious networks could be attempted only as a collaboration between historians and scientists. For the time being, therefore, while keeping in mind these preliminary thoughts, I will proceed with a qualitative approach, and introduce a few basic principles from network theory in order to test their applicability to religious networks. One salient observation in this endeavor is that religious networks are in constant motion, in an incessant process of transformation, adaptation, and realignment. According to social network analysis, networks are always growing, creating new paths, adding nodes and producing shortcuts. Two principles from network theory are of particular interest to us. First, a much-cited process is “small-world dynamics,” enabled through the “strength of weak ties,” a concept popularized long ago by Mark Granovetter (1973) and most

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:31:07, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.006

CULTS, CABOTAGE, AND CONNECTIVITY

97

effectively employed for antiquity by Irad Malkin’s A Small Greek World (2011) for the development of a Greek identity across vast distances throughout the Mediterranean. The basic idea is that local clusters tend to be densely interlinked, made of strong ties; just a small number of “weak,” long-distance links, themselves bound into clusters of strong ties, will quickly achieve greater connectivity by creating a “small-world” effect through network proximity and regardless of actual distance. These few, often random, “weak” ties are vital for the coherence of the local clusters because they facilitate the flow of information; simultaneously they enable the development of very large networks, consisting of many such clusters associated by a few random “weak” links. Networks are continuously expanding in this way (Watts and Strogatz 1998; Watts 2003). As a second principle, network growth is governed by the rule of “preferential attachment,” whereby central and/or already well-connected hubs tend to attract the majority of new nodes (Barabási and Albert 1999, 510 f.; Barabási 2003, 86). For the Bronze Age Aegean, this has been termed the “gravitational pull” of large centers. Such disproportionately well-connected hubs, moreover, tend to be attracted to one another, allowing for long-distance links and large-scale integration (Tartaron 2013, 207–208). Both these network properties—the small-world dynamics and the rule of preferential attachment—can initially assist us in trying to understand the agency of religious networks, ultimately, as I wish to argue, in the service of economic integration. It is, furthermore, helpful to think, with Knappett, Evans, and Rivers (2008), about the relation between costs and benefits involved in maintaining maritime connections across time and space (for an explanation of these basic principles of network theory and their relevance to antiquity, see Malkin 2011, 25–41; Tartaron 2013, 203–211; Collar 2013, 22–37). In what follows, I shall first step back a little and briefly discuss some recent approaches to religious networks of the Greco-Roman world, not necessarily seaborne. I will then turn to specifically maritime religious networks. To anticipate my main point, maritime religious networks can be seen to be grounded in cabotage mobility, often thought the smallest constituent unit of maritime connectivity in the Mediterranean (Tartaron 2013, 203; Horden and Purcell 2000, 365). Such “religious networks of cabotage” will emerge as a fundamental organizing principle of overlapping social, religious, and economic ties over long distances of time and space throughout the Greco-Roman Mediterranean. Far from being parochial, religious networks of cabotage are flexible forms of cognition through which to conceive of, organize, and benefit from a maritime space. They therefore deserve a special place in network analysis. Throughout the chapter, I will provide a series of examples that progressively illustrate the way religio-economic networks foster and build on

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:31:07, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.006

98

BARBAR A KOWALZIG

cabotage mobility. Starting from a group of cults of Artemis spanning a maritime region, the Euboian Gulf, in the first and most elaborate section I will show how such networks operate in a local milieu and are intrinsically bound up with the nature of seafaring, or “maritime knowledge.” The remaining sections will then look into different forms of network dynamics, adding an explicitly economic dimension; they will show how introducing the two simple mechanisms from network theory I have mentioned—small-world dynamics and preferential attachment—help us to see how these religious cabotage networks behave over time. While they virtually underlie most contexts of maritime connectivity, religious networks of cabotage are constantly transforming, intercepted, enlarged, and broken down again. Communities, states, and empires appear to use network patterns to their advantage and avail themselves of the possibilities and flexibilities of network behavior in the service of short-term crisis solution, medium-term economic change, longer-term economic transformation, and even economic growth. RELIGIOUS NETWORKS IN THE GRECO-ROMAN MEDITERRANEAN: THE QUALITATIVE APPROACH

Network parlance has been popular in the study of Greek and Roman history and religion (e.g., Malkin, Constantakopoulou, and Panagopoulou 2009; Taylor and Vlassopoulos 2015), though dataset-driven network analysis is only just beginning to be applied to religious phenomena (Collar 2009; 2013; Blakely 2015). The term “network” has been used for different forms of religious interconnectedness (Eidinow 2011; 2015), but I will focus on only one here. For the purposes of this chapter, “religious networks,” in a loose but not unhelpful sense, are cultic networks formed around a religious center, or simply clusters of cults without an obvious center, tied together by shared myth and ritual practice. The best-studied of these are networks of theoria, dubbed “pilgrimage” for want of a better translation. Here individual cities participate in a central cult by sending official delegations to a common festival, while replications of the central cult appear in the home city. A web of myths and rituals ties hub and participating localities together to form an extended network. The sanctuaries of Apollo and Artemis on Delos, of Apollo at Delphi, of the Great Gods on Samothrace, Zeus at Dodona, and Apollo Pythaieus at Asine in the Argolid are just some of many examples of such theoric cults. In all these cases, the central cult spreads to the members of the worshipping community, forming nodes within the resulting “network.” Reiterations of the same or similar cult, without necessarily having a single center, can also be seen as constituting a religious network, an example being the Apollo shrines tying together the Kopais basin in Boiotia.4 The diffusion of the cult of Asklepios

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:31:07, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.006

CULTS, CABOTAGE, AND CONNECTIVITY

99

might be thought of in such terms—its rapid spread from the later 5th century onwards, which, incidentally, could and perhaps should be investigated through network analysis. This might help to explain the dynamic underlying its peculiar and sudden popularity, where religious change is tied into the growing importance of healthcare provision in an increasingly interconnected ancient world (Scott 2017). The study of such broadly defined “religious networks,” entailing the geographical spread of a cult and its intertwining myths, rituals, and material culture, has provided detailed insights into how religious practice and imagination may affect the formation, interaction, transformation, and dissolution of religious, social, political, and economic communities, both within but especially beyond the polis; interstate relations are often enacted through shared religious practice (Kowalzig 2005; 2007; 2011; Rutherford 2004; 2007; 2009b; 2013b, Chapter 17; Malkin 1990; 2005; 2005; 2007; 2011). These studies lead us to think of networks as mechanisms of communal behavior and collective agency, where the strength and quality of ties determine the relationship between members in a community through contexts of common cult. So, for example, in Singing for the Gods (Chapter 2), I explored the “religious network,” that is to say the worshipping community, centered around the cult of Apollo and Artemis on the island of Delos in the Archaic and classical periods, which consisted of many islands and poleis in the Aegean Sea (Figure 5.1). Analyzing how this network forged, maintained, and transformed itself over time provided a key to understanding the political and economic dynamics of the Aegean world in a period that, among other things, saw the emergence of the Athenian Empire. The religious “network” in this case consisted of the participating communities who themselves worshipped iterations of the Delian gods, such as Apollo Delios and Artemis Delia, whose associated myths and rituals were tied into the central etiological myth that originally established the gods on Delos. These “ties” were spelled out in myth and embodied in ritual practice by delegations involving local civic choruses traveling across the sea to share worship of the Delian gods, merging local and central myths and rituals in their performances.5 In observing a cult community develop and disband over an extended period of time, this qualitative historical analysis of the Delian and other networks raised many of the major issues pertaining to network analysis without necessarily using the theory-specific terminology: among these are the formation, continuation, transformation, and resilience of cultic networks over time; hierarchy and centralization versus self-regulation; collective as opposed to individual agency; investment in interaction and the strength of ties; and so on. For Delos, the evidence is comparatively plentiful, and a full formal network analysis could be attempted with the same data, given the research time

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:31:07, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.006

100

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:31:07, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.006

IA

Delion

Karystos

50

MELOS

DELOS ‘Apollonion’

MYKONOS

THERA

Apollo, Artemis Eileithyia

Apollo Eileithyia Aigale Minoa AMORGOS Arkesine Apollo Delios Apollo Delios Artemis, Eileithyia ASTYPALAIA

NAXOS

Delion

IOS SIKINOS

DESPOTIKO

PAROS

Delion

IKARIA

CHIOS Delion?

Apollo Delios Artemis Delia, Leto, Eileithyia

Delion

SYROS

TENOS

Apollo Eleithyiades

AEGEAN SEA

ANDROS

PHOLEGANDROS 100

SIPHNOS

SERIPHOS

KYTHNOS

Prasiai Apollo Cape Soster Delios Apollo Apollo Koresia KEOS Iulis Apollo Poiessa Apollo Karthaia Apollo

Apollo Delios

A

TIC

AT

Marathon

Apollo Delios Artemis Leto

Phaleron

Athens

Delion

Apollo Dalios

SAMOS

Apollo Dalios

Apollo Dalios Artemis Leto NISYROS

KOS

Miletus

Panionion

Ephesos

Artemis Ephesia

KALYMNOS

5.1. The spread of Apollo Delios, Artemis Delia, and other divinities of the Delian pantheon in the Aegean island world.

0 km

IOT

BO

EU BO IA Eretria Amarynthos

Apollo Dalios

SYME

CULTS, CABOTAGE, AND CONNECTIVITY

101

and the technology. But the charm of formal network analysis for the religious networks of the ancient world may equally lie in creating predictive models for the many more cases where the evidence is scarce, yet using the parameters derived from examples well furnished with data. What then could formal network analysis bring to the field that cannot be supplied by other methods? The answer might lie in either “big data”—analyzing the religious connectivity of entire regions, or a massive cult center, such as Delos, over a very long period —or “excessively small data,” i.e., using the method for predictive purposes where evidence is lacking or insufficient. CULTS AND CABOTAGE

Within the qualitative analysis just outlined, the study of religious networks as understood here has been largely preoccupied with questions such as identity formation between the local and the central, with power relations among the various network members, and with social and political transformation through network behavior. Focusing on the specifically maritime character of religious networks presents different challenges and privileges a discrete dynamic, that is to say the economic dimension of networking processes. The nature of maritime religious networks in Greece is, I propose, closely bound up with the modalities of seafaring. It is a direct response to the patterns, and necessities, of navigation typical of the Mediterranean as they emerge in the wake of the Corrupting Sea—that is to say “cabotage” mobility. This short-haul coastal shipping from “port to port” (or cabo to cabo, as one etymology has it), the continuous movement of small traders and travelers uncontrolled by states and other institutions, according to The Corrupting Sea forms the consistent and uninterrupted background noise to life in the Mediterranean from the Iron Age to the premodern era (Horden and Purcell 2000, 123–172, 342–344). Cabotage in this model is the form of maritime mobility that most directly responds to the Mediterranean’s ecological and climatic fragmentation. Fractured and rugged coastlines, hundreds of islands, countless cliffs and promontories, deep inlets, and natural ports facilitate navigability, but simultaneously create strong and frequent local variations in prevailing winds and currents. Such a situation favors cabotage, i.e., movement in many short segments along the coasts, across narrow channels, from promontory to promontory, while producing countless possible itineraries, requiring flexibility and fostering opportunism in traveling. With this in mind, it is intriguing to note that the maritime religious networks that have been studied are overwhelmingly local or regional, constituting arenas of short- and medium-distance travel. This is the case where networks emanate from a cultic hub (such as Samothrace: Cole 1984, 43–44); but also in cases where no clear point of origin can be identified (such as the group of myths and cults in the western Mediterranean tied to the myth of

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:31:07, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.006

102

BARBAR A KOWALZIG

Artemision Histiaia

Kalapodi

EU

bo

ian

Gu

lf

BO

IA

Ta uri an

Eu

Ch ers on es os

SKYROS

Euripos Strait

Amarynthos BO

Aulis

to

A IOTI

ATTICA

Athmon Mounichia Myrrhinous

Karystos

Halai Brauron

ANDROS

Prasiai

AIGINA

KEOS TINOS

SYROS KYTHNOS

MYKONOS DELOS

5.2. Locations of Artemis’ shrines lining the Euboian Gulf and in Attica. A cult at Karystos is not securely attested; the cult of Artemis at Kalapodi may also have belonged to the group. (Map by Kate Morton.)

Herakles and Geryon: Kowalzig forthcoming). Even in a network as extensive as that of the cults of Apollo and Artemis on Delos, a local concentration can be clearly pinpointed. Delos’s “ties” to neighboring islands, like Keos, Naxos, Paros, and Amorgos, are much stronger than to places further afield (though note a few “weak” links, e.g., to Kos, and, with astounding “ripple effects,” to 5th-century Athens: Rutherford 2009b). In the case of Hellenistic Delos, this has a clear bearing on economic relations (Reger 2004). An interesting instance of a local maritime religious network is the string of Artemis cults lining the Euboian Gulf, and in particular its southern half (Figure 5.2). In this group, overlapping connections between individual shrines and their associated myths and rituals seem to be mapping out a cabotage element that is indelibly intertwined with actual practices of navigation, suggesting a function of such networks in seafaring itself. As usual in antiquity, the evidence for a “network” is far from complete. For example, a type of ritual may be shared between three cults, while myths leave out one of these three cults but instead include two additional ones. The combined evidence is fullest for c.600–300 BC, but this network was likely in operation over a long period

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:31:07, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.006

CULTS, CABOTAGE, AND CONNECTIVITY

103

of time. The connectivity of the Euboian Gulf intensifies in the period 1200–800 BC; the cults themselves all go back to at least the Geometric period, but archaeologically it is not possible to demonstrate their being interlinked this early. What follows is therefore largely a synchronic argument. Beginning with Aulis,6 by the fifty-meter-wide Euripos channel separating the northern from the southern Euboian gulf, there were cults of Artemis all along the northern coast of Attica, at Halai Araphenides,7 Brauron, and Myrrhinous;8 on the southern coast of Euboia, there is the important shrine at Amarynthos,9 and possibly close to the island’s southern tip, at Karystos, where recent archaeological research has revealed another cult possibly of Artemis.10 The shrines of Artemision at the strait between the northern tip of Euboia and Thessaly, and Artemis Mounichia in Piraeus to the south, may also belong to this group.11 The cluster has long attracted scholarly attention without ever having been investigated in its detailed geographical setting (Brulé 1987, 179–179; De Polignac 1995, 92–94; 1998, 26–28; Cole 2000, 477; also Hollinshead 1979). All these shrines are linked by myth, by a series of ritual practices, and by a shared topography, suggesting a local religious “network” forged through the multiplication of the cult’s features in different localities and in different media. Most conspicuously, they are tied to one another by a famous episode from the Trojan War cycle. It is at Aulis that the Greeks’ fleet gathers on the way to Troy; it was here that the lack of favorable winds for the journey led Agamemnon to sacrifice his daughter Iphigenia to overcome the lasting aploia. In many versions, Artemis snatched Iphigenia from the sacrificial altar, replacing her with a deer, and deposited her on the Crimea as priestess of Artemis, conceived as “Tauropolos.” Euripides’ Iphigenia in Tauris (c.414 BC) tells the story of how Iphigenia’s brother Orestes sails up to the Black Sea and brings sister and cult image back to the Euboian Gulf, a famous etiological legend establishing in Attica the cults and manner of worship of Artemis at Brauron, and of Artemis Tauropolos at Halai Araphenides, “as a neighbor to Karystos” (IT 1451) at the southern tip of Euboia, as mentioned above another suspected cult site of Artemis.12 Moving up the Euboian side of the gulf, Artemis’ shrine opposite Aulis, at Amarynthos, was also the scene of an alleged sacrifice by Agamemnon, associated in myth with another of her shrines at Myrrhinous, south of Brauron.13 It is impossible to say exactly when these individual cults became linked, but their connectivity seems to be largely taken for granted by the ancient sources from an early point onwards. Material ties during earlier periods of cult activity are difficult to demonstrate, though the pottery koine in the Euboian Gulf during the Late Bronze and Early Iron Ages (c.1200 BC onwards) includes material from all localities involved.14 But in the historical periods there is important evidence for shared ritual practices, featuring initiation rites for young women and gatherings of the

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:31:07, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.006

104

BARBAR A KOWALZIG

military youth. The famous krateriskoi, appearing from the 6th century onwards and showing girls engaged in ludic ritual activity thought to relate to “playing the bear” in girls’ transition rites at Brauron, have been recovered from several of these shrines, at Brauron, Halai, and Artemis’ cult on Mounichia in Piraeus.15 From the classical period onwards, ephebic ceremonies are attested at Halai and Eretria, and armed dances, pyrrhichai, feature in several of the cults.16 Connections have also been drawn between the ritual architecture of the cults at Brauron and Halai.17 Most significantly, however, all these are littoral shrines, situated in natural harbors (Aulis, Halai, Brauron), or on headlands overlooking the sea (Amarynthos, Mounichia, Artemision, Karystos). Several have distinctly identifiable spaces for the mooring of ships and others lie at small streams, or at the mouth of rivers leading inland.18 The cults’ geographical positions square well with our ideas about Artemis as a goddess mediating passageways, making boundaries permeable, between city and countryside, mountains and plains, and here between land and sea (Cole 2000, 475). Such coastal sanctuaries appear at straits or narrows, river deltas, and often monitor entries to, and exits from, a city’s port, where Artemis is the “guardian of harbors” (Call. Art. 39, 259: limenessi episokopos, limenoskope). Artemis does indeed have a strong interest in navigation in the Euboian Gulf. Literary sources refer to the harbors of Euripos as Artemis’ “favorite,” and describe the coasts of the Euboian Gulf as “dear to Artemis,” suggesting that the necessities of navigation tied these cults together.19 The intense clustering of myth and ritual surrounding Artemis within our group of coastal cults speaks to the importance of the Euboian Gulf in Aegean sailing patterns. A long time ago De Polignac (1995, 92–94; 1998, 26–28) briefly suggested that the shrines ringing the Euboian Gulf functioned as signposts or landmarks, and possibly as stations for ships en route to the northern Aegean. Indeed, the Euboian Sound was a vital passage to northern Greece and the Hellespont on the one hand, and the central Aegean island world and the eastern Mediterranean on the other. Sailing through the gulf, ships would avoid the dangerous Doro Strait between Euboia and Andros, the Aegean coasts of Euboia being among the most difficult to navigate, despite the harbors of modern Kyme and Kerinthos as possible way stations. The height of the Euboian landmass, by contrast, blocked the Etesian winds from the north, prompting ships to use the channel, further protected by the many sheltered bays in the gulf.20 However, the straits at Euripos were among the four most difficult passages in the Mediterranean. They are one of the few locations influenced by tides; the currents through the straits change several times a day at irregular intervals and at a speed of up to eight knots. This is a result of the different tidal states on either side of the straits, and the amassing of water by winds, or, as in the case of Agamemnon’s fleet, their absence (Euripides, Iphigenia at Aulis 9 ff.; 87–88,

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:31:07, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.006

CULTS, CABOTAGE, AND CONNECTIVITY

105

350–352).21 In fact, on closer examination, the episode of Agamemnon and Iphigenia at Aulis turns out to be a sort of navigational myth, portraying sailing conditions in the Euboian Gulf as intrinsically tied to Artemis’ favor. So the aploia (“inability to sail”), narrated in greatest detail in Aeschylus’ Agamemnon (184–204) and throughout Euripides’ two Iphigenia plays, is described in astonishing geographical specificity when the Euripos is “roaring with ebb and surge” (Aesch. Ag. 190–191); there are “light waves at the Euripus” (Eur. IA 813) at the “narrow anchorage” of Aulis (IA 81; 1496); the bay of Aulis is “not inundated,” forming the “protective wing” of the island of Euboia (IA 119–121). The winds are “lacking a breeze for the journey” (IA 352), the pompe, which is the typical word for the divine escort accompanying every sea voyage (see also IA 1319–1324, the breeze now granted by Zeus). Moreover, the social consequences of aploia are described in dramatic detail as “crossblowing, ship-detaining, long-continued” (Aesch. Ag. 147); aploia and winds are depriving the boats of all provisions, “causing famine” (Ag. 188, 193), leading to “harmful idleness,” and “causing delay” (Ag. 147, 196; Eur. IA 804). The bad anchorage (Ag. 192) leads to “waylessness” and “helpless desperation” (aporia, IA 89). It is Iphigenia’s sacrifice that makes it possible to depart at last (“there will be sailing [plous]” Eur. IA 92; 359). Once the sacrifice is completed, the goddess is asked to give plous (IA 1575), and having accepted the offering, Artemis does in fact grant plous (IA 1595–1596). There are indications, though, that this sacrifice may have tied all of our Artemises into the same “navigational” myth. In some traditions, for example, Agamemnon’s offering took place not at Aulis, but instead at Brauron; at Amarynthos, a different type of animal is mentioned, while Artemis at Mounichia is linked to Agamemnon’s hybris.22 More importantly still, the nature of this maiden sacrifice defines the rites performed in the different cults. Iphigenia was sent to Aulis as an adolescent girl on the verge of marriage, in some versions even on the pretext of being married to Achilles. The ritual imagery throughout the tragic plays is that of girls’ transition rites, even quite specifically that of the rituals at Brauron. Approaching the altar, Iphigenia “throws away her saffron robe” (Aesch. Ag. 239), as did the girls at the Brauronian arkteia to mark their imminent change of social roles.23 The sanctuary at Aulis is similarly described as a typical Artemisian cult of adolescent initiation, on a plain (Eur. IA 91) in the middle of a marshy area (leimwn Eur. IA 1462–1463), a meadow full of flowers (Eur. IA 1544). The term proteleia, used for Iphigenia’s sacrifice, typically designates the initiatory rites of a premarriage ceremony. If in Euripides the Greeks “offer the proteleia of Iphigenia to Artemis” (Eur. IA 433 protelizousin; cf. prothyma IA 1310) thinking that she will get married, this echoes the Aeschylean proteleia naon (Ag. 227), literally a prenuptial rite “on behalf of the ships”: we are left with the idea that the ritual practice of girls’ initiation rites, mythically

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:31:07, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.006

106

BARBAR A KOWALZIG

anticipated by the sacrifice of Iphigenia, ensures that there will be sailing, there will be plous. In light of these passages, entangling the cluster of cults in shared myth and ritual practice, Iphigenia’s sacrifice emerges as much more than a tragic story within the heroic cycle; it becomes an etiological myth establishing a rite that harnesses Artemis’ power and protection over human navigation in the entire Euboian Gulf, that important passage for sailing further into the Aegean. To be sacrificed “on behalf of the ships” is not a one-off occurrence on the occasion of the Trojan War, but rather the mythical background against which we should understand the initiatory rites of our network of Artemis cults, which in turn render the Euboian Gulf navigable. It is perhaps significant that human sacrifice, often maiden sacrifice, is frequently a precondition for a successful sea journey in myth.24 Taking a broader perspective, the network of Artemis cults in the Euboian Gulf can thus been seen as a response to Agamemnon’s story of aploia—one of the foundational myths of Greek civilization—as if Artemis’ role were to supply a breeze suitable for sailing the Euboian Gulf in a perpetual echoing or even re-enactment of Agamemnon’s experience in local religious ritual. Worship of this network of cults ensures that aploia turns into euploia (“smooth sailing”). Intriguingly, then, we might say that the mythical motif of aploia ties the southern Euboian Gulf into a knowledge network relating to conditions of navigation, and patterns of both local movement across the southern Euboian Gulf, and coastal navigation up and down the channel on the way to northern Greece, to Athens, the Saronic Gulf, and further afield into the central Aegean. Cult, myth, and ritual appear to reflect the maritime knowledge of a local seafaring world, but also to retain a link to a larger network, that of the Homeric tradition. While all these cults have local civic functions—such as forging civic roles for adolescent girls25—they simultaneously emerge as shareholders in a communal and networked maritime space. RELIGIOUS NETWORKS AND MARITIME KNOWLEDGE

So religious networks of cabotage and maritime knowledge seem to be inextricably intertwined: the mythical episode of aploia locally functions as maritime knowledge, and its pattern of distribution maps out the radius of such knowledge. Comparative anthropological studies, involving popular concepts such as Westerdahl’s “maritime cultural landscape,” suggest that in some societies, maritime knowledge is profoundly local; some 90% of local sailors and fishermen in coastal communities possess and share a deep and detailed knowledge of their immediate surroundings from traveling repeatedly and over many years from early childhood onwards to the same few destinations.26 While in some

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:31:07, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.006

CULTS, CABOTAGE, AND CONNECTIVITY

107

contemporary societies intricate awareness of local sailing conditions may have been more limited (e.g., a five-kilometer range), maritime knowledge in Greek antiquity seems to be “regional,” covering navigation conditions in the Euboian Gulf as here, or similarly in the Saronic Gulf or the Cycladic islands. One might think it to correspond to the sailing distance that can be covered in a day, along the lines of Justin Leidwanger’s recent model (2013b), where seafaring costs are quantified in GIS in order to understand the expanse of maritime connectivity in the light of environmental conditions. I suspect that regional religious networks such as those of Artemis in the Euboian Gulf are grounded in this kind of local maritime knowledge, mapping out the practice and perhaps even the trajectories of cabotage. The physical and environmental conditions of ancient seafaring, especially in Aegean Greece, have recently attracted much attention (e.g., Morton 2001; Arnaud 2005; Kowalski 2012). The details of the religious landscape pertaining to these patterns require much closer investigation, but it is clear that cults functioned as important landmarks and were part of a maritime belief system that took account of practical and psychological concerns alike. Anyone traveling the Mediterranean by boat will experience how Aegean landscapes become two-dimensional beyond a distance of roughly two kilometers. The visibility of shrines on headlands and promontories, on beaches in inlets, or at river deltas has a very practical function in piloting boats in the right direction, avoiding treacherous reefs, currents, and so on, and we should imagine just such a situation in the Euboian Gulf too (Morton 2001, 193–206, 310–317). The maritime knowledge inherent in such navigation shrines and their associated myths and rituals is communal, deep-seated, perhaps even “commonsense” knowledge, as opposed to “expert” knowledge, which, in the current anthropological literature, pertains to more adventurous travel and more distant destinations, mastered only by a select few and often surrounded by an aura of mystery (Tartaron 2013; Arnaud 2014; Broodbank 2000). By contrast, maritime knowledge in the Greek world seems to be created, processed, and maintained through communal religious activities—the telling of myth, the practice of public ritual and cult. Comparatively speaking, such collective maritime knowledge is difficult to access, hardly ever written down, and instead passed on orally from generation to generation and learned from early childhood onwards. More broadly, then, I would argue that in the Greek world, religious practice and imagination are vital in the creation, transmission, and preservation of maritime knowledge: they form mnemonic contexts for its continuation, its transformation, and, most importantly, its broad social diffusion. Cult is thus a significant factor in shaping the strength and wide social accessibility of this network.

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:31:07, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.006

108

BARBAR A KOWALZIG

As comparative evidence shows, the density of maritime knowledge in local communities is very high, reflecting the frequency of journeys in a given area, continuously “re-enacting” the ties, a steady flow of ships and knowledge, likely facilitated by worship at the shrines shared by almost all members of the community (Tartaron 2013, 265–270). Local maritime milieus are thus intensely networked areas encompassing large numbers of people, mapping out short-haul trajectories, and connecting the members of a maritime community to one another. From a network perspective, this may produce an interesting dynamic. These are extremely “strong” ties, forming an intense cluster of links binding together a local maritime landscape in a network that is both very stable and, as we shall see later, very vulnerable to change because of its superior connectedness. While strong, these ties are interestingly cheap: in principle, maritime links, given the greater risk involved in sailing, should be very costly to maintain. From what we have seen so far, we might conclude that inhabitants of local coastal worlds appear to cope with this problem by sharing the transaction costs of maintaining these links: by nourishing a common ritual and story world, they distribute its maintenance amongst participating communities. Local maritime religious networks seem to function as a social institution facilitating and stabilizing maritime connectivity, and act both as knowledge and as trust networks. THE STRENGTH OF WEAK TIES: SMALL WORLDS IN THE CLASSICAL AEGEAN?

In the previous section on the string of Artemis cults in the Euboian Gulf, I adopted a broadly synchronic perspective for the development of a basic model of religious connectivity by sea. I will now turn to the behavior of cultic cabotage networks in real time and in a diachronic perspective, allowing us to identify an economic dimension of a network dynamic. Just like cabotage mobility itself—in the words of Horden and Purcell a “Brownian motion” with ever-shifting, adapting forms of connectivity—religious networks are in a constant process of transformation, adding and subtracting new nodes and ties, just as network theory predicts. Small-scale cabotage networks, while enormously resilient and, once in place, difficult to dissolve—not least owing to their religious entrenchment, which also roots them in tradition and memory—are nevertheless constantly intercepted, enlarged, broken up, and recollapsed.27 Social groups, cities, leagues, and empires engage with network dynamics, experimenting with the possibilities of network behavior in order to offset short-term problems, but also effecting medium- or long-term economic change. If one observes religious cabotage networks develop and expand over time, continuously complemented as they are by additional cult sites and ritual links,

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:31:07, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.006

CULTS, CABOTAGE, AND CONNECTIVITY

109

interesting insights emerge. It is here that the tried and trusted principle of the “strength of weak ties” appears indicative of an emerging small-world system that can be seen to distribute the costs of long-distance economic ties. And further, the law of “growth and preferential attachment,” whereby new nodes gravitate toward association with already well-linked hubs, helps us to understand how adjacent local religious networks join to bring about increased maritime connectivity, and, consequently, raise the possibility of greater economic integration (see p. xxx above for both these principles). Seen from a network perspective and within a network dynamic, then, religious cabotage networks are much more than a reflection of the movement of “proletarians at sea”; they are instrumental in both mentally conceiving and practically facilitating the maritime interconnectedness of the Mediterranean as a whole. Once again, the cult of Artemis in the Euboian Gulf raises interesting suggestions. Euripides’ Iphigenia in Tauris forges a religious link between “Artemis” on the Crimea and the string of Artemis cults in the Euboian Gulf (Figure 5.2) when Iphigenia and Orestes travel with Artemis from the Black Sea to set up her cult at Brauron and Halai Araphenides. I have discussed elsewhere (2013) the detailed mechanism behind this process.28 The appearance of religious connectivity between Attica and the Black Sea in this play, conventionally dated to c.414 BC, should be seen in the context of long-standing Athenian attempts to secure sea routes into the Hellespont and the Euxine to ensure vital supply lines. Athens’ interests in Black Sea resources grew steadily during the Peloponnesian War in the 420s, and especially following the defeat after the Sicilian expedition of 415–413 BC, which precipitated the loss of the overland transportation route for grain from Euboia to Attica, and eventually of Euboia itself, one of Athens’ long-standing grain baskets. Under these circumstances, the connectivity of Attica’s northeastern harbors along the Euboian Gulf became more significant, leading to new links to the northern Aegean. It is with this new, uncertain economic context that the play engages in its forging of a link between Artemis in the Crimea and the cults in the Euboian Gulf (Kowalzig 2013, 207–210).29 The key to such a process lies in a growing convergence in myth and ritual between the cults of the Euboian Gulf and the Crimea unfolding over the course of the play. Once again the sacrifice of Iphigenia is an underlying mythical motif facilitating navigation. Euripides’ drama construes our Artemis, now bearing the epithet Tauropolos, as a maritime divinity in both myth and ritual, as a goddess counteracting danger at sea by steering and rescuing merchant ships in particular (Kowalzig 2013, 183–190). Additionally, we observe cultic ties to the Crimea being forged through patterns of religious commensuration: by collapsing cultural boundaries in rites to be shared by Greeks and Taurians, Artemis’ wellknown powers of “social integration” turn from civic integration to cross-

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:31:07, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.006

110

BARBAR A KOWALZIG

cultural economic mediation (Kowalzig 2013, 190–198). The incorporation in this way of Taurian Artemis into the Euboian network creates a sort of “small world,” where geographical distance is broken down through network proximity, thereby enabling the structuration of maritime travel in a situation of economic precariousness. Intertwining cults and myths configured in a network structure can thus support the flow of knowledge and information, while mutual religious ties may entail the suggestion, obligation, or even imperative of social and economic reciprocity; this construction of a small world enhances economic security in a crisis situation (Kowalzig 2013, 198–210). Approaching this case from a stricter network theory perspective, we may be able to go a step further. It constitutes an interesting instance of a “weak tie” being added to the cluster of maritime cultic ties in the southern Euboian Gulf, possibly triggering a “ripple effect” on the local network that may form the economic basis for Halai and Brauron’s prosperity and intense building activity in the later 5th century BC. At Halai, for example, it seems that the main temple was rebuilt lining a harbor that could accommodate a considerable number of ships (see notes 7 and 18). It is the interaction of myth and ritual across different localities—best, but probably not exclusively, observed in the Euripidean play—that facilitates this network proximity, as if this interplay allowed for telescoping in spatial rather than temporal terms. We may even wonder whether adding this long-distance tie to the preexisting network of myths and rituals does not collapse the set of links between Attica, Euboia, and the Crimea virtually into a cabotage network, where all cults behave as if they were local—as if “cabotage” were a way of mentally constructing religio-economic maritime connectivity. The intensity of mythical and ritual interaction makes up for geographical distance. Unlike the habitual cabotage ties along the Euboian Gulf, however, this “weak” link comes with the immense costs of creating and maintaining a long-distance tie to “foreign” lands, comparatively rare and less easily upheld. The investment in high-profile tragedy, as it were, is an expensive but intelligent choice given its vast audience among Athenians and foreigners, and the possibility of a firm place in mnemonic transmission. Though we will never know the details of the story’s success, let alone its long-term economic efficacy, the coast of northeast Attica never lost its Black Sea connections. PREFERENTIAL ATTACHMENT: MAXIMUM CONNECTIVITY IN THE CLASSICAL AEGEAN?

Let us now take this observation further to show how in this Aegean world adjacent local religious networks overlap to create a sort of maximum connectivity. It is here that the network dynamic of “preferential attachment” may

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:31:07, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.006

CULTS, CABOTAGE, AND CONNECTIVITY

111

prove a useful concept: weak ties gravitate toward already well-connected hubs. Evidence that is broadly contemporary to Euripides’ Iphigenia in Tauris (c.414 BC), that is to say the later Peloponnesian War and an increasingly critical state of the Athenian Empire, intriguingly suggests that the sanctuary of Apollo and Artemis on Delos, which boasted probably the most prolific religious network in the Aegean, may have played a role in tying the Euboian Gulf to the Black Sea, even specifically to the Crimea. Euripides’ Iphigenia in Tauris associates the choral dancing for the cults of Crimean Artemis on the Black Sea, Halai, and Brauron with the theoric choruses performing on Delos (Eur. IT 1094–1105). Delos had a privileged relationship to classical and Hellenistic Chersonesos in the Crimea, which was supposedly founded by settlers from Delos ((Skymnos) 822–830; Chankowski 2008, 108). Moreover, Delos was linked to the Black Sea through the semi-mythical tribute of the Hyperboreans to Apollo and Artemis, which famously traveled all the way from the Euxine to the gods of Delos, stationing at various Aegean localities on the way (Hdt. 4.33). Karystos, “facing Halai” in Euripides’ Iphigenia in Tauris and itself with a “treasury” (oikos) on Delos, was a stop-off point for this tribute; in later tradition, the harbor of Prasiai became this station, in the next bay south of Brauron. It featured a temple of Apollo Delios and is sometimes thought to be better positioned to connect Attica to Delos than the harbors in the Saronic Gulf.30 This port cult may have played a role in linking Brauron to the Delian network. Echoing the associations forged between the Crimea, the Euboian Gulf, and Delos in Iphigenia at Tauris, at Brauron itself there are a number of late 5th-century instances of Delian iconography among the finds.31 Inscriptions dating to the same period may even list the treasures of both Artemis and (an Attic iteration of?) Apollo Delios in a single document.32 All this raises the possibility that in the late 5th century, the new and “weak” Black Sea tie to Artemis in the Crimea drew these northeastern Attic harbors into the most densely interlinked maritime network of myths and rituals at the time of the Athenian Empire, that gravitating around the island of Delos. “Preferential attachment,” where an effective “weak tie” had initially to go through the Aegean hub, may have enabled this dynamic. The interaction of adjacent and overlapping networks poses a whole set of different questions on network dynamics in the ancient Mediterranean that we cannot even begin to address here. But our case does suggest that the broader religious networkedness—or perhaps degrees of economic integration?—of the sea is enabled through interacting sets of small worlds, i.e., clusters of strong ties undergoing massive change in connectedness through the addition of a few, weak links.

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:31:07, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.006

112

BARBAR A KOWALZIG

GLOBAL NETWORKEDNESS?

In trying to understand the effect of constantly added “weak ties” on smallworld systems, it is all the more intriguing to observe the long-term consequences of the Crimean goddess’s moving into the Attic orbit as Artemis Tauropolos and as a divinity whose powers of navigation became increasingly inseparable from the cross-cultural economic encounter. In later periods, this goddess would go on to command transcultural maritime and presumably economic networks, or at the very least a series of important ties. It would be interesting to investigate those in greater detail in their entirety.33 Artemis Tauropolos certainly continues to appear in seaborne contexts of ethnic diversity and economic importance, most significantly as one of the main deities of Amphipolis, high up on the acropolis and in central control of the Strymon river delta—note that Aeschylus refers to the “Strymonian winds” holding up the fleet at Aulis (Ag. 192)—surrounded by the economically resourceful, yet “barbarian,” Thracian hinterland. Alexander allegedly established her here as a divinity symbolizing his Hellenic Mediterranean empire; future Macedonian kings ensured her continuing prominence.34 Alexander, moreover, seems to have taken her as far as Phailaka in the Persian Gulf, ancient “Ikaros,” which cannot fail to evoke echoes of the Artemis Tauropolos at Nas, on the island of Ikaria.35 Much later, in the context of Mithridates Eupator’s economic and political expansion of the Pontic kingdom to the Greek cities around the Euxine in the late 2nd century BC, the goddess of Chersonesos assists Mithridates’ general Diophantos in driving off the Scythians in the Crimean hinterland, prompting his image to be put beside her altar and that of the goddess Chersonesos herself.36 And finally, it is unlikely to be a coincidence that the sanctuary of Diana at Nemi in the Alban Hills outside Rome, known in the literary sources as that of the “Scythian goddess,” carries dedications recalling the final Roman victory in the Mithridatic wars, bringing the Black Sea into the Roman orbit (Guldager-Bilde 2005). Whatever the complex story of Artemis Tauropolos may turn out to be, it has moved a long way from our initial set of cults around the Euboian Gulf where Artemis enabled the passage from land to sea, connecting local territories along and across the coasts of Attica and Euboia. The continuous expansion of the Tauropolos network is in line with the law of continuous “growth and preferential attachment” governing network behavior (Barabási and Albert 1999, 510–511), with ever more and ever more distant and seemingly random nodes adding themselves to the existing group. Even if we are at present inclined to avoid classifying Artemis Tauropolos’ far-flung distributions throughout the entire Mediterranean and the Near East as actively operating “networks” of maritime connectivity, we might still be justified to claim that

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:31:07, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.006

ne

LS

r Rho

U

Rive

GA

CULTS, CABOTAGE, AND CONNECTIVITY

L

U IG

Agathe

Ri

ve

rE

br

IA

Emporion

NS

Nikaia Massalia Tauroeis Rhode

Antipolis Olbia

CORSICA

NS

Alalia

IB

E

R

o

RIA

113

Hemeroskopeion (Dianium)

SARDINIA BALEARIC IS.

MEDITERRANEAN SEA

5.3. The cults of Artemis Ephesia along the Iberian coast and in southern France, as detailed by Strabo. (Map by Kate Morton.)

her diffusion and agency in larger politico-economic contexts ultimately takes its lead from the small-world dynamic in the Euboian Gulf. We might conclude that this dynamic has lasting cultural resonances in the appearance of her cult in ever-intensifying contexts of imperial religious and economic connectivity, from Athens to Alexander and the Macedonians kings to Mithridates VI to Rome. CONSTRUCTING CABOTAGE?

To return to my main claim—that religio-economic networks by sea fundamentally support cabotage connectivity—I would briefly like to address an example of what seems to be a more self-consciously created cabotage network that likely enables economic integration, or at least effects reciprocity and security. This instance is quite different from my original case study, where the network dynamic is a result of collective behavioral patterns. Irad Malkin has on various occasions discussed the diffusion of Artemis of Ephesos in the “Phokaian Mediterranean” centered around Massalia and its colonies, and the eastern coast of Spain (Figure 5.3; Malkin 1990; 2011, Chapter 6). The distances here are greater than in the Euboian Gulf, but remain

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:31:07, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.006

114

BARBAR A KOWALZIG

within the orbit of short- to medium-haul travel. The evidence for this network is almost entirely literary (Strabo 3.4.6–9; 4.1.4 ff.), and cannot be dated with certainty; however, for the purpose of my argument, this gives further weight to the all-pervasiveness of the cognitive patterns of cultic networks of cabotage in the Mediterranean longue durée. In this instance, a maritime religious network seems almost artificially produced, exploiting the specific cultic associations of a cabotage network in terms of the density and stability of its links. As with our Artemises around the Euboian Gulf, religious topographies of navigation are formed by assimilating the cults and making them appear part of the same visual and ritual network. According to Strabo (4.1.4), following an oracle, Artemis arrived with the Phokaians directly from Ephesos. A much-respected woman, later to become her priestess, was told in a dream to bring along an aphidryma (“sacred object”). Thus a xoanon (“archaizing or old image”) of the goddess traveled as the Phokaians’ hegemon (“leader”), referring to Artemis’ protective leadership in navigation and hoped-for delivery to the shore. Strabo is keen to emphasize that her cults were established at Massalia and its colonies in the greater Rhone delta and further to the east, and on the Iberian coast at Rhode, Emporion, and Hemeroskopeion. Strabo, as Malkin notes, is fascinated by the multiplication, the reiterations of the same cult in all the Phokaian cities, worshipping the same divinity, represented by cult images of identical shape, and strictly keeping to the nomima of the metropolis. For Hemeroskopeion and Emporion, Strabo singles out the rocky promontories on which her cults were set (ἐπὶ τῆι ἄκραι), landmarks indicating the entrance to ports, a liminal area between sea and coast . . . promising protection . . . distinctly identified with the community that established it . . . When connectivity is formally expressed in having the same goddess, situated in the same maritime, promontory position, and welcoming the same sailors in the same region . . . participants have become aware of the network as such. (Malkin 2011, 198)

It is well known that Massalia particularly, and perhaps the Phokaian cities in general, capitalized on their maritime connectivity to establish themselves as hubs for the trade, for example, in wine, metals, and pottery in the Western Mediterranean. Almost none of them had significant amounts of cultivable land much before the Roman period—something that also seems to be a topos in the literary sources (e.g., Str. 6.1.1; 4.1.5; Iust. 43.3.5 etc.). Massalia, whose soil, according to Strabo, was too poor to grow grain, did not begin intense cereal cultivation until its expansion probably no earlier than the late 3rd century (Dietler 2010, 115). Instead, the city, “trusting the sea,” from the 6th century onwards, benefited from the possibilities of its connectivity for the distribution by sea or river of its wine and pottery, and relied on the Iberians for its grain

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:31:07, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.006

CULTS, CABOTAGE, AND CONNECTIVITY

115

supply.37 According to the literary tradition, the Massalians founded their network of colonial cities as “fortresses” (epiteichismata) protecting the coast against incursions from inland barbarians who continuously waged war against them (Str. 4.1.5; Iust. Epit. 43.3.4–5.10). Agathe (Agde) and Olbia in the greater Rhone delta are among these, as are the cities further east, Tauroeis, Antipolis, and Nikaia, surrounded by the Sallyians and Ligurians of the southern Alps. Also belonging to this group are Hemeroskopeion, Emporion, and Rhode on the Iberian coast.38 All Phokaian cities were small and relatively isolated outposts in a sea of hinterland, potentially a source of risk but presumably also of lucrative business opportunities. Against this background, it is a distinct possibility that Ephesian Artemis, like Artemis Tauropolos, ties these coastal outlets into a network of economic interdependency and obligation across the sea. Moreover, again not dissimilar to Artemis Tauropolos, it is significant that she also takes on aspects of crosscultural commensuration, integrating coastal worlds with their hinterlands. In the case of the cult at Hemeroskopeion on the Iberian peninsula, we learn that the Massalians even taught the Iberians to worship their goddess “in the Greek manner” (Hellenisti, Strabo, 4.1.5).39 Strabo stresses the mixed nature of the city of Emporion (3.4.8), archaeologically confirmed from early on, while Livy knows of Iberians aiming to sell their goods at Emporion (Livy. 34.9.9). With a view to the apparently vital role of the Iberians in the seaborne economy of Massalia and its colonies, this suggests that knowledge of Artemis Ephesia’s rites enables the Iberians’ inclusion in the goddess’s network of maritime connectivity, tying them into the institutional arrangements that Greek religion seemingly has to offer in matters of trade. Meanwhile, for the Phokaians, the cult may have provided a gateway to the hinterland. In view of the argument presented earlier, we might suppose that fostering a common identity through strict and unremitting duplication of the cult seems a strategic response to the specific local economic situation (no cultivable land, surrounded by potentially hostile peoples), where survival depends on a functioning maritime network spreading the transaction costs of its maintenance among participants. Tight networks of religious identity tend to provide powerful imperatives of interaction and mutual obligation. Strabo’s aphidrumata of Artemis placed in cities interacting like “fortresses” might appear to express hostility toward the locals, while actually ensuring the success of the network by incorporating the economic potential of the “barbarian” hinterland. It is a pity that we cannot date this network, but this may actually be to our advantage: it supports the argument of a continuous network dynamic—albeit with different strengths and intensity of links at different times—in Mediterranean landscapes characterized by cabotage religiosity. Even if the tight integration of the religio-economic network of Artemis Ephesia in the West is not Archaic or classical but a tradition created by Strabo and/or his context with a view to

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:31:07, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.006

116

BARBAR A KOWALZIG

Rome’s 1st-century relations with the city that had now become Massalia, we might still be able to identify the same pattern of network behavior: if the tradition of Artemis Ephesia in the West emerged in the late Hellenistic and Roman period, Diana Ephesica, as she eventually appears on the Aventine, might then be seen as a “weak” tie commanding vital economic and even integrative power for the group of cities of Phokaian origin in the Roman period. CONCLUSIONS AND PERSPECTIVES

Networks have no beginning and no end, and this is no different in the specific case of overlapping religious and economic networks carried by the sea that I have begun to disentangle here. I have only used an exceedingly small body of data to make a fundamental argument that needs to be tested against a much broader set of examples. I could continue to dissect cultic networks in this manner; this would yield fascinating insights into individual cases of interaction, longevity, and path dependence across extended periods of time, and create a significantly more complex picture. But I hope that even with a small selection of cultic networks, I have been able to identify some patterns of their operation, and suggest that they had an active role in shaping an often economic dynamic that, while always changing, is ultimately founded on a few basic principles. These principles are, first, a grounding of maritime religious networks in the cognitive patterns of cabotage navigation and produced by local maritime knowledge which in its turn is expressed, and continues to live on, in sets of local cults, myths, and rituals. Cultic cabotage networks are local or regional, collective, and appear to be the basic way of organizing religious networks by sea. I propose that these also function as networks of economic knowledge and of extremely strong, since perpetually repeated, ties, whose transaction costs are shared among participating communities. Furthermore, a small-world dynamic with lasting consequences—“ripple effects”—seems to be common. The intrusion of a long-distance tie creates a weak, often costly, link transforming the connectivity of the existing network, as when a version of Artemis is “added” from the Crimea to the cabotage cluster of Artemis cults in the Euboian Gulf. Small-world dynamics generally collapse actual distances, but we might add that they can also alter spatial relationships within a given local geography. Apollo Pythaieus at Asine in the eastern Argolid, for example, draws on an area traditionally encompassing the coastal cities of the Argolic and Saronic Gulfs. When Argive Apollo Pythaieus adds himself, apparently hijacking the network, this does not break down geographical distance, but may well inaugurate a new maritime orientation of Argos in the course of its social and economic reformation during the 5th century BC (Kowalzig 2007, 110–128).

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:31:07, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.006

CULTS, CABOTAGE, AND CONNECTIVITY

117

A third, repeated dynamic lies in the development of new associations or shifts in emphasis of a networking deity. This is true of Artemis “Tauropolos” whose agency in transcultural economic encounter by sea by the 1st century BC seems to have lost the foregrounding of her navigational skills; instead she seems to have become an authoritative conduit for cross-cultural connections and even trade in the context of the Hellenistic and Roman empires. Fourth, inter-network integration also seems to be common, enabled among other things by the network mechanism of “preferential attachment.” As we have seen, the network of Artemis develops a link to the catchment area of Delian Apollo and Artemis, thereby integrating the Euboian Gulf into the more consequential network tying the central Aegean islands to imperial Athens. But this particular dynamic can just as easily become network competition. The “original” Artemis of Ephesos at Ephesos itself provides an example of this configuration, in her role as a hub of a consciously cultivated Ionian identity in Asia Minor. I have argued elsewhere that at the time of the early Athenian Empire in the first half of the 5th century, the Delian network, encompassing especially Athens and the Aegean islands, competed with that of Artemis of Ephesos, which sought to align the Ionians of the mainland in western Anatolia. At the height of empire in the 420s BC, the cult of Apollo and Artemis seems to have absorbed Ephesian Artemis’ catchment area, and it was the Delia that was restored and reinvented as a pan-Ionian festival including Athenians, islanders, and the cities of Asia Minor (Kowalzig 2007, 110–128). In the same vein, the aforementioned “Hyperborean tribute” coming from the Black Sea to Delos retains unmistakable traces of an earlier tie with the network of Artemis of Ephesos linking the cities and coastal islands of Asia Minor to the Euxine. The 5th-century reinterpretation of the Hyperboreans as servants of Apollo and as delivering their tribute through Athenian harbors (rather than across the northern Aegean islands) testifies to the reworking and reorientation of the network, investing in different ties, abandoning some and strengthening others (Kowalzig 2007, Chapter 2, 110–128).40 Incidentally, and this brings us back to my initial reflections on how cultic networks might be described through quantitative analysis, this entire complex set of interlinking networks in the Aegean might be much better understood through a full and formal network analysis: this could graphically map out the diachronic dynamics of emergent weak and strongly networked areas and their intersections. Intersecting networks increase zones of dense connectivity, and we could see clearly how interlacing cabotage networks might assist longdistance connectivity. To revert one last time to my central point, I maintain that cabotage religiosity is a fundamental organizing principle for religious networks in the Greek world and for the manner in which they interact, and often coincide, with economic

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:31:07, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.006

118

BARBAR A KOWALZIG

networks. While cabotage networks are a basic unit of religious and economic movement precisely because of the strong ties linking them, larger political entities such as poleis, empires, or perhaps even ethne regularly interfered in cabotage density in different ways, turning its efficacy to their advantage. Lycian Aperlae, for example, had a cabotage economy lasting for almost a millennium revolving around the murex shell; whichever larger nearby hub managed to gain access to this network would have invested very well (Hohlfelder 2000). Conversely, the possibility, recently raised by Justin Leidwanger (2013c), of a local cabotage community in late antique Cyprus investing in long-distance links without going through a medium-tier gateway suggests that cabotage communities can also produce effective and important distribution networks.41 The small-world dynamic, characterized by the strength of weak ties and repeatedly affecting our cultic cabotage networks, might, at least to a degree, account for a problem presented by The Corrupting Sea: Horden and Purcell understand that long-distance trade is no more and no less than an “outgrowth” of cabotage (Horden and Purcell 2000, 145, 366, 443), but do not offer a direct explanation of the relationship between localized exchange and the development of long-distance trade. Addressing this problem in the context of the trading networks of the Geniza Jews from Foustat in the Nile delta in the medieval period, in her recent book Jessica Goldberg points out how the Geniza merchants engage in local and regional exchange networks, “but this was interwoven with those zones that reliably produced agricultural surpluses,” “macro-ecologies” such as the Nile valley, the Ifrīqiyya plain, and parts of Sicily (Goldberg 2012, 23–24; 343–344). The efficacy of a small-world network explains the curious relation between the local density of links and the weak outlier tie; while cognitively speaking “collapsed,” these small worlds constitute what we call long-distance trade. The question remains why cabotage is so central, and so entrenched. Horden and Purcell (2000, 150, 155, 172) invoke the fundamental rhythms of connectivity in the Mediterranean, the continuous “background noise” of redistribution beyond the reach of corporate organization. Perhaps we can, after all, bring a little bit of history and change into this picture. James Whitley recently remarked on the duration of the Dark Ages in Greece, that is to say of a society without states and institutions, for much longer than in comparable cultures (Whitley 2017; forthcoming). Instead, immediately after the supposed demise of the Late Bronze Age centers, it is coastal life that seems to thrive, not least in enclosed maritime spaces, such as the Euboian Gulf, where this continues into the Early Iron Age.42 It is a curious coincidence that the little we know about LBA continuity of religion through the Dark Ages is often tied to later cults of Artemis mediating between local coastlines and hinterland (e.g., at Abai (Kalapodi) and Ephesos). In fact, several sites of our Artemises in the Euboian

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:31:07, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.006

CULTS, CABOTAGE, AND CONNECTIVITY

119

Gulf present archaeological traces dating back to the Late Bronze Age even though there is no sign of continuity of actual cult (Knodell 2013). Rooted as it is in the specific Mediterranean geography, cabotage mobility, as a coastal, communal, redistributive economy of interlaced regional networks, perhaps plays its part in conferring some stability on this world. If the Mycenaean palaces were not rebuilt, this may be a sign less of a power vacuum than of a perceived strength and endurance of this burgeoning local connectivity. We might, finally, speculate about the stability of the cabotage system, which was a communal, non-hierarchical network of production, distribution, and consumption within coastal worlds as the dominating social entities, rather than landed city states. Cabotage was most suited to the Mediterranean landscape, the most natural way of being. The “Dark Age” world was in no rush to develop lasting institutions and centralized powers. City states, long-distance travel, and cross-cultural contact as drivers of an economy of luxury and the exotic; the interlinkage of political and economic power; city states’ concern with their territory—all these may be seen to intercept, counteract, and manipulate this continuous “background noise.” However, if cultic networks of cabotage can be seen to underlie many large-scale Mediterranean networks throughout Greco-Roman antiquity, this is because cabotage is more than a form of economic mobility. It is a culture, and as such is prolific and changeable, constantly reinventing the networkedness of the Mediterranean in ever new configurations. Abbreviations AD AK FGrH IG IOSPE2

LIMC Rhodes-Osborne

SEG SIG3

Archaiologikon Deltion Antike Kunst F. Jacoby. Die Fragmente der griechischen Historiker. Leiden: Brill, 1923–. Inscriptiones Graecae. Berlin: Reimer, 1873–. B. Latyshev. Inscriptiones antiquae orae septentrionalis ponti Euxini Graecae et Latinae. St. Petersburg: Iussu et Impensis Societatis Archaeologicae Imperii Russici, 1916. Lexicon Iconographicum Mythologiae Classicae. Zurich, Munich, and Düsseldorf: Artemis, 1981–2009. P.J. Rhodes and R. Osborne (eds). Greek Historical Inscriptions: 404–323 BC. Oxford: Oxford University Press, 2003. Supplementum Epigraphicum Graecum. Leiden: Brill, 1923–. W. Dittenberger (ed.). Sylloge Inscriptionum Graecarum. 4 vols. 3rd edn. Leipzig: Hirzelium, 1915–1924.

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:31:07, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.006

120

BARBAR A KOWALZIG

NOTES 1.

2.

3.

4.

5.

6.

7.

8.

Call. Art.189–200; [Scylax] 47; Diod. Sic. 5.76.3–4; Str. 10.4.12–13; Paus. 2.30.3; 3.14.2; Ant. Lib. Met. 40; [Verg.] Ciris 294–309; Hsch. α 8533 s.v. Aphaia. See also Nonnus, Dion. 33.333–45; Aristoph. Vesp. 368; Ran. 1359 and scholia. Cf. Eur. Hipp. 146; 1120–1130; IT 127. Cults of Diktynna on sea-routes: e.g., Las: Paus. 3.24.9 (cf. at Sparta: Paus. 3.12.8; Liv. 34.38); Athens IG ii2 4688 (2nd century BC); 13163 (imperial); Astypalaia xii.3, 189 (3rd century BC). For the fusion of Diktynna–Britomartis see next note; Aigina’s economic motivations for settling Kydonia (Str. 8.6.16): Stefanakis 1999. At the Diktynnaion on the Rhodopos peninsula overlooking the bay of Chania (Hdt. 3.59); the port of Phalarsarna, Polyrhenia, and Lisos on the southern coast; a month name is attested at Aptera: Guarducci 1935; Boulotis 1986; Sporn 2001; 2002, esp. 277–280. A full discussion of Diktynna would include her overlaps with Britomartis, whose cults cover central and eastern Crete, and her name being a frequent epithet of Artemis. I omit these complications for the sake of clarity; they do not affect my point here. Though see Renfrew, Boyd, and Ramsey 2012; Rutherford 2013b on mobility in Mycenaean religion; Bendall 2014 on a possible “cultic tribute” traveling from the mainland to Ephesos. I have studied many of these networks at length in 2007 (Delos, Dodona, Asine, Kopais basin in Boiotia); 2005 (Samothrace). For theoric networks generally see now Rutherford 2013b. Such choral songs (paians) for performance on Delos are known to have been commissioned, for example, for Athens; for the islands of Euboia, Keos, Naxos, and possibly Paros and Kos (Pind. Pae. 5; Pae. 4; Bacch. 17; Pind. Pae. 12; fr. 140b; frr. 33a–d); and for many more unknown communities. For the combined archaeological and literary record of Aulis, Halai, and Brauron, Hollinshead 1979 is still useful. Aulis: Threpsiades, AD 17, 1961/2 [1963], B’ 137–144; Schachter 1981–1994, i.94–98; the temple is held classical, underneath which there is an unrelated Geometric, possibly earlier, structure; a Mycenaean settlement has been located nearby. Halai: Travlos 1988, 211–212, Figures 264–268. The sanctuary has now been published well by Kalogeropoulos 2013, assembling the materials from the excavations by Papadimitriou (1956–1957); Liangouras and Alexandri (1970s), and using Travlos’s unpublished drawings (Plates 5, 8–11). Ceramic finds at the site go back to the Bronze Age. Tile fragments may indicate a temple of c.470–460 BC; the currently visible foundations date to a (re)building c.430–410 BC, identified with the temple mentioned in Euripides’ Iphigenia in Tauris. It was part of a much larger sanctuary complex (Kalogeropoulos 213, i.221–227; ii.41–43), featuring a “long building” (?5th century, perhaps a stoa), surrounded by other buildings; a so-called “small temple” to the south (dated to the second half of the 5th century, but finds begin in Late Geometric). At 200 by 80 meters the sanctuary must have been one of the largest in Attica, but most of it has now disappeared in the sand. The site continues into the Roman period. Drinking and feasting seem to be significant features until the Hellenistic period; a military aspect emerges in the 4th century. For a synthesis see Kalogeropoulos 2013, i.469–524; ii. 62–70; on the cult’s broader religious and historical context see now McInerney 2015. Brauron’s archaeology is known largely through preliminary reports, listed in Ekroth 2003, 60 n. 3; synthesis in Papadimitriou 1963; Travlos 1988, 55–57 Figures 53–91; Ekroth 2003, 102–118. There are some 9th-century sherds, but the shrine itself developed from the late 8th century; great quantities of votives appear from 700 BC, possibly related to an architectural structure underlying the remains of a late 6th/early 5th-century Doric stone building; a new temple was built after 480 BC and the destruction by the Persians; in the last quarter of the 5th century the whole sanctuary underwent a major transformation, involving an open peristyle court with adjacent rooms and stoa. Unpublished inscriptions are said to refer to further buildings. For a recent study on the Mycenaean tombs at Brauron

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:31:07, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.006

CULTS, CABOTAGE, AND CONNECTIVITY

9.

10.

11.

12.

13.

14.

15.

16.

17.

121

see Papadopoulos and Kontorli-Papadopoulou 2014. Myrrhinous: IG ii2 1182, 19–21; Paus. 1.31.4; schol. Aristoph. Av. 872a. See Vivliodetis 2005, 39–40; 51–56; 131–143. Amarynthos: Str. 10.1.10; Knoepfler 1988 (citing relevant literary and epigraphic sources). The sanctuary, at the foot of Palaioekklesies hill near modern Amarynthos, has finally been identified definitely by dedications and stamped tiles (see Reber et al. 2017; Fachard 2017); recent work has revealed buildings from the 6th century onwards; notably a 4th-century portico surrounding the sacred area: AK 56, 2013, 100–107; 57, 2014, 127–133; 58, 2015, 143–151, and earlier, esp. AK 47, 2004, 2004–2016. Note the Artemisia: IG xii.9, 189=Rhodes-Osborne no. 73, later a pan-Euboian cult: Liv. 35.38.3; Paus. 1.31.5, already discussed in Nilsson 1906, 238–240. The site was an important local center in the LH IIIC period: AK 54, 2001, 144–159, esp. 152–153. Karystos: Crielaard following recent work at a sanctuary at Plakari (personal comment); reports: Crielaard 2013. At the opposite end of Euboia, at Histiaia, there was a cult of Artemis Proseoea, with a small temple Plut. Them. 8.2–3; Hdt. 7.176; IG xii.9 1190. Mounichia and Brauron are typically compared to one another on the grounds of similar mythical aitia and dedications, including the krateriskoi, and location by the sea: Papadopoulou 2014, esp. 111–115; Palaiokrassa 1991. The myth is first known from the Cypria: Procl. Chrest. 135–143 Severyns. The reason for Artemis’ anger varies: hybris towards Artemis: Procl. Chrest. 136–137; omission of sacrifice: Apollod. Epit. 3.21–22 (cf. 2.10); killing of a deer sacred to Artemis and boasting about it: Soph. El. 566–574; Call. Art. 263–264; Hyg. Fab. 98.1. For the full list of passages see Brulé 1987, 262 180–181 with nn. 14–16, 180–203 for a comprehensive discussion of the myth. Sacrifice at Amarynthos: see note 22 below. Myrrhinous: Artemis’ epithet is Kolainis, which Hell. FGrH 4 F 163 (= schol. Ar. Av. 872a) compares to the epithets Mounichia in Piraeus and Brauronia in Attika; cf. Paus. 1.31.4–5 who adds that neighbouring Athmon worships Artemis Amarysia (agon in IG ii2 1203, 4th century). Kydathenaion has a cult of Artemis Amarysia: IG i3 426, 67–69 (end of 5th century). Late Helladic III C: Sherratt 2006, 118–120, lists similarities between the pottery at Lefkandi, Amarynthos, Eretria, Aliveri, Psachna in central Euboea, as well as, for example, Kynos, Eleon, Anthedon, Lithosoros, and Kastro Volos on the mainland. Note also Aghia Irini, on Kea; possible similarities with the inland sites on Kalapodi, Eutresis, Orchomenos; cf. recently Knodell 2013, 200–203. LH IIIC Middle koine: Crielaard 2006, 282–284; for the period Proto-Geometric period koine: Lemos 2002, 212–217; for Late ProtoGeometric to early archaic: Mazarakis Ainian 1998, 187–188; 208. Kalogeropoulos 2013, i. 477–478, ii.63, now compares LH IIIC pottery from Halai to that at Kalapodi and the Early Iron Age pottery to that of Eretria (i.479). In addition, there are krateriskoi from the shrine of Artemis Aristoboule/Agrotera, the Athenian agora and the caves of Pan at Eleusis. See especially Kahil 1977; 1981; Scanlon 1990; Parker 2005, 234–235, also lists more recent finds. Kalogeropoulos 2013, i.272–266; ii.48–49; K90–101 adds non-figurative krateriskoi to the collection from Halai. Bronze mirrors dedicated at both Halai and Brauron bear similarities to one another (c.575–450 BC): Kalogeropoulos 2013, i.434–436; 500; ii.58, 66 M64–65. Several children’s statues in the museum of Chalkis resemble those at Brauron. Ephebes: Pyrriche and pyrrhichistai: Halai SEG 34, 103 (4th century); Amarysia IG xii.9, 191.58 (late 4th century); IG xii.9, 236.45–46 (c.100 BC); Histiaia: IG xii.9 1190. At the Amarysia, there was an armed procession: Str. 10.1.10; cf. IG xii.9, 189 = RhodesOsborne no. 73; contests in Σ Pind. Ol. 13.159, IG xii, 234, 236, both Hellenistic (with Chankowski 1993). Possible boys’ initiations at Halai: Kalogeropoulos 2013, i.103–112; ii. 19–25; ephebes running naval contests at Mounichia (2nd/1st century): IG ii2 1006, 1011, 1028, 1030; Meritt 1947, 170, no. 67 (+IG ii2 1009); Aulis: IG vii 2450; xii Suppl. 646, though these are 3rd-century AD. Kalogeropoulos 2013, i.505–506; ii.67–68.

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:31:07, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.006

122

BARBAR A KOWALZIG

18.

19.

20.

21. 22.

23.

24.

25. 26.

27. 28.

Aulis is generally thought the harbor of Mycenaean Thebes with the active ports of Mikro Vathi, Vathi Limen, and perhaps even Glypha. Hollinshead 1979, 21–23; 25–27, for the Greek fleet at Aulis, later a gathering point for military ships (e.g., Il. 2.303–304; Liv. 45.27.9). Halai provided space for a large number of ships along the long sandy coastline and anchorage at Cape Veleni (Kalogeropoulos 2013, i.37–39; ii.13–14); in Roman times it was a harbor for the marble transport from Marmari in Euboia: Str. 10.1.6. For Brauron as a harbor and stop-off point to Delos see Cole 1998, 29; it is called “coastal” by Euphorion ap. Σ Aristoph. Lys. 645a; it lies in the marshes by the Erasinos river; cf. the draining of the marshes at Eretria/Amarynthos (IG xii.9, 191). Aulis, Brauron, and Amarynthos had springs. Call. Art. 188 “the harbors of Euripus have been your favorites”; Statius Achill. 448–450: “Aulis climbing the Euboean sea with jutting cliffs and the embankment of her long chine, shores very dear to the mountain-ranging goddess”; Artemis’ shrine at Aulis is “on the opposite coast of Chalkis” (IA 1492–1493 antiporon). Interestingly, in 7th-century Stesichoros (PMG 215), Iphigenia turns into Hekate, goddess of the crossroads; cf. Hesiod fr. 23a, b MW, where Iphimede following her sacrifice becomes Artemis Enodia, often equaling Hekate, and Hekate “of Artemis.” Kalogeropoulos 2013, i. 130–132; 134–136; ii.26–28 sees the maritime aspect in the cult through the identification of Orestes and Pylades with the Dioskouroi in Eur. IT. Theophr. Vent. 28; Cary 1949, 73–74, for the route through the Euboian Gulf, the importance of which is further suggested by the recent underwater survey of the southern Euboian Gulf, where to date twenty-six shipwrecks have been found: Diamanti and Vlachaki 2015; Koutsouflakis 2013; Koutsoflakis et al. 2012. On the northern Euboian Gulf as a separate “small-world” system, see Tartaron 2013, 283; Knodell 2013, 286–287. A helpful survey of the coastlines of the Euboian Gulf, listing the many protected bays, can be found in Knodell 2013, 99–128. Morton 2001, 44–45, with bibliography; 85–105 on sailing conditions in straits generally. On the rapid wind changes see the Mediterranean Pilot 1916, 141–152. Brauron: schol. Aristoph. Lys. 645a; Phanodemos FGrH 325 F 14. Agamemnon supposedly sacrificed a ram without horns to Artemis (Kolainis) at Amarynthos: schol. Aristoph. Av. 872a, citing Euphronios and Callimachus fr. 200b Pf; cf. Ael. NA 12.34. Artemis at Mounichia “limenoskopos” and Agamemnon: Call. Art. 259, 263. For the saffron robe (krokotoi) see Aristoph. Lys. 644–645 and schol. 645 a and c; occurring also in the temple inventories of Artemis Brauronia in Athens: IG ii2 1518, 77 ff. (4th century), but see Linders 1972, 45. For the arkteia see Parker 2005, 232–248, with bibliography; Gentili and Perusino 2002; Sourvinou Inwood 1988, 132, on the Aeschylus passage. The arkteia at Mounichia: Papadopoulou 2014. E.g., to settle Lesbos, Smintheus’ daughter had to be sacrificed at sea (Conv. sept. sap. 20 = Mor. 163b–d; 984e); Menelaos sacrificed two boys to enable him to leave Egypt: Hdt. 2.119. See Wachsmuth 1967, 310–311, for fuller references. Human sacrifice and successful seafaring are tightly associated in Phoenician religion: Brody 1998, 82–3. And possibly even delineating territory: De Polignac 1998. Tartaron 2013, 265–270, summarizes the results so far of a compelling ethnographic project in the Saronic Gulf, involving the oral transmission of maritime knowledge; 107–124 on maritime knowledge in the ancient Aegean and the applicability of the maritime landscape approach to the Mediterranean (see Westerdahl 1992; 2011, with earlier bibliography). Five-kilometer range of maritime knowledge: Malkin, in discussion. On resilience and path-dependence of cabotage mobility see Horden and Purcell 2000, 128. Blake 2013; 2014 discusses path-dependence in southern Italy. Note that while the goddess of the Crimea for the Greeks was Artemis (Tauropolos), local sources refer to her as parthenos (“maiden”), picked up in Eur. IT. 1230; cf. Hdt. 4.103. Tauropolos was the goddess identified with Iphigenia’s sacrifice by the 5th century: e.g., Soph. Ajax, 172; Aristoph. Lys. 447.

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:31:07, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.006

CULTS, CABOTAGE, AND CONNECTIVITY

29.

30.

31.

32.

33.

34. 35.

36.

37.

123

For the importance of Attica’s northeastern harbors see Braund 2007a; for the role of Euboia after 411 BC see Garnsey 1988, 119 ff. The land route from Euboia through the port of Oropos was blocked after the Spartan occupation of Dekeleia in c.413 BC (Thuc. 7.28.1); in 411 BC Euboia revolted from Athens. See Moreno 2007, 117–126, on costly sea transport from Euboia to Athens traveling around Sounion; forts along the coast framed our Artemis cults: at Rhamnous, 5.2 kilometers north of the dangerous Cape Agia Marina (already in 450/446 BC); Thorikos, controlling the windy channel of Makronisos, was fortified in 413 or 410 BC; Sounion in c.413–411 BC. Paus. 1.31.2, with Beschi and Musti comm. Marathon and Phaleron had further cults of Apollo Delios: Matthaiou 2003; Peppas-Delmousou 1988, 328–329; cf. Cole 1998, 29, on Brauron as an important harbor and stop-off point to Delos. E.g., the relief “of the other gods” interpreted as showing the Delian triad (Brauron Museum no. 1180 (420–410 BC(?), see Venit 2003). The krateriskoi (note 15 above) typically feature a palm tree, associated with the birth of Apollo and Artemis on Delos. An altar, of c.400 BC, depicts the Delian triad as part of a procession of gods; the Delian triad figures on another votive relief: Van Straten 1995 no. R 74 = Brauron 1152. Artemis and Apollo appear jointly on several vases: e.g., Brauron nos. 32; 79. Apollo by himself: Brauron 425 (A 54) = Kahil 1963, 25 no. 54; 56 and Plate 14, 2. Votive pinakes showing Artemis Tauropolos riding a bull: Mitsopoulos-Leon 1997, 366 no. 1 = LIMC 700–701. See Peppas-Delmousou 1988, 329–332. A relief in the museum of Brauron, c.420 BC, with an unpublished inscription, lists the funds of Artemis, followed by the mention of “sacred monies of Apollo,” often held Delian Apollo: SEG 37.31 (with previous bibliography). The relief shows Artemis being approached by five male figures, thought the five treasurers of the “other gods”: PeppasDelmousou 1988, 330, with 339, Figure 2; Kahil 1990. See also the inventory SEG 37.30 (416/15BC), possibly referring to a transfer of the treasury of Brauron to Athens, equally mentioning funds of both Artemis and Apollo at Brauron: Peppas-Delmousou 1988, 330–346. Karystos, singled out as lying opposite the shrine of Halai by Eur. IT 1451, had an oikos on Delos; Herodotus links Karystos to the Hyperborean tribute: Hdt. 4.33.2; IG xi.2 144.A.88; the new finds at Amarynthos show dedications to the Delian triad: Reber et al. 2017. Graf 1979, Guldager Bilde 2003, and Kowalzig 2013, 193–196, all explore a number of the many Artemis cults with this title and/or a connection to Iphigenia spread across the entire Mediterranean, but none looks at the complete list. Alexander at Amphipolis: Diod. Sic. 18.4.5; Macedonian kings: Livy 44.44.4, mentioning a Thracian threat; SEG 31.614 (also mentioning Thracians; cf. 31.615); 33.499. Phailaka/Ikaros: Str. 16.3.2; Arr. Anab. 7.20.3–6 (Alexander’s cult foundation); Dionys. Per. 608–611; Ael. NA 11.9. SEG 38.1547 (late 4th/early 3rd century) with Roueché and Sherwin-White 1985: 4–6; Caubet and Salles 1984: no. 201, pp. 96, 125, 149, Figures 44, 64 (2nd century). Artemis Tauropolos at Nas on Ikaria dating back at least to the 6th century: Str. 14.1.19; Clem. Alex. Protr. 4.46.3; Arnob. Adv. Nat. 6.11; Papalas 1983; 1992, 27–31, on the archaeology. IOSPE i2, 352 (late 2nd century), with Braund 2007b; Saprykin 2007 on Mithridates’ economic policies. A couple of “inland” cults in northern Anatolia also seem to belong to this context, possibly in places relevant for Mithridates’ economic survival and part of a broader religious policy: see Guldager Bilde 2003. Large collective storage fields of underground pit granaries (silos) with a capacity of up to 10,000 liters each have been found close to the coast, chiefly in the Aude basin and Catalonia (e.g., at Mas Castellar at Pontós, near Emporion), which suggests grain production for export to the Greek cities or elsewhere from at least the 5th century until after the Roman conquest: Dietler 2007, 258. Str. 4.1.8 mentions the fertile soil in this area. For similar storage facilities in France see Morel 2006, 383.

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:31:07, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.006

124

BARBAR A KOWALZIG

38.

39. 40. 41.

42.

Archaeologically this policy might have happened from the late 5th century onwards. Agathe (Agde, late 5th century); Olbia (mid-4th century); Antipolis (Antibes, 3rd century) Tauroeis (late 3rd century); Nikaia (Nice, late 3rd/early 2nd century): Domínguez 2004. With the exception of Agde, these cities appear to have acquired strips of land not under native control only in Roman times: Dietler 2007, 252. The ritus graecus otherwise typically belongs to a Roman ritual context: Scheid 1995; 1998. Kowalzig 2007, Chapter 2, 110–128. In this light, one might speculate whether initiatives such as Peisistratos’s fostering of the link of Athens to Brauron were, among other things, designed to break through such coastal worlds, benefiting from their connectivity while focusing their economic energies inwards, towards the center, here the Athenian polis. E.g., see Crielaard 2006 (esp. 278); Dickinson 2006; Knodell 2013, 178–198; for the LHIIIC pottery koine see note 14 above; for Euboia as a dynamic EIA environment see the recent contributions in Mazarakis-Ainian, Alexandridou, and Charalambidou 2017.

REFERENCES

Arnaud, P. 2005. Les Routes de la navigation antique : itinéraires en Méditerranée. Paris: Errance. Arnaud, P. 2014. Ancient mariners between experience and common sense geography. In K. Geuss and M. Thiering (eds), Features of Common Sense Geography: Implicit Knowledge Structures in Ancient Geographical Texts, 39–68. Zurich: Lit-Verlag. Barabási, A.-L. 2003. Linked: How Everything Is Connected to Everything Else and What It Means for Business, Science, and Everyday Life. New York: Plume. Barabási, A.-L. and R. Albert. 1999. Emergence of scaling in random networks. Science 286 (5439), 509–512. Bendall, L. 2014. Gifts to the goddesses: Pylian perfumed oil abroad? In D. Nakassis, J. Gulizio and S.A. James (eds), KE-RA-ME-JA. Studies Presented to Cynthia W. Shelmerdine, 141–162. Philadelphia: INSTAP Academic Press. Blake, E. 2013. Social networks, path dependence, and the rise of ethnic groups in preRoman Italy. In C. Knappett (ed.), Network Analysis in Archaeology: New Approaches to Regional Interaction, 203–221. Oxford: Oxford University Press. Blake, E. 2014. Social Networks and Regional Identity in Bronze Age Italy. Cambridge: Cambridge University Press. Blakely, S. 2015. Human geography, GIS technology, and ancient mysteries: a case study from the island of Samothrace. Getty Research Journal 7, 133–141. Boulotis, Chr. 1986. Diktynna. In C. Ackermann and J.-R. Gisler, LIMC III 1, 391–394. Braund, D. 2007a. Black Sea grain for Athens? From Herodotus to Demosthenes. In V. Gabrielsen and J. Lund (eds), The Black Sea in Antiquity: Regional and Interregional Economic Exchanges, 39–68. Aarhus: Aarhus University Press. Braund, D. 2007b. Parthenos and the nymphs at Crimean Chersonesos: colonial appropriation and native integration. In A. Bresson, A. Ivantchik and J.-L. Ferrary (eds), Une koinè pontique: Cités grecques, sociétés indigènes et empires mondiaux sur le littoral nord de la mer Noire (VIIe s. a.C.–IIIe s. p.C.), 191–200. Bordeaux: Ausonius. Brody, A.J. 1998. Each Man Cried Out to His God: The Specialized Religion of Canaanite and Phoenician Seafarers. Atlanta: Scholars Press. Broodbank, C. 2000. An Island Archaeology of the Early Cyclades. Cambridge: Cambridge University Press. Broodbank, C. 2013. The Making of the Middle Sea: A History of the Mediterranean from the Beginning to the Emergence of the Classical World. London: Thames and Hudson.

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:31:07, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.006

CULTS, CABOTAGE, AND CONNECTIVITY

125

Brughmans, T. 2010. Connecting the dots: towards archaeological network analysis. Oxford Journal of Archaeology 29(3), 277–303. Brughmans T. 2013. Thinking through networks: a review of formal network methods in archaeology. Journal of Archaeological Method and Theory 20(4), 623–662. Brughmans, T., Collar, A., and Coward, F. (eds) 2016. The Connected Past: Challenges to Network Studies in Archaeology and History. Oxford: Oxford University Press. Brulé, P. 1987. Fille d’Athènes: La religion des filles à Athènes à l’époque classique: mythes, cultes et société. Paris: Les Belles Lettres. Cary, M. 1949. The Geographic Background of Greek and Roman History. Oxford: Oxford University Press. Caubet, A., and Salles, J.F. 1984. Le sanctuaire hellénistique B 6. In J.F. Salles (ed.), Failaka: Fouilles françaises, 1983, 73–156. Paris: de Boccard. Chankowski, V. 2008. Athènes et Délos à l’époque classique: recherches sur l’administration du sanctuaire d’Apollon délien. Athens: Ecole française d’Athènes. Cole, S.G. 1984. Theoi Megaloi: The Cult of the Great Gods at Samothrace. Leiden: Brill. Cole, S.G. 1998. Domesticating Artemis. In S. Blundell and M. Williamson (eds), The Sacred and the Feminine in Ancient Greece, 27–43. London: Routledge. Cole, S.G. 2000. Landscapes of Artemis. Classical World 93(5), 471–481. Collar, A. 2009. Network theory and religious innovation. In I. Malkin, C. Constantakopoulou, and K. Panagopoulou (eds), Greek and Roman Networks in the Mediterranean, 144–157. London: Routledge. Collar A. 2013. Religious Networks in the Roman Empire: The Spread of New Ideas. Cambridge: Cambridge University Press. Collar, A., Brughmans, T., Coward, F., and C. Lemercier. 2014. Analyser les réseaux du passé en archéologie et en histoire. Les nouvelles de l’archéologie 135, 9–13. Collar A., Coward, F., Brughmans, T., and Mills, B.J. 2015. Networks in archaeology: phenomena, abstraction, representation. Journal of Archaeological Method and Theory 22(1), 1–32. Crielaard, J.P. 2006. Basileis at sea: elites and external contacts in the Euboean Gulf region from the end of the Bronze Age to the beginning of the Iron Age. In S. Deger-Jalkotzy and I. S. Lemos (eds), Ancient Greece: From the Mycenaean Palaces to the Age of Homer, 271–297. Edinburgh: Edinburgh University Press. Crielaard, J.P. 2013. The Plakari Archaeological Project: preliminary report on the second field season (2011). Pharos 19(2), 35–56. De Polignac, F. 1995. Cults, Territory, and the Origins of the Greek City-State. Trans. J. Lloyd. Chicago: The University of Chicago Press. De Polignac, F. 1998. Divinités régionales et divinités communautaires dans les cités archaïques. In V. Pirenne-Delforge (ed.), Les Panthéons des cités: Des origines à la “Périégèse” de Pausanias: Actes du Colloque organisé à l’Université de Liège du 15 au 17 mai 1997, 23–34. Liège: Presses universitaires de Liège. Detienne, M., and Vernant, J.-P. 1991. Cunning Intelligence in Greek Culture and Society. Trans. Janet Lloyd. Chicago: The University of Chicago Press. Diamanti, E., and Vlachaki, F. 2015. 3D recording of underwater antiquities in the south Euboian Gulf. International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences XL-5(W5), 93–98. Dickinson, O.T.P.K. 2006. The Aegean from Bronze Age to Iron Age: Continuity and Change between the Twelfth and Eighth Centuries BC. London: Routledge.

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:31:07, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.006

126

BARBAR A KOWALZIG

Dietler, M. 2007. The Iron Age in the western Mediterranean. In W. Scheidel, I. Morris, and R.P. Saller (eds), The Cambridge Economic History of the Greco-Roman World, 242–276. Cambridge: Cambridge University Press. Dietler, M. 2010. Archaeologies of Colonialism: Consumption, Entanglement, and Violence in Ancient Mediterranean France. Berkeley and Los Angeles: University of California Press. Domínguez, A. J. 2004. Spain and France (including Corsica). In M.H. Hansen and T.H. Nielsen (eds), An Inventory of Archaic and Classical Poleis. Oxford: Oxford University Press. Eidinow, E. 2011. Networks and narratives: a model for ancient Greek religion. Kernos 24, 9–38. Eidinow, E. 2015. Ancient Greek religion: “embedded” . . . and embodied. In C. Taylor and K. Vlassopoulos (eds), Communities and Networks in the Ancient Greek World, 54–79. Oxford: Oxford University Press. Ekroth, G. 2003. Inventing Iphigeneia? On Euripides and the cultic construction of Brauron. Kernos 16, 59–118. Evans, T., Knappett, C., and Rivers, R. 2009. Using statistical physics to understand relational space: a case study from Mediterranean prehistory. In D. Lane, D. Pumain, S. van der Leeuw, and G. West (eds), Complexity Perspectives in Innovation and Change, 451–479. Berlin: Springer Methodos Series. Fachard, S. 2017. Recent research at the sanctuary of Artemis Amarysia at Amarynthos (Euboea). Archaeological Reports 63 (2016–2017), 167–80. Furtwängler, A. (ed.) 1906. Aegina: Das Heiligtum der Aphaia. Munich: Verlag der K.B. Akademie der Wissenschaften in Kommission des G. Franz’schen Verlags. Garnsey, P. 1988. Famine and Food Supply in the Graeco-Roman World: Responses to Risk and Crisis. Cambridge: Cambridge University Press. Gentili, B., and Perusino, F. (eds) 2002. Le orse di Brauron: Un rituale di iniziazione femminile nel santuario di Artemide. Pisa: Edizioni Ets. Goldberg, J. 2012. Trade and Institutions in the Medieval Mediterranean: The Geniza Merchants and Their Business World. Cambridge: Cambridge University Press. Graf, F. 1979. Das Götterbild aus dem Taurerland. Antike Welt 10(4), 33–41. Graham, S. 2006a. Ex Figlinis: The Network Dynamics of the Tiber Valley Brick Industry in the Hinterland of Rome. Oxford: Archaeopress. Graham, S. 2006b. Networks, agent-based models and the Antonine Itineraries: implications for Roman archaeology. Journal of Mediterranean Archaeology 19(1), 45–64. Graham, S. 2009. Behaviour space: simulating Roman social life and civil violence. Digital Studies/Le Champ numé rique 1.2, www.digitalstudies.org/ojs/index.php/digital_studies/ article/view/172/214 accessed 29 March 2017. Graham, S. 2014. On connecting stamps: network analysis and epigraphy. Les nouvelles de l’archéologie 135, 39–44. Graham, S., and S. Weingart. 2015. The equifinality of archaeological networks: an agent-based exploratory lab approach. Journal of Archaeological Method and Theory 22(1), 248–274. Granovetter, M.S. 1973. The strength of weak ties. American Journal of Sociology 78(6), 1360–1380. Guarducci, M. 1935. Diktynna. Studi e materiali di storia delle religioni 11, 187–203. Guldager Bilde, P. 2003. Wandering images: from Taurian (and Chersonesean) Parthenos to (Artemis) Tauropolos and (Artemis) Persike. In P.G. Bilde, J.M. Højte, and V.F. Stolba (eds), The Cauldron of Ariantas: Studies Presented to A.N. Šč eglov on the Occasion of his 70th Birthday, 165–183. Aarhus: Aarhus University Press.

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:31:07, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.006

CULTS, CABOTAGE, AND CONNECTIVITY

127

Guldager Bilde, P. 2005. What was Scythian about the “Scythian Diana” at Nemi?, www .academia.edu/1855901/What_was_Scythian_about_the_Scythian_Diana_at_Nemi accessed 29 March 2017 (translation of Что скифского было в “Скифской Диане” из Неми?, Archeologič eskie Vesti 12, 209–217). Hohlfelder, R.H. 2000. Cabotage at Aperlae in ancient Lycia. International Journal of Nautical Archaeology 29(1), 126–135. Hollinshead, M.B. 1979. Legend, cult, and architecture at three sanctuaries of Artemis. PhD thesis, Bryn Mawr College. Horden, P., and Purcell, N. 2000. The Corrupting Sea: A Study of Mediterranean History. Oxford: Blackwell. Ismard, P. 2015. Playing with scales in the classical city: the case of the Marathonian Tetrapolis. In C. Taylor and K. Vlassopoulos (eds), Communities and Networks in the Ancient Greek World, 80–97. Oxford: Oxford University Press. Kahil, L. 1963. Quelques vases du sanctuaire d’Artémis à Brauron. Antike Kunst Suppl. 1, 5–29. Kahil, L. 1977. L’Artémis de Brauron: Rites et mystère. Antike Kunst 20, 86–98. Kahil, L. 1981. Le “cratérisque” d’Artémis et le Brauronion de l’Acropole. Hesperia 50, 253–263. Kahil, L. 1990. Le relief des dieux du sanctuaires d’Artémis à Brauron: Essai d’interprétation. In J.-P. Descoeudres (ed.), Eumousia: Ceramic and Iconographic Studies in Honour of Alexander Cambitoglou, 115–120. Sydney: Meditarch. Kalogeropoulos, K. 2013. Το ιερό της Αρτέμιδος Ταυροπόλου στις Αλές Αραφηνίδες (Λούτσα). 2 vols. Athens: Γραφείον Δημοσιευμάτων της Ακαδημίας Αθηνών. Knappett, C. 2011. An Archaeology of Interaction: Network Perspectives on Material Culture and Society. Oxford: Oxford University Press. Knappett, C. (ed.) 2013. Network Analysis in Archaeology: New Approaches to Regional Interaction. Oxford: Oxford University Press. Knappett, C. 2014. Dossier: Analyse des réseaux sociaux en archéologie, Les nouvelles de l’archéologie 135, 5–61. Knappett, C., Evans, T., and Rivers, R. 2008. Modelling maritime interaction in the Aegean Bronze Age. Antiquity 82, 1009–1024. Knappett, C., Evans, T., and Rivers, R. 2011. The Theran eruption and Minoan palatial collapse: new interpretations gained from modelling the maritime network, Antiquity 85, 1008–1023. Knodell, A. 2013. Small-world networks and Mediterranean dynamics in the Euboean Gulf: an archaeology of complexity in Late Bronze Age and Early Iron Age Greece. PhD thesis, Brown University. Knoepfler, D. 1988. Sur les traces de l’Artémision d’Amarynthos près d’Érétrie. Comptes rendus/Académie des inscriptions et belles-lettres 132(2), 382–421. Koutsouflakis, G. 2013. Seafaring and commercial transport in the South Euboean Gulf (6th c. BC–14th c. AD). Unpublished PhD thesis, University of Athens. Koutsouflakis, G., Argyri, X., Papadopoulou, C., and Sapountzis, I. 2012. Υποβρύχια Επιφανειακή Έρευνα στο Νότιο Ευβοϊκό 2006–2008 (Underwater survey in the southern Euboean Gulf 2006–2008). Enalia: The Journal of the Hellenic Institute of Marine Archaeology 9, 40–69. Kowalski, J.M. 2012. Navigation et géographie dans l’antiquité gréco-romaine: La terre vue de la mer. Paris: Picard.

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:31:07, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.006

128

BARBAR A KOWALZIG

Kowalzig, B. 2005. Mapping out communitas: performances of theoria in their sacred and political context. In J. Elsner and I. Rutherford (eds), Pilgrimage in Graeco-Roman and Early Christian Antiquity: Seeing the Gods, 41–72. Oxford: Oxford University Press. Kowalzig, B. 2006. The aetiology of empire? Hero-cult and Athenian tragedy. In J. Davidson, F. Muecke, and P. Wilson (eds), Greek Drama III: Essays in Honour of Kevin Lee, 79–98. London: Institute of Classical Studies, University of London. Kowalzig, B. 2007. Singing for the Gods: Performances of Myth and Ritual in Archaic and Classical Greece. Oxford: Oxford University Press. Kowalzig, B. 2010. Musical merchandise “on every vessel”: religion and trade on the island of Aigina. In D. Fearn (ed.), Aegina: Contexts for Choral Lyric Poetry. Myth, History, and Identity in the Fifth Century BC, 129–171. Oxford: Oxford University Press. Kowalzig, B. 2011. Identità greche tra modelli religiosi ed economici: il caso di Egina. In L. Breglia, A. Moleti and M.L. Napolitano (eds), Ethne, identità e tradizioni: la “terza” Grecia e l’Occidente, 1–18. Pisa: ETS. Kowalzig, B. 2013. Transcultural chorality: “Iphigenia in Tauris” and Athenian imperial economics in a polytheistic world. In R. Gagné and M.G. Hopman (eds), Choral Mediations in Greek Tragedy, 178–210. Cambridge: Cambridge University Press. Kowalzig, B. forthcoming. Cross-cultural trade cum grano salis: religion and trade beyond the pillars of Herakles. Leidwanger, J. 2013a. Between local and long-distance: a Roman shipwreck at Fig Tree Bay off SE Cyprus. Journal of Roman Archaeology 26(1), 191–208. Leidwanger J. 2013b. Modeling distance with time in ancient Mediterranean seafaring: a GIS application for the interpretation of maritime connectivity. Journal of Archaeological Science 40(8), 3302–3308. Leidwanger, J. 2013c. Opportunistic ports and spaces of exchange in late Roman Cyprus. Journal of Maritime Archaeology 18, 221–243. Leidwanger J. 2014. Maritime networks and economic regionalism in the Roman eastern Mediterranean. Les nouvelles de l’archéologie 135, 32–38. Lemos, I.S. 2002. The Protogeometric Aegean: The Archaeology of the Late Eleventh and Tenth Centuries BC. Oxford: Oxford University Press. Linders, T. 1972. Studies in the treasure records of Artemis Brauronia found in Athens. PhD thesis, University of Stockholm. Malkin, I. 1990. Missionnaires païens dans la Gaule grecque: La France et la Méditerranée. In I. Malkin (ed.), La France et la Méditerranée: Vingt-sept siècles d’interdépendance, 42–52. Leiden: Brill. Malkin, I. 2005. Networks and the emergence of Greek identity. In I. Malkin (ed.), Mediterranean Paradigms and Classical Antiquity, 65–74. London: Routledge. Malkin, I. 2007. Ethnicité et colonisation: Le réseau d’identité grecque en Sicile. Pallas 73, 181–190. Malkin, I. 2011. A Small Greek Word: Networks in the Ancient Mediterranean. Oxford: Oxford University Press. Malkin, I., Constantakopoulou, C., and Panagopoulou, K. 2009. Preface: networks in the ancient Mediterranean. In I. Malkin, C. Constantakopoulou, and K. Panagopoulou (eds), Greek and Roman Networks in the Mediterranean, 1–11. London: Routledge. Matthaiou, A.P. 2003. Ἀπόλλων Δήλιος ἐν Ἀθήναις. In D. Jordan and J. Traill (eds), Lettered Attica: A Day of Attic Epigraphy: Proceedings of the Athens Symposium, 8 March 2000, 85–93. Athens: Canadian Archaeological Institute at Athens.

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:31:07, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.006

CULTS, CABOTAGE, AND CONNECTIVITY

129

Mazarakis-Ainian, A.I. 1998. Oropos in the Early Iron Age. In M. Bats and B. D’Agostino (eds), Euboica: l’Eubea e la presenza euboica in Calcidica e in Occidente: atti del convegno internazionale di Napoli 13–16 novembre 1996, 23–29, 179–215. Naples: Centre Jean Bérard. Mazarakis-Ainian, A.I., Alexandridou, A., and Charalambidou, X. (eds) 2017. Aristeia: Regional Stories towards a New Perception of the Early Greek World. Acts of an International Symposium in honour of Professor Jan Bouzek. Volos, 18–21 June 2015. Volos: Thessaly University Press. McInerney, J. 2015. “There will be blood . . . ”: the cult of Artemis Tauropolos at Halai Araphenides. In K.F. Daly and L.A. Riccardi (eds), Cities Called Athens: Studies Honoring John McK. Camp II, 289–320. Lewisburg: Bucknell University Press. Meritt, B.D. 1947. Greek inscriptions. Hesperia 16(3), 147–183. Mitsopoulos-Leon, V. 1997. Tonstatuetten im Heiligtum der Artemis von Brauron. In B. Petrakos (ed.), Epainos Ioannou Papadimitriou, 357–378. Athens: Athens Archaeological Society. Morel, J.P. 2006. Phocaean colonisation. In G.R. Tsetskhladze (ed.), Greek Colonisation: An Account of Greek Colonies and Other Settlements Overseas, 358–428. Leiden: Brill. Moreno, A. 2007. Feeding the Democracy: The Athenian Grain Supply in the Fifth and Fourth Centuries BC. Oxford: Oxford University Press. Morton J. 2001. The Role of the Physical Environment in Ancient Greek Seafaring. Leiden: Brill. Musti, D., and Beschi, L. 1987. Pausania: Guida della Grecia, Libro I. L’Attica. 2nd edn. Rome: Fondazione Lorenzo Valla. Nilsson, M.P. 1906. Griechische Feste von religiöser Bedeutung, mit Ausschluss der attischen. Leipzig: Teubner. Palaiokrassa, L. 1991. To hiero te¯s Artemidos Mounichias. Athens: Athens Archaeological Society. Papadimitriou, J. 1963. The sanctuary of Artemis at Brauron. Scientific American 208(6), 110–122. Papadopoulos, Th.I. and Kontorli-Papadopoulou, L. 2014. Vravron: The Mycenaean Cemetery. SIMA 142, Uppsala: Åströms Förlag. Papadopoulou, C. 2014. Transforming the surroundings and its impact on cult rituals: the case study of Artemis Mounichia in the fifth century. In C. Moser and C. Feldman (eds), Locating the Sacred: Theoretical Approaches to the Emplacement of Religion, 111–127. Oxford: Oxbow Books. Papalas, A.J. 1983. The temple of Artemis Tauropolos in Icaria. Archaeological News 12, 8–13. Papalas, A.J. 1992. Ancient Icaria. Wauconda: Bolchazy-Carducci Publishers. Parker, R. 2005. Polytheism and Society at Athens. Oxford: Oxford University Press. Peppas-Delmousou, D. 1988. Autour des inventaires de Brauron. In D. Knoepfler and N. Quellet (eds), Comptes et inventaires dans la cité grecque: Actes du colloque international d’épigraphie tenu à Neuchâtel du 23 au 26 septembre 1986 en l’honneur de Jacques Tréheux, 323–346. Geneva: Librairie Droz. Reber, T., Knoepfler, D., Karapaschalidou, A., Krapf, T., and Theurillat, T. 2017. Die Grabungen in Amarynthos. École suisse d’archéologie en Grèce: Rapport annuel 2017, 10–15. Reger, G. 2004. Regionalism and Change in the Economy of Independent Delos. Berkeley: University of California Press. Renfrew C., Boyd, M., and Ramsey, C.B. 2012. The oldest maritime sanctuary? Dating the sanctuary at Keros and the Cycladic Early Bronze Age. Antiquity 86, 144–160. Rhodes, P.J., and Osborne, R. (eds) 2003. Greek Historical Inscriptions: 404–323 BC. Oxford: Oxford University Press.

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:31:07, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.006

130

BARBAR A KOWALZIG

Roueché, C., and Sherwin-White, S.M. 1985. Some aspects of the Seleucid Empire: the Greek inscriptions from Failaka, in the Arabian Gulf. Chiron 15, 1–39. Rutherford, I. 2004. Xορος εἷς ἐκ τῆσδε τῆς πόλεως (Xen. Mem. 3.3.12): song-dance and state-pilgrimage at Athens. In P.J. Wilson and P. Murray (eds), Music and the Muses: The Culture of “Mousike¯” in the Classical Athenian City, 67–90. Oxford: Oxford University Press. Rutherford, I. 2007. Theoria and theatre at Samothrace: the Dardanos by Dymas of Iasos. In P.J. Wilson (ed.), The Greek Theatre and Festivals: Documentary Studies, 279–293. Oxford: Oxford University Press. Rutherford, I. 2009a. The Koan–Delian ritual complex: Apollo and “theoria” in a sacred law from Kos. In L. Athanassaki, R.P. Martin, and J.F. Miller (eds), Apolline Politics and Poetics, 655–687. Athens: Hellenic Ministry of Culture. Rutherford, I. 2009b. Network theory and theoric networks. In Malkin et al. 2009, 24–38. Rutherford, I. 2013a. Mycenaean religion. In M. Salzman (ed.), The Cambridge History of Religions in the Ancient World, Vol. 1, 256–279. Cambridge: Cambridge University Press. Rutherford, I. 2013b. State Pilgrims and Sacred Observers in Ancient Greece: A Study of Theo¯ria¯ and Theo¯roi. Cambridge: Cambridge University Press. Saprykin, S. 2007. The unification of Pontos: the bronze coins of Mithridates VI Eupator as evidence for commerce in the Euxine. Black Sea Studies 6, 195–208. Scanlon, T.F. 1990. Race or chase at the Arkteia of Attica? Nikephoros 3, 73–120. Schachter, A. 1981–1994. Cults of Boiotia. 4 vols. London: Institute of Classical Studies. Scheid, J. 1995. Graeco ritu: a typically Roman way of honoring the gods. Harvard Studies in Classical Philology 97, 15–31. Scheid, J. 1998. Nouveau rite et nouvelle piété: Réflexions sur le ‘ritus Graecus’. In F. Graf (ed.), Ansichten griechischer Rituale: Geburtstags-Symposium für Walter Burkert, Castelen bei Basel, 15. bis 18. März 1996, 168–182. Stuttgart: Teubner. Scott, C. 2017. Asklepios on the move. Health, Healing and Cult in Classical Greece. PhD dissertation, New York University. Sherratt, S. 2006. The pottery in a wider context. In D. Evely (ed.), Lefkandi IV: The Bronze Age: The Late Helladic IIIC Settlement at Xeropolis, 218–31. London: British School at Athens. Sinn, U. 1988. Der Kult der Aphaia auf Aegina. In R. Hägg, N. Marinatos, and G. Nordquist (eds), Early Greek Cult Practice: Proceedings of the Fifth International Symposium at the Swedish Institute at Athens, 26–29 June, 1986, 149–159. Stockholm: Aström. Sourvinou-Inwood, C. 1988. Studies in Girls’ Transitions. Aspects of the Arkteia and Age Representations in Attic Iconography. Athens: Kardamitsa. Sporn, K. 2002. Heiligtümer und Kulte Kretas in klassischer und hellenistischer Zeit. Studien zu antiken Heiligtümern 3. Heidelberg: Verlag Archäologie und Geschichte. Sporn, K. 2001. Auf den Spuren der kretischen Diktynna. In S. Böhm and K.-V. Von Eickstedt (eds), Ithakê: Festschrift für J. Schäfer zum 75. Geburtstag am 25. April 2001, 225–233. Würzburg: Ergon. Stefanakis, M. I. 1999. The introduction of coinage in Crete and the beginning of local minting, in A. Chaniotis (ed.), From Minoan Farmers to Roman Traders: Sidelights on the Economy of Ancient Crete, 247–268. Stuttgart: Steiner. Tartaron, T. F. 2013. Maritime Networks in the Mycenaean World. Cambridge: Cambridge University Press.

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:31:07, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.006

CULTS, CABOTAGE, AND CONNECTIVITY

131

Taylor, C., and Vlassopoulos, K. (eds) 2015. Communities and Networks in the Ancient Greek World. Oxford: Oxford University Press. Travlos, N.J. 1988. Bildlexikon zur Topographie des antiken Attika. Tü bingen: Wasmuth. US Hydrographic Office under the Authority of the Secretary of the Navy. 1916. Mediterranean Pilot: From Cape Matapan (Greece) Eastward, the Mediterranean Archipelago, and the Southern Shore of the Mediterranean Sea, Eastward to Ras Ashdir (Libia), Vol. 4. Washington: Government Printing Office. Van Straten, F.T. 1995. Hierà Kalá: Images of Animal Sacrifice in Archaic and Classical Greece. Leiden: Brill. Venit, M.J. 2003. A reconsideration of the “Relief of the Gods” from Brauron. Antike Kunst 46, 44–55. Vernant, J.-P. 1983. Hestia-Hermes: the religious expression of space and movement in ancient Greece. In Myth and Thought among the Greeks, 157–196. London: Routledge and Kegan Paul (first published Paris: Maspero, 1965). Vernant, J.-P. 1990. The society of gods. In Myth and Society in Ancient Greece, 101–119. New York: Zone Books (first published Paris: Maspero, 1974). Wachsmuth, D. 1967. Πóμπιμoς ὁ δαíμων: Untersuchung zu den antiken Sakralhandlungen bei Seereisen. Berlin: Freie Universität. Watts, D.J. 2003. Six Degrees: The Science of a Connected Age. New York: W.W. Norton & Company. Watts, D.J. 2004. The “new” science of networks. Annual Review of Sociology 30, 243–270. Watts, D.J., and Strogatz, S.H. 1998. Collective dynamics of “small-world” networks. Nature 393(6684), 440–442. Westerdahl, C. 1992. The maritime cultural landscape. International Journal of Nautical Archaeology 21(1), 5–14. Westerdahl, C. 2011. The maritime cultural landscape. In A. Catsambis, B. Ford, and D.L. Hamilton (eds), The Oxford Handbook of Maritime Archaeology, 733–762. Oxford: Oxford University Press. Whitley, J. 2017. Afterword. Regional stories towards a new perception of the early Greek World. In Mazarakis-Ainian et al. 2017, 695–700. Whitley, J. forthcoming. The re-emergence of political complexity. In I. Lemos and A. Kotsonas (eds), A Companion to the Archaeology of Early Greece and the Mediterranean. Oxford: Wiley-Blackwell. Williams, D. 1983. Aegina, Aphaia-Tempel. V. The pottery from Chios. Archäologischer Anzeiger (1983), 155–186.

Downloaded from https://www.cambridge.org/core. Universiteit Leiden / LUMC, on 07 Nov 2018 at 19:31:07, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.006

CHAPTER SIX

SHIPWRECKS AS INDICES OF ARCHAIC MEDITERRANEAN TRADE NETWORKS* Elizabeth S. Greene

In civilizations without boats, dreams dry up, espionage takes the place of adventure, and the police take the place of pirates. —Michel Foucault

Boats, according to Foucault (1986, 27), are the conveyors of dreams and adventure, the great instrument of economic development, and the greatest reserve of the imagination. Linking communities through their cargos and routes, sunken ships offer tantalizing clues into the social, intellectual, and economic connections forged by the sea. Scholars who investigate shipwrecks and ancient trade inevitably hear variants of the following questions: (1) how can we tell from its archaeological remains where a ship began its journey, what direction it was headed, and what its final destination was (or rather would have been)? And (2) how can we use shipwrecks to map ancient trade routes? There are no simple answers to the two questions. Ancient shipwreck publications often include the creation of a hypothetical “trade network” that does one of two things. The first option involves a map that connects, with a series of arrows, the points of origin of cargo goods, galley * This chapter owes a particular debt of gratitude to Justin Leidwanger, with whom I have had many long discussions on the topic of shipwrecks and networks. I am grateful to Carl Knappett and the participants in the 2013 conference for stimulating papers that provided ample food for thought, and particularly to Barbara Mills for her suggestion to explore shipwrecks as ego networks. Shawn Graham’s advice on Gephi visualizations has been invaluable. All errors remain my own.

132 Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.007

SHIPWRECKS AS INDI CES OF TRADE NETWORKS

133

wares, and personal items discovered aboard the ship. With some reference to winds, currents, and sailing patterns, these dots and arrows are described as reflecting the ship’s last voyage or the broader network to which it belonged.1 The location of the shipwreck is plotted as a point somewhere along this path, though of course the position of this dot is a matter of chance and hardly reflects the same type of locational data as a port of call where goods may have been produced, exchanged, and consumed. A second exercise involves the plotting of locations of wreck sites, anchors, and cargo scatters from a discrete chronological period. By connecting these dots on a map, “trade routes” are identified, highlighting the degree of frequency with which ships passed particular ports or coasts. While such approaches invoke certain network concepts and make for attractive illustrations, they are not entirely satisfying as explanations for the role of one or multiple wrecks within a framework of communities linked by these vessels. This contribution approaches a small dataset of Mediterranean shipwrecks dated to the 7th and 6th centuries BC as a means of considering how to model maritime networks from shipwreck evidence (Figure 6.1). Although the number of known shipwrecks from this period is relatively small, the wrecks have been, on the whole, more intensively observed than sites from later dates—such as the Roman period—which are far greater in number, but these numbers often come with only cursory detail (Leidwanger 2017). On their surface, shipwrecks would appear to be an ideal resource for the determination of network activity, their cargos marking connections between various trading centers or nodes. But if the origins of cargo elements, personal items, or hull timbers signify nodes of some fixed (if often imprecise) geographical position, what is the shipwreck? Is it simply another node in the network picture? Is it a mobile node? If so, how and where do we conceptualize, map, and meaningfully reflect such a moving target? Is a shipwreck really an edge that links two nodes? And if so, how can we transform an archaeological site into a line that spans geographic locales without inhabiting them? Is the spatial positioning of a wreck site even meaningful in a network structure? Or do shipwrecks simply bridge communities through the copresence of cargo elements in their archaeological assemblages without marking nodes at all? How can we use individual and multiple shipwrecks to define the structure of maritime connectivity and the ways that regions were joined by journeys across the sea? At this stage an example of the complexity posed by any individual shipwreck would seem convenient. Leaving aside the standard dilemmas of comparing archaeological data across sites (e.g., different levels of publication detail, uneven levels of compositional analysis, different conventions of ceramic attribution), I turn to the Pointe Lequin 1A wreck (Figure 6.1), discovered in 1985 off the coast near Marseilles, and excavated from 1986 to 1993 (Long,

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.007

134

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.007

Bon Porté

Grand Ribaud F

Dattier

La Love

S ITALY

IONIA

Km

0

250

500

750

Kepçe Burnu Kekova Adası Pabuç Burnu CYPRUS ˘ Çaycagız Koyu

Akio

6.1. Map of the Mediterranean with Archaic shipwreck sites discussed in this chapter. (Map by Thomas Kocjan.)

Pointe Lequin 1A

Miet 3

Marseilles

Jules Verne 7 & 9 & César 1

Gela

ETRURIA Giglio

N AFRICA

Cala Sant Vicenç

IBERIA

1,000

SHIPWRECKS AS INDI CES OF TRADE NETWORKS

135

Miro, and Volpe 1992; Krotscheck 2008; 2015). The shallow-water wreck dating to the late 6th century contains an impressive array of cargo. Along with more than 700 black-figure Athenian cups, the ship held about 1,500 “Ionian cups” and other black gloss vessels. Stopping here, we might imagine a voyage that took sailors from an east Greek port such as Massalia’s “mother city” at Phokaia to Athens to Massalia, where the ship sank somewhat short of its destination. But the picture grows more complicated. The cargo includes amphoras identified with Miletus, Samos, Klazomenai, Chios, Lesbos, the northern Aegean, Athens, Corinth (and its colony Corcyra), Massalia, and Etruria. Suddenly we have many more dots to place on our map, potentially explained through a system of cabotage—of the sort which Horden and Purcell (2000) suggest served as a primary mechanism of maritime shipping before the age of steam—that brought the ship to multiple ports over the course of a longer journey. How much of the journey can be tied to the eastern Mediterranean comes into question with recent chemical analyses that suggest a production area for the wreck’s “Ionian cups” in Sicily or southern Italy.2 This shift in nodes of association now raises the question whether the ship ventured to the eastern Mediterranean at all, or if its cargo of east Greek jars was collected at some more western center. An alternative scenario centers on network hubs where, as Pascal Arnaud (2005) has discussed for the Roman world, larger ports served as central hubs from which goods continued along shorter, more localized routes between smaller secondary ports. In such a model, the ship might have loaded Aegean amphoras and Attic finewares somewhere in southern Italy before heading to Massalia. In spatial network terms, and perhaps appropriately for the late 6th century, we might use our cargo to suggest a system of long-distance ties between central “hubs” like Athens and Massalia, to which other centers were linked through a shorter-haul, perhaps denser, network of local or regional ties. But can one shipwreck speak for such network communities? How relevant is the site’s location? If geography is a key factor, then Massalia is arguably an important node because of its proximity to the wreck site. Where are the other key nodes? Would we imagine the same set of ties for a ship carrying precisely the same cargo that sank in the Straits of Messina or off the shores of Samos? Should it matter in our understanding whether a ship sinks early or late in its voyage? How do we systematically translate a mixed cargo such as the contents of the Pointe Lequin 1A shipwreck—or even a uniform cargo—into network links? Is there some quantification or ranked importance within the cargo? Should we consider the copresence of cargo elements in different proportions? Do 700 Athenian cups create a stronger network tie than fifty Milesian amphoras or two Etruscan jars? Now we might envision a relationship that highlights the connection between southern Italy and Athens, both featuring strongly in the cargo by virtue of relative numbers and value. But where

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.007

136

ELIZABETH S. GREENE

does this link fit within the context of a ship that sank off the coast of France, with Massaliot goods comprising a nearly insignificant percentage of preserved cargo? Can we set up a network that combines evidence from shipwreck cargos with the distribution of similar goods on land? In this volume, Brughmans’s analysis of the distribution of Eastern Sigillata and Lawall and Graham’s mapping of transport amphoras shed light on the distributions of single classes of pots of known origins, but help less to interpret the copresence of markers from multiple producers at a single moment of their distribution rather than at sites of production or consumption. To what degree does a single shipwreck speak for the broader network community in which it participated? Is the process of analyzing one vessel assemblage tantamount to building a full network? Is one wreck equivalent to one snapshot or one pulse of that economic network? What is the relevance of other similarly dated shipwrecks and their cargos in the network picture? While habitation sites are frequently deemed to have an association based on the shared presence of imported or luxury items (see Arthur, Imperiale, and Muci in this volume), we cannot assume a formal connection between two shipwrecks simply because they share one or more common cargo items. Many questions remain to be asked, but suffice it to say that no single process of “dot placing” provides a persuasive explanation of the network in which any particular ship participated. Also worthy of note are environmental patterns such as maritime hazards, winds, and currents, alongside hull construction and technological developments that help us to understand when and how quickly voyages could be conducted.3 Historically particularized data including known economic, social, political, or colonial relationships between possible agents and their associated goods might add a further layer of complexity to the archaeological picture, difficult as these relationships might be to graph.4 In what follows, I consider various ways of thinking about shipwrecks in the context of trade networks, ranging from abstract and metaphorical to representational, with the assistance of network visualization software. SHIPWRECKS, MIDDLE GROUNDS, AND OTHER SPACES

As a means of bypassing the specific dots and lines or nodes and edges that mark a social and spatial network, I turn first to a construct used by Irad Malkin (2011; 2002) and Carla Antonaccio (2013) to discuss colonial relationships in the Archaic Mediterranean. For them, the notion of “middle grounds” characterizes regional clusters of networks where Greeks founded colonies or lived in emporia or mixed settlements. Initially formulated by Richard White (1991) in a discussion of cultural contact between French settlers and Native Americans in the Great Lakes region during the 17th–19th centuries, the physical and metaphorical “middle ground” serves as both a space and a process of

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.007

SHIPWRECKS AS INDI CES OF TRADE NETWORKS

137

accommodation and understanding. As utilized by Malkin and Antonaccio, the middle ground is a field of interaction for Greeks and indigenous groups in colonial contexts. For Malkin, the middle ground functions as a system of weak ties that allows for the collapse of difference between cultural groups in proximity, leading to the adoption, adaption, and production of hybrid culture. As he explains, maritime cities were initially linked by “a ‘many-to-many’ network that radiated from each node to the Mediterranean at large” (Malkin 2011, 157); in this way, the sea becomes the “virtual center” of multiple links between multiple spaces and cultures, driven by the variety of connections rather than a single parent city or network hub. Antonaccio proposes panHellenic sanctuaries as middle grounds, while Malkin places these spaces in coastal zones on the periphery of communities; he also uses the term to discuss physical objects, such as the 5th-century lead tablet from Pech Maho with its bilingual inscriptions in Etruscan and Greek that describe the sale of one or more boats at Emporion. As an object, the tablet functions as a “multiethnic middle ground,” in which multiple ethnicities from multiple regions cooperate alongside each other in a commercial transaction (Malkin 2011, 166). For shipwrecks, the vessel itself might function as a middle ground similar to a sanctuary or contractual tablet. The boat is a space where objects and people set off together to form something new; the sea serves as a backdrop or a landscape for this middle ground, participating in its creation, since the boat’s transformative potential arises only after it has left the land. A variant of this middle ground might be the “other space” or “heterotopia” proposed by Foucault (1986) in an approach to the history of human emplacement in space and time. The conceptual heterotopia combines several spaces and several slices of time, in which real sites and real cultures may be represented, contested, and inverted simultaneously; they are both isolated and penetrable. Like middle grounds, heterotopias work well with colonial models, offering opportunities for hybridization through the melding of the illusory and the real. As the heterotopia par excellence, Foucault (1986, 26–27) offers the image of a boat, “a floating piece of space, a place without a place, that exists by itself, that is closed in on itself and at the same time is given over to the infinity of the sea.” By virtue of their movement through space, time, and culture, the ship and archaeological wreck are perhaps unmappable, but function as spaces of contact and common understanding. As such, it may be helpful to see the ship, perhaps alongside the swath of sea through which it regularly sails, as a “middle ground,” an “other space,” or a convergence of links radiating to and from different directions. I lay out these possibilities without diagrams, as middle grounds, other spaces, and the moving multidimensional shapes created by joined lines make for difficult graphs, especially when we speak of a ship whose cargo encapsulates in one moving target a series of objects and relations that link much of the Mediterranean.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.007

138

ELIZABETH S. GREENE

ZONES OF INTERACTION

Is there a way to depict visually these imaginary middle grounds represented by shipwrecks? For the purpose of experimentation, I turn to a mid-6th-century wreck in the eastern Mediterranean, at Pabuç Burnu, about thirty-five kilometers southeast of Bodrum or ancient Halicarnassus (Greene, Lawall, and Polzer 2008). Excavated by the Institute of Nautical Archaeology in 2002 and 2003, the wreck represents a moderate-sized merchant ship carrying wine and olive oil in amphoras, along with a possible secondary cargo of foodstuffs and other perishables. The ship’s primary cargo was carried in more than 200 amphoras, the majority of which Mark Lawall has attributed by form and fabric to the region around Halicarnassus, including fabrics resembling production at nearby Knidos and Rhodes. The remaining jars on the wreck have been connected in decreasing numbers to the southeast Aegean region more generally, along with Klazomenai and Lesbos. Coarse wares from the ship reflect a similarly limited geographic range, with the overwhelming majority displaying parallels in form and fabric to vessels from the southeast Aegean. Both the amphoras and coarse wares raise a secondary problem in the explanation of network links for shipwreck cargos: that is, a basic geographic uncertainty for the origins of jars that share a similar regional form, in this case what Lawall has termed a southeast Aegean koiné.5 Although we make best guesses for the origins of different cargo elements, sometimes a dot on a map actually signifies a broader region. More unique in form, and so easier to attribute, a sole Klazomenian amphora, two Lesbian amphora toes, and a northern Ionian cup expand our zone northward if their presence should be seen as significant to the ship’s broader economic network, whether as remnants of a specific voyage or as merchandise obtained at a diverse market center. The cargo and construction of the Pabuç Burnu shipwreck suggest that it was engaged in the regional exchange of processed agricultural products (Greene, Lawall, and Polzer 2008). A “goods to market” model provides one explanation for the short-distance voyage undertaken by the vessel in its final journey, highlighting long-term and ongoing relationships between buyer and seller that reduced their dependence upon formal institutions designed to lower the transaction costs of doing business with more distant trading partners. Leidwanger (2013, 3306–3307) has used the wreck as a case study to test geographic modeling of environmental and technological parameters, and determines that the cities likely represented by the major types of amphoras were connected by no more than two or three days of sailing in each direction. Such regional connections are confirmed historically by the relationship that Herodotus (1.144) records among Halicarnassus, Knidos, and Rhodes in the Dorian Hexapolis, where religious connections might also reflect mercantile ties, much as Kowalzig

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.007

SHIPWRECKS AS INDI CES OF TRADE NETWORKS

139

elucidates in her contribution to this volume. But how do we move from one shipwreck, occupying a particular geographic location at some arbitrary point along its presumed final path, to the larger community linked by its association with the goods carried aboard the vessel? The simplest visualization of the ship’s activity network might be represented by a zone of interaction marked by the different communities whose wares were discovered on board (Figure 6.2). In the case of the Pabuç Burnu shipwreck, the primary zone of interaction can be marked by an oval spanning roughly from Halicarnassus to Rhodes; the inclusion of a few rare amphora finds from Lesbos and Klazomenai expands the oval to a larger part of the east Aegean. The size of the zone varies depending on the degree of importance allotted to less frequently represented cargo items on board the vessel. For the Pabuç Burnu wreck, the location of the site falls within the zone of interaction, at whatever scale it is drawn, and this would seem to be a logical assumption for most vessels. A broader eastern Mediterranean sphere of activity is visible through the shipwreck at Kekova Adası, a mid-7th-century site located off the Lycian coast of Turkey (Greene, Leidwanger, and Özdaş 2011; 2013; Aslan 2015). Surveyed extensively in 2008 and 2010, the wreck’s assemblage is composed primarily of basket-handle amphoras, large jars with looping handles typically associated with Cyprus and the Levantine coast. The remaining cargo comprises southeast Aegean and Corinthian A amphoras that together amount to less than onequarter of the total number of jars in the assemblage. Two coarse ware mortaria —mixing or measuring bowls similar in form to those observed at Pabuç Burnu, but made of a probably Cypriot fabric—might, along with a large pile of riverine ballast stones also potentially from the island, suggest a point of origin for the shipwreck. An assortment of fragmentary cooking pots hints at the ship’s capacity for longer voyages requiring meals for the sailors. The primary cargo of the Kekova Adası shipwreck has a likely origin in eastern Cyprus, carried in tandem with amphoras from Corinth and Miletus or some other city in the southeast Aegean. Its sphere of operation (Figure 6.3) is larger than that of the Pabuç Burnu vessel, encompassing a region from Salamis to Corinth. Miletus and the shipwreck site both fall within a zone marked by Salamis and Corinth at its edges. Is a zone of operation at this scale meaningful when it now spans much of the eastern Mediterranean and the Aegean? Does the oval on the map delineate routes or seafaring communities? And to what degree can single wrecks speak for increasingly larger zones of maritime interaction? Does it help to increase the sample size? A total of three Archaic wrecks in the eastern Mediterranean stand out for their primary cargos of basket-handle amphoras: the shipwreck at Kekova Adası, as well as wrecks at Kepçe Burnu (east of Pabuç Burnu in the Gulf of Gökova) and Çaycağ ız Koyu (near the modern town of Marmaris). These two

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.007

140

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.007

Akio

Chios hios

Lesbos sbos

hode Rhodes

Pabuç Pabuç B Burnu

Kepçe SE EA AEGEAN Burnu Bu ˘ Koyu Çaycagız uss Halicarnassus

Klazomenai

IONIA

0

200

Kekova Adası

100

300

Cyprus

400

6.2. Detail map of the Aegean indicating zones of interaction for the shipwreck at Pabuç Burnu. (Map by Thomas Kocjan.)

Laconia

Corinth

Athens

N AEGEAN

Km 500

141

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.007

Akio

Chios

Lesbos

Rhodes

Pabuç Burnu

Kepçe SE AEGEAN Burnu Çaycagız ˘ Koyu Halicarnassus

Klazomenai

IONIA

0

200

Kekova Adası

100

300

Cyprus

400

6.3. Zones of interaction for the shipwrecks at Kekova Adası, Kepçe Burnu, and Çaycağ ız Koyu. (Map by Thomas Kocjan.)

Laconia

Corinth

Athens

N AEGEAN

Km 500

142

ELIZABETH S. GREENE

shallow-water sites with extremely fragmentary cargos were surveyed in 2010 (Greene, Leidwanger, and Özdaş 2013). The Kepçe Burnu wreck can be dated through its ceramics to the second half of the 7th century; at Çaycağ ız Koyu, fragments of fifty to sixty probably Cypriot basket-handle jars alongside more than two dozen ceramic mortaria date somewhat later, probably to the turn of the 6th century. Viewed alongside the shipwreck at Kekova Adası, the two additional wrecks add a smaller zone to the map (Figure 6.3). The picture suggests a clear link between Cyprus and the Aegean, occasionally extending as far as Corinth on the Greek mainland and bridged by ships carrying baskethandle jars. The location of the wrecks strengthens their connection to the Turkish coast, though in each case the basket-handle amphoras that mark their cargos are strongly tied to Cyprus. In contrast to the three basket-handle amphora wrecks off the Turkish coast, the distribution of these containers at sites of consumption for the early Archaic period reveals a very different picture (Figure 6.4). Basket-handle amphoras probably have their typological origin in Cyprus with the earliest well-dated examples from late 8th-century tombs at Salamis (Karageorghis 1974, 115). Only in the 7th century does the type appear consistently outside Cyprus: at sites in Cilicia as well as the northern and southern Levant and as far south as Egypt (Sagona 1982, 106–108; Lehmann 1996, 443–445).6 A community or zone of interaction for the early Archaic basket-handle jars reveals the importance of Cyprus and the Levantine coast in consumption, a very different sphere from what we imagine from the three shipwreck sites. Archaeological evidence reveals a vibrant system of economic exchange between the Levant and East Greece in the Iron Age (Fantalkin 2006), despite rather poor representation in the record of known shipwrecks.7 This discrepancy highlights some of the problems of utilizing shipwrecks—moving spaces of transport rather than fixed spaces of consumption—for modeling networks of interaction for particular cargo items. There is always the problem of archaeological data collection. Scattered amphora sherds, including the distinctive thick cylinders that are the best markers of this type of container, are rarely described comprehensively in published site reports. If these zones of interaction do not offer a means of visualizing all communities connected by particular cargos, they can nonetheless help us to picture areas in which certain peoples and certain objects were connected by the sea. What happens when the shipwreck falls outside the zone of interaction best delineated by the cargo it carried? In the case of the Pointe Lequin 1A wreck (Figure 6.5) off the coast of France, the ship’s location only falls within a secondary zone of interaction imagined for the vessel, and only at the very edge of that zone. Based on its best-represented cargo items—Athenian fine wares and “Ionian cups,” probably manufactured in south Italy or Sicily—the primary zone of interaction is fairly limited; only with the expansion of the oval to include the less represented cargo items—Greco-Massaliot, southeast Aegean, and Ionian

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.007

SHIPWRECKS AS INDI CES OF TRADE NETWORKS

Kepçe Burnu

143

Kekova Adası

˘ Çaycagız Koyu

Minimum Number 1 (or unknown) 2-4 5 - 10 11 - 50 51 - 100+

Km

0

250

500

6.4. Zone of interaction for the shipwrecks at Kekova Adası, Kepçe Burnu, Çaycağ ız Koyu mapped alongside a distribution zone for early Archaic basket-handle amphoras found at sites of consumption. Inset depicts a reconstructed basket-handle amphora from the shipwreck at Kekova Adası. (Map by Justin Leidwanger; drawing by Lana Radloff.)

amphoras—does the oval include the wreck site as well as much of the rest of the Mediterranean. At what scale are these zones most meaningful? Can such a wide zone reflect the ship’s complete network or the space traversed by the mobile middle ground, even if the ship itself may have done little more than collect its cargo at some western Mediterranean outpost for travel toward Marseilles? If so, the information that the zone’s oval provides is necessarily limited, as the relative strength of ties between the shipwreck and neighboring communities cannot be determined. But what role does the geographic location of a shipwreck play in this system of cargo relationships? What role should be played by cargo elements beyond reflecting the broad dimensions of our area? How can proportions of cargo items factor into visualizations? Is there a better way than cartographic space to conceptualize shipwrecks and their roles in linking communities?

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.007

1,000 250

500

750

CYPRUS

N AFRICA

6.5. Zones of interaction for the Pointe Lequin 1A shipwreck. (Map by Thomas Kocjan.)

Km

0

IONIA IONI S ITA ITALY ETRURI RIA IIA A ETRURIA Pointe Lequin 1A IBERIA

Marseilles arseilles

144 Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.007

SHIPWRECKS AS INDI CES OF TRADE NETWORKS

145

SOCIAL NETWORK GRAPHING OF SHIPWRECKS: THE SHIP AS EGO-NETWORK

Social network graphs—created with programs like Gephi (Bastian, Heymann, and Jacomy 2009)—provide a means of visualizing communities marked by shipwrecks and their cargos. Considering shipwrecks within the framework of social network analysis raises its own set of challenges, some of which are addressed at the beginning of this chapter and elsewhere in this volume. To what degree do shipwrecks function as nodes within a broader network of economic interaction? How meaningful is one shipwreck in the identification of Mediterranean connectivity more broadly? And if we expand the picture to include multiple shipwrecks, how do we connect two sites that (almost certainly) never had any direct connection to each other? A shipwreck is clearly a site and as such should lend itself well to being represented as a node, but can a shipwreck also be an edge—or a bundle of edges—linking communities to each other? If we transform shipwrecks from nodes into edges, what nodes remain and how can these be used to detect meaningful communities? In short, what data are useful to include in network visualization?8 In the construction of tabular data for inclusion in the present network explorations, I begin at the level of the individual shipwreck. Conceptualizing a shipwreck from a network perspective raises the question whether to consider the wreck as a community itself, with external ties embedded in the single community of links visible together at this discrete site, or whether to treat the wreck as one node or edge within a broader community of wrecks from a single period. For the former approach, the network connections might best be described as ego-centered, or the network as comprising an individual and his immediate contacts (Everett and Borgatti 2005). The resulting network should reveal a social world from the perspective of the ego. Linton Freeman (1982, 291) explains, if the social units are persons, those in the ego network are the ones that have the greatest impact on ego’s attitudes, norms, values, goals and perceptions of the world . . . In some sense, then, an ego network is a map of ego’s personal social world. It shows something about how that ego is tied in to the larger human society.

If the units are archaeological sites, then the material reflected in the ego network offers the clearest identifiers of the ego’s origins, relationships, and connections to its sociocultural and technological environment. In his utilization of the ego network for analysis of the 14th-century AD site of Kelbey’s Ridge 2, Saba, in the northeastern Caribbean, Angus Mol (2014) advocates the inclusion of different types of relational datasets into a single ego network or centered graph. As Mol, Hoogland, and Hofman describe (2015,

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.007

146

ELIZABETH S. GREENE

278), “a site ego-network model connects a site to its own archaeological record or intra- and inter-site ties regardless of the scale of these interactions. In other words, by applying an ego-network approach, it becomes theoretically possible to abstract and explore a site as a multi-levelled, relational entity.” With the application of the ego-network approach to the Kelbey’s Ridge 2 site, Mol is able to productively map certain aspects of micro-scale and inter-site relationships based on a broad range of datasets including structures, hearths, burials, and objects of local or imported production. By extension, on a single shipwreck site we might find the copresence of structures (the hull itself along with recognizable usage areas), equipment associated with the hull (anchors, steering oars, pumps, etc.), cargo, galley wares, and personal items belonging to crew or passengers. While the information available about each category of objects may not be parallel in terms of material, origin, or construction methods, resulting in a diverse and multimodal network, the chance to look at the different components of this ego network together in a single visualization might draw out connections that are not otherwise clear, and reveal specific aspects of the particular relationships that tie one wreck into the broader spectrum of Mediterranean seafaring. To visualize a wreck as a self-contained ego network,9 I divided the Pabuç Burnu site’s artifacts and features into a series of functional groups: cargo, galley wares, ship and ship’s equipment (including hull timbers, anchors, and fishing weights). Each group includes available information, as relevant, regarding form, production region, material, etc. The diverse data raise a question of how to connect features and artifacts such as hull remains (whose attributes include construction type and wood species) and other equipment associated with the sailing and mooring of the vessel (anchors and tackle), tools (fishing weights or maintenance devices, each distinct in material and form), alongside cargo and galley wares (with attributes of forms, fabrics, regional production zones) and the related but distinct organic remains. The graph in Figure 6.6 collects a number of these categories to produce an ego network according to a ForceAtlas 2 layout. This is a force-directed layout that spatializes the network by simulating a physical system in which nodes repulse each other like charged particles, while edges attract their nodes like springs (Jacomy et al. 2014, 2); in this layout, better-connected nodes are drawn toward the center of the graph; more distant nodes are located on the graph’s periphery. Nodes are colored according to their category (cargo, galley wares, ship and equipment, and types of object or regional forms); they are sized according to degree centrality, or the number of connections they have to other nodes in the network.10 A first look at an ego network produced from the site data reveals connections through fabric types between cargo and galley wares, most clearly visible in the Halicarnassian and Tan Fabric Southern Ionian categories. These two fabric types also comprise the majority of the amphoras represented in the

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.007

SHIPWRECKS AS INDI CES OF TRADE NETWORKS

147

6.6. Ego network for the shipwreck at Pabuç Burnu, including cargo, galley items, the ship and its equipment, and regional ceramic forms marked in shades of gray.

cargo. Such intra-site connections might offer a sense of the primary operational sphere of the vessel, but in this case they are limited by a difficulty in confidently linking the Tan Fabric Southern Ionian jars to a particular city’s production; Greene, Lawall, and Polzer (2008) note certain similarities in color and inclusions to fabrics from Hellenistic Rhodes, but do not specify a single point of origin for these amphoras. According to Leidwanger’s (2013) calculation of sailing distances, much of Rhodes would have been within a day’s sail of Halicarnassus and Pabuç Burnu, suggesting a rather small zone of primary operation for the vessel. The attempt to construct such graphs reveals some of the limitations of imbalanced archaeological data. At Pabuç Burnu, we face the challenge of interpreting connections between forms and fabrics plausibly associated with specific sites and those that can only be connected to broader regions. A different sort of example comes from the Archaic shipwreck at Giglio, Italy (Bound 1991), where looting of the site makes it nearly impossible to distinguish between cargo items and galley wares since the context and quantity of individual objects and categories of artifacts are uncertain. While connections through regional forms and fabrics between cargo containers and galley wares are clear on the Pabuç Burnu ego-network graph, the

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.007

148

ELIZABETH S. GREENE

6.7. Ego network for the shipwreck at Pabuç Burnu expanded to include Archaic wrecks that share construction elements and ship’s equipment, cargo items, and galley wares. If known, regional or typological identifiers are provided for ceramics, e.g., Lesbian, Klazomenian, Halicarnassian, Southern Ionian Slipped, etc.

hull and equipment branches stand at a distance from other sections of the network. In part, this is a function of the nature of the data. Nili Liphschitz (2005) has identified the wood used in the laced hull remains from the ship as Pinus nigra, a species native to Turkey, Cyprus, Crimea, west Caucasia, the Balkans, the south Carpathians, and western Syria. Liphschitz concludes from her analysis of the hull remains that the ship was built in western Anatolia, the most intensive region of Pinus nigra growth in Turkey, but this designation is far broader than the production regions suggested by the ceramic types, so we cannot link the hull remains to a single specific region of ceramic production. For shipwrecks with cargo, galley wares, or other tools and equipment that

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.007

SHIPWRECKS AS INDI CES OF TRADE NETWORKS

149

have been more closely sourced, the relationships between the different components or features of the site will likely be clearer.11 Beyond the level of the single site, extending the ego network offers the potential to consider external connections through hull construction, for example where the laced technique is represented in timbers from eight vessels with preserved wooden remains from the 7th and 6th centuries.12 The graph in Figure 6.7 extends the ego network to include Archaic shipwreck sites on which there are wooden remains with evidence of laced construction or a stone anchor stock. Such similarities suggest parallel technology rather than common geographic origin. While joinery and anchor typology might follow broad trends or availability, these features rarely can be associated with specific cities or regions. The graph reveals an expansion of the Pabuç Burnu ego network to include wrecks discovered around the Mediterranean, and a ship that appears from its cargo to be highly regional in nature demonstrates much wider connections in construction technology, shipboard equipment, cargo, and galley wares, hinting at the general similarities among the various ships that sailed the Archaic Mediterranean. Notable on the graph is the network proximity of the Pabuç Burnu wreck through these elements to western sites including those at Bon Porté, Giglio, and Gela, all modest traders whose international (from a western perspective) cargo may reflect goods gathered from regional hubs for distribution of a smaller scale. Links between shared components expand the ego network widely, but the multilevel or multimode relationships quickly become difficult to track in a jumble of artifact types, origins, materials, and construction methods.13 VISUALIZING SHIPWRECKS AND COMMUNITIES

The ego network offers a lens through which to look at connections among the components of a single wreck alongside links to other sites through a variety of aspects that are particular to the ego. It is less helpful as a path toward identifying the broader communities that link multiple wrecks in a particular period or region. To enable this broader view, fourteen Archaic shipwreck assemblages have been compiled, all dated between 700 and 500 BC (see Figure 6.1): Pabuç Burnu, Kekova Adası, Kepçe Burnu, Çaycağ ız Koyu (Turkey); La Love, Bon Porté, Pointe Lequin 1A, Dattier, Miet 3, Grand Ribaud F (France); Giglio, Gela (Italy); Cala Sant Vicenç (Spain); Akio (Greece).14 Some of these wrecks have been excavated completely; others were the subject of survey alone. For this reason a few caveats should be noted about the data. Stone anchor stocks, a common feature of Archaic wrecks, are frequently identified in survey, though we cannot always assume a direct correlation between the wreck and a nearby anchor stock. Wooden hull remains, which normally survive only as a result of being buried in deep sand, are unlikely to be recorded in unexcavated

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.007

150

ELIZABETH S. GREENE

wrecks. Major cargo items of ships should generally be the most visible and therefore the most easily detected in both excavated and unexcavated (surveyed) sites. Weaker ties based on a one or two scattered finds are probably less well represented in the surveyed and partially excavated sites (Kekova Adası, Kepçe Burnu, Çaycağ ız Koyu, Grand Ribaud F, and Akio), while heavily looted sites (e.g., Giglio) pose interpretive problems as a result of loss of contexts and complete assemblage profiles. How, then, can shipwrecks be connected logically to other wrecks and to communities through their artifacts? As a path toward associating shipwrecks and communities, it seems best to focus on the geographic locations of wrecks and the cities or regions with which shipboard items—whether cargo or galley wares—can be associated. The resultant network structure is visualized as a two-mode graph, containing both shipwreck sites and cargo origins.15 For Archaic shipwrecks, sourceable cargos and galley wares are generally ceramics.16 Identifiable galley wares should offer insight into the cultural background of the sailors and stayed aboard the ship for multiple journeys; cargos were surely more ephemeral and, with the exception of scattered detritus from earlier ventures, generally speak for the goods carried on the ship’s final voyage. Both types of object provide information about the ship’s social and economic orbit. While the presence of an amphora or cup from a particular city does not necessarily signify that the vessel made a stop at that location to pick up or drop off cargo, some network connection is plausible, through a direct or indirect link. Certain common forms can be clearly connected to specific regions, such as amphoras from Klazomenai and Chios; other forms, including southeast Aegean (e.g., Samian, Milesian, and Halicarnassian), Etruscan, and Massaliot amphoras, or the “Ionian cups” that feature on wrecks in both the eastern and western Mediterranean, can be connected only to broader regions, rather than specific cities. For the purposes of graphing, I assign node designations to ceramic types; these include specific ancient cities when possible and broader regions when identifications are too uncertain to allow for more than a general area. For the latter, I select geographic positions that represent the approximate center point of the region. For each shipwreck in the sample, node and edge tables were created that included shipboard artifacts (cargo, galley wares, and personal items) for which geographic origins were available in published reports. Because the analysis focused only on geographic information (shipwreck locations and cargo origins), features such as hull remains and anchors for which such data are insufficiently specific are not included. Edge tables link shipwreck and object origins through undirected ties weighted on a scale of one to five. The scale is relative and based on the proportion of shipboard items associated with a particular region, on the assumption that greater numbers of objects suggest stronger ties. In the case of the Pabuç Burnu shipwreck, for example,

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.007

SHIPWRECKS AS INDI CES OF TRADE NETWORKS

151

6.8. Geographic network comprising shipwreck sites and cargo origins using Gephi’s ForceAtlas2 layout; shades of gray reflect communities determined by Gephi’s modularity filter.

the Pabuç Burnu–southeast Aegean edge (marked by the presence of more than 200 Halicarnassian, Tan-Fabric Southern Ionian, and Southern Ionian slipped amphoras and galley wares) received a weight of five, while the Pabuç Burnu–Lesbian edge (based on the presence of two amphora toes) received a weight of only one.17 The graph in Figure 6.8 depicts a two-mode network comprising shipwrecks and cargo origins using a ForceAtlas2 layout. Node size on the graph is determined according to betweenness centrality, or their importance in the network based on their position along paths between many other node pairs (Collar 2015, 17). The Pointe Lequin 1A, Grand Ribaud F, Cala Sant Vicenç, and Gela wrecks, all carrying diverse cargos from a range of origins across the Mediterranean, are notable for their high centrality in contrast to the Akio, Kepçe Burnu, Çaycağ ız Koyu, and Miet 3 wrecks, which demonstrate links only to small communities. From the perspective of cargo origins, Corinthian, Etrurian, Massaliot, north Ionian, Aegean, and Attic wares participate as more broadly connected nodes than the more isolated Iberian, Cypriot, and Laconian wares, which seem more peripheral to the network as a whole. Node and edge shading in the graph is determined by Gephi’s modularity algorithm, which identified four communities. This algorithm divides a network into a series of sub-units or communities comprising interconnected nodes (Blondel et al. 2008; Graham 2014). The communities indicated by this shading highlight the strongest groupings of nodes. The small sample size of Archaic

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.007

152

ELIZABETH S. GREENE

shipwrecks and cargo origins, as well as their connection through geographic location, results in a sensible identification through modularity of a community structure that links shipwrecks, cargos, and their regions of interaction.18 Within the four communities identified through the modularity algorithm, we see an assortment of different maritime zones ranging from small to large; in order of geographic size, these are: (1) A small and local community of the Akio wreck and the north Aegean, based on the shipwreck at Akio and its sole known cargo of north Aegean amphoras. Weak network ties link the north Aegean with the Cala Sant Vicenç and Pointe Lequin 1A wrecks. The location of the Akio site hints at certain ties to Attica. (2) An eastern Mediterranean community in which commodities from the southeast Aegean and Cyprus are connected through shipwrecks at Pabuç Burnu, Kekova Adası, Kepçe Burnu, and Çaycağ ız Koyu. Within this community, the southeast Aegean demonstrates a high centrality, with cargo links to an additional four wrecks discovered farther west in the Mediterranean (Gela, Grand Ribaud F, Giglio, and Pointe Lequin 1A). (3) A community in which Corinthian, Etruscan, and Laconian wares are connected through their presence on the Giglio, Bon Porté, and Grand Ribaud F shipwrecks; weaker ties link the Miet 3 and La Love wrecks to this network of Corinth, Etruria, and Laconia. (4) The broadest geographic community in which Attic, south Italian, north Ionian, Massaliot, Punic or North African, and Iberian wares are associated through the Pointe Lequin 1A, Gela, Cala Sant Vicenç, and Dattier shipwrecks. The cargo items on these wrecks and their network community hint at the largest scale of international trade in the Archaic Mediterranean. Cities such as Athens, Massalia, Emporion, Carthage, and perhaps northern Ionian cities like Klazomenai or Mytilene, and cities in Magna Graecia like Gela or Locri, served as nodes in this larger system of transportation that allowed goods to circulate more widely from central hubs. We cannot say with certainty whether the ships themselves participated in such broad networks directly, or whether they simply reflect local vessels that collected their wares from well-connected ports of trade. Nonetheless, the presence of such diverse cargos on these wrecks speaks for the connections that allowed these goods to be gathered at one or many centers.

These connected communities are perhaps most easily visualized through the application of Gephi’s MultiMode Networks Projection filter, which collapses two-mode networks into one-mode networks. In these graphs, we effectively transform shipwrecks into the edges that link geographic nodes to explore how the communities described above are connected through shipwreck edges, a cleaner visualization of connected communities than those containing the

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.007

SHIPWRECKS AS INDI CES OF TRADE NETWORKS

153

6.9. ForceAtlas2 graph of a network community represented by cargo origins connected by shipwreck edges.

6.10. GeoLayout graph of a network community represented by cargo origins connected by shipwreck edges.

added noise of the individual assemblages. Two graphs are used here to show the same data in two contrasting layouts: ForceAtlas (Figure 6.9) and geographic (Figure 6.10).19 The single-mode graph now highlights the clusters of regions linked by seaborne traffic. An alternative approach collapses the network into a one-mode graph in which objects are used as the edges to join shipwrecks (Figure 6.11). This new graph displays links among shipwrecks through their shared cargos, offering some clues to the different clusters or communities in which Archaic ships operated. These are, broadly speaking: a regional eastern Mediterranean cluster, a regional cluster around Italy and France, and the more international circuit reflecting the increasing globalization of the Archaic world through direct voyages or the influential role of central ports of trade.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.007

154

ELIZABETH S. GREENE

6.11. One-mode graph in which shared cargos are used as the edges to join shipwrecks. The graph uses a ForceAtlas layout; nodes are sized according to betweenness centrality. Connected communities are indicated by similar shading.

Limited by the sample size of shipwreck data, the communities seen in these visualizations are in no way indicative of the only maritime connections in the Archaic Greek world, but present, I believe, a sample of the various scales of interaction similar to those discussed by Tom Tartaron in this volume. These range from the coastscapes of everyday life, to the maritime small worlds of habitual experience, to the regional or intracultural maritime sphere, and the interregional or intercultural maritime sphere (see also Tartaron 2013). Through network graphing and modularity exploration, we can begin to gain a sense of the different types of community that linked the Mediterranean through shipborne traffic. Measurements of centrality offer a sense of the cities and regions that played greater and lesser roles in tying together communities of different scales.20 CONCLUSIONS: SHIPWRECK COMMUNITIES AND SOCIAL RELATIONSHIPS

From the standpoint of distribution within an emerging agricultural economy, the wreck assemblages offer solid archaeological evidence for expanding relationships between the Greek world and the broader Mediterranean, with

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.007

SHIPWRECKS AS INDI CES OF TRADE NETWORKS

155

spheres of interaction that range from local links or island communities (e.g., Broodbank 2000; Knappett, Evans, and Rivers 2008) to the large-scale globalization or “Mediterraneanization” models that have been proposed by Ian Morris (2003), Tamar Hodos (2009) and others for the Iron Age (Knappett 2016; Sherratt 2016). Returning to the “zones of interaction” with which I began this chapter, we can use our network communities to represent a series of overlapping zones that hint at a few of the multifaceted connections of the Archaic world (Figure 6.12). Human relational networks might provide another possible way forward, as an interpretive frame or an additional facet of a network visualization. The literary record preserves evidence of exchange of what Lin Foxhall (1998) calls “semi-luxuries,” or processed agricultural goods like wine, olive oil, or other fashionable food products that traveled over the seas. Sappho’s brother Charaxus (Strabo, Geog. 17.1.33) exports Lesbian wine to Naucratis. Iliad 7 records the commerce of wine between men of similarly high status on the shores of Troy. From the regions of Samos and Corinth, Herodotus (4.152) tells of Colaeus, the Samian whose travels to the west gained him sufficient profit to dedicate a six-talent bronze cauldron at the Heraion. For Corinth, Dionysius of Halicarnassus (Rom. Ant. 3.46) recounts the tale of Demaratus the Bacchiad who made his fortune through transporting Corinthian pottery westward to Etruria.21 These human relationships further texture the communities depicted by Gephi’s community visualizations as we consider how Demaratus might have participated in some network interaction that linked Corinth and Etruria through the shipboard transport of fine wares. Archaeologically, the presence of mortaria and basket-handle amphoras alongside bronze and ivory in the Royal Tombs at Salamis on Cyprus suggests a similar level of elite participation in the distribution and consumption of processed agricultural goods and other preferential foods in an eastern Mediterranean network. Members of such elite families might be best equipped to initiate ventures of international trade, forging horizontal bonds of friendship between different regions. This multifaceted exchange reveals a large-scale mobility of people, customs, and commodities, as well as a web of relationships between those who had sufficient agricultural surplus to export, capital to fund a venture, and desire to forge connections. Existing personal relationships, and the possibility of building new ones, could have been far more important to our Archaic sailors than distance, sailing speed, or the pursuit of profit. Mapping such networks of human relations provides yet another challenge, and surely yet another set of overlapping spheres. Like Foucault’s boat, the ship is real and illusory, mappable and unmappable. I have tried here to look at how shipwrecks—dynamic, multifaceted interactions of systems—might be approached as material markers of exchange networks, and how they might be used to conceptualize network communities. As simple dots

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.007

156

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.007

La Love

S ITA ITALY

Km

0

250

500

750

Kepçe pçe Burnu nu Kepç Kekova Adasi K si CYPRUS YPRU Çayc ayca yccagiz Çaycagiz K Ko Koyu

IONIA ION

6.12. Network communities and overlapping zones of interaction in the Mediterranean. (Map by Thomas Kocjan.)

Dattier Bon Porté Miet 3 Pointe Lequin 1A Grand Ribaud F

Jules Verne 7 & 9 & César 1 Marseilles

Gela

ETRURIA RIA IA G Giglio

N AFRICA

Cala Sant Vicenç

IBERIA RIA

1,000

SHIPWRECKS AS INDI CES OF TRADE NETWORKS

157

on a map, wrecks serve only a limited function; instead they may comprise small relational spaces themselves, serving as the bundles of edges that link communities within maritime landscapes of contrasting scales. The circulating blood of the network itself, they are unmappable “other spaces” or “middle grounds,” floating in and encapsulating the Mediterranean in which they sail. NOTES 1.

2.

3. 4.

5.

6.

7.

8.

9. 10. 11.

12.

13.

See, for example, Pulak (2008, 298), and Vichos (1999, 93) on hypothetical routes based on cargo origins for the Late Bronze Age shipwrecks at Uluburun and Point Iria respectively. Seland (2015) offers a series of conceptual suggestions for modeling connectivity. See Krotscheck (2008; 2015), Krotscheck, Ferguson, and Glascock (2009), for the results of analysis of the so-called Ionian B2 cups, which seem to have been manufactured somewhere in south Italy or Sicily. See, for example, Morton (2001) on the marine environment generally; Leidwanger (2013) on the role of winds; and various chapters in Harris and Iara (2011) on technology. Reed (2003) and Tandy (2004) use literary and historical evidence to offer perspectives on the status and social relationships between individuals who conducted business on the sea in the Archaic Greek world. Studies such as these might inform the archaeologically visible networks discussed here. Also focused on the historical evidence, Malkin (2011) takes a network approach to colonial ventures in the Archaic world, focusing on identity and hybridity. Lawall (2003) suggests that the typical paradigm of associating specific amphora shapes with specific cities of production in the southeast Aegean during this early period should probably be abandoned. In the Aegean, only a few examples have been reported and these are generally late: on Rhodes, Kommos, and Miletus: Jacopi (1929, tav. IV; 1931, tav. VIII); Johnston 1993, 370; 2005, 358 and 372); Niemeier (1999, 389–392). On the history and distribution of the basket-handle amphora, see Leidwanger (2007). Future survey activity along the Levantine coast would surely change the current status quo. Already, targeted exploration by Ballard et al. (2002) has resulted in the discovery of two 8th-century shipwrecks in deep water west of Ashkelon. For the purpose of this project, Gephi was selected based on its Macintosh-friendliness, its relative ease of use, and its still powerful analytical components. It is also among the more commonly used programs to date in the field. I am grateful to Shawn Graham for advice on using the software and for his assistance in helping me to brainstorm ideas for modeling and implementation. I am grateful to Barbara Mills for suggesting this path of analysis. For more extensive definitions of betweenness centrality and other archaeological applications of key network terminology, see the glossary in Collar et al. (2015), 17–25. The Late Bronze Age shipwreck at Uluburun (Pulak 2008), for example, with wellsourced ceramics, metals, and organics in its cargo, would make an excellent candidate for an ego-network approach to a shipwreck. See Polzer (2005), 99. Although there are no published hull remains from the Pointe Lequin 1A shipwreck, lacing on the the preserved rudder (Long, Miro, and Volpe 1992; Krotscheck 2008, 73–74) may allow a connection. Laced repairs on the César 1 wreck allow its inclusion here, but the primary mode of construction was based on mortise-andtenon joinery. Hull remains are, by definition, exclusively discovered at shipwreck sites, on land or underwater, but anchors and other elements of a ship’s equipment might be found on land as well. A number of stone anchor stocks have been dedicated in sanctuary contexts (see, for example, Gianfrotta 1977), through which the Pabuç Burnu’s ego network might

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.007

158

ELIZABETH S. GREENE

14.

15.

16.

17. 18.

19.

20.

21.

again be expanded beyond the site level to suggest a broader community of connections. A similar expansion of the graph could track the presence of particular amphora types or vessels in settlement sites and tombs on land. Pabuç Burnu (Greene et al. 2008); Kekova Adası, Kepçe Burnu, Çaycağ ız Koyu (Greene, Leidwanger, and Özdaş 2013); La Love, Bon Porté, Pointe Lequin 1A, Dattier, Miet 3, Grand Ribaud F (Long, Pomey, and Sourisseau 2002); Giglio (Bound 1991), Gela (Panvini 2001); Cala Sant Vicenç (Nieto and Santos 2009); Akio (Koutsouflakis and Kourkoumelis 2006). The list is not meant to represent an exhaustive catalogue of Archaic Greek shipwrecks. The Archaic wrecks at Place Jules Verne (7 and 9) (Pomey 1995; 2002), and César 1 (Pomey 2001) preserve only hull remains, but lack cargo; the Jules Verne 7 and César 1 wrecks are constructed primarily with mortise-and-tenon joinery, but have lacing at their endposts and in repairs. The two Phoenician wrecks at Mazarrón (Negueruela 2014; Negueruela et al. 1995) and Bajo de la Campaña (Polzer 2014) are also omitted. The predominantly Phoenician cargo of the Mazarrón wrecks does not offer clear links to other wrecks in the sample; no comprehensive publication of cargo or origins is yet available for the Bajo de la Campaña wreck, which also seems to be predominantly Phoenician. See Borgatti and Everett (1997, 243–245) for discussion of one-mode and two-mode networks. One example is the preferential assignment of graduate students to faculty members as research assistants. A two-mode network would regard faculty and graduate students as separate entities; a one-mode network would see both as a single entity (i.e., people). Seland (2016) productively utilizes a two-mode network to investigate the interrelationships between ports and commodities in the Periplus of the Erythraean Sea. For the Archaic wrecks that comprise the dataset for this paper, only the ceramic items were well associated with points of origin. Although metals and stone from other shipwrecks have been effectively sourced (see Pulak 2008 on the Uluburun shipwreck), this is not the case for wrecks from the 7th and 6th centuries. See Greene, Lawall, and Polzer (2008) for the relative numbers of object type and fabric category. When the modularity algorithm was run after the application of Gephi’s MultiMode Networks Projection filter, it generated uncertain communities that appear far less meaningful to me than those described here. See http://blog.ouseful.info/2012/11/09/drugdeal-network-analysis-with-gephi-tutorial for discussion of a very different sort of twomode network upon which Gephi’s modularity filter reveals meaningful community divisions. A similar study to mine might be run on a larger sample size of shipwrecks, perhaps drawing on a specific period from Parker’s (1992) catalogue of sites. See note 19 above and Graham (2014). In the creation of these one-mode graphs, I did not run the modularity filter a second time after running the MultiMode transformation. The compression of data in the transformation process made the new communities less meaningful in any sort of network terms. There are many different approaches to the production of Gephi graphs for discerning networks of shipwrecks, which reflect the data used to create them. Leidwanger (2017) takes a broadly similar approach to the one detailed here, but categorizes his data slightly differently. The current project is designed to be exploratory, opening paths for future study. Salmon (1984, 106, 114) discusses the archaeological evidence for Corinthian trade with Etruria.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.007

SHIPWRECKS AS INDI CES OF TRADE NETWORKS

159

REFERENCES

Antonaccio, C. 2013. Networking the middle ground. Archaeological Review from Cambridge 28(1), 213–236. Arnaud, P. 2005. Les routes de la navigation antique: Itinéraires en Méditerranée. Paris: Editions Errance. Aslan, E. 2015. Kekova Adası 2012–2013 yılı sualtı araş tırmalarında bulunan amphoraların tipolojik değ erlendirmesi. Olba 23, 321–369. Ballard, R.D., Stager, L., Master, D., Mindel, D., Whitcomb, L.L., Singh, H., and Piechota, D. 2002. Iron Age shipwrecks in deep water off Ashkelon, Israel. American Journal of Archaeology 106, 151–168. Bastian, M., Heymann, S., and Jacomy, M. 2009. Gephi: an open source software for exploring and manipulating networks. Third International AAAI Conference on Weblogs and Social Media (ICWSM 2009), San Jose, CA, 2009. www.aaai.org/ocs/index.php/ICWSM/ 09/paper/view/154. Blondel, V.D., Guillaume, J.-L., Lambiotte, R., and Lefebre, E. 2008. Fast unfolding of community hierarchies in large networks. arXiv 0803.0476. http://works.bepress.com /lambiotte/4. Borgatti, S.P., and Everett, M.G. 1997. Network analysis of 2-mode data. Social Networks 19, 243–269. Bound, M. 1991. The Giglio Wreck: A Wreck of the Archaic Period (c.600 BC) off the Tuscan Island of Giglio. An Account of Its Discovery and Excavation: A Review of the Main Finds. Enalia Supplements 1. Athens: Hellenic Institute of Marine Archaeology. Broodbank, C. 2000. An Island Archaeology of the Early Cyclades. Cambridge: Cambridge University Press. Collar, A., Coward, F., Brughmans, T., and Mills, B.J. 2015. Networks in archaeology: phenomena, abstraction, representation. Journal of Archaeological Method and Theory 22, 1–32. Everett, M., and Borgatti, S.P. 2005. Ego network betweenness. Social Networks 27, 31–38. Fantalkin, A. 2006. Identity in the making: Greeks in the eastern Mediterranean during the Iron Age. In A. Villing and U. Schlotzhauer (eds), Naukratis: Greek Diversity in Egypt. Studies on East Greek Pottery and Exchange in the Eastern Mediterranean, 199–208. British Museum Research Publication 162. London: British Museum. Foucault, M. 1986. Of other spaces. Trans. J. Miskowiec. Diacritics 16, 22–27. Foxhall, L. 1998. Cargoes of the heart’s desire: the character of trade in the Archaic Mediterranean world. In N. Fisher and H. van Wees (eds), Archaic Greece: New Approaches and New Evidence, 303–7. London: Duckworth. Freeman, L.C. 1982. Centered graphs and the structure of ego networks. Mathematical Social Sciences 3, 291–304 Gianfrotta, P. 1975. Le ancore votive di Sostrato di Egina e di Faillo di Crotone. La parola del passato 30, 311–318. Graham, S. 2014. On connecting stamps: network analysis and epigraphy. Les nouvelles de l’archéologie 135, 39–44. Greene, E.S., Lawall, M.L., and Polzer, M.E. 2008. Inconspicuous consumption: the sixthcentury B.C.E. shipwreck at Pabuç Burnu, Turkey. American Journal of Archaeology 112, 685–711. Greene, E.S., Leidwanger, J., and Özdaş , H. 2011. Two early Archaic shipwrecks at Kekova Adası and Kepçe Burnu, Turkey. International Journal of Nautical Archaeology 40(1), 60–69.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.007

160

ELIZABETH S. GREENE

Greene, E.S., Leidwanger, J., and Özdaş , H. 2013. Expanding contacts and collapsing distances in early Cypro-Archaic trade: three case studies of shipwrecks off the Turkish coast. In M.L. Lawall and J. Lund (eds), The Transport Amphorae and Trade of Cyprus, 21–34. Gösta Enbom Monographs, Vol. 3. Aarhus: Aarhus University Press. Harris, W.V., and Iara, K. (eds) 2011. Maritime Technology in the Ancient Economy: ShipDesign and Navigation. Journal of Roman Archaeology Supplement 84. Portsmouth: Journal of Roman Archaeology. Hodos, T. 2009. Colonial engagements in the global Mediterranean Iron Age. Cambridge Archaeological Journal 19(2), 221–241. Horden, P., and Purcell, N. 2000. The Corrupting Sea: A Study of Mediterranean History. Oxford and Malden: Blackwell. Jacomy M., Venturini, T., Heymann, S., and Bastian, M. 2014. ForceAtlas2: a continuous graph layout algorithm for handy network visualization designed for the Gephi software. PLoS ONE 9(6), e98679. doi:10.1371/journal.pone.0098679. Jacopi, G. 1929. Scavi nella necropoli di Jalisso, 1924–1928. Clara Rhodos 3. Rhodes: Istituto Storico-Archeologico. Johnston, A. 1993. Pottery from Archaic Building Q at Kommos. Hesperia 62(3), 339–382. Karageorghis, V. 1974. Excavations in the Necropolis of Salamis, Vol. 3. Nicosia: Department of Antiquities. Knappett, C. 2016. Globalization, connectivities and networks: an archaeological perspective. In T. Hodos (ed.), The Routledge Handbook of Archaeology and Globalization, 29–41. London: Routledge. Knappett, C., Evans, T., and Rivers, R. 2008. Modelling maritime interaction in the Aegean Bronze Age, Antiquity 82, 1009–1024. Koutsouflakis, G.B., and Kourkoumelis, D. 2006. Ναυαγιο υστερης αρχαικης περιοδου στο νοτιο Ευβοικο. Archaiologika Analekta ex Athenon 39, 83–104. Krotscheck, U. 2008. Scale, structure, and organization of Archaic maritime trade in the western Mediterranean: the “Pointe Lequin 1A.” PhD dissertation, Stanford University. Krotscheck, U. 2015. Pointe Lequin 1A: wine cups and economic networks in the western Mediterranean. Ancient West & East 14, 169–189. doi:10.2143/AWE.14.0.3108192. Krotscheck, U., Ferguson, J.R., and Glascock, M.D. 2009. 15. Annex 2. Results of the Neutron Activation Analysis (NAA) of ‘Ionian B2’ cups from Cala Sant Vicenç and Emporion. In X. Nieto and M. Santos (eds), El vaixell grec arcaic de Cala Sant Vicenç, 323–327. Monografies del CASC 7. Barcelona: Museu d’Arqueologia de Catalunya. Lawall, M.L. 2003. Ilion before Alexander: amphoras and economic archaeology. Studia Troica 12, 197–244. Lehmann, G. 1996. Untersuchungen zur späten Eisenzeit in Syrien und Libanon: Stratigraphie und Keramikformen zwischen ca. 720 bis 300 v.Chr. Münster: Ugarit-Verlag. Lehmann, G. 2002. Iron Age. In A. Kempinski (ed.), Tel Kabri: The 1986–1993 Excavation Seasons, 178–222. Tel Aviv: Institute of Archaeology, Tel Aviv University. Leidwanger, J. 2007. The Cypriot transport amphora: notes on its development and distribution. Skyllis 7, 24–31. Leidwanger, J. 2013. Modeling distance with time in ancient Mediterranean seafaring: a GIS application for the interpretation of maritime connectivity. Journal of Archaeological Science 40, 3302–3308. Leidwanger, J. 2017. From time capsules to networks: new light on Roman shipwrecks in the maritime economy. American Journal of Archaeology 121, 591–619.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.007

SHIPWRECKS AS INDI CES OF TRADE NETWORKS

161

Liphschitz, N. 2005. Dendroarchaeological investigations: 398 and 404: Pabuç Burnu shipwreck. Unpublished report on file. Long, L., Miro, J., and Volpe, G. 1992. Les épaves archaïques de la Pointe Lequin (Porquerolles, Hyères, Var). In M. Bats, G. Bertucchi, G. Conges, and H. Treziny (eds), Marseille grecque et la Gaule, 199–234. Etudes Massaliètes 3. Lattes: A.D.A. M. é ditions. Long, L., Pomey, P., and Sourisseau, J.-C. (eds) 2002. Les Etrusques en mer: Epaves d’Antibes à Marseille. Aix-en-Provence: Musées de Marseille, Edisud. Malkin, I. 2002. A colonial middle ground: Greek, Etruscan and local elites in the Bay of Naples. In C. Lyons and J. Papadopoulos (eds), The Archaeology of Colonialism, 151–181. Los Angeles: Getty Research Institute. Malkin, I. 2011. A Small Greek World: Networks in the Ancient Mediterranean. Oxford: Oxford University Press. Mol, A.A.A. 2014. The Connected Caribbean: A Socio-material Network Approach to Patterns of Homogeneity and Diversity in the Pre-colonial Period Caribbean. Leiden: Sidestone Press. Mol, A.A.A., Hoogland, M.L.P., and Hofman, C.L. 2015. Remotely local: ego-networks of late pre-colonial (AD 1000–1450) Saba, north-eastern Caribbean. Journal of Archaeological Method and Theory 22, 275–305. Morris, I. 2003. Mediterraneanization. Mediterranean Historical Review 18, 30–55. Morton, J. 2001. The Role of the Physical Environment in Ancient Greek Seafaring. Leiden: Brill. Negueruela Martinez, I. 2014. The Phoenician ships of Mazarrón. In J. Aruz, Y. Rakic, and S. Graff (eds), Assyria to Iberia: At the Dawn of the Classical Age, 243–246. New York: Metropolitan Museum of Art. Negueruela, I., Pinedo, J., Gómez, M., Miñano, A., Arellano, I., and Barba, J.S. 1995. Seventh-century BC Phoenician vessel discovered at Playa de la Isla, Mazarron, Spain. International Journal of Nautical Archaeology 24, 189–197. Niemeier, W.-D. 1999. “Die Zierde Ioniens”: ein archaischer Brunnen, der jüngere Athenatempel und Milet vor der Perserzerstörung. Archäologischer Anzeiger 1999(3), 373–413. Nieto, X., and Santos, M. 2009. El vaixell grec arcaic de Cala Sant Vicenç. Monografies del CASC 7. Barcelona: Museu d’Arqueologia de Catalunya. Panvini, R. 2001. The Archaic Greek Ship at Gela (and Preliminary Exploration of a Second Greek Shipwreck). Trans. B.E. McConnell. Palermo: Salvatore Sciascia Editore. Parker, A.J. 1992. Ancient Shipwrecks of the Mediterranean and the Roman Provinces. BAR International Series 580. Oxford: Tempus Reparatum. Polzer, M.E. 2005. Hull remains from the Pabuç Burnu shipwreck and early transition in Archaic Greek shipbuilding. M.A. thesis, Texas A & M University. Polzer, M.E. 2014. The Bajo de la Campaña shipwreck and colonial trade in Phoenician Spain. In J. Aruz, Y. Rakic, and S. Graff (eds), Assyria to Iberia: At the Dawn of the Classical Age, 230–242. New York: Metropolitan Museum of Art. Pomey, P. 1995. Les épaves grecques et romaines de la place Jules-Verne à Marseille. Comptes rendus des séances de l’Académie des inscriptions et belles-lettres, 459–484. Pomey, P. 2001. Les épaves grecques archaïques du VIe siecle av. J.-C. de Marseille: Epaves Jules-Verne 7 et 9 et César 1. In H. Tzalas (ed.), Tropis VI. Proceedings of the 6th International Symposium on Ship Construction in Antiquity, Lamia, 1996, 425–437. Athens: Hellenic Institute for the Preservation of Nautical Tradition. Pomey, P. 2002. Epaves Jules-Verne 9 et Jules-Verne 7. In L. Long, P. Pomey, and J.-C. Sourisseau (eds), Les Etrusques en mer: Epaves d’Antibes à Marseille, 121–123. Aixen-Provence: Musées de Marseille, Edisud.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.007

162

ELIZABETH S. GREENE

Pulak, C. 2008. The Uluburun shipwreck and Late Bronze Age trade. In J. Aruz, K. Benzel, and J.M. Evans (eds), Beyond Babylon: Art, Trade, and Diplomacy in the Second Millennium B.C., 289–305. New York: Metropolitan Museum of Art. Reed, C.M. 2003. Maritime Traders in the Ancient Greek World. Cambridge and New York: Cambridge University Press. Sagona, A.G. 1982. Levantine storage jars of the 13th to 4th century B.C. Opuscula Atheniensia 14, 73–110. Salmon, J.B. 1984. Wealthy Corinth: A History of the City to 338 BC. Oxford: Oxford University Press. Seland, E.H. 2015. Writ in water, lines in sand: ancient trade routes, models and comparative evidence. Cogent Arts and Humanities 2. http://dx.doi.org/10.1080/23311983 .2015.1110272. Seland, E.H. 2016. The Periplus of the Erythraean Sea: a network approach. Asian Review of World Histories 4(2), 191–205. Sherratt, S. 2016. A globalizing Bronze and Iron Age Mediterranean. In T. Hodos (ed.), The Routledge Handbook of Archaeology and Globalization, 602–617. London: Routledge. Tandy, D. 2004. Trade and commerce in Archilochus, Sappho, and Alkaios. In R. Rollinger and C. Ulf (eds), Commerce and Monetary Systems in the Ancient World: Means of Transmission and Cultural Interaction: Proceedings of the Fifth Annual Symposium of the Assyrian and Babylonian Intellectual Heritage Project, Held in Innsbruck, Austria, October 3rd–8th 2002, 183–196. Stuttgart: Franz Steiner Verlag. Tartaron, T.F. 2013. Maritime Networks in the Mycenaean World. Cambridge: Cambridge University Press. Vichos, Y. 1999. The Point Iria wreck: the nautical dimension. In W. Phelps, Y. Lolos, and Y. Vichos (eds), The Point Iria Wreck: Interconnections in the Mediterranean ca. 1200 B.C. Proceedings of the International Conference, Island of Spetses, 19 September 1998, 77–98. Athens: Hellenic Institute of Maritime Archaeology. White, R. 1991. The Middle Ground: Indians, Empires, and Republics in the Great Lakes Region, 1650–1815. Cambridge: Cambridge University Press.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:38, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.007

CHAPTER SEVEN

NETLOGO SIMULATIONS AND THE USE OF TRANSPORT AMPHORAS IN ANTIQUITY* Mark L. Lawall and Shawn Graham

INTRODUCTION

A substantial challenge to network studies using amphoras is, perhaps surprisingly, presented by the amphoras themselves. Both our scholarly ancestors and some of our contemporary colleagues have postulated that Greek amphora stamps should be ideal, straightforward data points for quantified studies of ancient shipping (Rostovtzeff 1941, 774–776; Davies 2001, 27–28; cf. Lawall 2005). This is incorrect. Many amphora classes never used stamps and so disappear from such data; others used stamps for certain periods only; still others stamped only a fraction of their jars. Reconciling these and other complications to the amphora record is challenging to say the least. The alternative—treating amphoras like any other ceramic data—encounters the same problems of quantification that impact pottery studies in general (Orton, Tyers, and Vince 1993, 166–181) and, in addition, must contend with the impact of the scholarly love of stamps at the expense of non-stamped sherds. This is not to say that quantified data related to amphoras are entirely inaccessible, but they are not readily at hand from many centers; nor is the most desirable kind of data necessarily obvious. The first part of this chapter gives an overview of the

* The data presented here are the results of my research on the amphora fragments from the Sanctuary of Demeter and Kore on Acrocorinth. I thank Nancy Bookidis and Ronald Stroud for the invitation and opportunity to study this material.

163 Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:21, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.008

164

MARK L. LAWALL AND SHAWN GRAHAM

positive and negative elements of the archaeological data for networks related to Greek amphoras. Compounding problems with the archaeological data is the difficulty of defining the network or networks to be studied and with what goals. This might seem obvious: select a center, count the jars from each exporter found there, use those data as a measure of the strength of connection between the consuming center and each exporter (e.g., Bechtold 2008, 37–39). Or, as has been done recently for Roman Cyprus, use statistically significant correlations to identify chains or groups of exporters supplying a given site or region (e.g., Kaldeli 2013). The former case, however, has difficulty taking account of the indirect nature of attested ancient trading voyages, and the latter case documents correlations without demonstrating actual links between exporters. Indeed, a basic challenge here is to find ways of using the archaeological data from the perspective of network analysis to arrive at qualitative—not necessarily quantitative—conclusions about amphora-related trade in antiquity. The second part of this chapter, therefore, explores how generic network models can be reinterpreted or reimagined for this purpose. Our goal here is not to determine the specific network configurations that created the extant archaeological record. Instead, we explore the relative impact of different variables and different combinations of variables in generic models. We are particularly interested in the ways in which multiple underlying network structures could have resulted in the multifaceted patterning that exists in the archaeological record of transport amphoras—the distributions of types, the evidence for production, choices of shape, amphora marking systems, and so on. The simulations used here, however, do not depend on the simulated agents being ancient traders, potters, or amphoras; they are simply generic agents with certain assigned characteristics, and could just as easily pertain to widgets, shoes, or tennis rackets. By comparing the outcomes of multiple simulation runs with existing features of the archaeological record, we can raise the possibility that similar characteristics of networks as used in the simulations were also at play in antiquity. Such proposals, of course, are only to be taken as hypotheses. We are by no means at the point of proving the operational features of ancient amphora-related networks. Ultimately, in terms of the archaeology of ancient economies, we are trying to get beyond the equally unproven but also quite uninformative hypothesis that an amphora, for example from Chios, found in Syracuse, is to be equated with “commercial ties” or “a trade route” linking Chios and Syracuse. THE ARCHAEOLOGICAL SETTING

To set the stage, we begin with some general impressions of amphora production and patterns of distribution that have emerged from current research. For

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:21, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.008

SIMULATIONS A ND THE USE OF AMPHOR AS

165

production, we have the kiln sites themselves; substantive publications of amphora kiln sites for Archaic through Hellenistic Greece include those at Kerkyra (Preka-Alexandri 1992), Peparethos and Ikos (Garlan and DoulgériInzessiloglou 1990), Thasos (Garlan 2004–2005), Mende (AnagnostopoulouHatzipolychroni 2004 [2006]), and Kos (Kantzia 1994). One can consider their locations relative to urban centers and to ports, as well as to the arable land of the given region. The amphora shapes produced at such workshop sites show a wide range of approaches to the basic task of “make me an amphora!” In most cases, potters spread over a given region shared a single regional style. At times these regions were small enough that one shape could be equated with one polis, but this is the exception rather than the rule as was assumed for so long (Lawall 2011, 50–52 for regional styles, 46–50 for city-specific types). For the packaging process, moving the amphoras from kiln sites to vineyards or olive oil factories and the like for filling and then on to exporters is very poorly attested in the archaeological data. The far greater evidence comes from textual sources, especially papyri from Egypt (Mayerson 2000; Kruit and Worp 2000, Dzierzbicka 2005, 37–41; Gallimore 2010, 177–183). Here we read of vineyards ordering jars from potters in advance, and of empties—both new and used—being gathered from far and wide for the season’s bottling needs. The next stage of activity is the assembly of cargos, the evidence for which comes primarily from shipwrecks (Gibbins 2001) with some additional testimony from the Attic orators (Dem. Against Lakritos). Data likely to be relevant to networks of amphora use here include the numbers of jars of each type found in the cargo, numbers of different workshops contributing to the cargo, points of origin of the crew’s equipment, and the position of the wreck relative to the likely points of origin of that cargo. Onward distribution patterns range from the very localized circulation within the Aegean basin of many amphora types in the Archaic and early classical periods to the more dispersed patterns of the later classical era (Lawall 2013). Certain ports, like Athens, seem to be better connected to long-distance, trans-regional shipping than others (Lawall 2005, 210–215). The data from consumption sites are by far the most obviously useful for network analyses, yet they are also problematic, as noted earlier. On the positive side, there are data from many sites now on the numbers of amphoras identified by type and date. In some cases we have multiple “samples” from each period. Figure 7.1 shows multiple samplings from the Demeter Sanctuary at Corinth and then, on the right, the Demeter Sanctuary late 4thcentury phase compared with a contemporary drain fill also at Corinth (McPhee and Pemberton 2012). There is a clear dominance of local jars through the 5th century at the Demeter Sanctuary and the increasing diversity of sources thereafter. This same phenomenon repeats elsewhere in the Aegean, especially in a contrast between the late Archaic period and into the 5th century

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:21, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.008

166

MARK L. LAWALL AND SHAWN GRAHAM

7.1. Left: distribution of amphora types from phases of activity at the Sanctuary of Demeter and Kore on Acrocorinth; right: comparison of type frequencies between the Sanctuary of Demeter and Kore in the 4th century BC with the fill of Corinth Drain 1971-1 as reported by McPhee and Pemberton (2012).

on the one hand, and the 4th century on the other. And yet such observations depend on quantified data that are, as noted above, problematic. By way of illustration, consider two contexts from the Athenian Agora and the frequency distribution of the stamp types in each deposit compared with the distribution of diagnostic rims and toes (Figure 7.2). The Rhodian material, so dominant among the stamped handle fragments, nearly disappears from significance when one counts only rims and toes. Unstamped amphora fragments are often discarded for lack of storage space; perhaps the Rhodian fragments were all discarded since the more important handle bits were already retained. Post-

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:21, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.008

SIMULATIONS A ND THE USE OF AMPHOR AS

167

7.2. Comparison between (left) frequency distributions of stamped amphora handles from Athenian Agora deposit Q 8–9:1 and the Middle Stoa Building Fill (the latter as reported by Grace 1985) and (right) frequency distribution of amphora types based on counts of rims and toes in the same deposits.

excavation discard certainly reduces the available data (Rotroff 2006, 9–12): for one context at the Athenian Agora, well F11:2, excavated in 1934, ten original containers of pottery are recorded as having been reduced to one tin in 1970 (Agora Notebook section B, pages 1089–1090). Such actions are necessary on nearly every major excavation, and such situations do not require that we abandon any notion of quantification. But they do require that we seek a wide range of methods for interpreting the surviving material.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:21, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.008

168

MARK L. LAWALL AND SHAWN GRAHAM

NETWORK MODELS

In this second part of the chapter, we turn to a preliminary consideration of how some generic network models can be reinterpreted and modified as tools for exploring how the archaeological record relating to amphoras may have emerged.1 We cannot hope to explore networks related to all of the many actors or agents in an amphora’s life. In what follows we focus our attention on two areas: first, the selection of shapes by the consumer, and potters’ responses to those selections; second, the potential impact of familiarity of shape and prior social connections in the growth of distribution networks.

Mimicry Model We began with Netlogo’s mimicry model, found in the Netlogo Model Library bundled with the program (Wilensky 1999; 1997). This model is framed as a representation of the dynamics of butterflies and birds, which may seem a rather distant analogy from amphora making. However, what is of interest is the underlying dynamics. We first describe how the model works, and then show its applicability to reframing our understanding of consumer behavior. In this model, birds eat butterflies of two types: monarchs (initially colored red), which make the bird sick, or viceroys (initially blue), which are tasty. Over time, reproducing butterflies create mutants of slightly different colors, and this rate of mutation is one variable in the model. Birds remember a certain number of experiences over time, so that they will avoid colors they associate with monarchs. Hence the more a viceroy mutant looks like a monarch, the more likely it is to survive and reproduce. Bird memory span and capacity are both variables; birds can remember and avoid only a limited number of “bad”tasting colors. Eventually all of the tasty viceroy butterflies are red and hence appear to be monarchs, and the birds give up. We can apply this model to amphora production by reconceptualizing the narrative. There are two classes of amphoras, X and O, each of which might adopt different forms over time. Amphoras of type O are “viceroys” in the simulation code, and X are “monarchs.” Due to the limitations of the 2D visualization of the simulation, we represent vessel “forms” with colors, with red for type X and blue for type O initially. Consumers (birds in the original model) find one form desirable and it continues to circulate, with undesirable forms “falling out” (“being eaten,” in the original model); think of them as being smashed in disgust, never to circulate again. Acceptable circulating forms “reproduce” in this model and can mutate into new forms. This mutation rate can be set at a low number to simulate the conservatism of potters. Consumers can remember a certain number of desirable shapes and the length of these

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:21, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.008

SIMULATIONS A ND THE USE OF AMPHOR AS

169

7.3. Screen image of Netlogo mimicry model with networked sharing turned off, high memory duration. A color version of this figure is available at https://github.com/shawngraham/ maritimenetworks.

memories can also be adjusted. Without including any other influences, the X producers gradually change the preferred form of the pottery from one that is often smashed (blue) to one that stays in circulation (red). All amphoras end up with one common shape. The next step was to introduce networks of consumers that share information regarding tastes in pottery. To this end, we inserted the option to create either a preferential attachment or a small-world network for the consumer population. In a preferential-attachment model—that is, where the likelihood of forming a connection to a new node depends on the number of connections a node already has (the rich get richer)—most nodes only have one or very few links to others, but a few are linked to many other nodes. By contrast, the smallworld network has many possible links between clusters of nodes and also some shortcut links to more distant nodes. Sliders here allow us to compare quickly the results of different memory size, memory duration, and mutation rate settings. Sharing knowledge over the network can be turned on or off. For a preliminary sense of the impacts of the variables in play here, we can explore the results of a “neutral” run, where we start with the model set for a high memory duration, low mutation rate (conservatism of potters), and low memory size (Figure 7.3). With the networking turned off, the global distribution of “forms” (represented by color) converges on a color value close to the red of type X, 15.5–16.5. Netlogo maps the color palette against a range of values from 0 to 140; in this schema, red are values around 15. This happens after around 400–500 ticks. “Ticks” are simply cycles of the simulation, which are not significant in themselves but only in comparison with other runs of the

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:21, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.008

170

MARK L. LAWALL AND SHAWN GRAHAM

7.4. Screen image of Netlogo mimicry model with addition of small-world network: sharing turned on, low memory duration. A color version of this figure is available at https://github .com/shawngraham/maritimenetworks.

model. The difference between fast runs and slow runs can be of interest. With knowledge sharing turned on under small-world network conditions, 700 or more ticks are needed to settle on an average color around 16. With the preferential-attachment model network, we arrive at the same point with the average color just over 16 by around 300 ticks. With these settings, then, the small-world network slows down the achievement of homogeneity; while both the preferential-attachment network and the absence of any networked sharing of knowledge both result in fairly rapid homogenization. Such results can then be compared with the situation when memory duration is low but the other variables are kept the same. In this case, the absence of networking again leads to convergence, albeit more slowly and with greater change in the average color. The preferential-attachment model does not quite achieve that degree of convergence, but is heading in that direction. In the small-world network, the colors appear to be staying farther apart for even longer (Figure 7.4). Again, small-world networking in this mimicry model seems to lead toward a state of heterogeneity. A similar result is achieved when one aggregates all of the data from 100 runs of the models after 1,500 ticks each (Figure 7.5). This figure is a scatterplot, where the points are put into hexagonal bins to better represent areas of density. Thus the darker cells in these graphs represent concentrations of results of the average color after 1,500 ticks with fifteen being red and 105 being blue as the starting arrangement. Without networked sharing of knowledge, and regardless of the other settings, the model tends to average out with a color close to red. With a preferential-attachment network operating there is either a lack of significant

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:21, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.008

(a)

(b)

(c)

7.5. Aggregate results of mimicry model with (a) no network, (b) preferential-attachment network, and (c) small-world network (x axis = monarch colors, y axis = viceroy colors). Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:21, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.008

172

MARK L. LAWALL AND SHAWN GRAHAM

change to the colors—so the x-axis value is close to 15 and the y-axis is just over 100—or a flipping toward the inverse. For the small-world model, however, there is pressure toward convergence, but a much greater diversity of final end points. In this mimicry model, however, the reasons (or triggers) for “form” changes do not vary. Form change here is a matter of random mutation. Prior knowledge either prevents the destruction of the amphora shape encountered or it allows that destruction. The nature of that knowledge can be changed deliberately in the model but there are limited options for the results: destroy the amphora, let it continue to circulate and perhaps mutate in a random fashion when it reproduces. Furthermore, the shapes being encountered by each consumer depend on the random movements of the amphoras. A second model shifts the focus to the process of choosing a shape on the part of potters rather than the decisions of the consumers.

Stylistic (Linguistic) Change Model Netlogo’s language change model (Troutman and Wilensky 2007) starts with two competing grammars distributed over a given social network. In the original form, this model limits the network structure to a preferentialattachment network. Individual agents “speak” to each other using different grammars, allowing an agent to change its grammar if a certain amount of a particular grammar is heard, given an individual’s threshold for change. In this way the model explores convergence toward a common grammar. The model allows the grammar change trigger to occur in three different ways: (1) adopt the grammar of a randomly chosen neighbor, (2) adopt the grammar used by some threshold-level number of their neighbors (if, say, 50% of the neighbors use grammar 1, then the grammar 0 user will switch to grammar 1), or (3) the level of use of one grammar or the other depends on built-up experience (the more you hear of one grammar, the more you will use that grammar). We reconceptualized this model to understand “grammar”’ as “shape.” The “speakers” of a certain grammar are considered as potters making their chosen amphora shape. The model then becomes an exploration of the degree to which a potter’s social situation might influence stylistic convergence. The only significant change made to the model code was the addition of options for the structure of the network: as in the case of the mimicry model we added the possibility of a small-world network instead of the default preferential-attachment network. Since we are interested in the effect that patterns of social connection have on the adoption (or not) of an amphora shape, we distribute the different grammars equally across both kinds of network. With the trigger for a shift in shape based on that of a randomly chosen neighbor, and with a preferentialattachment network, one shape wins out very quickly (Figure 7.6). Under the

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:21, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.008

SIMULATIONS A ND THE USE OF AMPHOR AS

173

7.6. Screen image of Netlogo modified language change model with preferential-attachment network.

7.7. Screen image of Netlogo modified language change model with small-world network.

same conditions but with a small-world network, there is more fluctuation from one shape to the other, but the end result is again the complete use of one shape relatively quickly (Figure 7.7). On the other hand, when the change of shape depends on the shape used by some threshold number of the agent’s neighbors, then the results are more varied.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:21, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.008

174

MARK L. LAWALL AND SHAWN GRAHAM

(a)

Mean [state] of nodes 1.2 1 0.8 0.6 0.4 0.2 0 0

50

100

150

200

250

200

250

Mean [state] of nodes

(b) 1.2 1 0.8 0.6 0.4 0.2 0 0

50

100

150

7.8. Aggregate results of Netlogo modified language change model (top) using preferentialattachment network and threshold model for adoption, and (bottom) using small-world network and threshold model for adoption.

Given a preferential-attachment network, a new shape will come to dominate if the threshold value is set between 0 and 10% of one’s neighbors. With 60% and higher as the threshold, the new shape will be suppressed and disappear. But with between roughly 20% and 50% as the threshold values, multiple shapes coexist. With a small-world network, coexistence of forms only occurs in a narrow band of threshold values around 40%. When considered in aggregate, there is a much wider range of average states resulting when using the preferential-attachment network model as contrasted with the narrower range of results from the small-world model (Figure 7.8). In other words, regardless of the threshold setting for numbers of influential neighbors or links in the network, a small-world network will rarely allow for the coexistence of multiple shapes; too many potters “know” of the success of one shape and come to adopt it. A similar association between a tendency for the coexistence of multiple shapes and preferential-attachment networks arises when the trigger for change is the

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:21, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.008

SIMULATIONS A ND THE USE OF AMPHOR AS

175

Mean [state] of nodes

(a) 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0

50

100

150

200

250

200

250

mean [state] of nodes

(b) 1.2 1 0.8 0.6 0.4 0.2 0 0

50

100

150

7.9. Aggregate results of Netlogo modified language change model (top) using preferentialattachment network and rewards model for adoption, and (bottom) using small-world network and rewards model for adoption.

built-up exposure to another shape, the rewards model (Figure 7.9). In a smallworld network, by definition, the intensive local clumping combined with some long-distance links gives many chances to build up exposure to new shapes. There is a rapid shift to the new shape regardless of the initial state. In the preferentialattachment model, where few nodes have multiple links, the rate of change is much more gradual and there is a much greater likelihood of coexisting shapes. DISCUSSION OF MIMICRY AND LANGUAGE CHANGE MODELS IN TERMS OF THE ARCHAIC THROUGH HELLENISTIC AMPHORA RECORD

With the first model we presented, the mimicry model, changes in shapes depended on survival and mutation, and survival depended in part on the network nature of the consumers. Small-world networks of consumers slowed the mutation/adaptation process of the amphora shapes; regardless of the

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:21, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.008

176

MARK L. LAWALL AND SHAWN GRAHAM

settings, more amphora shapes could become acceptable to them through their networked knowledge, and hence those shapes would not get smashed. On the other hand, preferential-attachment networks of consumers had less access and more rigidly structured access to knowledge, so adaptation converged more quickly and directly on certain shapes. The second model considered the network to be the amphora producers themselves. In this case, rapid convergence on one shape in a preferentialattachment network only occurred when the trigger for change was the random selection of a neighbor’s shape. Otherwise, when the triggers for change were more complex, preferential-attachment networks of “potters” provided the more likely scenario for maintaining multiple, coexisting shapes. Small-world networks were more likely to end with one surviving shape. Significantly, this combination of simulations raises the possibility that different network structures pertaining to different agents offer the best fit to archaeologically attested patterns. Here, small-world networks of consumers as they relate to amphora shapes lead to heterogeneity, but small-world networks of potters choosing their shapes result in homogeneity. To see how such results might play out in the archaeological record, we return to the indications of regions in the Aegean—of varying sizes, each with its own regional amphora type—and the degree to which such regions also defined zones of amphora circulation. On the basis of the results of the “language change model,” rapid convergence on one form, such as we have within these regions of style, can be achieved by purely random interaction and sharing of knowledge (i.e., a lack of network structure), small-world networking, or preferential-attachment networking depending on the conditions for knowledge transmission. Convergence on one shape, as clearly happens within each region, is achieved under many conditions of interaction. Moving to the Aegean as a whole, we have a situation of multiple, coexisting shapes. This situation is most likely to occur with a preferential-attachment network and when the trigger for change is either some threshold number of contacts using the new shape or some rule based on built-up experience of the new shape. Coexistence of shapes occurs under far fewer conditions in a smallworld network. For the border areas between regional styles, one can imagine that those who made the choices about amphora shape—a complex group in itself, presumably—had multiple contacts of various strengths within their respective regions (small world) but only fixed, specific links across regional boundaries. With the mimicry model, with its actors being consumers smashing or sending on various amphora shapes, we can start to think about the merchants or other users of amphoras. If there was no networking between these merchants/consumers, then there would likely be a convergence toward one preferred amphora type. This situation only appears in certain regions,

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:21, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.008

SIMULATIONS A ND THE USE OF AMPHOR AS

177

particularly in the late 6th century, at those sites where the local type dominates the amphoras in circulation to the near exclusion of all others. With preferential-attachment networking among the merchants/consumers, a few types might come to be preferred, precisely the situation encountered even in a general paradigm of regional circulation of amphoras. One often encounters more than one quite common type, and the others are sparsely represented. Small-world networking would predict a lack of structure to import patterns, just a general mass of imports of various types. Such a situation is rare, though it may characterize very late classical and early Hellenistic Athens. The two simulation models, mimicry and language spread, fit neatly together when one overlays the effects of small-world models within regions of amphora production and the effects of preferential-attachment models on the interactions between regions. Thus, for our final series of simulations, we explore the impacts of certain additional variables on the operation of preferential-attachment models. PREFERENTIAL ATTACHMENT

Netlogo’s preferential-attachment model (Wilensky 2005) expands links from two nodes, adding a node with each step and adding a link (“edge”) from that node to a pre-existing node. The chance of a link being made in this model is proportional to the number of links it already has, a mechanism that explains network formation observed in various domains. Success breeds success. Runs of this basic model result in the rapid achievement of a skewed distribution of network sizes with very few hubs controlling very large networks. The basic model usefully illustrates that, all other factors being negligible, networks will grow to focus on a very limited number of hubs. But what of the impact of other factors? We altered this basic model to take into account not just wellconnected others, but also preferences for shapes, and preferences for preexisting social connections. Each agent has its own shape representing the amphora that agent makes; the range of possible shapes is limited by the pre-included list of possible node icons in Netlogo. The first step was to add the possibility of a preference for a certain amphora shape in addition to the possibility of preferring a “popular” hub. A sliding scale was inserted into the model such that the choice of attachment could depend more or less on familiarity of shape (100 = no shape bias; 0 = complete shape bias). Second, we added a sliding scale for the influence of preexisting social connections (0 = no friendship bias; 100 = complete friendship bias).2 Each agent has a randomly generated list of “known” others, from which a network is then stitched together at random. We added the ability to track both the distributions of numbers of links created and the number of agents without links.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:21, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.008

178

MARK L. LAWALL AND SHAWN GRAHAM

7.10. Screen image of Netlogo preferential-attachment model set for no social bias and no shape bias.

The contrast between the impact of shape bias and the impact of pre-existing friendship bias is quite striking. We start with shape bias turned off entirely, so the only competing influences are preferential attachment and pre-existing friendships. With friendship bias at zero a strongly hierarchical network encompassing most agents develops relatively slowly (Figure 7.10). As friendship bias becomes more important, the graph of degree distribution becomes far less steep, the network is created with more speed, and, interestingly, more agents are left out, especially when friendship bias and preferential attachment bias are roughly coexisting. We can see this play out as we move from values of 20 to 50 to 100 for friendship bias (Figure 7.11). By contrast, when shape bias is activated (at a value of 75, so relatively little shape bias) and friendship bias is turned off entirely, we see, early on, one large clustered network and smaller, unconnected other networks (Figure 7.12). As more and more weight is placed on shape bias (e.g., shape bias at 50, then 20, then 0) we see a tendency toward small, unconnected networks (Figure 7.13). Throughout the process, very few agents are left out entirely. When shape bias is left fully in effect and friendship is added in as well (shape bias is set at 0, friendship at 20) there is a continuing tendency toward rapid achievement of a steady state as we had with friendship alone, but also a continuation of the low number of outlying, unconnected nodes. The friendship bias can then be increased (from 50, to 80, to 100), keeping shape bias at 0 (Figure 7.14). Only when the friendship slider is nearer to 80% and higher does the number of outliers again rise quite high; the same is true when the shape effect is lessened, for example such that both sliders are at 50%.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:21, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.008

SIMULATIONS A ND THE USE OF AMPHOR AS

179

7.11. Screen image of Netlogo preferential-attachment model set for friendship bias at 50, no shape bias.

7.12. Screen image of Netlogo preferential-attachment model set for no friendship bias, shape bias at 75.

Within the preferential-attachment model, therefore, three major effects emerge with the addition of pre-existing friendships and/or bias toward familiar shapes. Pre-existing friendships lessen the severe hierarchy of the network, allowing more nodes to have more connections, speeding up the network formation process, but also making the exclusion of nodes more likely. Shape

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:21, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.008

180

MARK L. LAWALL AND SHAWN GRAHAM

7.13. Screen image of Netlogo preferential-attachment model set for no friendship bias, maximized shaped bias (0).

7.14. Screen image of Netlogo preferential-attachment model set for friendship bias at 50, maximized shape bias.

bias on its own creates a very fragmented system within the preferentialattachment model, though the speed of formation is still better than that seen with no additional influences. Shape bias also tends to bring a greater proportion of the nodes into the system. When shape and friendship are combined, under most conditions, one gets the best of both worlds: a single, coherent network, rapidly created, with very few outliers.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:21, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.008

SIMULATIONS A ND THE USE OF AMPHOR AS

181

DISCUSSION OF PREFERENTIAL-ATTACHMENT MODEL AND THE AMPHORA RECORD

These results set up some potentially interesting hypotheses for the Aegean amphora trade. The periods and regions characterized by narrowly regional circulation can be considered in terms of networks in which a bias toward familiar shape plays a very important role. Linking such regions together, however, such that there is substantial transregional shipping, as is seen in the later 5th and early 4th centuries BC in the Aegean, requires some additional layer of networking (here referred to as pre-existing friendships) beyond simply the emerging popularity of one or more hubs. By contrast, one might consider the later Hellenistic pattern, in which Rhodes, Delos, Ephesus, and perhaps Athens seem increasingly dominant, in terms of networks that are even more dominated by pre-existing ties or simply by the principle of preferential attachment. Such networks seemed to exclude many nodes and create the more dramatically hierarchical patterning. In the Hellenistic period, indeed, some highly localized amphora shapes never leave their very narrow region of production (Panagou 2015); some whole regions —such as the north coast of the Aegean—nearly disappear from the record. Amphora specialists tend to place great importance on the distinctive shape of Hellenistic Rhodian amphoras, but perhaps the island’s well-attested interests and activity in building maritime networks were far more important (Gabrielsen 1997). CONCLUDING OBSERVATIONS

The models used for various simulations in this paper depend neither on the specific geography of the ancient Greek world nor on specific aspects of ancient behavior (distributions of amphora production sites, practices of assembling cargos, etc.). The relationships being modeled are deliberately generic while at the same time introducing specific kinds of connection and behavior (small-world network, preferential-attachment network, adoption of familiar forms, etc.). The various iterations of these models often gave similar results such that it would be dangerous to project backwards from such a result to the specific model that created it. But in other cases, a certain range of results only pertained in certain specific conditions. Hence, when patterns are present in the archaeological record—in our case for amphora-related behavior—that seem to resemble such particular results, one should ask whether the specific conditions of the model needed to produce those patterns might have held true in antiquity. For example, do we have evidence for a dominant importance of preferentialattachment networking or pre-existing ties in the late Hellenistic amphora trade, as opposed to either the absence of any significant networking or the importance of small-world networks? One could explore the specific geography of a trade axis featuring Alexandria, Rhodes, Delos, and Athens; elements of royal and

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:21, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.008

182

MARK L. LAWALL AND SHAWN GRAHAM

aristocratic patronage; broader political, military, and commercial interests for all parties involved; and the institutions shaping amphora use from production to discard. The generic simulations used here, therefore, help to narrow our questioning to seek evidence for specific kinds of network with specific characteristics.3 NOTES 1. 2. 3.

All models discussed here, along with color versions of figures, may be found at https:// github.com/shawngraham/maritimenetworks. See the code, lines 44–74, for how this is operationalized. For the repository, see github. com/shawngraham/maritimenetworks. See also Graham and Weingart 2015 for another approach using agent-based modeling as a laboratory for understanding the equifinality of archaeological networks.

REFERENCES

Anagnostopoulou-Hatzipolychroni, E. 2004 [2006]. Σωστική ανασκαφή στην αρχαία Μένδη. Το αρχαιολογικό έργο στη Μακεδονία και Θράκη 18, 133–140. Bechtold, B. 2008. Observations on the Amphora Repertoire of Middle Punic Carthage. Carthage Studies 2. Ghent: Ghent University Press. Davies, J.K. 2001. Hellenistic economies in the post-Finley era. In Z.H. Archibald, J.K. Davies, V. Gabrielsen, and G.J. Oliver (eds), Hellenistic Economies, 11–62. London: Routledge. Dzierzbicka, D. 2005. Wineries and their elements in Graeco-Roman Egypt. Journal of Juristic Papyri 35, 9–91. Gabrielsen, V. 1997. The Naval Aristocracy of Hellenistic Rhodes. Studies in Hellenistic Civilization 6. Aarhus: Aarhus University Press. Gallimore, S. 2010. Amphora production in the Roman world: a view from the papyri. Bulletin of the American Society of Papyrologists 47, 155–184. Garlan, Y. 2004–2005. En visitant et revisitant les ateliers amphoriques de Thasos. Bulletin de correspondance hellénique 128–129, 269–329. Garlan, Y., and A. Doulgéri-Intzessiloglou. 1990. Vin et amphores de Péparéthos et d’Ikos. Bulletin de correspondance hellénique 114, 361–389. Gibbins, D. 2001. Shipwrecks and Hellenistic trade. In Z. Archibald, J.K. Davies, V. Gabrielsen and G.J. Oliver (eds), Hellenistic Economies, 273–312. London: Routledge. Grace, V.R. 1985. The Middle Stoa dated by amphora stamps. Hesperia 54, 1–54. Graham, S., and Weingart, S. 2015. The equifinality of archaeological networks: an agent-based exploratory lab approach. Journal of Archaeological Method and Theory 22, 248–74. Kaldeli, A. 2013. Early Roman amphorae from Cyprus as evidence of trade and exchange in the Mediterranean. In M. Lawall and J. Lund (eds), Transport Amphorae and Trade of Cyprus, 123–132. Gösta Enbom Monograph 3. Aarhus: Aarhus University Press. Kantzia, Ch. 1994. Ενα κεραμικό εργαστήριο αμφορέων του πρώτου μισού του 4ου αι. π. Χ. In Γ´ Επιστημονική συνάντηση για την ελλενιστική κεραμική 1991, 323–354. Athens: Archaeological Receipts Fund. Kruit, N., and Worp, K.A. 2000. Geographical jar names: towards a multi-disciplinary approach. Archiv für Papyrusforschung und verwandte Gebiete 46, 65–146.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:21, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.008

SIMULATIONS A ND THE USE OF AMPHOR AS

183

Lawall, M.L. 2005. Amphoras and Hellenistic economies: addressing the (over)emphasis on stamped amphora handles. In Z.H. Archibald, J.K. Davies, and V. Gabrielsen (eds), Making, Moving, and Managing: The New World of Ancient Economies, 323–31 BC, 188–232. Oxford: Oxbow Books. Lawall, M.L. 2011. Transport amphoras and regional styles in the Classical Aegean. In M. Lawall and P. van Alfen (eds), Caveat Emptor: A Collection of Papers on Imitations in Ancient Greco-Roman Commerce, Marburger Beiträge zur Antiken Handels-, Wirtschafts- und Sozialgeschichte 28, 45–88. Lawall, M.L. 2013. Patterns in the production and distribution of transport amphoras in the 5th century BC: an archaeological perspective on economic change. In A. Slawisch (ed.), Handelsund Finanzgebaren in der Ägäis im 5 Jh. v. Chr., 103–120. Byzas 18. Istanbul: Ege Yayinlari. McPhee, I.D., and Pemberton, E.G. 2012. Late Classical Pottery from Ancient Corinth: Drain 1971-1 in the Forum Southwest. Corinth 7.6. Princeton: American School of Classical Studies at Athens. Mayerson, P. 2000. The meaning and function of Ληνόϲ and related features in the production of wine. Zeitschrift für Papyrologie und Epigraphik 131, 161–165. Orton, C., Tyers, P., and Vince, A. 1993. Pottery in Archaeology. Cambridge: Cambridge University Press. Panagou, T. 2015. Patterns of amphora stamp distribution: tracking down export tendencies. In E.M. Harris, D. Lewis, and M. Woolmer (eds), The Ancient Greek Economy: Markets, Housholds and City-States, 207–229. Cambridge: Cambridge University Press. Preka-Alexandri, K. 1992. A ceramic workshop in Figareto, Corfu. In F. Blondé and J. Perrault (eds), Les ateliers de potiers dans le monde grec aux époques géométrique, archaïque et classique, 41–52. Bulletin de correspondance hellénique Supplement 22. Athens: Ecole française d’Athènes. Rostovtzeff, M.I. 1941. The Social and Economic History of the Hellenistic World, 3 vols. Oxford: Oxford University Press. Rotroff, S. 2006. Hellenistic Pottery: The Plain Wares. Athenian Agora 33. Princeton: American School of Classical Studies at Athens. Troutman, C., and Wilensky, U. 2007. NetLogo language change model. http://ccl .northwestern.edu/netlogo/models/LanguageChange. Center for Connected Learning and Computer-Based Modeling, Northwestern University. Wilensky, U. 1997. NetLogo mimicry model. http://ccl.northwestern.edu/netlogo/mod els/Mimicry. Center for Connected Learning and Computer-Based Modeling, Northwestern University. Wilensky, U. 1999. NetLogo, http://ccl.northwestern.edu/netlogo. Center for Connected Learning and Computer-Based Modeling, Northwestern University. Wilensky, U. 2005. NetLogo preferential-attachment model. http://ccl.northwestern.edu /netlogo/models/PreferentialAttachment. Center for Connected Learning and Computer-Based Modeling, Northwestern University.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:30:21, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.008

CHAPTER EIGHT

LESSONS LEARNED FROM THE UNINFORMATIVE USE OF NETWORK SCIENCE TECHNIQUES An Exploratory Analysis of Tableware Distribution in the Roman Eastern Mediterranean* Tom Brughmans

IF ALL YOU HAVE IS A HAMMER, EVERYTHING LOOKS LIKE A NAIL

What determines the usefulness of particular formal network methods for scholars studying ancient maritime connectivity? Is it the convenient representation of entities such as islands, ports, and the past interactions between them as dots and lines? Or is it the good fit between the past phenomena of trade, transportation, and communication, with their abstraction as network concepts? In this chapter I argue that although these reasons might be sufficient to lead scholars to consider using formal network methods in their research, they are not sufficient to motivate the adoption of specific network techniques. Through a practical example I illustrate how it is unhelpful for scholars of ancient maritime connectivity to use formal network methods as a hammer to hit every nail we can find, just because we can (Doreian 1988: 290–291). Specific formal network methods should preferably be selected in light of their ability to lead to insights that other approaches cannot offer. In this chapter I present a case study concerning the changing distribution patterns of ceramic tableware in the Roman eastern Mediterranean using exploratory network analysis techniques, which serve here as an example of * I would like to thank Justin Leidwanger and Carl Knappett for hosting the inspiring workshop that led to the current volume; Philip Bes, Rinse Willet, and Jeroen Poblome for the creation of and access to the ICRATES database of tablewares in the Roman East used in the case study; and the Algorithmics group at the University of Konstanz and in particular Ulrik Brandes and Viviana Amati for discussions on this paper.

184 Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:29:22, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.009

LESSONS LEA RNED F ROM NETWORK SCIENCE TECHNIQUES

185

the use of formal network methods for the study of maritime connectivity. I first briefly introduce a definition of network science that reveals how it differs from other formal methods, and therefore where its potential for leading to new knowledge lies. I subsequently introduce the case study by setting out its research aims from the perspective of this definition of network science, which will inform the selection of specific formal techniques. However, this case study led to largely uninformative results, or at least its results did not reveal anything that could not have been achieved using other formal methods. Nevertheless, in these early days of the archaeological and historical application of network science, it is crucial for uninformative experiments to be published as such. It will allow the small but growing community of scholars interested in these techniques to avoid known pitfalls and will help us work toward the development of new techniques specifically designed to address these challenges that are offered by the study of past maritime connectivity. Moreover, the ability for network methods to reproduce results obtained through other methods should be seen as part of its contribution to our discipline. I therefore conclude this chapter by drawing on the lessons learned through an uninformative experiment, and offer a number of guidelines for scholars who wish to evaluate the potential of specific network science techniques to address research questions concerning past maritime connectivity. WHAT IS NETWORK SCIENCE?

This chapter takes the perspective of network science in order to assess the potential of formal network methods for studying past maritime connectivity (for a more elaborate definition of network science for the study of the past, see Brughmans, Collar, and Coward 2016; Collar et al. 2015). Network science is the study of network models (Brandes et al. 2013). In such models, past phenomena are abstracted into network concepts, which are in turn represented by network data (Figure 8.1). The past phenomena we are

abstraction past phenomenon

representation network concept

network data

network model abstraction trade between individuals

representation flow of goods social entities interact; flow of individual 1 individual 2 goods takes place

8.1. The network model. (After Brandes et al. 2013; reproduced with permission from Collar et al. 2015)

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:29:22, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.009

186

TOM BR UG HMAN S

interested in are therefore distinct from the network representation we use to study them. Specifically, the nature of past maritime connectivity which we aim to understand is not directly represented by nodes, edges, and networks. Instead, we abstract this complex past phenomenon of maritime connectivity using network concepts, where we describe the entities we are interested in, the way in which they relate, and how the relationships affect the system as a whole. For example, we are interested in Roman traders, related to each other by the ability to perform commercial transactions, and the specific way in which all traders are related will affect the flow of goods between pairs of traders. These concepts are useful analytical categories if they are clearly and unambiguously described, in which case they allow for the discussion and reproduction of analytical results between scholars. Network concepts are subsequently represented by network data, where entities are represented as nodes, their relationships as edges, and the combination of the sets of all nodes and edges as a network representation of the system. This definition of network science is very similar to that of other formal methods like geographic information systems (GIS): the past phenomena we are interested in are abstracted using analytically useful concepts, which are represented by data. What sets network science apart, however, is the nature of network data (see Brandes et al. 2013 for a detailed definition of network data). Unlike in other data types used with other formal methods—e.g., spatial data used in GIS—in network data we explicitly formulate dependence assumptions about how relationships affect each other’s existence: the presence of one edge can determine the presence or absence of any other edge in the network. This definition of network data represents the fact that, when we draw nodes and edges, we do not just do this because we can, but because we assume that these mean something. We assume that they represent opportunities of interaction between pairs of nodes that affect the opportunities of other nodes in the network and the existence of other edges in a specific and defined way. This chapter argues that the true advantage of a networks perspective, qualitative or quantitative, lies in the formulation of such dependence assumptions and the analysis of their effects. Through the following case study, I illustrate how the formulation of dependence assumptions is completely determined by the particular research context and aims, and the selection of specific network science techniques necessarily happens in light of this. However, I also illustrate that some network science techniques are limited to representing or exploring dependence assumptions and not testing them; in such cases they might not lead to insights that other techniques more commonly used in the study of the past can.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:29:22, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.009

LESSONS LEA RNED F ROM NETWORK SCIENCE TECHNIQUES

187

CASE STUDY: EXPLORATORY NETWORK ANALYSIS

Research Questions In this case study I analyze the large-scale distribution of four ceramic tablewares produced and circulated in the eastern Mediterranean between c.150 BC and AD 200, with a particular focus on the period 25 BC–AD 150 when all four tablewares circulated in the eastern Mediterranean. The four wares in question are Eastern Sigillata A, Eastern Sigillata B, Eastern Sigillata C, and Eastern Sigillata D (hereafter abbreviated as ESA, ESB, ESC, and ESD respectively). Of these four, only ESA was distributed on a supra-regional scale. Moreover, it reached this wide distribution before most of the eastern Mediterranean was incorporated into the Roman Empire and maintained it for centuries. This fact forces one to look for a complex combination of contributing factors rather than accepting a political framework as the sole explanation for the observed distribution patterns. The past phenomenon I aim to better understand through this case study centers on the differences between the distributions of ESA, ESB, ESC, and ESD, and the overarching research question can therefore be summarized as follows: what processes gave rise to these strong differences in tableware distributions? Widely ranging hypotheses exist that aim to explain these patterns. These hypotheses often consist of a complex mix of contributing factors including state involvement, consumption “pulling forces,” commercial “piggyback” trade, closeness to large-scale agricultural production, and traders’ social networks (e.g., Abadie-Reynal 1989; Lewit 2011; Reynolds 1995; and Bes 2015 for an overview of the available data and competing hypotheses). It is not possible to explore any of these factors in any detail within the scope of the current chapter. It suffices for the moment to state that through an exploratory network analysis of a tableware dataset I aim to understand better the distribution patterns under scrutiny, and whether the results of this formal network approach suggest that some hypotheses are more likely than others. The most common approach to studying the distributions of different data types is to compare the similarities and differences in their distributions. Underlying this approach is an assumption that similarities and differences in artifact distribution patterns reveal something about the human behavior that led to them. More specifically, it is assumed that data types with a similar distribution pattern have a higher probability of having been distributed through processes of similar types. Such an assumption should be tested as a hypothesis, but that is outside the scope of the current chapter, and it cannot be tested through an exploratory network analysis. The latter merely aims to provide a representation of the observed tableware distribution patterns as network data and analyze the outcome, but making this assumption explicit

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:29:22, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.009

188

TOM BR UG HMAN S

is important when deciding on a suitable method for describing artifact similarity. The assumption can be formulated more explicitly as a network science dependence assumption: the presence of a tableware form at a site is dependent on the presence of other tableware forms at this site. This assumption is a reflection of the idea that the distribution patterns of a set of tableware forms, as represented by network data patterns in a similarity network, will be similar when these forms are distributed through similar processes. The selection of formal network methods in this case study is aimed at exploring the dataset in light of this dependence assumption, but not at testing it (for the testing of this dependence assumption using formal network methods, see Brughmans and Poblome 2016). Three more specific research questions can be formulated in light of this chapter’s aim and dependence assumption: • What differences can be observed in the distribution patterns of different tablewares and forms? • Are forms’ distributions always more similar to those of the same ware? • Were similar processes responsible for the distribution of forms of the same ware?

Data The large-scale distribution pattern of these four tablewares is reflected in the ICRATES (Inventory of Crafts and Trade in the Roman East) database, a resource which will be used in this case study. The ICRATES database is compiled mainly from published sources, supplemented with unpublished data from the Boeotia Survey in Greece and from the site of Amata in Jordan. Each entry in the database concerns a single tableware sherd, the site where it was found, the publication in which it was mentioned, the standardized fabric and form attributed by the authors (sometimes corrected or standardized by project members), and more information when available (for a detailed description of the database, see Bes 2015). The database currently includes over 33,000 individually recorded tableware sherds. Of these, 8,073 sherds of the four tablewares under scrutiny in this chapter are dated between 150 BC and AD 200. Table 8.1 shows the standard typo-chronological reference used to date each tableware form, as well as each ware’s possible region of production. To explore the distribution pattern in periods of twenty-five years, I assumed a normal probability distribution of a sherd being deposited in each year of its form’s date range by applying a method developed by Fentress and Perkins (1988). Probabilities for each twenty-five-year time-slice were calculated per form, and the observed number of sherds per form was multiplied by this probability for each time-slice. This results in a “probable” number of sherds

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:29:22, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.009

LESSONS LEA RNED F ROM NETWORK SCIENCE TECHNIQUES

189

table 8.1. Typo-chronological references and (possible) region of production for major eastern tablewares.

Ware

Abbreviation

Typo-chronological reference

Eastern Sigillata A

ESA

Hayes 1985

Eastern Sigillata B

ESB

Hayes 1985

Eastern Sigillata C

ESC

Eastern Sigillata D

ESD

Meyer-Schlichtmann 1988; Hayes 1972; 1985 Hayes 1985

Region of production Coast between Tarsos (TUR) and Latakia (SYR) Maeander Valley in western Asia Minor (TUR). Possibly Aydin (ancient Tralleis) Pergamon and surrounding region Cyprus (probably the western part)

Number of sites per 25-year period (n = 8073) 160 140 Number of sites

120 100 ESA ESB ESC ESD

80 60 40 20 BC 0B 50 C –2 5B 25 C –1 B 1– C 25 25 AD –5 0 50 AD –7 75 5AD –1 10 00A D 0– 1 12 25A 5– D 1 15 50A 0– D 1 17 75A 5– D 20 0A D –5

75

75

C 10

0–

0B 10

5– 12

15

0–

12

5B

C

0

8.2. Number of sites a certain ware is attested at, per twenty-five-year period (n = 8,073).

per form per twenty-five-year time-slice, allowing for differences in distribution patterns per form and per ware to be explored through time. Figure 8.2 shows the distribution pattern for the case study: during most of the period under investigation, ESA is attested at far more sites than the other three wares. Tableware forms derived from the standard typologies are here treated as conventions that were constructed by individual ceramicists or groups of scholars, and reflect their decisions and assumptions, but mainly their need to reduce a mass of information into analytically useful categories. The patterns emerging from the distributions of these forms will therefore, at least in part, reflect such academic decisions and assumptions. Tableware forms are not considered to represent categories that were conceived as distinct by people in the past.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:29:22, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.009

190

TOM BR UG HMAN S

(a)

(b) Site A

Site B

Site C

Form A

2

0

0

Form B

10

3

Form C

0

8

Site A

Site B

Site C

Site A

-

1

1

6

Site B

1

-

2

4

Site C

1

2

-

(c) Form A

Form B

Form C

Form A

-

1

0

Form B

1

-

2

Form C

0

2

-

8.3. Three matrices representing site assemblages: (a) tableware forms and their quantities at sites, (b) sites and the number of forms they have in common with other sites, (c) forms and the number of sites where a pair of forms is copresent.

Similarity of Forms’ Distributions: The Brainerd–Robinson Coefficient The aim, research questions, and dependence assumption introduced above suggest that an exploratory analysis of networks per twenty-five-year period that represent the similarity of tableware forms’ distributions would be a valuable first step toward a better understanding of the dataset and how it can be interpreted in light of the dependence assumption. In this case study I discuss a method for creating and analyzing similarity networks, paying particular attention to the issue of the robustness of network analytical results: how sensitive are the results derived from network analysis techniques to our decisions of what constitutes the core of the network (in the sense of a subnetwork with particularly high similarity values)? There are many ways of describing the similarity of artifact distributions (Doran and Hodson 1975: 135–157; Shennan 1997: 222–234; for different approaches to similarity networks, see Östborn and Gerding 2014). Many such measures can be derived from and manipulated as a matrix, as in Figure 8.3a, which represents a list of tableware forms and their quantities attested at different sites. From this matrix we can derive two further matrices depending on the focus and aim of analysis: Figure 8.3b shows sites and the number of forms they

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:29:22, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.009

LESSONS LEA RNED F ROM NETWORK SCIENCE TECHNIQUES

191

have in common with other sites, while Figure 8.3c shows tableware forms and the number of sites where a pair of forms are both present. In the exploratory network analysis I focus on the latter representation since I aim to explore similarities and differences between forms, rather than between sites. In this case study I use a similarity measure commonly employed in archaeology: the Brainerd–Robinson (BR) coefficient (Brainerd 1951; Robinson 1951; Cowgill 1990; Shennan 1997, 233–234). Rather than considering absolute numbers as in the presence/absence technique, the BR coefficient considers proportions. Although it is commonly used to compare the similarity of pairs of site assemblages, I will use it to compare the similarity of pairs of forms’ distributions. This measure as used in this case study therefore compares the proportion of all sherds of a pair of forms found in different sites. For every pair of forms, this measure sums up the absolute difference between proportions per site and subtracts this from 200 (the maximum possible difference for a pair of forms’ distributions). This provides a numerical similarity value between 0 and 200 where 0 indicates no similarity and 200 complete similarity, using the following equation: p X S ¼ 200  jPik  Pjk j k¼1

Eq. 1. Brainerd–Robinson coefficient where P is the percentage representation of site k in the distribution patterns of forms i and j. For example, the distributions given in Figure 8.3a can be expressed as proportions (Figure 8.4a) for which a BR similarity matrix representing the similarity of forms’ distributions can be computed (Figure 8.4b). In this case study, the BR coefficient is calculated using a script written in R by Matt Peeples (2011b). Using the BR coefficient for forms’ distributions rather than sites’ distributions can only be justified if one understands what this equation does with the data and interprets the results in light of this. One must not forget that the BR coefficient employed in this way does not use the proportion of a form in sites’ assemblages at all. Most crucial is the difference between this approach and the presence/absence technique. Notice how in Figure 8.3c forms A–B have a presence/absence value of 1, and B–C a value of 2, while in Figure 8.4b forms A–B have a BR value of 106, and B–C a lower value of 94. Although one should be cautious not to overinterpret this small difference in the BR values, it is nevertheless clear that these two different approaches to comparing forms’ distributions reveal very different things. Forms A–B are only copresent on one site but their distributions are marginally more similar than those of forms B–C.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:29:22, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.009

192

TOM BR UG HMAN S

(a)

(b) Form A

Form B

Form C

Form A

200

106

0

31%

Form B

106

200

94

33%

Form C

0

94

200

Site A

Site B

Site C

Form A

100%

0%

0%

Form B

53%

16%

Form C

0%

67%

A

B 106

C 94

(c)

8.4. (a) The same matrix as Figure 8.3a but showing percentages of forms’ distributions rather than absolute numbers (e.g., 100% of all sherds of Form A are found on Site A); (b) BR coefficients of the same matrix, representing the similarity of forms’ distributions; (c) network representation of (b).

One drawback of using the BR coefficient for creating similarity networks is that it does not calibrate the results based on the number of sherds attested for each form; rather it emphasizes the diversity of sites where a form is attested. This means that often forms for which few sherds are attested and which often have a very limited distribution are directly compared to forms for which a large number of sherds are included in the dataset, leading to very high BR values for the former and possibly rather low values for the latter. However, the impact of this is immediately evident from the results, and in this case study the number of sites where a form is present is considered more indicative of the wideness of its geographical distribution than is the volume of that form attested at that site.

Creating and Exploring Similarity Networks The matrix of BR values given in Figure 8.4b can be represented as a network (Figure 8.4c). The diagonal of the matrix, representing perfect similarity of a forms’ distribution to itself, is disregarded in the exploratory network analysis. In these one-mode similarity networks, forms are connected to each other if the similarity of their distributions as expressed by the BR coefficient is greater than zero. The meaning of the relationships of these networks should be derived from our discussion of the meaning of the BR values above: a relationship between a pair of tableware forms indicates that they are copresent on at least one site, and gives a measure for the similarity of the proportions of their respective total distributed volume at sites where they are copresent. It is important to note that these

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:29:22, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.009

LESSONS LEA RNED F ROM NETWORK SCIENCE TECHNIQUES

193

networks are undirected (since the BR similarity matrices are symmetric), which implies that questions concerning the directionality of flows of goods cannot be addressed with this approach to creating similarity measures. This is not considered problematic for this case study, since my interest here is merely in the similarity of distributions. Tableware similarity networks are created per twenty-five-year period excluding self-loops (i.e., relationships from one form to itself are excluded since these always have complete similarity and a BR value of 200). These networks will be explored as a whole, but also using different “thresholds” on the BR values (i.e., removing edges with a BR value lower than a certain threshold value; my motivation for using thresholds is discussed in more detail below). The networks are also explored by grouping forms together according to their wares. This approach provides insights into how similar or different the distribution patterns of wares are. A range of network analysis measures to explore these networks is selected in light of the research questions and dependence assumption introduced above, and their interpretation within this particular research context should be described. Since in this case study I do not assume that indirect similarity (i.e., a path of two) allows for flows between two nodes, no path-based measures (e.g., closeness and betweenness centrality) are used to explore the similarity networks: I focus on nodes and their direct neighbors. Various measures are used to identify the structural features of networks as a whole and to compare them with each other (from here on referred to as “global measures”): number of nodes, number of edges, connected components, network clustering coefficient, heterogeneity, average degree, and density. The number of nodes is the number of individual forms— with a standard chronology that falls within a certain twenty-five-year period —that have been identified by archaeologists on sites and included in the ICRATES dataset. Connected components are groups of forms whose distributions are directly or indirectly similar to each other but completely dissimilar to forms in other groups. The network clustering coefficient is an indication of the tendency for forms to cluster together, which is suggestive of very similar distributions. The average degree is the average number of forms with which a given form has similarities in distribution patterns. The density is a normalized version of average degree. Heterogeneity indicated the existence of forms with similarities to far more other forms in terms of distribution, i.e., hubs. Just two node-based measures are used, hereafter referred to as “local measures”: node clustering coefficient and degree. The node clustering coefficient is a measure of the density of connections between one node’s direct neighbors; in this case study, it reflects the degree distribution similarity of the set of forms neighboring each form. The node degree is the number of forms that have a similar distribution to the node in question (for the technical definition of these measures, see the glossary in Collar et al. 2015).

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:29:22, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.009

194

TOM BR UG HMAN S

Network Structure and Sensitivity Analysis The general structure of the created networks per twenty-five-year period will be explored and a sensitivity analysis will be performed to see how this structure changes when one uses different thresholds of similarity values to create subnetworks. There are several reasons why it might be useful to cut off a chunk of the networks to analyze subnetworks: to identify and analyze nodes connected by the strongest relationships (i.e., the highest BR values between tableware forms); to make a dense network more sparse for easier visual exploration; or to test whether the analytical results one draws from a formal analysis of the whole network also hold true for subnetworks. Threshold values either could be selected with reference to a certain theory or they could be determined arbitrarily using some quantitative approach. The latter approach has been used by many archaeological network analysts, either by selecting a minimum edge value (e.g., Golitko et al. 2012) or by comparing the distribution of similarity values of the observed network with those of simulated networks (e.g., Östborn and Gerding 2014; Peeples 2011a). I believe the strong variations in the number of forms per site and the fragmentary nature of this dataset require one first to analyze the observed networks completely, and only then to analyze subnetworks derived with arbitrary threshold values in evaluating the robustness of the analytical results of the complete networks; that is, we should not attach any interpretive value to arbitrary thresholding but merely use it as an exploratory and sensitivity analysis tool. For this case study I do not believe it would be useful or even possible to argue for theoretical claims about which specific threshold value represents a strong similarity between forms’ distributions and which do not, other than the extreme BR values of 0 and 200. The “complete networks” referred to below are networks per twenty-fiveyear period of all forms and the similarity links between them of whatever strength. Below I discuss the distribution of the BR values of all complete networks to argue for the selection of possibly useful—but nevertheless completely arbitrary—threshold values. Two different thresholds are then used to create subnetworks from these complete networks: a threshold on the mean BR value and on the mean plus the standard deviation. I also explore how the ranking of each node according to certain network measures (degree and clustering coefficient) changes with these changing thresholds. Such a sensitivity analysis is crucial since it influences the interpretation of network patterns.

Distribution of BR Values Histograms representing the distribution of BR values for each twenty-fiveyear period summarize a wealth of information and can be used as an aid in deciding what threshold values to use (Figures 8.5, 8.6). Each histogram in

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:29:22, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.009

LESSONS LEA RNED F ROM NETWORK SCIENCE TECHNIQUES

200

1000 500

12000

100–125AD

8000

6000

2500

2000

190

170

150

130

90

190

170

150

130

190

170

130

2000 1500 1000 500

Brainerd-Robinson coefficient

Brainerd-Robinson coefficient

150–175AD

190

170

150

130

110

90

70

50

30

10

190

170

130

150

110

0 10

190

170

130

150

110

0 90

1000

0 70

1000

90

2000

3000

70

3000

4000

50

4000

5000

30

5000

Frequency of values

3000

Frequency of values

7000

6000

110

125–150AD

9000 7000

110

Brainerd-Robinson coefficient

Brainerd-Robinson coefficient

75–100AD

150

10

190

170

130

150

10

90

0 110

2000

0

110

4000

2000

Brainerd-Robinson coefficient

90

6000

90

4000

8000

70

6000

10000

50

8000

12000

30

10000

190

170

150

Frequency of values

14000

12000

0

50

70

10

190

170

150

10 14000

70

5000

16000

50

10000

130

50–75AD

16000

30

Frequency of values

15000

110

Brainerd-Robinson coefficient

25–50AD

20000

90

130

0 110

2000

0

1–25AD

70

4000

Brainerd-Robinson coefficient

25000

50

6000

1000

Brainerd-Robinson coefficient

30

8000

50

2000

10000

30

3000

190

170

150

130

110

90

70

50

30

10

4000

90

1000

5000

70

2000

6000

50

3000

7000

30

4000

Frequency of values

8000

5000

10

70

25–1BC

6000

30

50

50–25BC 14000

10

30

10

170

150

10

190

Brainerd-Robinson coefficient

9000 Frequency of values

Brainerd-Robinson coefficient

175–200AD 1200 Frequency of values

1000 900 800 700 600 500 400 300 200 100 0

1000 800 600 400 200

Brainerd-Robinson coefficient

190

170

130

150

110

70

50

30

10

190

170

150

130

110

90

70

50

10

30

0 90

Frequency of values

1500

7000

0

Frequency of values

2000

Brainerd-Robinson coefficient

75–50BC

Frequency of values

2500

0

190

170

150

130

90

110

70

0 50

100

0 30

20

130

40

300

90

60

400

110

80

500

70

100

600

50

120

30

140

Frequency of values

700 Frequency of values

160

3000

Brainerd-Robinson coefficient

Frequency of values

100–75BC

125–100BC 800

10

Frequency of values

150–125BC 180

195

Brainerd-Robinson coefficient

8.5. Distribution of BR values of form–form similarity matrices per period. Full line: mean value; dashed line: mean + standard deviation.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:29:22, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.009

TOM BR UG HMAN S

125–100BC

Frequency of values

400 300 200 100

600 400 200

200

180

200 200

180

160

120

140

200

180

160

140

120

100

150 100 50

150–175AD

200

180

160

140

120

100

80

60

Brainerd-Robinson coefficient

Brainerd-Robinson coefficient

Brainerd-Robinson coefficient

40

200

180

160

200

200

0 140

0 20

0 180

100

160

100

120

200

300

80

300

400

100

400

60

500

500

40

600

125–150AD

250

20

600

700

80

60

20

180

160

140

120

100

200

Brainerd-Robinson coefficient

Frequency of values

800 Frequency of values

700

100

400

100–125AD

75–100AD

140

600

Brainerd-Robinson coefficient

900

120

80

60

40

20

200

180

160

140

120

100

800

0

Brainerd-Robinson coefficient

100

1200 1000

200

0 80

80

1400

800

0

80

60

20

50–75AD

1600

40

200 60

200

Brainerd-Robinson coefficient

Frequency of values

Frequency of values

400

40

300

200

180

160

120

140

80

100

60

40

20

200

180

160

140

120

100

80

60

40

1000

60

400

25–50AD

1000

20

500

0

12000

40

600

100

1–25AD

20

700

40

Frequency of values

800

500

1200

600

160

25–1BC

900

Brainerd-Robinson coefficient

800

140

20

200

50–25BC

0 20

175–200AD

50 45 40 35 30 25 20 15 10 5 0

Frequency of values

70 60 50 40 30 20 10

Brainerd-Robinson coefficient

200

180

160

120

140

80

60

40

20

200

180

160

140

120

100

80

60

40

20

0 100

Frequency of values

Brainerd-Robinson coefficient

600

500 450 400 350 300 250 200 150 100 50 0

Brainerd-Robinson coefficient

Frequency of values

180

20

Brainerd-Robinson coefficient

75–50BC

Frequency of values

160

0 140

50

0

Brainerd-Robinson coefficient

Frequency of values

100

10

120

20

80

30

200 150

100

40

250

60

Frequency of values

50

200

180

160

140

120

80

100

60

40

20

0

60

120

5

300

70

80

10

80

60

20 15

350

40

Frequency of values

Frequency of values

25

100–75BC

90

40

150–125BC

30

100

196

Brainerd-Robinson coefficient

8.6. Distribution of BR values of form–form similarity matrices per period, excluding values between 0 and 10. Full line: mean value; dashed line: mean + standard deviation.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:29:22, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.009

LESSONS LEA RNED F ROM NETWORK SCIENCE TECHNIQUES

197

Figures 8.5 and 8.6 draws on all BR values for a single network (forms dated to a twenty-five-year period), where the sum of the number of occurrences (y-axis) of a certain BR value (grouped per 10 on x-axis) is displayed as a bar. Figure 8.5 includes all of the values for each network and immediately shows the heavy skew toward low BR values and the long tail. Indeed, for all networks a high number of BR values are 0 or no more than 10. This is due to the high number of sites that are included in this analysis (between 59 in 150–125 BC and 162 in AD 1–25). As a result, the mean BR value is low for all networks (solid vertical line in Figures 8.5, 8.6, Table 8.2). This high variation has a strong impact on the results of the standard deviation as well, which will be pushed quite high due to the existence of outliers (dashed vertical line in Figures 8.5, 8.6, Table 8.3). The inter-quartile range could be considered more informative of typical observations in such cases (Shennan 1997, 44). However, due to these distributions’ great spread, the inter-quartile range itself is very low for many of the periods (Table 8.2). Using the third quartile as a threshold value would, for most networks at least, not serve my purpose of bringing more visual clarity to the exploration of a dense network. A high threshold would be more suitable for that purpose, which is why I decided to use two threshold values: the mean, and the mean plus standard deviation (Table 8.3). This decision implies that pairs of forms which are connected by BR values higher than the mean plus standard deviation should be considered rather extreme outliers in the exploratory network analysis. In Figure 8.6 I remove all BR values lower than 10, which reveals that for the networks up to AD 75 the histogram trend is not very smooth either: there is greater difference in the proportions of the variation for these networks, and outliers are more frequent than for networks of later periods. The threshold values plotted on these distributions show a further distinction: the mean lies between 22 and 31 for networks up to AD 75 and between 16 and 12 for later networks. These differences might be indicative of the decrease in the diversity of ESC forms after AD 75. Before this date ESC has a high diversity of forms but many of these forms are copresent and have a limited distribution, which would result in high BR values. It will be interesting in this case study to explore whether this distinction is also reflected in some of the exploratory network measures. These results suggest that the distributions of tableware forms are not very similar, given the low BR values. However, the results also suggest there is great variety among BR values larger than 10, and that a larger number of forms have a more similar distribution in the periods up to AD 75. This might be a result of the decrease in the distribution of ESA around this time. In the rest of this section I explore the variance in BR values in more detail by using threshold values and grouping forms according to wares.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:29:22, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.009

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:29:22, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.009

Minimum First quartile Median Mean Third quartile Maximum

0 0 0 22 33 167

0 0 6 30 59 200

0 0 0 22 30 200

0 0 0 31 40 200

0 0 0 27 23 200

0 0 0 26 22 200

150–125 BC 125–100 BC 100–75 BC 75–50 BC 50–25 BC 25–1 BC 0 0 0 23 15 200

1–AD 25

table 8.2. Summary statistics of BR coefficients for the complete networks per period.

0 0 20 28 20 200

AD 25–50 0 0 0 23 18 200

AD 50–75 0 0 0 16 13 200

0 0 0 15 12 200

AD AD 75–100 100–125

0 0 0 14 10 200

AD 125–150

0 0 0 14 1 200

0 0 0 12 1 200

AD AD 150–175 175–200

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:29:22, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.009

Mean 22 St. dev. 37 Mean + St. dev. 59

30 44 74

22 39 61

31 54 85

27 52 78

26 51 77

150–125 BC 125–100 BC 100–75 BC 75–50 BC 50–25 BC 25–1 BC 23 49 73

1–AD 25 28 54 82

AD 25–50 23 48 71

AD 50–75

16 37 54

15 33 48

AD AD 75–100 100–125

14 34 48

AD 125–150

14 37 51

12 31 43

AD AD 150–175 175–200

table 8.3. Summary statistics of BR coefficients of complete networks per period. The mean and mean + standard deviation are suggested as thresholds for exploring the networks.

200

TOM BR UG HMAN S

Global Network Measures Results of the global network measures are first discussed for the complete networks, followed by the networks with a threshold on the mean BR value, and finally the networks with a threshold on the mean plus standard deviation. This is followed by a discussion of the effects of the threshold on the number of nodes in the networks. Complete network: the network density remains rather high throughout time, although it is highest in the first few periods (Table 8.4, Figure 8.7). This is possibly due to the high similarity in distributions of ESC and ESA forms. Heterogeneity, on the other hand, increases through time, suggesting that forms with a similar distribution to many other forms and skewed degree distributions occur more frequently in the later periods. This further suggests that the decreasing density creates a pattern of well-connected nodes and less well-connected nodes. The groups of forms with a similar distribution become smaller over time. The clustering coefficient remains very high throughout all periods, suggesting that forms in general have a distribution that shows similarities with many other forms’ distributions. Threshold mean: density is much lower when thresholding on the mean BR value (Table 8.5, Figure 8.8). The density of networks decreases over time. The heterogeneity shows a similar increasing pattern as in the complete network. However, now the values are slightly higher, and there is a peak with many hubs in AD 25–50. This suggests that this threshold emphasizes differences in node degrees more. It seems that hubs are a strong pattern in these networks. The clustering coefficient is invariably high. Nodes with this threshold are still very much clustered with other nodes with similar distributions. Threshold mean plus standard deviation: the density is again lower with this threshold, which is to be expected (Table 8.6, Figure 8.9). However, the general trend of decreasing density over time remains, suggesting that this is a robust pattern. The heterogeneity measure is very interesting. Although it still shows a similar trend as before, this is now much exaggerated: the threshold reveals a much stronger difference between nodes in terms of degree as time moves on, in particular between 50 BC and AD 100 when hubs are most common. This seems to be caused by one or two nodes that bridge dense cliques, mainly between ESC and a combination of ESA/ESB/ESD. The clustering coefficient shows a similar trend as with other thresholds, although slightly lower overall. Total number of nodes (forms): the networks of BR similarity values described above do not include isolated forms (i.e., forms whose distribution similarity with other forms’ distributions as described by the BR coefficient is 0). However, only a marginal number of forms per period are not included in the networks (min 0, max 6; Table 8.4). The largest number of isolated forms is

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:29:22, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.009

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:29:22, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.009

Nodes 17 Nodes (incl. 17 isolates) Connected 1 components Average degree 8 Clustering 0.78 coefficient Density 0.5 Heterogeneity 0.43

66 66

1

31.3 0.84

0.48 0.44

37 37

1

20.4 0.85

0.57 0.34

0.44 0.45

42.8 0.84

1

99 100

0.42 0.52

45.2 0.84

1

109 111

0.4 0.54

53.9 0.83

1

137 137

150–125 BC 125–100 BC 100–75 BC 75–50 BC 50–25 BC 25–1 BC

0.36 0.55

58.7 0.81

1

166 167

AD 1–25

table 8.4. Global network measures for the complete networks per twenty-five-year period.

0.38 0.54

53.3 0.82

1

143 146

AD 25–50

0.38 0.54

54.7 0.83

1

146 149

AD 50–75

0.37 0.57

39.2 0.79

2

107 111

0.39 0.58

34.7 0.81

1

91 97

AD AD 75–100 100–125

0.38 0.56

21.5 0.82

1

58 64

AD 125–150

0.31 0.64

9.8 0.76

1

33 36

0.29 0.59

10.1 0.78

1

36 39

AD AD 150–175 175–200

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:29:22, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.009

Nodes 17 Nodes (incl. 17 isolates) Connected 2 components Average degree 4.5 Clustering 0.659 coefficient Density 0.279 Heterogeneity 0.442

64 66

1

19.2 0.792

0.304 0.461

37 37

1

12.3 0.856

0.342 0.422

0.314 0.495

29.2 0.832

1

94 100

0.264 0.601

27.5 0.788

2

105 111

0.257 0.644

33.7 0.827

1

132 137

150–125 BC 125–100 BC 100–75 BC 75–50 BC 50–25 BC 25–1 BC

0.232 0.662

37.5 0.791

2

163 167

1–AD 25

0.241 0.721

33.7 0.788

3

141 146

AD 25–50

0.239 0.64

34 0.784

2

143 149

AD 50–75

0.235 0.601

24 0.751

2

103 111

0.259 0.623

23 0.771

1

90 97

AD AD 75–100 100–125

table 8.5. Global network measures for the networks per twenty-five-year period with a threshold on the mean BR value.

0.261 0.594

14.6 0.771

1

57 64

AD 125–150

0.222 0.639

6.6 0.777

1

31 36

0.197 0.544

6.9 0.726

1

36 39

AD AD 150–175 175–200

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:29:22, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.009

Nodes 14 Nodes (incl. 17 isolates) Connected 2 components Average degree 3.9 Clustering 0.833 coefficient Density 0.297 Heterogeneity 0.414

61 66

2

9.9 0.713

0.164 0.515

33 37

4

6.4 0.601

0.201 0.572

0.233 0.771

18.9 0.807

3

82 100

0.201 0.817

18.3 0.806

7

92 111

0.201 0.877

23.1 0.822

7

116 137

150–125 BC 125–100 BC 100–75 BC 75–50 BC 50–25 BC 25–1 BC

0.175 0.962

25.1 0.799

8

144 167

1–AD 25

0.191 0.901

24.4 0.828

11

129 146

AD 25–50

0.154 0.946

20.4 0.802

8

133 149

AD 50–75

0.126 0.758

12 0.696

4

96 111

0.134 0.597

11.3 0.702

2

86 97

AD AD 75–100 100–125

0.129 0.59

6.9 0.712

3

54 64

AD 125–150

table 8.6. Global network measures for the networks per twenty-five-year period with a threshold on the mean + standard deviation BR value.

0.148 0.688

3.9 0.8

5

27 36

0.127 0.44

3.8 0.718

5

31 39

AD AD 150–175 175–200

204

TOM BR UG HMAN S

1.000 0.900 0.800 0.700 0.600 0.500

Clustering coefficient Density

0.400

Heterogeneity 0.300 0.200 0.100

0– C 7 75 5BC –5 0 50 BC –2 5 25 BC –1 B 1– C 25 A 25 D –5 0A 50 D –7 75 5AD –1 10 00A 0– D 1 12 25 5– AD 1 15 50 0– AD 1 17 75 5– AD 20 0A D

0B

10

5B 12

5–

0–

12

15

10

C

0.000

8.7. Global network measures per twenty-five-year period for the complete network.

1.000 0.900 0.800 0.700 0.600 0.500 0.400

Clustering coefficient Density Heterogeneity

0.300 0.200 0.100

15

0– 1 12 25 5– BC 10 10 0B 0– C 75 75 BC –5 0 50 BC –2 5B 25 C –1 B 1– C 25 25 AD –5 0 50 AD –7 5 75 A –1 D 10 00 0– AD 1 12 25 5– AD 1 15 50 0– AD 1 17 75 5– AD 20 0A D

0.000

8.8. Global network measures per twenty-five-year period for the network with a threshold on the mean similarity value.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:29:22, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.009

LESSONS LEA RNED F ROM NETWORK SCIENCE TECHNIQUES

205

1.000 0.900 0.800 0.700 0.600 0.500

Clustering coefficient

0.400

Density Heterogeneity

0.300 0.200 0.100

15

0– 1 12 25 5– BC 10 10 0B 0– C 75 75 BC –5 0 50 BC –2 5B 25 C –1 B 1– C 25 A 25 –5 D 0 50 AD –7 75 5A –1 D 10 00 0– AD 1 12 25 5– AD 1 15 50 0– AD 1 17 75 5– AD 20 0A D

0.000

8.9. Global network measures per twenty-five-year period for networks with a threshold on the mean + standard deviation value.

90 80 70 06 50

ESA ESB

40

ESC

30

ESD

20 10

D

D

0A

17

5–

20

D

5A

0A 15

0–

17

D 5A 5– 12

12 0–

15

AD

D

00 –1

75

10

D

5A –7

50

0A

AD

–5 25

BC

25 1–

25

–1

C

C 50

–2

5B

0B

BC

–5 75

75

0B 10

0– 10

5–

12

15

0–

12

5B

C

C

0

8.10. Number of nodes (forms) per ware for the complete networks.

in the second part of the period of study (AD 25 onwards) and the effect is stronger for the last four periods, which have fewer nodes: more forms in these periods have no similarity to other forms’ distributions. The period between 50 BC and AD 100 has the highest number of forms (Figures 8.10–8.12). However, the thresholds do not strongly affect the number of forms. A threshold on the mean BR value only decreases the number of forms for

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:29:22, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.009

206

TOM BR UG HMAN S

90 80 70 06 50

ESA ESB ESC ESD

40 30 20 10

50

25

–2

5B C –1 BC 1– 25 AD 25 –5 0A D 50 –7 5A 75 D –1 00 10 AD 0– 12 12 5A D 5– 15 15 0A D 0– 17 17 5A D 5– 20 0A D

0B C

BC

–5

75

75

0B C

0– 10

5– 12

15

0–

10

12

5B C

0

8.11. Number of nodes (forms) per ware for the networks with a threshold on the mean similarity value. 90 80 70 06 50

ESA ESB ESC ESD

40 30 20 10

D

D

0A

5A 17

5–

20

D 0–

17

D

0A 15

5A 5– 12

10

0–

12

00 –1

75

15

AD

D

D

5A

0A 50

–7

AD

–5 25

BC

25 1–

5B

–1 25

–2 50

–5

0B

C

C

BC 75

0–

75

0B 10

10 5–

12

15

0–

12

5B

C

C

0

8.12. Number of nodes (forms) per ware for the networks with a threshold on the mean + standard deviation similarity value.

a few periods, while the mean plus standard deviation threshold decreases the number of forms for all periods only marginally. The biggest difference is for the periods between 25 BC and AD 75, and for ESA forms in particular. Up to AD 75 and after AD 150 the number of ESA forms is most affected by using a threshold on the mean plus standard deviation value. This suggests that although most wares always have significantly similar distributions to some forms (mostly of the same ware), a large proportion of ESA forms have an overall extremely low similarity to all forms. This is not the case in the period AD 100–150, when ESA forms show significant similarity with at least one other form, while a large proportion of ESD forms show no significant similarity to any other forms.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:29:22, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.009

LESSONS LEA RNED F ROM NETWORK SCIENCE TECHNIQUES

207

Local Network Measures: Ranking Clustering Coefficient and Degree The previous section discussed the sensitivity of global network measures to thresholds; this next section focuses on its impact on local network measures’ results by tracing the changes in the ranking of node clustering coefficient and degree. The ranking was created as follows. Node clustering coefficient and degree measures were calculated. The results were ordered from high to low and nodes were given a rank (equal values were given an equal rank). The number of places each node changed in the ranking was counted as it was traced from the complete network to that with a threshold on the mean, and further to that with a threshold on the mean plus standard deviation (this was only done for nodes that are present in the mean plus standard deviation network). In order to compare the proportional change of node rankings across different periods, the results were normalized by dividing them by the maximum possible number of changes in ranking a node could undergo in shifting from the complete network to the network with a threshold on the mean plus standard deviation (and multiplied by 100 to express this as a percentage). Boxplots of these results for clustering coefficient and for degree were created. Clustering coefficient (Figure 8.13): the networks of all periods show strong similarities in their sensitivity to changing thresholds. The inter-quartile range in almost all cases is limited to 0–35/45%, while some few nodes see up to 80% change in ranking. In the periods 125–100 BC and AD 1–25, one node sees 100% change (ESC form M200ST27 and ESD form EAAP28 respectively). The period AD 25–50 knows the highest range of change. The networks of the

100

* *

80

60

40

20

D

D 17

5–

20

0A

5A

D 15

0–

17

0A

D 15

5A 5– 12

10

0–

12

00

5A

–1 75

–7 50

AD

D

D 0A

AD 25

–5 25

1–

BC

C 5B

–1 25

C –2 50

0B

BC

–5 75

75

0B

0– 10

10 5–

12

15

0–

12

5B

C

0 C

Proportion of change in node ranking (in%)

Clustering Coefficient sensitivity analysis

8.13. Boxplot of the proportion of change in node ranking of the clustering coefficient.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:29:22, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.009

208

TOM BR UG HMAN S

80 70 60 *

50

* * *

40 30 20 10

10

0–

10

12

5– 12

0– 15

0B C 75 BC 75 –5 0B C 50 –2 5B C 25 –1 BC 1– 25 AD 25 –5 0A D 50 –7 5A 75 D –1 00 10 AD 0– 12 5A 12 D 5– 15 0A 15 D 0– 17 5 17 AD 5– 20 0A D

0 5B C

Proportion of change in node ranking (in%)

Degree sensitivity analysis 90

8.14. Boxplot of the proportion of change in node ranking of the degree.

period 150–125 BC are very different from all others because of the limited number of nodes, and changing thresholds have little effect in the case of the clustering coefficient results. This sensitivity analysis therefore indicates that changes in the clustering coefficient ranking of nodes are common. However, nodes only rarely change their ranking dramatically by over 50%. The vast majority of nodes remain within 50% of their ranking position. Degree (Figure 8.14): the networks of all periods show strong similarities in their sensitivity to changing thresholds. The inter-quartile range in almost all cases is limited to 15–35%, while some nodes see a change of up to 80% in the first period only. A few outliers are identified with only 50–55% change. Two periods show a different pattern. In the period 125–100 BC change in degree is slightly higher, with an inter-quartile range of 15–45%. The most dramatic difference is seen in the networks of 150–125 BC, with most nodes changing their rank on the order of 0–65%. This is no doubt the result of the small number of nodes in this network. This sensitivity analysis therefore indicates that changes in the degree rankings are common. Almost every node is affected by changing thresholds. However, the change in their ranking overall is very limited, rarely more than 35%. An exception to this is the period 150–125 BC, which has significant change but also a low number of nodes.

Detailed Description per Twenty-Five-Year Period The appendix to this chapter includes a more detailed description of the similarity networks for the twenty-five-year periods from 25 BC to AD 150

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:29:22, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.009

LESSONS LEA RNED F ROM NETWORK SCIENCE TECHNIQUES

209

in which all four tablewares circulated on a large scale in the eastern Mediterranean. This description is performed on the level of individual forms as well as for each ware, and takes into account the results of the sensitivity analysis of the suggested threshold values’ impact. RESULTS OF THE EXPLORATORY NETWORK ANALYSIS CASE STUDY

The distributions of tableware forms are, in general, not very similar, but there is great variation in the BR values. The sensitivity analysis suggests that the exploratory network measures overall show the same general trends for the complete networks as for the networks with a threshold applied: decreasing density, high clustering coefficient, peak in heterogeneity, similar trend in the number of nodes. This suggests these trends are quite robust and should inform the discussion of the exploratory network analysis results. Unsurprisingly, however, the effect of applying a threshold on the mean plus standard deviation BR value had the strongest impact on the results: for example, concerning the decrease in the number of ESA forms included. The proportional change of nodes in rankings as a result of changing thresholds is common but never dramatic. The change for the clustering coefficient is slightly higher on average, 0–45%. Change for the degree measure is 15–35%. The period 150–125 BC shows strong sensitivity, especially for the degree measure. Overall this sensitivity analysis indicates that the clustering coefficient and degree rankings of all periods are sensitive to changing thresholds, but not dramatically, suggesting that the most persistent patterns will be reflected in all thresholds. This analysis also reveals that the outliers and networks with a high degree of change should be addressed in more detail, which was accomplished through the exploratory network analysis. It should be clear that the sensitivity analysis performed here is by no means conclusive regarding the robustness of the exploratory network analysis results. One could argue that the sampling bias in the ICRATES dataset is not directly addressed, something that could be done by removing randomly selected chunks of data, performing the analysis again, and comparing results, or by comparing the observed networks with simulated random networks based on the same numbers of nodes and edges. This sensitivity analysis merely served to explore the dataset in its network form and to calibrate any possible interpretive statements about pairs or groups of forms that show particularly high similarity. A more in-depth evaluation of the sampling bias in the ICRATES dataset falls outside the scope of this chapter (but see Willet 2012, 43–58). The technical description of similarity networks can be summarized by focusing on the general trends. From 25 BC onwards, ESA distribution becomes less internally homogeneous. Although tableware produced in Italy

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:29:22, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.009

210

TOM BR UG HMAN S

also becomes widely distributed around this time (Italian Sigillata), this is nevertheless also a local eastern phenomenon. ESD, ESB, and ESA show increasing similarities in their distribution patterns, especially between 25 BC and AD 25. However, one should not forget that during this period ESA was still present at a much larger number of sites than any of the other wares. The higher similarity between wares in this period should also be seen in light of ESA being the dominant ware on many sites with diverse assemblages. After that the similarity of ESB and ESD to ESA forms’ distributions decreases, while ESA maintained an unrivaled breadth of distribution. There is another increase of similarity between these wares’ distributions from AD 75 until AD 125. As well as a higher similarity to ESB and ESD, it is important to note that the period between AD 75 and AD 125 also marks the decrease of the breadth of ESA distribution. ESC, on the other hand, retains its internal homogeneity until its corpus of forms decreases from AD 75 onwards. From then on it becomes increasingly similar to ESB and to some extent also ESA, although this trend stops around AD 125. ESC remains very dissimilar to ESD throughout all periods. In addition to these general trends, a detailed description of the similarity of distribution patterns of different forms identified those forms whose distributions are more similar to those of other wares (see appendix). In light of the dependence assumption introduced above, this suggests that the processes that gave rise to these distributions should not be seen as exclusively warespecific. It is true that one persistent pattern is the higher volume and diversity of a ware close to its production area, and the hypothesis that a wide range of a ware’s forms in the region around its production area was distributed through similar processes is not unlikely. However, focusing exclusively on the core areas of a ware’s distribution will not allow one to understand such exceptional cases that suggest similar distribution processes for different wares, and to what extent different wares were considered different products in the past. Sadly, but importantly, the exploratory analysis performed here also added a sobering note to our ability to answer such interesting questions using our current knowledge of tableware distributions. Since forms were considered modern analytical constructs, the differences in their distributions can hardly be considered a direct reflection of differences in past human behavior. Moreover, the strong variability in the number of forms and sherds per form recorded at different sites did not allow me to use this form-based evidence to perform a statistical goodness-of-fit test to evaluate hypotheses of distribution processes, unless the sampling bias is more thoroughly quantified, an exercise that was not possible within the framework of this chapter.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:29:22, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.009

LESSONS LEA RNED F ROM NETWORK SCIENCE TECHNIQUES

211

DISCUSSION AND CONCLUSIONS

The exploratory network analysis provided interesting insights into the patterns in the dataset used and how these should be interpreted in light of our assumptions about the meaning of similarity in tableware forms’ distributions. The use of formal network methods was therefore informative, but in light of the aim of this chapter I would also like to highlight how it was uninformative, in the sense that the same results could have been obtained using formal methods more familiar to archaeologists and historians. In this final section I discuss the advantages of using formal network methods in this case study, and I also take a skeptical stance about the methods used, and how these can be formulated as a series of suggestions to help evaluate the potential of formal network methods for studying maritime connectivity. The advantages of the method used in the case study presented here should be seen in light of the fact that exploratory network analysis is a form of exploratory data analysis; that is, summarizing the main characteristics and specific patterns in a dataset by visual and statistical means (Tukey 1977). A combination of statistical techniques and visualization were used to represent and explore the distribution patterns of different wares and forms. This allowed me to identify which wares and forms had similar or different distribution patterns, and therefore allowed me to answer the first two research questions. Second, the representation of data patterns as networks accompanied by a dependence assumption also aids a critical evaluation of the implications of formulating specific dependence assumptions. Network data representations offer a powerful tool for considering theories on past maritime connectivity and how they could be reflected in the data available to us. Third, my critical stance toward what I call “uninformative” results here should actually be seen as informative in light of a practice that is not common enough in archaeology. The ability of one formal method to reproduce the results of other formal methods is an important advantage. The results might not be new or surprising, but we increase our confidence in them if they are revealed through multiple different approaches. Finally, thanks to the formulation of a dependence assumption, the network patterns of interest could be interpreted as representations of the outcomes of particular past distribution processes. For example, in this case study distribution patterns of forms and wares that were similar were interpreted as being the result of similar distribution processes. The value of this kind of result lies in its ability to offer a data representation of an “expected outcome” that can be compared with the outcomes of simulated processes in a confirmatory analysis (e.g., this is done for the patterns identified in this case study in Brughmans and Poblome 2016). It therefore allowed me to provide an answer to the third research

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:29:22, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.009

212

TOM BR UG HMAN S

question in the form of formulating a hypothesis, but it did not allow me to prove this hypothesis since it was not tested. This leads me to the main critical point I want to make in this paper using the present case study: exploratory network analysis was not the only method that would have allowed me to address the first two research questions, and it does not succeed in answering the third research question. This limitation results from the exploratory network methods employed only drawing in part on what makes network science distinct from other approaches: the ability to represent and analyze dependence assumptions of how relationships affect each other. Network methods were not used to test hypotheses about how relationships influence one another. As a result this approach did not reveal anything about the processes driving the distribution of tableware forms, and whether different forms were distributed through similar processes. In this case study, network methods were used to represent a dataset as network data that was subsequently visually explored. However, a visual exploration of the BR similarity matrices (e.g., Figure 8.4), without representing them as networks, proved equally useful for identifying patterns of interest. Network methods were also used to offer summary statistics describing the changing structure of the similarity networks (Tables 8.4–8.6). However, the results of these measures mainly revealed a few general trends which were also evident from the histograms with distributions of BR values per period (Figures 8.5, 8.6). Moreover, the use of network methods in this case study reveals a further challenge of a technical nature. A wide variety of network techniques exists, each with different data requirements, implemented in different software packages, with different advantages and limitations. However, these network techniques are still less well known to archaeologists and historians than other types of exploratory data analysis. As a result, they require more specialized technical knowledge if they are to be implemented, interpreted, and evaluated critically by both the scholars employing them and those assessing their use. All of the advantages mentioned above are good reasons for considering the exploratory network analysis in this case study informative, but none of them are exclusive to formal network methods. It should be clear that I do not wish to argue against the use of formal network methods motivated by these advantages. I merely wish to emphasize that this is not where their unique contribution should be sought when at present the potential of formal network methods for the study of past maritime connectivity is still ill-defined. By focusing strongly in this case study on the motivations behind the use of formal methods and by critically discussing the implementation and interpretation of the selected methods in much technical detail, a few lessons can be learned that can aid those evaluating the use of formal network methods within specific research contexts. First, formal network methods are more

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:29:22, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.009

LESSONS LEA RNED F ROM NETWORK SCIENCE TECHNIQUES

213

than just metaphors, and they should not be considered methodological “black boxes.” Although a wide variety of network techniques exists—each with different data requirements, different software packages, and different advantages and limitations—these technical details will need to be scrutinized and these formal network methods will inevitably become more common in archaeological research. Existing network science techniques were originally developed to address specific types of research question and function according to formalized rules. Archaeologists and historians may apply these techniques when addressing similar research questions, in which case the data requirements of the chosen techniques need to be met and the interpretation of the obtained results may not violate the techniques’ rules. However, many of the research questions that commonly confront us in the study of ancient maritime connectivity might require the development of new network science techniques. Second, the particular needs and challenges of a specific research context will dominate the selection of formal methods. I have argued that the unique contribution of network science lies in its ability to deal with dependence assumptions, which are formal treatments of our theories about past maritime connectivity. There can therefore be no such thing as a standard set of network analysis techniques that can be used in every research context, at least not yet in the case of the study of past maritime connectivity. Each technique used should be selected for its potential ability to do something necessary in light of the research aims that other techniques cannot do, and its functioning and interpretation will depend entirely on that research context. A third lesson follows from this: the ability of a formal network method to provide unique insights should be evaluated in light of how it allows for the representation or analysis of dependence assumptions. The clear and unambiguous formulation of dependence assumptions within the framework of a particular research context is therefore a crucial step in any evaluation of the use of formal network methods. Formal network methods are many and varied, and their potential for advancing our knowledge of past maritime connectivity is still ill-defined. In this chapter I have argued that the process of exploring this potential will require us to draw explicitly on our theories about past connectivity to motivate the selection of the most appropriate network methods. In doing so I did not aim to argue against the creative exploratory use of formal methods that could well lead to new insights without a clear sense as to the method’s unique contribution. I merely play devil’s advocate to explore how network methods can make unique future contributions to our discipline. In the words of the prominent social network analyst Patrick Doreian (1988, 310–311), “we do not need to use a lot of hammers nor decide prematurely on a few hammers and use only those. We need to select structural hammers on the basis of their ability to find appropriate structure.”

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:29:22, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.009

214

TOM BR UG HMAN S

APPENDIX: DESCRIPTION OF SIMILARITY NETWORKS PER TWENTY-FIVE-YEAR PERIOD In the following descriptions “very strong similarity” refers to a BR value higher than the mean plus standard deviation; “strong similarity” refers to a BR value higher than the mean; a “homogeneous distribution” refers to forms of the same ware which are similarly distributed among sites.

25–1 bc ESA: has a homogeneous distribution but much less so than in previous periods. A number of forms’ distributions are very similar to each other (EAA11, EAA12, EAA12/32, EAA13A, EAA13B, EAA21/22, EAA22/43, EAA22A, EAA22A–B, EAA22B, EAA2–3, EAA24, EAA24–25, EAA25, EAA26–27, EAA26A, EAA26A–D, EAA26B, EAA26C, EAA26D, EAA27, EAA28, EAA28–30, EAA29, EAA3, EAA42, EAA4A, EAA4A–B, EAA4B, EAA5A, EAA5A–B, EAA5B, EAA7, EAA8). All other forms’ distributions are not very similar to each other. Some of the latter (EAA21/22, EAA22/43, EAA22A, EAA22A–B, EAA22B, EAA2–3) show a strong similarity with some ESB forms (EAA16, EAA21). Some other forms’ distributions are more similar to ESD than to ESA or ESB (EAA102, EAA22B, EAA26–27, EAA26A–D). Of these only EAA22B is very similar to ESA forms. EAA26–27 is similar to many ESD forms, because this form is only present at Paphos. ESA forms’ distributions are dissimilar to those of ESC forms. ESB: not very internally homogeneous. One form (EAA21) shows strong similarity to ESA forms. Two forms (EAA2, EAA22) are very similar to ESC forms due to their presence on Assos and Ephesus. ESC: overall very homogeneous. A few forms are less similar due to limited distribution, but not more similar to ESA or ESD. No forms show strong similarity with ESA. M-SSu25 is very similar to many ESD forms, because it is present on Paphos, the only sites with evidence of a number of ESD forms. ESD: a group of ESD forms is pretty homogeneous, mostly because some are only present on Paphos. Three forms are dissimilar to all others (EAAP18A–B, EAAP33–34, EEP44–X45). Two forms are more similar to ESA (EAAP37A–B, EAAP37B).

ad 1–25 ESA: not very homogeneous, less so than in the previous period, but still showing slightly more similarity among ESA forms’ distributions than with forms of other wares. A number of forms are very similar to each other (EAA12, EAA42, EAA43, EAA44, EAA45, EAA46, EAA47, EAA4A, EAA4A–B, EAA4B, EAA5A, EAA5A–B, EAA5B; these are less than in the previous period, although there is some overlap). All other forms are not very similar to each other. Some of the latter (EAA12/32, EAA21/22, EAA30/33–34, EAA33/36) show a strong similarity with some ESB forms (EAA21, EAA30–31). Some other forms are more similar to ESD than to ESA or ESB (EAA26–27, EAA26A–D, EAA45). Of these only EAA45 is very similar to ESA forms. EAA26–27 is similar to many ESD forms, because this form is only present at Paphos. ESA forms are dissimilar to ESC forms. ESA forms show most similarity with ESB and ESD forms, not ESC. ESB: rather homogeneous distributions of forms. One form (EAA30–31) shows strong similarity to ESA forms. Two forms (EAA2, EAA22) are very similar to ESC forms due to their presence on Assos and Ephesus. ESB is very dissimilar from ESD.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:29:22, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.009

LESSONS LEA RNED F ROM NETWORK SCIENCE TECHNIQUES

215

ESC: overall very homogeneous. A few forms are less similar due to limited distribution, but not more similar to ESA or ESD. No forms show strong similarity with ESA. M-SSu25 is very similar to many ESD forms, because it is present on Paphos, the only sites with evidence of a few ESD forms. Two ESB forms (EAA2, EAA22) are very similar to ESC forms due to their presence on Assos and Ephesus. ESD: less homogeneous internally than in the previous period. A group of ESD forms is pretty homogeneous, mostly because some are only present on Paphos. Three forms are dissimilar to everything (EAAP44–X45, EAAP4B/P6, EAAP5/P6). ESD forms show much stronger similarity to ESA forms than in the previous period. One form is more similar to ESA (EAAP23B).

ad 25–50 ESA: not very homogeneous internally, even less than in the previous period, but still slightly more homogeneous internally than with other wares. ESA forms show most similarity with ESB and ESD forms, not ESC. But that similarity is also less striking than in the previous period. There are no forms very similar with ESD. One ESB form (EAA30–31) shows strong similarity to ESA forms, because of its presence in Jerusalem. Two ESC forms (EAAL6, EAAL9A) show strong similarity to some ESA, because they are virtually exclusive to Jerusalem. ESB: very internally homogeneous. One ESB form (EAA30–31) shows strong similarity to ESA forms, because of its presence in Jerusalem. One ESC form (M-ST22) is very similar to ESB; it is present on Assos and Ephesus. ESB is rather dissimilar from ESD. ESC: very internally homogeneous. One ESC form (M-ST22) is very similar to ESB; it is present on Assos and Ephesus. Two ESC forms (EAAL6, EAAL9A) show strong similarity to some ESA, because they are virtually exclusive to Jerusalem. ESC and ESD are very dissimilar, with the exception of ESC form EAAL15 which is most similar to ESD and not at all to ESC, because it is only present at Lepcis Magna. ESD: not very homogeneous, similar to the previous period. ESD forms show much less similarity to ESA than in the previous period. ESC and ESD are very dissimilar, with the exception of ESC form EAAL15, which is most similar to ESD and not at all to ESC, because it is only present at Lepcis Magna.

ad 50–75 ESA becomes slightly more internally similar. Similarities to ESD increase slightly, in particular forms EAA39, EAA41, EAA45 due to their presence on Paphos and Amathous. Similarities with ESB increase slightly, especially form EAA35/40, which is only recorded for Athens. ESA is still very dissimilar from ESC. ESB: only slightly less internally homogeneous than in the previous period. Shows slightly more similarity to ESD and ESA than in the previous period. In particular forms 58 early, 60, 60 early, 64/65, 65, 70 and 70 early are similar to ESA form EAA35/40 (only found in Athens) because they are all found together in Athens. Almost all ESB forms show strong similarity with one or more ESC forms. Just one form (30–32) shows strong similarity with ESD because it is only found on Berenice. ESC: ESC becomes slightly less internally homogeneous. It is still very dissimilar from ESA and ESD. Some forms show similarities with ESB (EAAL20,M-SN11a, MSN33a, M-SSa27a, M-ST22, M-ST24b), even though they have a wide distribution. ESD becomes slightly more similar to ESB and ESA, and it is not very internally homogeneous. EAAP5/P6 ESD is similar to ESB because it is only found on Berenice.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:29:22, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.009

216

TOM BR UG HMAN S

EAAP10 and EAAP28 are similar to ESA, both are copresent on Berenice, Amathous, Corinth, Panayia, Paphos, Tarsos. They have a wide distribution and are attested at sites with a high diversity.

ad 75–100 ESA becomes slightly less internally homogeneous. Similarities to ESD increase slightly again, in particular for ESA forms 111, 36, 39, 40A–C, 40C, 41, 51, 53, 54, 62. Similarities to ESB increase again, in particular with ESA forms 35/40, 37A–B, 40A–C, 65. These are copresent at Athens and all but 35/40 have a wide distribution. Although overall ESA distribution is quite different from ESC, a few ESA forms (117, 33/36, 34/37, 38–39, 40A–C, 40C) show significant similarities with ESC. ESB: almost all ESB forms are very similar to just a few ESA forms (35/40, 40A–C, 65) and to just a few ESC (M-SN33a, M-SN33d, M-SSa27a, M-SSa27a–c); few are very similar to ESD. ESC becomes even less internally homogeneous. It remains dissimilar to ESD. It becomes slightly more similar to ESA (through EAAL6, EAAL9A, M-SB9, M-SSa27a–c, M-ST8a) and ESB (through M-SN11a, M-SN33a, M-SN33d, M-SSa27a, MSSa27a–c, M-ST24b). ESD: almost all ESD forms (with the exception of p10/p11, p11/p12, p19, which are very dissimilar to ESD and show way more similarity to ESB) have strong similarities with ESA but less so with ESB (only p28, which has a wide distribution on sites with diverse assemblages like Jerusalem, Athens, Corinth) and are extremely dissimilar to ESC.

ad 100–125 ESA becomes slightly less internally homogeneous. It remains similar to ESB (through 35/ 40, 40A–C, 40C, 58/60, 65 (mostly similar to the previous period)) and ESD (through 40A–C, 40C, 51, 52, 53, 54, 62 (mostly similar to the previous period)), and shows slight similarity to some ESC forms, but not much more than in the previous period. ESB: similarity of most ESB with just a few ESA (35/40, 40A–C, 58/60, 65 (similar to previous period)) and ESC forms. Some ESB (62b, 74a, 77) are particularly similar to ESA. Like in the previous period still quite dissimilar to ESD. ESC (LRP2, LRP3, M-SN33d, M-SSa27a, M-SSa27a–c) is very similar to ESB, but less so to ESA (only ESA 40A–C, 40C are similar, present at Athens and Gortyn). It is very dissimilar to ESD. ESD is very dissimilar to ESC and rather dissimilar to ESB. Most ESD (with the exception of p10/p11, p11/p12, p30a–b, found on sites with extremely limited assemblages) is similar to a few ESA forms (40C, 51, 52, 62).

ad 125–150 ESA is less internally homogeneous. It becomes less similar to ESB and ESD and is completely dissimilar to ESC. Only one ESB form (77, widely distributed but highest proportion in Athens) is significantly similar to one ESA form (58/60, only found in Athens). Many ESD forms are similar to ESA 52 and 54. ESB becomes internally very homogeneous. As in the previous period, many ESB forms are similar to ESC forms (LRP1, M-SN33d, M-SSa27a, M-SSa27a–c). As in the previous periods, ESB is not extremely similar to ESD.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:29:22, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.009

LESSONS LEA RNED F ROM NETWORK SCIENCE TECHNIQUES

217

ESC becomes less internally homogeneous. It is very similar to ESB, but not at all to ESD and ESA. ESD is dissimilar to ESB and ESC but most forms are very similar to two ESA forms (52, 54). As in the previous period, with the exception of p10/p11, p11/p12, p30a–b, found on sites with extremely limited assemblages.

REFERENCES

Abadie-Reynal, C. 1989. Céramique et commerce dans le bassin Egéen du IVe au VIIe siècle. In V. Kravari, J. Lefort, and C. Morrisson (eds), Hommes et richesses dans l’empire byzantin. Tome I, IVe–VIIe siècle, 143–159. Paris: P. Lethielleux. Bes, P. 2015. The Distribution of Terra Sigillata and Red Slip Ware in the Roman East: A Chronological and Geographical Study. BAR International Series. Oxford: Archaeopress. Brainerd, G.W. 1951. The place of chronological ordering in archaeological analysis. American Antiquity 16(4), 301–313. Brandes, U., Robins, G., McCranie, A., and Wasserman, S. 2013. What is network science? Network Science 1(1), 1–15. doi:10.1017/nws.2013.2. Brughmans, T., Collar, A., and Coward, F. (eds) 2016. The Connected Past: Challenges to Network Studies of the Past. Oxford: Oxford University Press. Brughmans, T., and Poblome, J. 2016. Roman bazaar or market economy? Explaining tableware distributions in the Roman East through computational modelling. Antiquity 90(350), 393–408. Collar, A., Coward, F., Brughmans, T., and Mills, B.J. 2015. Networks in archaeology: phenomena, abstraction, representation. Journal of Archaeological Method and Theory 22, 1–32. doi:10.1007/s10816-014–9235-6. Cowgill, G.L. 1990. Why Pearson’s R is not a good similarity coefficient for comparing collections. American Antiquity 55(3), 512–521. Doran, J.E., and Hodson, F.R. 1975. Mathematics and Computers in Archaeology. Cambridge, MA: Harvard University Press. Doreian, P. 1988. Using multiple network analytic tools for a single social network. Social Networks 10(4), 287–312. doi:10.1016/0378–8733(88)90001–9. Fentress, E., and Perkins, P. 1988. Counting African red slip ware. In A. Mastino (ed.), L’Africa Romana: Atti del V Convegno di studio Sassari, 11–13 dicembre 1987, 205–214. Sassari: Universitá degli Sudi di Sasari, Dipartimento di Storia. Golitko, M., Meierhoff, J., Feinman, G.M., and Williams, P.R. 2012. Complexities of collapse: the evidence of Maya obsidian as revealed by social network graphical analysis. Antiquity 86, 507–523. Hayes, J.W. 1972. Late Roman Pottery. London: British School at Rome. Hayes, J.W. 1985. Sigillate orientale. In Enciclopedia dell’arte antica. classica e orientale. Atlante della forme ceramiche II: ceramica fine romana nel bacino mediterraneo (tardo ellenismo e primo imperio), 1–96. Rome: Istituto della Enciclopedia italiana. Lewit, T. 2011. Dynamics of fineware production and trade: the puzzle of supra-regional exporters. Journal of Roman Archaeology 24, 313–332. Meyer-Schlichtmann, C. 1988. Die pergamenischen Sigillata aus der Stadtgrabung von Pergamon. Mitte 2. JH v. Chr.–Mitte 2. JH n. Chr. Berlin: de Gruyter. Östborn, P., and Gerding, H. 2014. Network analysis of archaeological data: a systematic approach. Journal of Archaeological Science 46, 75–88.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:29:22, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.009

218

TOM BR UG HMAN S

Peeples, M.A. 2011a. Identity and social transformation in the Prehispanic Cibola world: A.D. 1150–1325. PhD dissertation, Arizona State University. Peeples, M.A. 2011b. R script for calculating the Brainerd–Robinson coefficient of similarity and assessing sampling error. www.mattpeeples.net/BR.html. (accessed March 28, 2015) Reynolds, P. 1995. Trade in the Western Mediterranean, AD 400–700: The Ceramic Evidence. BAR International Series 604. Oxford: Archaeopress. Robinson, W.S. 1951. A method for chronologically ordering archaeological deposits. American Antiquity 16(4), 293–301. Shennan, S. 1997. Quantifying Archaeology. 2nd edn. Edinburgh: Edinburgh University Press. Tukey, J.W. 1977. Exploratory Data Analysis. Reading: Addison-Wesley. Willet, R. 2012. Red slipped complexity: the socio-cultural context of the concept and use of tableware in the Roman East (second century BC–seventh century AD). Unpublished PhD thesis, Katholieke Universiteit Leuven.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:29:22, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.009

CHAPTER NINE

AMPHORAS, NETWORKS, AND BYZANTINE MARITIME TRADE Paul Arthur, Marco Leo Imperiale, and Giuseppe Muci

U

ntil not that many years ago archaeological research on the Byzantine economy after the 6th and 7th centuries was almost nonexistent. Now, a greater sensitivity for a period that had traditionally been seen as one of unremitting decline has led to an increase in archaeological attention, and the resultant identification of indicators of trade and consumption, from sites to artifacts. These new data, furthermore, have led to ever more nuanced considerations about the nature and scale of exchange in the early medieval Mediterranean by historians and archaeologists. One need only consider the magisterial study by Michael McCormick that analyses the abundance of European and Mediterranean exchange through an impressive analysis of the various types of source material and items of trade, often previously underestimated, such as holy relics and slaves. His networks clearly show that Mediterranean connectivity in the 8th and 9th centuries had by no means ceased, although movement and traffic had certainly lessened since late antiquity. Peregrine Horden and Nicholas Purcell (2000, 160–172) come to the same conclusion despite their pessimism about identifying archaeological remains of the 7th and 8th centuries. They, as others, cite the oft-quoted voyage of the pilgrim St. Willibald of Wessex, as a clear, though unique, example of movement along the early 8th-century Mediterranean seaways (Noble and Head 1995). We are still a long way from the superabundant archaeology of more than 120 major published assemblages upon which Søren Sindbæk was able to base 219 Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:27:17, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.010

220

PAUL ARTHUR, MARCO LEO IMPERIALE, AND G IU SEPPE MUCI

his inspired studies on interaction in the Baltic and the North Sea. Nonetheless, the growing archaeological interest in the early medieval Mediterranean is detecting a series of sites and artifact types that may be used to illustrate shifting networks of interaction and exchange, suggesting optimism about future possibilities. At present, though until now rarely presented in their relative quantities, we can rely ever more on the increasing publication of ceramics, particularly transport amphoras, soapstone vessels, glass, coins, volcanic lava quern stones, and certain items of metalwork. It is worth noting the similarity in the range of artifacts that were objects of exchange both in the North Sea and in the Mediterranean, including soapstone vessels, lava querns, and ceramic transport containers. Much of the most recent archaeological work on early medieval trade in the Mediterranean and beyond has been presented in the book edited by Sauro Gelichi and Richard Hodges (2012). Both curators are also responsible for major excavations of two important early medieval trading sites. At Comacchio, on the Adriatic coast between Ravenna and Venice, Gelichi has brought to light remains of what might plausibly be identified as an early medieval emporium, mediating exchange between the Po valley and the greater Mediterranean and, more specifically, between the Lombards and the Byzantine world (Gelichi, Calaon, and Grandi 2012). At Butrint, on the southern coast of Albania, Hodges has unearthed a Byzantine fort perched on the edge of Slav territory, which likely also performed the functions of an emporium as well as a station along the Adriatic and Aegean seaway (Hodges, Bowden, and Lako 2004). Both of these locations are characterized by the presence of Byzantine globular amphoras, which appear to be typical of many Mediterranean sites in and around the 8th century. Indeed, the transport amphora is one of the most explicit archaeological items to indicate exchange in antiquity and the early Middle Ages, as it was employed to ship various foodstuffs, though principally wine and olive oil. During the Roman period ceramic amphoras were produced throughout the empire, from the Levant to Brockley Hill, north of London, from Mauretania to the northern coast of the Black Sea. The variety of forms is stunning: apart from differences between containers that could signal diverse contents, many were specific to geographical areas, though these were sometimes imitated elsewhere. Indeed, in some cases amphora specialists are able to hypothesize provenance on form alone, while in others cases, petrological analysis is required. An idea of the variety in form is shown, for instance, by Dominique Pieri’s illustration of the principal late Roman amphora types of the eastern Mediterranean.1 This remarkable number of individual forms gradually disappeared toward the second half of the 7th century to make way for the monopoly of what was basically a single shape, that of the so-called globular amphora, with a carrying

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:27:17, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.010

A MP H O R A S , N E T WOR KS , A ND BYZ A NT I NE MA R I TI M E T R A DE

221

9.1. Major eastern Roman and Byzantine amphora forms between the 3rd and 13th centuries. A dominance of globular forms appears evident between the 6th and 9th centuries.

capacity of about twenty-five to thirty liters (Figure 9.1).2 Only lately has it become an important object of analysis in attempts to chart early medieval Mediterranean trade. Enrico Zanini (2010) has stressed the advantages of these amphoras, particularly in the changing conditions of the early Middle Ages. Even full, they could be handled by a single person, and were adapted to being carried by differing means, whether on the small ships engaged in coastal cabotage, on carts or barges for inland and river transport, or indeed by pack animals. Although they appear to have been substantially wine containers, the possibility that some contained olive oil, or even fish sauces and other commodities, cannot be excluded. During the 8th century they became the standard amphoras throughout the Byzantine Empire as well as in erstwhile Byzantine territories in Italy, in part dependent on Papal Rome (Arthur 1993; De Rossi 2005). Sites and areas of globular amphora manufacture have been recognized in the Bay of Naples, at Otranto in southern Apulia, in eastern Sicily, in Tunisia, on Crete, in the Aegean (possibly on Kos, near Ephesus, and elsewhere), and at Chersonesos in the Crimea; the vessels are frequent elsewhere, too, including Rome, Ravenna/Classe, and Constantinople.3 Though various types of the globular amphora exist, they may largely have been the result of the idiosyncrasies of individual potters and workshops, rather than any attempt at manufacturing differing vessel types.4 Their known distribution very evidently coincides with the limits of Byzantine territory in the first half of the 8th century, with some outliers that may represent preferential extraterritorial exchange. We will suggest explanations for the presence of the few examples known outside the empire later,

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:27:17, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.010

222

PAUL ARTHUR, MARCO LEO IMPERIALE, AND G IU SEPPE MUCI

apart from noting that although Carthage appears on the map, perhaps as a center of manufacture, it drops out when overrun by the Arabs around AD 700. A large number appear in and around the Crimean peninsula, including the Sea of Azov and the Don area.5 One may suspect that their large number in that area reflects not only local production in the Crimea, but also the impressive amount of publication of archaeological contexts. In order to examine the distribution of globular amphoras within the structure of Byzantine exchange in and around the 8th century, we have constructed a network model based on the known distribution of a number of roughly contemporaneous Byzantine artifact types across the Mediterranean and the degree of affiliation of the sites in which these artifacts have been recognized. We have chosen objects from different production areas, with supra-regional distributions, that should represent largely seaborne multidirectional exchange. As Sindbæk (2013, 71) notes, “the fragmentary archaeological evidence presents researchers with the task of reconstructing the broken links of a ruined network from observable distributions and patterns of association in the archaeological record.” This fragmentation, particularly acute for the early Middle Ages, makes particularly difficult the study of commercial dynamics through the creation of a “traditional” network, which explores the relationships between individuals or actors of the same category (e.g., ports and harbors). Given the current state of knowledge, it is impossible to formulate entire reconstructions of commercial networks in terms of directional market flows from the site of production to the most peripheral sites of occurrence for a specific item. Instead, examination of shared material-culture attributes can lead to the definition of links between nodes, even though we are not yet in a position to add absolute and relative quantities of these artifacts to our analysis because of the state of publication. A useful tool to explore this type of dataset can be found in the technique of “affiliation networks”: affiliation or bimodal networks are two-mode networks that allow one to study the dual perspectives of the actors (e.g., sites) and the events (e.g., occurrences), unlike one-mode networks, which focus on only one item at a time. In other words, they look at collections or subsets of actors rather than at directed ties between pairs of actors. Connections between actors of one of the modes are built on links established through the second mode: using this kind of graph one can delineate the relationships between actors and events and visualize how events create ties among actors, and vice versa. The use of affiliation networks allows us to connect data that could not otherwise be clearly evidenced in a single system or through more traditional methods, and thereby facilitates our role in historical interpretation. Thus, in the words of Sindbæk (2013, 82), we have selected objects that seem to be particularly indicative for the identification of “currents in a pool of

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:27:17, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.010

A MP H O R A S , N E T WOR KS , A ND BYZ A NT I NE MA R I TI M E T R A DE

223

9.2. Bipartite network graph of selected Byzantine artifacts and their site associations.

overlapping material” in so far as they are identifiable to areas of production and consumption. The first graph that we illustrate (Figure 9.2), produced using the opensource software Gephi,6 shows the bipartite model obtained through the association of artifacts and sites. In our attempt at building the affiliation network we started collecting distribution data on the occurrence of artifacts on different sites. The data were encoded into two simple spreadsheets: one indexes all nodes, which in our case are both sites and artifacts, the second one lists edges, that is to say the connections between objects and sites based on their reciprocal occurrence. Using these input datasets we were able to produce a rectangular bimodal matrix composed of n lines by m columns, where n is the number of actors and m that of the events to which they are linked; the

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:27:17, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.010

224

PAUL ARTHUR, MARCO LEO IMPERIALE, AND G IU SEPPE MUCI

information on the presence/absence relationships form the elements of this matrix. In the graphical representation of this single undirected network each vertex can represent either a site or a find, while edges indicate the occurrence of an artifact on a site. Obviously sites can only be connected to artifacts, and artifacts can only be connected with sites where they have been documented (Brughmans 2010). Depending on their relative proximity, artifacts and sites tend to cluster, with sites displaying more complete assemblages grouping in the center of the graph, while sites having one or a few links to artifacts clearly move toward the borders of the graph. It is important to stress that the marginality of these nodes is directly related to the main circulation area of the object of our analysis—the Byzantine territories around the Mediterranean—while these sites could have played more central roles in other commercial circuits. This bipartite network can be split into two distinct single-mode networks (sites only/artifacts only) based on relations connecting nodes of two different types.7 Figure 9.3 shows the previous dataset rearranged as a unimodal network in which sites are connected by edges expressing affiliations based on the cooccurrence of artifacts; the thickness and tone of the edges are a representation of the strength of affiliation.8 The coexistence of certain artifacts on the same sites may be read as an indication of the similarity or dissimilarity of these artifacts’ distribution networks. The algorithm applied in drawing the graph rests on a number of basic rules, the most important of which are that the distance between vertices expresses the strength or number of their ties, and that vertices that are related are closer than vertices that are not related (Brughmans 2010). This first rough approximation provides a preliminary view of the topology and structure of this complex system and allows us to determine the relative centrality of those sites that can be viewed as nodal points within the network (Preiser-Kapeller 2013). In order to make the interpretation of the network clearer and to focus on what is probably a more robust exchange network and range of affiliations, the data representation has been remodeled by excluding those nodes whose affiliation is based on the presence of no more than a single shared attribute (Figure 9.4). As Tom Brughmans (2010, 288) states, “one of the most enlightening ways of exploring complex networks is through visual inspection”: what clearly emerges from our network model is a heavy clustering of sites connected by strong affiliations. The higher the number of common artifacts, the stronger the attractions connecting two nodes. The dense core falls within the area under Byzantine political control around the beginning of the 8th century. Asia Minor remains largely uninvolved in the network, almost certainly because of a dearth of published information, with the evident exception of Ephesus, seat of a major trade fair, that was assessed a lump duty of a hundred gold pounds in 787 (Foss 1979). Much of the Black Sea is also largely absent in the network, perhaps because of a similar lack of evidence, although the

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:27:17, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.010

9.3. Affiliation network of sites based on selected 8th-century artifacts.

225 Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:27:17, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.010

226

PAUL ARTHUR, MARCO LEO IMPERIALE, AND G IU SEPPE MUCI

9.4. Network displaying only sites connected by strong affiliations (sites within Byzantine territories are circled).

important and well-published trading town of Chersonesos stands out for its affiliation with Constantinople and sites as far away as Italy and Sicily. Instead, the available information for the Aegean, and even more for Italy, leads to a fairly substantial number of nodes and connections (Figure 9.5). Unfortunately, having to deal with weighted affiliation graphs means that many statistical measurements usually adopted by social network analysis, such as betweeness and closeness centrality, may be misleading. On the other hand, valued data stored in the edges can be used to compute eigenvector centrality (Bonacich 1972; Borgatti and Halgin 2011). In this case it can be read as an index of the relative importance of a node within the network: this value is characterized with different shades on the map, with the darker nodes with higher values frequently corresponding to main sites or hypothetical hubs. The next map (Figure 9.6) shows the distribution of all globular amphoras in the affiliation network. As with the preceding illustration, this highlights the coincidence of the network with the extension of Byzantine territory in the first half of the 8th century. In Italy, the higher degrees of affiliation link Rome,

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:27:17, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.010

A MP H O R A S , N E T WOR KS , A ND BYZ A NT I NE MA R I TI M E T R A DE

227

9.5. Geographic projection of the early medieval sites’ affiliation network (Byzantine territories around the 8th century are hatched).

Naples, and Sicily on the Tyrrhenian coast, and Otranto and Ravenna/Classe and Comacchio on the Adriatic coast. In part, this coincides with recent scholarship suggesting that “[a]n economic picture is starting to emerge for the late Byzantine period, which involves two major regions of exchanges: a Tyrrhenian region which is dominated by products from Campania and Latium [to which we would add Sicily], and an Adriatic region which is more dependent on supplies from the Eastern Mediterranean” (Bruno and Cutajar 2013, 26, n. 25; Vaccaro 2013). The illustration also shows the location (marked by stars) of kiln sites producing globular amphoras across the empire, from the Black Sea, through the Aegean, to Italy. It is quite striking that in a period of economic recession and increasing self-sufficiency of the disparate areas of the Mediterranean and the Black Sea, amphora forms, instead of becoming increasingly individual, gravitate ever more toward the common globular type. This might be explained by imitation of a successful model or, one might say, the spread of an idea or ideal of container type. Instead of illustrating amphora diffusion from one or more production centers, which will be possible once we know more about individual clay fabrics, the network analysis of the globular amphora illustrates the transmission or flow of a unitary idea of form or of a small range of archetypes. It underlines

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:27:17, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.010

228

PAUL ARTHUR, MARCO LEO IMPERIALE, AND G IU SEPPE MUCI

9.6. Globular amphoras distribution within the network (Byzantine territories around the 8th century are hatched).

contact between disparate and distant areas of the greater Byzantine world despite the economic and political difficulties of the time. However, rather than being empirical, with similarity based on simple imitation, we suggest that the production of such a standard model was the result of trade regulation, perhaps by Byzantine kommerkiarioi. The recent work by Charikleia Diamanti and Andrei Opaiţ on official stamps present on globular amphoras of type LRA 13 et similis, of 7th-century date, suggests attempts at imperial control of production and distribution (Diamanti 2010; Opaiţ and Diamanti 2014). The amphora stamps replicate coin iconography, but with the indication of the eparch, the urban prefect of Constantinople, who supervised state distribution, production, and weights and measures. Furthermore, Van Alfen (2015) has noted the metrological standardization, based on a system of volume–weight capacity, within the early 7th-century Yassıada shipwreck’s cargo of these new jars.9 The later piriform amphoras from the 11th-century Serçe Limanı shipwreck appear to break down into vessels that followed interrelated weight capacity systems based on three or five litrai (Van Doorninck 1993; 1995; 2015). Wine may have been included in the annona or state-supplied food at Constantinople in the 7th century, and state-regulated supply to key areas,

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:27:17, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.010

A MP H O R A S , N E T WOR KS , A ND BYZ A NT I NE MA R I TI M E T R A DE

229

particularly for the army, almost certainly continued after that date (Mundell Mango 2000, 190). In such a context, the attention of the Byzantine state to weights and measures would not be at all surprising. Indeed, it has been suggested that the early medieval globular amphora was that known in the texts as amphora megarica, which clearly relates to the term megarikon, a Byzantine unit of measurement (Bakirtzis 1989, 73; Villa 1994, 414).10 Alongside gold solidi, some 100 amphora megaricae were cited as annual tax payment of the monastery of St. Stephen on the island of Capri to Theodorus (or Theodosius), Duke of Naples (719–730), when the city was still under Byzantine control. It must have referred to globular amphoras, the only ones circulating in any quantity in the Bay of Naples at the time. The fact that the amount due to be paid was calculated in the number of vessels suggests that each and every vessel could be envisaged as representing and containing a fixed quantity of wine, perhaps the approximately twenty-six liters standard for Roman amphora measure, which closely fits the roughly twenty-five- to thirty-liter carrying capacity of the globular amphoras (Parkin and Pomeroy 2007, 359). The amphoras from Capri were thus explicitly used for tax payments, and standardization of the form and carrying capacity would only have aided this and other similar fiscal dues in wine. Although it is still too early to clearly chart the distribution of globular amphoras to see how far they may reflect imposed movement of goods (dirigisme), it seems highly likely that the production at Misenum by the sea was sited on papal property within the Duchy of Naples, and that the amphoras largely served papal requirements of wine in Rome and elsewhere within a distributive network of Byzantine inception. Indeed, excavations at the Crypta Balbi site in Rome have shown how the range of amphoras—and ceramics in general—changed dramatically at the beginning of the 8th century, from Mediterranean-wide importation of various amphora types to a regional network of mainly globular vessels substantially limited to Rome, the Bay of Naples, Calabria, Sicily, and Sardinia (Saguì, Ricci, and Romei 1997, 39–44). Interestingly, there is yet little indication of globular amphoras having penetrated territories outside the empire or its closely allied states after the 7th century. Some examples are known from Lombard Campania, mainly from Benevento, which had strong commercial contacts with Byzantine and, later, an autonomous though still Byzantinophile, Naples. Similarly, they appear at Lombard sites along the Po valley, likely through the mediation of Byzantine Comacchio. A few were distributed in Khazar territory in southern Ukraine, likely through Byzantine emporia such as Chersonesos. Furthermore, early 8th-century globular amphoras also appear to have been present in Ummayad Beirut (Reynolds 2003). The relative rarity of globular amphoras outside Byzantine territory, if not reflecting the state of research, might be due to their role within a specifically

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:27:17, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.010

230

PAUL ARTHUR, MARCO LEO IMPERIALE, AND G IU SEPPE MUCI

Byzantine-controlled exchange system. Exchange between Byzantine and Lombard areas appears in written trade agreements, such as the pact stipulated in 715 between Byzantium and the Lombards at Comacchio, or the early 9thcentury Pactio Sicardi, which regulated trade between the Byzantinophile Duchy of Naples and Lombard Benevento (Hartmann 1904; Bluhme 1868). Sites such as Naples, Comacchio, Otranto, Butrint, and Chersonesos must have played a role as hubs or communication points in interlocking exchange networks of distinct political and cultural entities. As Raymond Lopez (1959, 73) once wrote, Byzantium “channelled and supervised all external trade through special international fairs held periodically at a small number of checkpoints just inside the border. No alien merchant was allowed to trespass beyond the fairs, or to trade except in the presence of a representative of the central government.” The role of such sites as nodal points or hubs needs further investigation, which may include their position as centers of craft production, their social relationships, and in some cases their privileged religious status and function as early medieval mints. Naples agrees on all counts, while Comacchio certainly was a bishopric and center of craft production. Intriguingly, it appears to have produced glassware, an example of which has been found in the Lombard center of Cividale del Friuli. Butrint and Chersonesos are other commercial gateways, in whose foreign hinterlands we might in the future expect to appear globular amphoras and other artifacts that we have examined. Contrasting with the development of market sites around the North Sea and the Baltic, our evidence seems to stress the role of a central power, Byzantium, in the establishment of nodes, apparently resulting in a “more connected world” compared with early Viking Age Scandinavia of the 8th to 9th centuries (Sindbæk 2007, 129). The globular amphora seems largely to have disappeared during the course of the 9th century, at a time of economic development throughout Byzantium. It was perhaps in the second half of the century that a whole series of new amphora forms appeared on the market, from Asia Minor and the Black Sea to Greece and Italy. Whether or not some of them came from the same workshops as the previous globular vessels is hard to say, but it is striking that essentially one recognizable form, the globular amphora, gave way to many differing types. As in earlier Roman times, a period of more complex economic production and exchange was characterized by a plethora of commercial container types whose forms differed substantially from one area to another. In the lower Adriatic, where a later 7th- or 8th-century kiln site for globular amphoras is known at Otranto, the dominating form by the second half of the 9th century was a ribbed vessel with peaked handles and a narrow neck (Arthur 1992; Arthur and Auriemma 1996). It could not differ more from the earlier globular containers. Its distribution appears to have concentrated in and around the Adriatic, with the majority of vessels being known at Otranto and its

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:27:17, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.010

A MP H O R A S , N E T WOR KS , A ND BYZ A NT I NE MA R I TI M E T R A DE

231

9.7. Affiliation network of sites based on 10th- to 11th-century selected artifacts.

hinterland—perhaps one of the main areas of manufacture—and on a number of sites along the other shore of the Adriatic like Corinth, Durres, and Butrint (Reynolds 2004), and the underwater contexts investigated by the Liburna project (Volpe, Turchiano, and Leone 2011; Leo Imperiale 2014, 330–333). It seems to have reached sites in the northern Adriatic, such as Torcello and Ferrara, and strongly supports the vision of “intra-Adriatic port hopping” (Dorin 2012). The entire group, in sum, might be considered a western Byzantine amphora, with a general morphology that was transmitted through interlocking trade networks, one sited in the Adriatic, and another perhaps between the Ionian Sea and Sicily. Nonetheless, the group’s distribution overlaps with the distribution of other different amphora forms, often coming from production sites in Greece and Asia Minor, such as the Günsenin I and II types (Günsenin 1990). The affiliation network appears less eloquent than that of earlier Byzantine times also because of the current paucity of data, though we intend to add to the number of diagnostic materials represented in the network in the near future (Figure 9.7). Nonetheless, two distinct clusters can be distinguished in the

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:27:17, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.010

232

PAUL ARTHUR, MARCO LEO IMPERIALE, AND G IU SEPPE MUCI

graph, which appear to coincide with a more Adriatic circuit, projecting east toward Corinth and Athens, and an area that gravitates around Constantinople, even extending into the Russian hinterland and as far away as Scandinavia largely through fluvial routes. Tyrrhenian coastal centers seem only marginally touched by these networks of Byzantine artifacts, which will change during the course of the 12th century. As exchange and contacts increased during the course of the 9th and 10th centuries, amphora forms diversified and diverged from a common type, as if greater political unity and stability allowed greater independence in manufacturing decisions. John Haldon wrote of “the ability of the state in the 7th and 8th centuries to implement a relatively full control over its tax-base,” perhaps because of its need, particularly in such times of political and economic difficulty, to be more attentive over the fiscal balance; he continues, “the government in the extraction and redistribution of resources is especially clear in terms of taxation” and, thus, of the administrative infrastructure (Brubaker and Haldon 2011, 465). Perhaps we are seeing the difference between a more controlled system of dirigisme in place during the centuries of severe economic and financial crisis, followed by a certain measure of deregulation and enterprise as the role of market forces in economic integration was in the ascendant, and exchangeable surplus was increasing (Laiou and Morrisson 2007, 89). This may well have had an impact upon container production, form, capacity, and distribution (with its relative contents), with variety being a result of reduced trade restrictions and an increase in independent suppliers. Amphoras (and wooden barrels) from the 10th century seem perhaps more suited to large-scale maritime shipping, and appear in a variety of new and diverse forms, in the context of an increasingly complex system of interlocking trade networks. NOTES 1.

2. 3.

4.

5.

A typology by Simon Keay (1984) of late Roman amphoras found in Spain documented ninety-three types, many of which could be subdivided into varieties. Further types have since joined the list. Few other amphora forms appear to have existed during the 8th and much of the 9th centuries, as shown by the evidence at Istanbul: see Hayes (1992). Bay of Naples, at Misenum and probably Ischia: Arthur 1993; De Rossi 2005; Otranto: Leo Imperiale 2004; Crete: Yangaki 2005; Kos: Didioumi 2014; Lipsi: Papavassiliou, Sarantidis, K., and Papanikolaou 2014; Pesochnoiy Bay, Sevastopol: Borisova 1969; Quarantine Bay, Sevastopol: Zolotarev 1982; Rome: Romei 2001; Ravenna/Classe: Cirelli 2009; Yenikapı, Istanbul: Kocabaş 2008. Vroom (2012), for instance, has divided the globular vessels at Butrint into three groups, though they cannot be defined as different amphora types, but variants. We think that the same may be said for De Rossi’s subdivision of the amphora from Misenum. Garver (1993), class 4, where other examples are also cited; for the Crimea, for instance, see Parshina, Teslenko, and Zelenko (2001).

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:27:17, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.010

A MP H O R A S , N E T WOR KS , A ND BYZ A NT I NE MA R I TI M E T R A DE

6.

7.

8.

9. 10.

233

Gephi was mainly used for visualization and measurement on the networks. Other software included Sci2, for calculations on matrixes and networks, and ArcGIS 10, especially useful for spatial data extrapolation and networks’ georeferenced visualization. We performed this operation in Sci2, using the exported rectangular matrix from Gephi (File > export > graph file, save as *.net file (Pajek)). To execute an affiliation we used the function “Extract Reference Co-occurence (Bibliographic Coupling) Network” in the “Data Preparation” menu. The resulting single-mode network was then cleaned up to remove unnecessary nodes by selecting Preprocessing > Networks > Delete Isolates. We used the ForceAtlas representation algorithm implemented in Gephi, especially useful to spatialize small-world/scale-free networks and focus on the qualitative aspect of data, meaning “being useful to explore real data.” It is based on a linear–linear mode (attraction and repulsion proportional to distance between nodes), so that the rendering of the datasets highlights relations such as clustering between nodes. See also van Alfen (1996). For the role of the state in the Byzantine economy see Oikonomides (2007). De Rossi (2005) suggests that the term, deriving from Megara in Attica, as home to Byzas, mythical founder of the city of Byzantium, by extension provided a time-honoured assurance of measure.

REFERENCES

Ahrweiler, H. 1978. Les ports byzantins (VIIe–IXe siècles). In La navigazione mediterranea nell’alto medioevo. Atti XXV settimana di studio del centro italiano di studi sull’alto medioevo, 259–283. Spoleto: Presso La sede del Centro. Ardizzone Lo Bue, F. 2012. Anfore in Sicilia (VII–XII sec. d.C.). Palermo: Torri del Vento. Arthur, P. 1992, Amphorae for bulk transport. In F. D’Andria and D.B. Whitehouse (eds), Excavations at Otranto, Vol. 2, 197–217. Galatina: Congedo. Arthur, P. 1993. Early medieval amphorae, the Duchy of Naples and the food supply of Rome. Papers of the British School at Rome 61, 231–244. Arthur, P., and Auriemma, R. 1996. A search for Italian wine: Middle Byzantine and later amphorae from southern Puglia. INA Quarterly 23(4), 14–17. Bakirtzis, c. 1989. Byzantine amphorae. In V. Derochè and J.-M. Spieser (eds), Recherches sur la céramique byzantine, 73–77. Bulletin de correspondance hellénique Suppl. 18. Bluhme F. (ed.) 1868. Pactio Sicardi, MGH, Leges, IV. Hannover: Hahn. Bonacich, P. 1972. Factoring and weighting approaches to status scores and clique identification. Journal of Mathematical Sociology 2, 112–120. Borgatti, S.P., and Halgin, D.S. 2011. Analyzing affiliation networks. In J. Scott and P.J. Carrington (eds), The Sage Handbook of Social Network Analysis, 417–433. London: Sage. Borisova V.V. 1969. Srednevekovaya goncharnaya pech. Soobshcheniya Hersonesskogo Muzeya, 42–46. Simferopol. Brubaker, L., and Haldon, J.F. 2011. Byzantium in the Iconoclast Era, c.680–850: A History. Cambridge: Cambridge University Press. Brughmans, T. 2010. Connecting the dots: towards archaeological network analysis. Oxford Journal of Archaeology 29(3), 277–303. Bruno, B., and Cutajar, N. 2013. Imported amphorae as indicators of economic activity in early medieval Malta. In D. Michaelides, P. Pergola and E. Zanini (eds), The Insular System of the Early Byzantine Mediterranean. Archaeology and History, 15–29. British Archaeological Reports 2523. Oxford: Archaeopress.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:27:17, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.010

234

PAUL ARTHUR, MARCO LEO IMPERIALE, AND G IU SEPPE MUCI

Cirelli, E. 2009. Anfore globulari a Classe nell’alto Medioevo. In G. Volpe and P. Favia (eds), Atti del V Congresso di Archeologia Medievale (Foggia-Manfredonia, 30 settembre–3 ottobre 2009), Vol. 1, 563–568. Florence: All’Insegna del Giglio. Curta, F. 2011. The Edinburgh History of the Greeks, c.500–1050: The Early Middle Ages. Edinburgh: Edinburgh University Press. De Rossi, G. 2005. Indicatori archeologici della produzione e diffusione del vino della Baia di Napoli in età altomedievale. In G. Volpe and M. Turchiano (eds), Paesaggi e insediamenti rurali in Italia meridionale fra tardoantico e Altomedioevo, Atti del Primo Seminario sul Tardoantico e l’Altomedioevo in Italia meridionale (Foggia 12–14 febbraio 2004), 541–549. Bari: Edipuglia. Diamanti, C. 2010. Stamped late Roman/proto-Byzantine amphoras from Halasarna of Kos. Rei Cretariæ Romanæ Fautorum Acta 41, 1–8. Didioumi, S. 2014. Local pottery production in the island of Cos, Greece from the early Byzantine period: a preliminary report. In N. Poulou-Papadimitriou, E. Nodarou, and V. Kilikoglou (eds), LRCW4. Late Roman Coarse Wares, Cooking Wares and Amphorae in the Mediterranean: Archaeology and Archaeometry, 169–180. British Archaeological Reports Series 2616, Oxford: Archaeopress. Dorin, R.W. 2012. Adriatic trade networks in the twelfth and early thirteenth centuries. In C. Morrisson (ed.), Trade and Markets in Byzantium, 235–279. Washington, DC: Dumbarton Oaks Research Library and Collection. Foss, C. 1979. Ephesus after Antiquity, Cambridge: Cambridge University Press. Garver, E.L. 1993. Byzantine amphoras of the ninth through thirteenth centuries in the Bodrum Museum of Underwater Archaeology. Unpublished MA thesis, Texas A & M University. Gelichi, S., Calaon, D., and Grandi, E. 2012. History of a forgotten town: Comacchio and its archaeology. In S. Gelichi and R. Hodges (eds), From One Sea to Another: Trading Places in the European and Mediterranean Early Middle Ages. Proceedings of the International Conference (Comacchio 27th–29th March 2009), 169–205. Turnhout: Brepols. Gelichi, S., and Hodges, R. (eds) From One Sea to Another: Trading Places in the European and Mediterranean Early Middle Ages. Proceedings of the International Conference (Comacchio 27th–29th March 2009). Turnhout: Brepols. Günsenin, N. 1990, Les amphores Byzantines (Xe–XIIIe siècles): Typologie, production, circulation d’apres les collections Turques. Unpublished PhD dissertation, Université Paris I (Panthé on–Sorbonne). Hartmann, L.M. (ed.) 1904. Pactum. Zur Wirtschaftsgeschichte Italiens im frü hen Mittelalter. Gotha: Analekten. Hayes, J.W. 1992. Excavations at Saraçhane in Istanbul, Vol. 2: The Pottery. Princeton: Princeton University Press. Hodges, R. 2012. Adriatic sea trade in a European perspective. In S. Gelichi and R. Hodges (eds), From One Sea to Another: Trading Places in the European and Mediterranean Early Middle Ages. Proceedings of the International Conference (Comacchio 27th–29th March 2009), 207–234. Turnhout: Brepols. Hodges, R., Bowden, W., and Lako, K. (eds) 2004. Byzantine Butrint: Excavations and Surveys 1994–1999. Oxford: Oxbow. Horden, P., and Purcell, N. 2000. The Corrupting Sea: A Study of Mediterranean History Malden: Blackwell. Keay, S.J. 1984. Late Roman Amphorae in the Western Mediterranean, a Typology and Economic Study: The Catalan Evidence. British Archaeological Reports Series 196. Oxford: B.A.R.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:27:17, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.010

A MP H O R A S , N E T WOR KS , A ND BYZ A NT I NE MA R I TI M E T R A DE

235

Kocabaş , U. (ed.) 2008. The Old Ships of the New Gate/Yenikapı’nın Eski Gemileri. Istanbul: Ege Yayınları. Laiou, A.E., and Morrisson, C. 2007. The Byzantine Economy. Cambridge: Cambridge University Press. Leo Imperiale, M. 2004. Otranto, cantiere Mitello: un centro produttivo nel Mediterraneo bizantino. In S. Patitucci Uggeri (ed.), La Ceramica Altomedievale in Italia, 327–342. Quaderni di Archeologia Medievale VI. Florence: All’Insegna del Giglio. Leo Imperiale, M. 2014. Ceramiche e commerci nel Canale d’Otranto tra X e XI secolo. Riflessioni sulla cultura materiale bizantina tra Salento e Albania meridionale. In G. Tagliamonte (ed.), Ricerche archeologiche in Albania: Atti dell’incontro di studi (Cavallino-Lecce 29–30 aprile 2011), 327–342. Rome: Aracne. Lopez, R. 1959. The role of trade in the economic readjustment of Byzantium in the seventh century. Dumbarton Oaks Papers 13, 67–85. McCormick, M. 2001. Origins of the European Economy: Communications and Commerce, A.D. 300–900. Cambridge: Cambridge University Press. Mundell Mango, M. 2000. The commercial map of Constantinople. Dumbarton Oaks Papers 54, 189–207. Negrelli, C. 2007. Produzione, circolazione e consumo tra VI e IX secolo: del territorio del Padovetere a Comacchio. In F. Berti, M. Bollini, S. Gelichi and J. Ortalli (eds), Genti nel Delta da Spina a Comacchio. Uomini, territorio e culta dall’Antichità all’Alto Medioevo, 437–471. Ferrara: Corbo. Negrelli, C. 2012. Towards a definition of early medieval pottery markers: amphorae and other vessels in the Adriatic region between 7th and 8th c. In S. Gelichi and R. Hodges (eds), From One Sea to Another: Trading Places in the European and Mediterranean Early Middle Ages. Proceedings of the International Conference (Comacchio 27th–29th March 2009), 393–415. Turnhout: Brepols. Noble, T.F.X., and Head, T. (eds) 1995. Soldiers of Christ: Saints and Saints’ Lives from Late Antiquity and the Early Middle Ages. University Park: Pennsylvania State University Press. Oikonomides, N. 2007. The role of the Byzantine state in the economy. In A. Laiou (ed.), The Economic History of Byzantium from the Seventh through the Fifteenth Century, 973–1058. Washington, DC: Dumbarton Oaks Research Library and Collection. Opaiţ, A., and Diamanti, C. 2014. Imperial stamps on early Byzantine amphorae: the Athenian Agora examples. Rei Cretariæ Romanæ Fautorum Acta 43, 55–61. Papavassiliou, E., Sarantidis, K., and Papanikolaou, E. 2014. A ceramic workshop of the early Byzantine period on the island of Lipsi in the Dodecanese (Greece): a preliminary approach. In N. Poulou-Papadimitriou, E. Nodarou, and V. Kilikoglou (eds), LRCW4. Late Roman Coarse Wares, Cooking Wares and Amphorae in the Mediterranean: Archaeology and Archaeometry, 159–168. British Archaeological Reports Series 2616, Oxford: Archaeopress. Parkin, T.G., and Pomeroy, A.J. 2007. Roman Social History: A Sourcebook. London: Routledge. Parshina, E., Teslenko, I., and Zelenko, S. 2001. Tavrika pottery centers from VIII–X centuries A.D. In Sea Trade in North Black Sea Region, 52–81. Kiev: Stylos. Portale, E.C., and Romeo, I. 2001. Contenitori da trasporto. In A. Di Vita (ed), Gortina, V.3, 1, I-II. Lo scavo del Pretorio (1989–1995), 260–410. Padova: Bottega d’Erasmo. Preiser-Kapeller, J. 2013. Mapping maritime networks. Working paper, September 2013. Reynolds, P. 2003. Pottery and the economy in 8th century Beirut: an Umayyad assemblage from the Roman imperial baths (BEY 045). In C. Bakirtzis (ed.), VIIe Congrès

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:27:17, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.010

236

PAUL ARTHUR, MARCO LEO IMPERIALE, AND G IU SEPPE MUCI

International sur la Céramique Médiéval en Méditerranée (Thessaloniki, 11–16 Octobre 1999), 725–734. Athens: Caisse des recettes arché ologiques. Reynolds, P. 2004. The Medieval amphorae. In R. Hodges, W. Bowden, and K. Lako (eds), Byzantine Butrint: Excavations and Surveys 1994–1999, 270–277. Oxford: Oxbow. Romei, D. 2001. Anfore. In Roma dall’antichità al Medioevo. Archeologia e storia nel Museo Nazionale Romano, Crypta Balbi, 503–507. Milan: Electa. Saguì, L., Ricci, M., and Romei, D. 1997. Nuovi dati ceramologici per la storia economica di Roma tra VII e VIII secolo. In G.D. D’Archimbaud (ed.), La céramique médiévale en Méditerranée: Actes du 6e congrès, 35–48. Aix-en-Provence: Narrations Editions. Sindbæk, S. 2007. Networks and nodal points: the emergence of towns in early Viking Age Scandinavia. Antiquity 81, 119–132. Sindbæk, S. 2013. Broken links and black boxes: material affiliations and contextual network synthesis in the Viking world. In C. Knappett (ed.), Network Analysis in Archaeology: New Approaches to Regional Interaction, 71–94. Oxford: Oxford University Press. Vaccaro, E. 2013. Sicily in the eighth and ninth centuries A.D.: a case of persisting economic complexity? Al-Masaq: Islam and the Medieval Mediterranean 25(1), 34–69. Van Alfen, P.G. 1996. New light on the 7th-c. Yassi Ada shipwreck: capacities and standard sizes of LRA1 amphoras. Journal of Roman Archaeology 9, 189–213. Van Alfen, P.G. 2015. The restudy of the LR2a and LR2b (LR13) amphoras from the seventhcentury Yassıada shipwreck: preliminary evidence for standardization. In D.N. Carlson, J. Leidwanger, and S.M. Kampbell (eds), Maritime Studies in the Wake of the Byzantine Shipwreck at Yassıada, Turkey, 17–34. College Station: Texas A & M University Press. Van Doorninck, F.H., Jr. 1993. Giving good weight in eleventh-century Byzantium: the metrology of the Glass Wreck amphoras. INA Quarterly 22(2), 8–12. Van Doorninck, F.H., Jr. 1995. The piriform amphoras from the 11th-century shipwreck at Serçe Limanı: sophisticated containers for Byzantine commerce in wine. Graeco-Arabica 6, 181–189. Van Doorninck, F.H., Jr. 2015. The metrology of the pirifom amphoras from the eleventhcentury Byzantine ship at Serçe Limanı: new designs but an old system. In D.N. Carlson, J. Leidwanger, and S.M. Kampbell (eds), Maritime Studies in the Wake of the Byzantine Shipwreck at Yassıada, Turkey, 35–54. College Station: Texas A & M University Press. Villa, L. 1994. Le anfore tra tardoantico e medioevo. In S. Lusuardi Siena (ed.), Ad mensam: Manufatti d’uso da contesti archeologici fra tarda antichità e medioevo. Udine: Del Bianco. Volpe, G., Turchiano, M., and Leone, D. 2011. Il progetto Liburna: ricerche archeologiche subacque in Albania (campagne 2007–2010). Annuario della Scuola Archeologica di Atene e delle Missioni Italiane in Oriente 89(3), 251–286. Vroom, J. 2012. From one coast to another: early medieval ceramics in the southern Adriatic region. In S. Gelichi and R. Hodges (eds), From One Sea to Another: Trading Places in the European and Mediterranean Early Middle Ages. Proceedings of the International Conference (Comacchio 27th–29th March 2009), 353–392. Turnhout: Brepols. Wickham, C. 2005. Framing the Early Middle Ages: Europe and the Mediterranean, 400–800. Oxford: Oxford University Press. Wickham, C. 2012. Comacchio and the central Mediterranean. In S. Gelichi and R. Hodges (eds), From One Sea to Another: Trading Places in the European and Mediterranean Early Middle Ages. Proceedings of the International Conference (Comacchio 27th–29th March 2009), 503–510. Turnhout: Brepols. Yangaki, A.G. 2005. La céramique des IVe–VIIIe siècles ap. J.-C. d’Eleutherna. Athens: Editions of the University of Crete.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:27:17, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.010

A MP H O R A S , N E T WOR KS , A ND BYZ A NT I NE MA R I TI M E T R A DE

237

Yangaki, A.G. 2007. Amphores crétoises de forme globulaire: Remarques préliminaires. In M. Bonifay and J.-C. Treglia (eds), LRCW 2: Late Roman Coarse Wares, Cooking Wares and Amphorae in the Mediterranean: Archaeology and Archaeometry, BAR International Series 1662, 767–774. Oxford: Archaeopress. Zanini, E. 2010. Forma delle anfore e forme del commercio tardoantico: spunti per una riflessione. In S. Menchelli, S. Santoro, M. Pasquinucci, and G. Guiducci (eds), LRCW 3: Late Roman Coarse Wares, Cooking Wares and Amphorae in the Mediterranean: Archaeology and Archaeometry, 139–148. BAR International Series 2185. Oxford: Archaeopress. Zmaić, V. 2012. Late Byzantine amphorae from eastern Adriatic underwater sites. Skyllis 13 (1), 81–88. Zolotarev, M.I. 1982. Hersoneskiiy goncharnyiy kompleks VIII–IX vv. In Antichnaya Drevnost’ i Srednie veka, 145–148. Sverdlovsk: s.n.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:27:17, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.010

CHAPTER TEN

NAVIGATING MEDITERRANEAN A R C H A E O L O G Y ’S M A R I T I M E NETWORKS Barbara J. Mills

T

he mediterranean basin seems like an optimal place for studying archaeological social networks. The interaction of different people, groups, and things has been a major part of regional-scale grand narratives from Braudel (1949) to Horden and Purcell (2000), and, more recently, Broodbank (2013). What these volumes bring out are the many ways in which people were connected, their historical contingencies, and their dependencies on materials that circulated throughout the region. Network approaches to these data have been explored by Broodbank (2000) and Davis (1982), both of whom used proximal point analysis, a form of spatial network analysis based on graphs. More recent works have applied a range of formal network methods to the region, resulting in exciting new insights (e.g., Blake 2014; Collar 2013; Brughmans, Isaksen, and Earl 2012; Fulminante 2012; Graham 2006; Iacono 2016; Isaksen 2008; Knappett, Evans, and Rivers 2008; 2011; Leidwanger 2011; 2013). These complement the metaphorical use of network thinking that has played a big part in other syntheses of Mediterranean archaeology (e.g., Malkin 2011). The multi-scalar character of Mediterranean networks makes the region one of the most interesting for archaeological applications of network method and theory. Such multi-scalar qualities can be approached in terms of spatial scales as well as in the kinds of ties and different transportation modes. Yet the multiplicity of people and projects conducting work in the region has created many challenges (Leidwanger et al. 2014). Taken together, the chapters in this

238 Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:26:52, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.011

NA VIGA TING MA RIT I ME NET WORKS

239

volume demonstrate that network method and theory provide insights into the multilayered nature of social interactions in the past, but must be balanced with critical evaluation of how networks are constructed and interpreted, as with any other area of the world. My aim in this chapter is twofold. First, I address one of the major questions posed by the editors, whether maritime networks are fundamentally different from terrestrial networks. Second, I bring out several cross-cutting themes that thread through the individual chapters. One of these is the ways in which networks are constructed by each of the authors, and how those open up new ways of thinking about connectivity within the broader region of the Mediterranean basin. I conclude with some final thoughts on the questions that still remain to be addressed from the overview, provided by the volume editors in their compelling and comprehensive introduction. MARITIME VERSUS TERRESTRIAL NETWORKS

What could be more different than the ways that people made their livelihood by traversing the seas versus the land? Like all dichotomies, there are some dimensions along which they are very different while there are others that are quite the same, suggesting that it is more of a continuum. Although there are a multitude of ways in which maritime and terrestrial networks vary, I will argue that there are many overarching processes that require us to think about Mediterranean networks as comprising both. In addition, as network approaches are used to address archaeological questions, it is important to keep in mind the idea that research should be driven by the questions asked, not the methods, as Brughmans’s chapter reminds us. But first, some contrasts, or at least juxtapositions. Broodbank (2013) notes how sea travel is risky business. While there are risks in any kind of movement and in any environment, deaths at sea are much higher than among those whose living is based exclusively on terrestrial environments. In a short period of time the topography of the sea can change from calm to catastrophic. Hundreds or even thousands of lives can and have been lost in a single storm. The survival rates for those at sea were even worse for people who are forced to travel, whether as refugees or as slaves. The heterogeneity and dynamism of political power in the circum-Mediterranean region suggests that there were frequent movements of people willing to risk their lives to move to safer social settings, just as there are today. At least one study (Lenski 2008) has shown that slaves were present in both Greece and Roman Italy, comprising at least 20% of the latter (see also Cameron 2008). And piracy has been documented from as early as the Bronze Age, which also involved the capture of boats with slaves (e.g., Wiener 2013). Overloaded ships with human cargo are more prone to disaster, as we are often reminded in the recent news.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:26:52, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.011

240

BARBARA J . MILLS

What such risks and high death rates suggest in terms of social networks is the importance of transmission of information about shipbuilding, maintenance, and navigation. This would have been the case whether by small canoes, long boats, or sailing ships since learning about the winds, tides, and rocky coasts— among other threats—requires a high degree of skill. Landscape learning (Rockman and Steele 2003) is part of all successful movement, whether at sea or on land, but the risks may have made transmission about seafaring more canonical or institutionalized. Tartaron (this volume) refers to this as “maritime habitus,” and as Kowalzig (this volume) brings out, this knowledge was probably intertwined with religious life. Being able to read the stars to navigate is just one example—and was probably more important at sea than on land— although once learned it became part of a community of practice (sensu Lave and Wenger 1991). Second, ships were and are containers, technologies for moving goods and people in relatively constrained spaces over short and long distances. The social significance of working in these closed spaces created opportunities for the establishment of new social networks, bridging the heterogeneity represented by what were often diverse origins of their crews. When people depend on each other in risky environments and work in close quarters, the probability of learning about each other’s societies, including each other’s languages, increases. The heterogeneity that emerged from people with different social backgrounds learning from each other likely contributed to network transformation as ideas and technologies were shared that bridged different social backgrounds. There are, of course, barriers to transmission with social status and class. But in the context of a risky economic voyage, co-operation should have been highly valued over competition. This would have resulted in a degree of cosmopolitanism (Meskell 2009) on board as well as in ports. Maritime networks were almost certainly heterogeneous and multi-scalar in the ways that they connected people with each other. Third, ships may carry larger loads and convey them faster than most overland routes. Moreover, they can connect islands as well as portions of continental coasts that are less easily reached by land. Again, we must be mindful that all maritime networks are ultimately connected to terrestrial networks, but for islands in particular maritime networks would have brought these spaces into more regular contact with other islands and the surrounding continents. The quantities of materials and people that could be transported and exchanged, often on routes with multiple ports of call, would have accelerated the rate of interaction, created opportunities for the formation of new social networks, and ultimately altered the structure of regional connectivities. This is, in a nutshell, one of the central arguments of this volume and each of the chapters takes a slightly different perspective on just how archaeologists can approach the topic of maritime connections.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:26:52, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.011

NA VIGA TING MA RIT I ME NET WORKS

241

And finally, archaeologically speaking, whole ships and their contents can and have been recovered whereas we rarely have the entire contents of land-based caravans, or what can be carried on an animal, much less by an individual. In the area that I work in, the North American Southwest, domesticated animals were limited to dogs and turkeys—neither of which were beasts of burden. We do not have any examples of whole shipments (to extend the maritime-based term) from terrestrial contexts that can be argued to be a single event. In some archaeological contexts there may be storerooms, and there are assemblages that were catastrophically buried at one time, but these are rare (Mediterranean archaeology, of course, has iconic examples). Maritime archaeology does not depend on shipwrecks, but where shipwrecks are present the networks that emerge from the analysis of this source of data are key to looking at how things interacted with things as much as people with things. Thus, as Greene (this volume) brings out, the networks of objects recovered from shipwrecks add a new dimension to the ways in which maritime interactions can and should be looked at through the entanglement of people and things (see also Hodder 2012; Hodder and Mol 2015). Overarching the differences between maritime and terrestrial networks are several themes that come out of the chapters in this volume, including who became more connected to whom at different points in the Mediterranean past, the power of the sail and its transformative effects on Mediterranean society, the cosmopolitanism of crews and ports, and how specific memory practices worked to ensure the intergenerational transmission of seagoing routes. These themes, however, are not necessarily constrained by maritime contexts. Each one encapsulates broader topics that cross-cut human networks around the world, including the ways in which people and things connect to each other and how archaeologists make those connections, the role of technological innovations in transforming social practices, the role of culture contact and social heterogeneity in creating new social spaces, and the embeddedness of memory work within religious practices to facilitate transmission of traditional knowledge. Thus, while there are empirical and contextual differences between thinking about networks at sea and those on land, they are not as contrastive as they initially appear. Such tensions between highly contextual discussions and comparative studies are commonly found in archaeology. By focusing on the Mediterranean region, this volume might be expected to tilt toward the former. But there is much more in each of the chapters that makes them valuable for understanding connectivities in other situations, including those that are more exclusively land-based. My aim in the remainder of this chapter, then, is to take a more comparative, thematic perspective, highlighting the particular contributions of each of the chapters with respect to these overarching themes.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:26:52, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.011

242

BARBARA J . MILLS

Agent-Based Theory and Modeling Genetic Flows

Landscape Modeling

Interaction and Networks

Materiality/ Thing Theory

Evolutionary/ World System Theory

Settlement Modeling

Culture Flows Complexity Theory and Modeling

10.1. Kristian Kristiansen’s representation of the Third Science Revolution in archaeology. (After Kristiansen 2014, Figure 1, reproduced courtesy of the author.)

ANALYZING NETWORKS IN MEDITERRANEAN ARCHAEOLOGY

Kristian Kristiansen (2014) has recently suggested that there is a new paradigm in archaeology that centers on the concept of interaction and networks, which he calls the Third Science Revolution. Although not all archaeologists would agree that it is a paradigm shift, or that the axes depicted in his representation (Figure 10.1) are inclusive of all contemporary approaches in archaeology, the centrality of networks to a wide range of archaeological approaches is compelling. To me, this reflects the importance of thinking relationally, a concept that has theoretical and methodological ramifications for all archaeology. Knappett (2011) has been quite explicit about how thinking in network terms is transformative to archaeological practice by focusing on the interactions rather than on the entities. Such relational thinking is at the core of formal network modeling, but it also permeates other approaches, as Kristiansen’s depiction shows. A second important point coming out of Kristiansen’s representation of the Third Science Revolution is that the polar ends of each axis of variation are at different analytical scales. So when one looks at flows of things, they are arranged from small (e.g., genetic flows) to large (e.g., culture flows), just as

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:26:52, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.011

NA VIGA TING MA RIT I ME NET WORKS

243

formal modeling ranges from agent-based to larger-scale complexity approaches. Scale is dependent on the contexts that we study, on the materials subjected to analysis, and on the actual methods themselves. And finally, Kristiansen’s overview also suggests that we are at a point when archaeologists are free to draw upon multiple approaches in looking at interaction. Such intersections are not only possible (e.g., Boivin 2008), but are also at the forefront of archaeological research. Some of this has been in the works for quite a long time but the pace has surely quickened, supporting Kristiansen’s thesis for at least a shift in praxis, if not in paradigm. The chapters as a whole are a cross-section of approaches that can be found in network applications in archaeology more broadly (for some recent reviews see Brughmans 2010; 2012; Collar et al. 2015; Mills 2017), and each could be characterized along one or more of Kristiansen’s dimensions. The approaches range from the metaphorical to the more formal. And within the formal approaches there is a range from those relying on empirical data to those that are more model-based. Of the chapters in this volume, Tartaron is the most skeptical of formal approaches because of the complexity of maritime networks and factors that have not yet been incorporated into network analyses such as winds, currents, and tides (but see Leidwanger 2011). Instead, his chapter is more historical and its strength is in looking at the local scale, how it worked in the past, and how knowledge is transmitted today. It takes a group of coastal cities over time, then draws on ethnographic examples to understand memory transmission and conservatism. These coastal areas bring the scale to the local, or what he terms “small worlds.” These are not the same as the small worlds of Duncan Watts and Steven Strogatz (1998; see also Watts 2003), however, which emphasize how occasional short-circuiting increases overall connectivity within the network. Watts’s idea of small worlds is one that has been used by others in Mediterranean archaeology as a model for what happens with increasing long-distance voyaging to increase overall connectivity (e.g., Malkin 2011). Tartaron’s usage of “small worlds” instead refers to the strong ties that make up the local networks of coastal towns, an important consideration in itself because of the way it sets up interactions of other kinds and especially the transmission of maritime information. Although it is correct that any model does not take into account all of the factors that must have been significant in the past, it is also correct that archaeologists who take a modeling approach generally do not expect models to account for all variability. The value of modeling is in showing the relationships between different variables, the interactions that result when some of the variables are nonlinear, and what might happen because of random processes. Archaeologists who create models—whether network models or not—generally start simple and then add variables in a stepwise fashion. A large number of parameters produces results, but understanding which variables are driving

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:26:52, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.011

244

BARBARA J . MILLS

those results is made much more difficult if the model is too complex, especially at the outset. Simulation modeling is one example of complexity approaches (Figure 10.1; see also Barton 2014). Evans’s and Rivers’s contributions employ formal models, building on prior collaborations (e.g., Evans, Knappett, and Rivers 2009; Knappett, Evans, and Rivers 2008; 2011; Rivers and Evans 2013), unpacking some of the assumptions, and making the case for those models that might be best for understanding Mediterranean connectivity. Evans emphasizes how even such a simple variable as distance may be variously measured and analyzed, resulting in different models and families of models. The important message here is that there is not just one way to formally construct spatial networks, but rather families of models that create similar results. The choices that are made in using one method over another should be examined in terms of how much variation results within distinct families of models to assess the robustness of the model. Rivers’s contribution builds on a paradox noted in their previous work. Based on exchange, the period after the Middle Bronze Age eruption that destroyed Akrotiri, the major Minoan settlement on the island of Thera, seems to have been characterized by greater connectivity in the southern Aegean network. But using network logic, taking out a central node in the network such as Thera would be expected to break up the network and make it less connected. Instead, through the use of ariadne simulations, Rivers shows how this model, with its cost–benefit assumptions, can account for the increased amount of post-eruption exchange among the remaining islands. What is most germane here is that there is a back-and-forth between empirical data recorded by archaeologists and the application of the models. This is one way to finetune or tweak the simulated network models, which offers a new direction in network studies (see also Östborn 2016; Östborn and Gerding 2016). A simulation approach is also used in this volume by Lawall and Graham, who combine empirical data on amphoras with formal models, but at the scale of individual producers or consumers as agents. Here, there is explicit attention to the more general aspects of amphora assemblage variability, such as their heterogeneity or homogeneity. Using different model assumptions in their simulations (e.g., preferential attachment, small worlds), and varying the networks according to producers versus consumers, they are able to help show how heterogeneity in the underlying structure of social networks produces different outcomes in terms of the number of vessel shapes. Their work has many implications for how amphora diversity may be interpreted, and especially how different network structures affect archaeological assemblages. Between the formal simulation models and historical case studies are empirical studies of social networks. Several of the chapters in this volume take such approaches, applying method and theory from social network analysis (SNA)

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:26:52, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.011

NA VIGA TING MA RIT I ME NET WORKS

245

(see Greene; Arthur, Leo Imperiale, and Muci; Brughmans, this volume). When assessing the application of SNA to empirically derived data there are many different considerations, but foremost is determining what constitutes a tie. Is it similarities in the contents of assemblages? If so, how similar do they have to be to say that a tie is present? Is it based on production or provenance of the materials, on distribution, or on shared consumption? Compositional data, stamps, and other direct ways of determining where things were produced are helpful for constructing network ties and can even add a direction to the ties. These are ties that are based on communities of production. If looking at shared forms across assemblages, then we are more likely to be seeing consumer choices or communities of consumption (Mills 2016). These are issues that pertain to any archaeological analyses of interaction patterns, but SNA helps to bring out the assumptions more explicitly. Another consideration in applying SNA to empirical data is the scale at which a node or vertex is defined. In most of the simulated and empirical networks discussed in this volume, the settlement is the scale at which different linkages are discussed. However, the Mediterranean is one of the areas of the world where there are a large number of shipwrecks available for inclusion, and Greene’s contribution skillfully uses these within a formal SNA approach. Shipwrecks were in motion until the time that they sank; as such they illustrate the complexity of ways in which networks can be and were constructed, and how the snapshot of a shipwreck might compare with the longer-term accumulations from terrestrial contexts. There are fundamental differences between shipwrecks and terrestrial assemblages. What we have with shipwrecks are the failures—the sunken ships—and while there may be some degree of randomness in terms of their origins and contents, there may be less randomness in where they sank. But perhaps more importantly is that shipwreck cargos on the whole represent distribution networks and their intersections. Both the contributions of Greene and of Arthur, Leo Imperiale, and Muci illustrate another theme that emerges in SNA in archaeology more generally: how altering the mode of analysis results in different perspectives. Modes refer to whether actors or events or both are used in the analyses, with one-mode analyses, or actor-to-actor, being one of the most common in SNA applications. Most archaeological networks do not actually represent true actor-toactor relations, however, because we rarely have direct evidence for this level of interaction. Only with historical networks can direct links between actors be determined and so archaeologists out of necessity are conceptually reliant upon affiliation (or two-mode) networks to begin their analyses. Then, actor-toactor (or one-mode) analyses are derived from shared participation in a group or community, use of a common kind of artifact, or use of a common resource, all of which are affiliation networks. The latter might be an obsidian source or a clay source, so most compositional data used for archaeological network

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:26:52, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.011

246

BARBARA J . MILLS

construction is essentially a two-mode network. But as both Greene and Arthur, Leo Imperiale, and Muci show, one can transform two-mode networks into one-mode networks and vice versa (for this and other details on SNA see Borgatti, Everett, and Johnson 2013; Brandes et al. 2013; Wasserman and Faust 1994). Greene’s innovative approach combines ego (individual shipwreck cargos), one-mode (geographic locations of shipwrecks), and two-mode networks (shipwrecks and cargo provenance) to show how each one provides insights into different kinds of community. Her analysis of shipwrecks as ego networks (see also Mol, Hoogland, and Hofman 2015) looks at the ways in which the origins of the hull, equipment, and other ship contents are connected. Galley wares are appropriately treated separately from cargo wares and show some differences in origin, as might be expected given the diverse origins of their crews. Her analyses make possible the connections between different sets of objects within the ship, as well as a basis for comparisons of ships to each other. But most significant is the way in which ships are seen as linkages or edges between different nodes, and the scales at which clusters of nodes reveal nested or segmented communities within the ancient Greek world. Paul Arthur and his colleagues focus on Byzantine amphoras and their implications for changing economic networks. They use a two-mode analysis first to show how sites (or actors) are linked through shared artifacts, then a one-mode network to look at the strongest ties. The network of strong ties is concentrated in the area of Byzantine control, although they note that there are missing data for some nodes at the periphery. Amphoras during this period are largely dominated by a single globular form with a capacity of roughly twentyfive to thirty liters. This shape and size may have been popular because of its flexibility for different transportation modes, including being able to be carried by a single person when full. But what the authors argue is that the standardized form and capacity were ultimately because of a state-supported standard unit of measurement. The network diagrams show which areas were more central in the amphora circulation and the potential limits of a state-controlled taxation and exchange system. This was followed by a period of greater heterogeneity in which the authors posit that there might have been deregulation along with greater economic certainty and extension of trade networks. Another of the empirical examples, by Brughmans, is more intractable owing in part to the nature of the question asked and the mismatch of formal network methods to that question. His research question was concerned with distribution patterns, as shown by the co-occurrence of pottery types (Eastern Sigillata A–D) and their forms from Roman eastern Mediterranean contexts. As he concludes in a self-critique, formal network methods did not result in any greater insights than could be found through other quantitative means given the question that was posed. Part of the problem is in the framing of the

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:26:52, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.011

NA VIGA TING MA RIT I ME NET WORKS

247

research question. Rather than looking exclusively at distribution, the percentages of different types of sigillata are more reflective of the choices that consumers make. Communities of consumption strongly affect the relative proportions of different wares and forms within different assemblages (Mills 2016). A key contribution of Brughmans’s chapter, however, is in the incorporation of sensitivity tests, which are critical when using empirically derived data from archaeological contexts. As the chapters show, there are a variety of ways to look at how Mediterranean networks were constructed, both theoretically and empirically. Each of them is rich in interpretive detail. I now turn to how the case studies relate to three additional overarching themes: (1) the impact of technological innovations on network dynamics, (2) the creation of heterogeneous social spaces, and (3) networks of memory and maritime knowledge transmission. TECHNOLOGICAL INNOVATION AND NETWORK DYNAMICS Do significant changes in maritime technology, such as the innovation and widespread adoption of the sail between the late 4th and the 3rd millennium, correspond to new networks? (Leidwanger and Knappett, this volume)

Tartaron notes that life near the sea tends to be conservative, “with limited scope for experimentation and innovation.” Yet there are major technological transformations that have occurred over the course of the premodern Mediterranean world that must have had widespread implications for social networks. How quickly did they spread and what was the extent of their impacts? Perhaps the most impactful was the innovation and adoption of the sail during the Bronze Age. This was not the only way that transMediterranean networks were established, and we are finding that even as early as the Paleolithic, and surely by the Neolithic, there were multiple routes for the dispersion of people as well as domesticated crops and animals (Broodbank 2006; Howitt-Marshall and Runnels 2016). While most scholars seem to think that the sail must have been important for the transformation of existing networks and the establishment of new ones, it seems appropriate here to ask, for whom? How much did these new technologies at sea affect daily life on land? Was it the ability to control goods and the emergent power that goes along with it that fundamentally altered life for those living around the sea? Or was there really no “tipping point” at all, but rather a cumulative effect without the kind of dramatic change that we tend to think of with the “small-world” model with its intense rewiring? Are we biased by events well known at the other end of the temporal continuum, i.e., the widespread globalization that ushered in the modern era in Europe, with new kinds of ships and their abilities to travel long distances? I think that it is both possible and desirable to look at these networks, but like Timothy Brook’s

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:26:52, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.011

248

BARBARA J . MILLS

(2007) discussion of 17th-century life in Vermeer’s Hat, I think it will have to be through the objects of everyday life, some of which were more mundane than others. Ultimately, understanding the impact of such an important innovation as the sail on maritime networks will require looking at areas that are not well represented in this volume, especially North Africa and the Levant. Although some of the shipwrecks carried goods and crew that circulated through these areas, this volume focuses more on the northern Mediterranean, rather than along the coast of Africa. And perhaps even more importantly, the impact will have to be viewed at a more continental scale; that is, on the ways that daily lives were transformed for people living inland. CULTURE CONTACT, NETWORK HETEROGENEITY, AND NEW SOCIAL SPACES At different times, a maritime voyage from Mycenae to Dimini or to Knossos might be an intracultural or a cross-cultural journey. (Tartaron, this volume)

As the above quote indicates, places on the Mediterranean landscape were socially more homogeneous at some times and heterogeneous at others. Heterogeneity was introduced by maritime trade, whether multiethnic, multilingual, or multiobject. Single vessels may have had crews with diverse identities and cargos from far-flung locations. When such heterogeneity emerged for each sub-area of the Mediterranean is a contextual question that can only be addressed with empirical research. Some may have been driven by sumptuary behavior, such as by Minoans during the Bronze Age, but most scholars agree that it would have been quite difficult for one polity to control the entire Mediterranean transport network. Instead, there was a diversity of short-, medium-, and long-distance transport methods, and each of these had implications for the diversity of people and goods that traversed the inland sea and stopped, and perhaps settled, in coastal ports. At sea, longer-distance transport entailed crews with multiple origins, the Late Bronze Age Uluburun shipwreck being one of the best-known examples (Pulak 1998). Longer-distance voyages were less expensive than short-haul transport, but multiple kinds of transport were necessary to move goods safely and to multiple consumer markets. Arnaud (2005) has shown how these different scales resulted in segmented sailing, or the movement of goods from ships of different types to deliver products to consumers. These would have crossed social and political boundaries as ships originating from one place made journeys involving multiple ports and polities. And on land, ports contained enclaves of traders and sailors who were connected to inland markets as well as

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:26:52, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.011

NA VIGA TING MA RIT I ME NET WORKS

249

ships. Some of these ports became “hubs” in the maritime networks, places that then attracted more people and goods, resulting in greater cosmopolitanism. Greene (this volume) uses the concept of “middle ground,” developed by Richard White (1991; see also Malkin 2011; Antonaccio 2013). This concept refers to social spaces that are between, as well as the process of finding common ground in situations of highly heterogeneous and seemingly incommensurate worldviews. Indigenous archaeologists in the Americas have found this concept especially useful because it does not see culture contact within simplistic frameworks of accommodation, assimilation, or acculturation (e.g., chapters in Ferris, Harrison, and Wilcox 2014). Instead it recognizes that there are deep historical trajectories that create long-term pathways through adoption of foreign items that individuals and groups see as beneficial on their own terms. Greene also uses Foucault’s heterotopia to refer to the heterogeneous spaces (including boats) in which people from different backgrounds find themselves and make new spaces. This process of hybridization also fits the Mediterranean model, not just for ships, but also for the ports. What is especially important in applying these concepts is not just their descriptive accounts, but also the ways in which such spaces are transformative as power relationships are restructured. The transformative power of heterogeneous communities, whether as parts of colonialism or not, have been shown to be highly catalytic in SNA. For example, in the Southwest US, migration followed pre-existing networks to some extent, building on historical relationships. Yet these situations were also powerful in changing the entire topology of social networks (Mills, Clark et al. 2013a; Mills, Roberts et al. 2013b; Mills et al. 2015; Peeples and Haas 2013). Weak ties were concentrated in the areas where migrants moved into, which were largely on the edges of more densely settled areas, but relatively quickly were transformed in most cases into strong ties through intermarriage. Exceptions were where power differentials may have been more pronounced, and in those cases the separate identities of different groups were maintained over longer, rather than shorter, time periods. Such parallels are present in the Mediterranean as ports became settled by people originally recruited from, or at least originating in, more distant places. In network terms, these heterogeneous spaces should be characterized first by weak ties with diverse connections. Arnaud’s (2005) secondary ports would seem to be an excellent place in which these weak-tie relationships might be initially found (see also Malkin 2011). Like the southwestern US example, these secondary ports were at the edges of more densely settled primary ports, which set the stage for later development of the strong ties that knit these communities so tightly together.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:26:52, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.011

250

BARBARA J . MILLS

MEMORY, TRANSMISSION, AND RELIGIOUS RITUAL Under what conditions do maritime networks manifest a form of “memory,” continuing to inform movement through the physical and social landscape despite significant political and cultural change? (Leidwanger and Knappett, this volume)

The above question, posed in Chapter 1, points to the importance of memory work in the construction and maintenance of maritime networks. Networks do not necessarily form anew, even though some ships may be blown off course, but are built on previous interconnections. History matters, and networks, social and spatial, must be contextualized in terms of how knowledge about maritime routes was transmitted. Several of the chapters in this volume (those by Tartaron and Kowalzig in particular) address how maritime networks were intertwined with religious networks, and underlying this connection is the role of memory and transmission. Both terrestrial and maritime networks require mnemonic devices for successfully traversing distances. Constellations and individual stars and planets are well-known navigational tools. Tartaron compares the Mediterranean to other well-known seafaring traditional knowledge systems in India. The Pacific is another place where expert knowledge was transmitted that facilitated longdistance seafaring. Details of native navigational systems demonstrate that not only were they complex systems, especially in open waters, but that they also involved learning sequences of places and stars through detailed mnemonics. In the Pacific, these are well documented to have been embedded in ritual songs and to have involved ritual specialists (e.g., Goodenough 1953; Goodenough and Thomas 1987). I am reminded by this ethnographic example of how traditional knowledge of trails in the North American Southwest also is embedded within songs as mnemonics. They can be performed en route and shrines and topographic features pointed out, and the songs repeated at home to initiate and reinforce travelers’ experiences (Darling and Lewis 2007). Memory and religious ritual were integral features of maritime networks and their transmission. Kowalzig (this volume) suggests that religion had a dynamic and creative role in “carrying, structuring, and managing” economic connections, serving to help counteract and benefit from “the unpredictability of this maritime world.” Religious networks may have even stabilized seafaring mobility through shared participation in knowledge about both land and sea. She points to sanctuaries and their mythological roles as equivalent to nodes and ties within networks. As brought out in her work, then, religious networks of people are constructed through communities of practice and shared ideological ties. The connections between religious and economic networks are through the mariners themselves and the knowledge that they transmit through significant places on the landscape, mythological figures, and shared participation in cultic movements.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:26:52, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.011

NA VIGA TING MA RIT I ME NET WORKS

251

As Kowalzig also argues, polytheism is itself a network, with relations between the gods. And cults have their own subnetworks by virtue of shared shrines, sanctuaries, pilgrimage sites, myths, and rituals. Kowalzig additionally points out that it may be the combination of strong and weak ties that most effectively spreads or diffuses religious innovation and participation (see also Collar 2007; Malkin 2011). She brings out how transport networks and religious networks varied in terms of their spatial coverage, ranging from short- to long-distance. Clusters of local, strong ties were created through “interacting sets of small worlds,” which were themselves linked and often transformed by weaker links of longer distances. Kowalzig’s conclusion that maritime networks are overwhelmingly local (short-distance) or regional (medium-distance) is especially interesting because it provides the context in which strong ties may have been more likely to facilitate the spread of religious knowledge without diminishing the impact of less frequently encountered weak ties that resulted from longer-distance interactions. In this way her interpretations are complementary to Tartaron’s in this volume by emphasizing the strong ties present in short- and medium-distance interactions. The correlation of social and spatial distances in his chapter focuses on the primacy of the local scale, such as ports and their immediately surrounding areas. The local scale is where strong kinship and friendship ties are present, which Tartaron calls the “true fabric of Mycenaean life” (see also Broodbank 2000). It is the medium-distance ties that Kowalzig finds most useful in explaining the diffusion of religious movements, such as the “string of cults” of Artemis in the Euboian Gulf. She attributes these to the practice of cabotage trade. Cabotage, often interpreted as “port-to-port,” has become a key concept in understanding maritime networks (Horden and Purcell 2000; see also Blakely 2016; Malkin 2011; Mazurek 2016). In other recent work, Mazurek (2016) suggests that cabotage facilitated the long-distance spread of the cult of Isis from Egypt to Hellenistic Greece through a series of short hops. In both cases, cabotage created heterogeneity and a space of shared culture contact within cultic practices. Thus, while the individual length of the path between ports may have been short, cabotage, just like the longer-distance journeys, may have promoted the diffusion of cultic practices that were then reinterpreted at the local level. Cultic practices call out places on landscapes, including shrines that were pivotal navigational tools. These would have been especially critical for locations with difficult passages. A single explanation for the diffusion of religious innovations is not necessary or desirable, but it is clear that the diversity of ties, both strong and weak, characterized the transmission of religious knowledge. And it is also apparent that much of the power of this knowledge was in its ability to balance the local and the distant within mythic narratives that strengthened navigational success.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:26:52, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.011

252

BARBARA J . MILLS

CLOSING REMARKS

All of the chapters in this volume are linked by their attention to maritime networks, but the diversity of approaches is striking. This is generally representative of archaeology today, as depicted in Kristiansen’s diagram (Figure 10.1). There is a healthy eclecticism and skepticism in the different ways that interaction can be modeled, analyzed, and interpreted. Given the highly interdisciplinary nature of network approaches today, this variation should be expected. A single convergent approach would not be as interesting or as compelling as coming at the problem from multiple perspectives. There is an advantage to working across disciplinary lines that I hope will continue in the future. One very positive cross-cutting approach in the current set of chapters is that almost all of them pay attention to the dynamics of networks, i.e., their change over time. This dynamism is not always present in network approaches applied to non-archaeological data but is one of the great contributions of archaeological analyses to network studies more generally. The degree of control over provenance and the additional information that historical Mediterranean networks provide make this one of the best areas in the world for understanding the dynamics of networks. Yet there are certain questions that are not as widely addressed in this volume and which may be seen as potential next steps. For example, power and centrality are closely linked in many analyses of social networks, as Tartaron’s chapter reminds us. Greene uses concepts like the middle ground and heterotopia that encompass relationships of power within colonial programs. The Byzantine control over globular amphoras brings together economic and political power from a single artifact class, as demonstrated by Arthur, Leo Imperiale, and Muci. While only a few of the chapters explicitly address the dynamics of power relationships, control over people, ships, and their cargo must have been one of the most significant factors in determining who was connected to whom. As Chapter 1 points out, landscapes of power are a common theme in terrestrial networks and thus should also apply to maritime networks. A second question, raised by Kowalzig but only touched upon in the other chapters, is that of path-dependency. To what extent did historical networks and practices channel future networks? Broodbank’s (2013, 597) long view of Mediterranean networks points out that historical Roman administrative units trace the “phantom outlines and evolutionary histories of older maritime connections.” How did previous ties affect the ways in which future trade and interactions of other kinds were structured? I am thinking here about sinks or path-dependency in terms of limiting possible pathways taken in each subsequent sailing. “Paths” play a key part in

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:26:52, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.011

NA VIGA TING MA RIT I ME NET WORKS

253

network models: they help describe the structures that channel information/ideas and goods through the network. Making greater use of different ways of measuring path dependency might be more than a metaphoric use of SNA. Finally, Chapter 1 pointed out that there is a wide range of practice in Mediterranean archaeology, depending on the time periods, materials, and training of researchers. Yet there have been some very rewarding collaborations and this interdisciplinary goal should be pursued. For example, the formal simulation models discussed by Lawall and Graham might be used to help interpret the empirical network observations on amphora homogeneity and heterogeneity in the contribution by Arthur, Leo Imperiale, and Muci. As another example, a collaboration that looks at how transmission moves through social networks in highly risky occupational settings, such as at sea, might entail ethnography, network science, and Mediterranean archaeology. Tartaron’s ethnoarchaeological recording of expert knowledge brings to life how this transmission process works today. The effects of the loss of individuals with substantial knowledge could be modeled much like the larger-scale spatial modeling that Knappett and his colleagues (2008; 2011) have done when island nodes are catastrophically removed (see also Evans and Rivers, this volume). In the long run, it will be difficult to study Mediterranean archaeological networks without having networks of Mediterranean scholars working together to build models, databases, and historical narratives. Fortunately, this volume represents intersecting networks of collaboration that take us closer to thinking about the interconnectivity of the Mediterranean world. REFERENCES

Antonaccio, C. 2013. Networking the middle ground. Archaeological Review from Cambridge 28(1), 213–236. Arnaud, P. 2005. Les routes de la navigation antique: Itinéraires en Méditerranée. Paris: Errance. Barton, M. 2014. Complexity, social complexity, and modeling. Journal of Archaeological Method and Theory 21(2), 306–324. Blake, E. 2014. Social Networks and Regional Identity in Bronze Age Italy. Cambridge: Cambridge University Press. Blakely, S. 2016. Beyond Braudel: network models and a Samothracian seascape. In C. Concannon and L.A. Mazurek (eds), Across the Corrupting Sea: Post-Braudelian Approaches to the Ancient Eastern Mediterranean, 17–38. New York: Routledge. Boivin, N. 2008. Material Cultures, Material Minds: The Role of Things in Human Thought, Society and Evolution. Cambridge: Cambridge University Press. Borgatti, B., Everett, M.G., and Johnson, J.C. 2013. Analyzing Social Networks. Los Angeles: Sage. Brandes, U., Robins, G., McCranie, A., and Wasserman, S. 2013. What is network science? Network Science 1(1), 1–15.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:26:52, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.011

254

BARBARA J . MILLS

Braudel, F. 1949. La Méditerranée et le monde méditerranéen à l’époque de Philippe II. Paris: Armand Colin. Broodbank, C. 2000. An Island Archaeology of the Early Cyclades. Cambridge: Cambridge University Press. Broodbank, C. 2006. The origins and early development of Mediterranean maritime activity. Journal of Mediterranean Archaeology 19(2), 199–230. Broodbank, C. 2010. “Ships a-sail from over the rim of the sea”: voyaging, sailing and the making of Mediterranean societies c.3500–800 BC. In A. Anderson, J.H. Barrett, and K.V. Boyle (eds), The Global Origins and Development of Seafaring, 249–264. Cambridge: McDonald Institute for Archaeological Research. Broodbank, C. 2013. The Making of the Middle Sea: A History of the Mediterranean from the Beginning to the Emergence of the Classical World. London: Thames & Hudson. Brook, T. 2007. Vermeer’s Hat: The Seventeenth Century and the Dawn of the Global World. London: Bloomsbury Publishing. Brughmans, T. 2010. Connecting the dots: towards archaeological network analysis. Oxford Journal of Archaeology 29(3), 277–303. Brughmans, T. 2012. Thinking through networks: a review of formal network methods in archaeology. Journal of Archaeological Method and Theory 20(4), 623–662. Brughmans, T., Isaksen, L., and Earl, G. 2012. Connecting the dots: an introduction to critical approaches in archaeological network analysis. In M. Zhou, I. Romanowska, Z. Wu, P. Xu, and P. Verhagen (eds), Proceedings of Computer Applications and Quantitative Techniques in Archaeology Conference, 2011, Beijing, 359–369. Amsterdam: Amsterdam University Press. Cameron, C. 2008. Introduction: captives in prehistory as agents of social change. In C. Cameron (ed.), Invisible Citizens: Captives and Their Consequences, 1–24. Salt Lake City: University of Utah Press. Collar, A.C.F. 2007. Network theory and religious innovation. Mediterranean Historical Review 22(1), 149–162. Collar, A.C.F. 2013. Religious Networks in the Roman Empire: The Spread of New Ideas. Cambridge: Cambridge University Press. Collar, A., Coward, F., Brughmans, T., and Mills, B.J. 2015. Networks in archaeology: phenomena, abstraction, representation. Journal of Archaeological Method and Theory 22 (1), 1–32. Darling, J.A., and Lewis, B.V. 2007. Songscapes and calendar sticks. In S.K. Fish and P.R. Fish (eds), The Hohokam Millennium, 131–139. Santa Fe: School for Advanced Research Press. Davis, J.L. 1982. Thoughts on prehistoric and archaic Delos. Temple University Aegean Symposium 7, 23–33. Evans, T., Knappett, C., and Rivers, R. 2009. Using statistical physics to understand relational space: a case study from Mediterranean prehistory. In D. Lane, D. Pumain, S. van der Leeuw, and G. West (eds), Complexity Perspectives in Innovation, 451–479. Dordrecht: Springer. Ferris, N., Harrison, R., and Wilcox, M.V. (eds) 2014. Rethinking Colonial Pasts through Archaeology. Oxford: Oxford University Press. Fulminante, F. 2012. Social network analysis and the emergence of central places: a case study from Central Italy (Latium Vetus). BABesch 87, 27–53. Goodenough, W.H. 1953. Native Astronomy in the Central Carolines. Philadelphia: University Museum, University of Pennsylvania.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:26:52, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.011

NA VIGA TING MA RIT I ME NET WORKS

255

Goodenough, W.H., and Thomas, S.D. 1987. Traditional navigation in the Western Pacific: a search for pattern. Expedition 29(3), 3–14. Graham, S. 2006. Networks, agent-based models and the Antonine Itineraries: implications for Roman archaeology. Journal of Mediterranean Archaeology 19(1), 45–64. Hodder, I. 2012. Entangled: An Archaeology of the Relationships between Humans and Things. New York: John Wiley & Sons. Hodder, I., and Mol, A. 2015. Network analysis and entanglement. Journal of Archaeological Method and Theory 23(4), 1066–96. Horden, P., and Purcell, N. 2000. The Corrupting Sea: A Study of Mediterranean History. Oxford: Oxford University Press. Howitt-Marshall, D., and Runnels, C. 2016. Middle Pleistocene sea-crossings in the Eastern Mediterranean? Journal of Anthropological Archaeology 42, 140–153. Iacono, F. 2016. From networks to society: pottery style and hegemony in Bronze Age Southern Italy. Cambridge Archaeological Journal 26(1), 121–140. Isaksen, L. 2008. The application of network analysis to ancient transport geography: a case study of Roman Baetica. Digital Medievalist 4. www.digitalmedievalist.org/journal/4/ isakse. Knappett, C. 2011. An Archaeology of Interaction: Network Perspectives on Material Culture and Society. Oxford: Oxford University Press. Knappett, C., Evans, T., and Rivers, R. 2008. Modelling maritime interaction in the Aegean Bronze Age. Antiquity 82, 1009–1024. Knappett, C., Evans, T., and Rivers, R. 2011. The Theran eruption and Minoan palatial collapse: new interpretations gained from modelling the maritime network. Antiquity 85, 1008–1023. Kristiansen, K. 2014. Towards a new paradigm? The Third Science Revolution and its possible consequences in archaeology. Current Swedish Archaeology 22, 11–40. Lave, J., and Wenger, E. 1991. Situated Learning: Legitimate Peripheral Participation. Cambridge: Cambridge University Press. Leidwanger, J. 2011. Maritime archaeology as economic history: long-term trends of Roman commerce in the northeast Mediterranean. Unpublished PhD dissertation, University of Pennsylvania. Leidwanger, J. 2013. Modeling distance with time in ancient Mediterranean seafaring: a GIS application for the interpretation of maritime connectivity. Journal of Archaeological Science 40, 3302–3308. Leidwanger, J., Knappett, C., Arnaud, P., Arthur, P., Blake, E., Broodbank, C., Brughmans, T., Evans, T., Graham, S., Greene, E.S., Kowalzig, B., Mills, B., Rivers, R., Tartaron, T.F., and van de Noort, R. 2014. A manifesto for the study of ancient Mediterranean maritime networks. Antiquity+. http://journal.antiquity.ac.uk/ projgall/leidwanger342. Lenski, N. 2008. Captivity, slavery, and cultural exchange between Rome and the Germans from the first to the seventh century CE. In C. Cameron (ed.), Invisible Citizens: Captives and Their Consequences, 80–109. Salt Lake City: University of Utah Press. Malkin, I. 2011. A Small Greek World: Networks in the Ancient Mediterranean. Oxford: Oxford University Press. Mazurek, L.A. 2016. Material and textual narratives of authenticity? Creating cabotage and memory in the Hellenistic Eastern Mediterranean. In C. Concannon and L.A. Mazurek (eds), Across the Corrupting Sea: Post-Braudelian Approaches to the Ancient Eastern Mediterranean, 39–64. New York: Routledge.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:26:52, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.011

256

BARBARA J . MILLS

Meskell, L. (ed.) 2009. Cosmopolitan Archaeologies. Durham, NC: Duke University Press. Mills, B.J. 2016. Communities of consumption: cuisines as networks of situated practice. In A.P. Roddick and A.B. Stahl (eds), Knowledge in Motion: Constellations of Learning across Time and Place, 248–270. Amerind Studies in Anthropology (SAA-Amerind Series). Tucson: University of Arizona Press and Amerind Foundation. Mills, B.J. 2017. Social network analysis in archaeology. Annual Review of Anthropology 46, 379–397. Mills, B.J., Borck, L., Clark, J.J., Haas, W.R., Jr., Peeples, M., and Roberts, J.M., Jr. 2015. Multiscalar perspectives on social networks in the Prehispanic Southwest. American Antiquity 80(1), 3–24. Mills, B.J., Clark, J.J., Peeples, M., Haas, W.R., Jr., Roberts, J.M., Jr., Hill, B., Huntley, D.L., Borck, L., Breiger, R.L., Clauset, A., and Shackley, M.S. 2013. The transformation of social networks in the late Prehispanic U.S. Southwest. Proceedings of the National Academy of Sciences 110(15), 5785–5790. Mills, B.J., Roberts, J.M., Jr., Clark, J.C., Haas, W.R., Jr., Huntley, D.L., Peeples, M., Trowbridge, M., Borck, L., and Breiger, R.L. 2013. Late Prehispanic social dynamics in the Southwest U.S., A.D. 1200–1500. In C. Knappett (ed.), Network Analysis in Archaeology, 181–202. Oxford: Oxford University Press. Mol, A., Hoogland, M.L.P., and Hofman, C.L. 2015. Remotely local: ego-networks of late pre-colonial (AD 1000–1450) Saba, north-eastern Caribbean. Journal of Archaeological Method and Theory 22(1), 275–305. Östborn, P., and Gerding, H. 2016. Brick makers, builders and commissioners as agents in the diffusion of Hellenistic fired bricks: choosing social models to fit archaeological data. Journal of Greek Archaeology 1, 233–270. Peeples, M.A., and Haas, W.R., Jr. 2013. Brokerage and social capital in the prehispanic U.S. Southwest. American Anthropologist 115(2), 232–247. Pulak, C. 1998. The Uluburun shipwreck: an overview. International Journal of Nautical Archaeology 27, 188–224. Rivers, R., and Evans, T.S. 2013. What makes a site important? Centrality, gateways and gravity. In C. Knappett (ed.), Network Analysis in Archaeology: New Approaches to Regional Interaction, 125–150. Oxford: Oxford University Press. Rockman, M., and Steele, J. (eds) 2003. The Colonization of Unfamiliar Landscapes: The Archaeology of Adaptation. New York: Routledge. Wasserman, S., and Faust, K. 1994. Social Network Analysis: Methods and Applications. Cambridge: Cambridge University Press. Watts, D.J., and Strogatz, S.H. 1998. Collective dynamics of “small-world” networks. Nature 393(6684), 440–442. Watts, D. 2003. Small Worlds: The Dynamics of Networks between Order and Randomness. Princeton: Princeton University Press. White, R. 1991. The Middle Ground: Indians, Empires, and Republics in the Great Lakes Region, 1650–1815. Cambridge: Cambridge University Press. Wiener, M. 2013. Realities of power: the Minoan thalassocracy in historical perspective. In R.B. Koehl (ed.), AMILLA: The Quest for Excellence. Studies Presented to Guenter Kopcke in Celebration of His 75th Birthday, 149–173. Philadelphia: INSTAP Academic Press.

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:26:52, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.011

INDEX

Aegean amphoras, 135, 139 Aegean Bronze Age, 8, 61–62, 82–83, 89–90 geography and distance, 62–63 Kalamianos, 79–82, 80–81f maritime communities in, 88–89 multi-scalar framework, 67–72 multi-scalar network model, 72–73, 74f Mycenaean anchorages and harbors, 67–68 regional/intra-cultural maritime interaction sphere, 75–76 Saronic Gulf, 73, 76–82, 77f, 78f, 80–81f scalar issues, 63–67, 64f affiliation networks, 222, 225–227f, 231f, 232, 245 Agamemnon (Aeschylus), 105 agency, 41, 44, 96, 97, 99, 113, 117 Agent-Based Models (ABMs), 22–23, 30 agricultural production, 187 amphora megarica, 229 amphoras Aegean amphoras, 135, 139 archeaological setting and, 164–167, 166–167f basket-handle amphoras, 139, 142 Byzantine maritime trade, 219–232, 228f, 231f co-occurrences of amphoras, 9 Corinthian A amphoras, 139 Hellenistic Rhodian amphoras, 181 location of shipwrecks and, 12 Massaliot amphoras, 150 Tan Fabric Southern Ionian amphoras, 146 Antonine Itineraries, 23 Aperlae, 118 aphidrumata, 115 aploia, 106 Apollo, 98, 99, 100f, 102, 116, 117 ariadne model, 30f, 30 Artemis, 99, 102, 103, 104–105, 109, 112–113, 115, 117 artifacts, 2, 3, 8, 13, 94, 135. see also amphoras; tableware distribution in Roman Eastern Mediterranean Athenian Agora, 166 Athenian cups, 135 Attic finewares, 135 Average Weighted Distance, 36

bipartite network, 223f, 224 Brainerd-Robinson coefficient distribution of, 194–197, 195–196f, 198–199t overview of, 190f, 192f, 194, 212 Braudel, Fernand, 5 British rule on Cyprus, 11 Bronze Age. see also Aegean Bronze Age Early Bronze Age (EBA) Cyclades, 7, 77 Late Bronze Age (LBA), 7, 39, 103, 118 Middle Bronze Age, 7, 39, 42, 71 prehistoric record, 7, 9 Bronze Age Eastern Mediterranean, 39 Bronze Age Ugarit, 71 Bronze and Iron Age Mesopotamia, 39 Brownian motion, 11, 108 bulk artifact distributions, 13 bulky commodities, 3 Byzantine kommerkiarioi, 228 Byzantine maritime trade, 219–232, 221f, 225–227f, 228f, 231f cabotage, 113f, 116 Cala Sant Vicenç shipwreck, 151 Christianity, 87, 88 clustering coefficients, 200, 207–208, 208f coast-hugging, 15 coastscape, 73, 84, 86–87 common sense knowledge, 107 connectivity cabotage connectivity, 97, 113f, 116 hyper-connected modern world, 1 maximum connectivity, 110–111 past maritime connectivity, 213 Corinthian Gulf coast, 68 The Corrupting Sea (Horden, Purcell), 5, 72, 94, 101, 118 cosmopolitanism, 240, 241, 249 cross-cultural maritime interaction. see maritime interaction cultic practices, 96, 101–106, 102f, 251 cultural interaction, 13, 15, 16, 66 Cycladic longboat, 62 “Dark Age” world, 119 Demaratus the Bacchiad, 155

257 Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:05:26, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.012

258

I N DEX

Demeter Sanctuary, 165 deterrence function, 26, 45 Diana Ephesica, 116 Dionysius of Halicarnassus, 155 Directed Proximal Point Analysis (Directed PPA), 44, 50, 51 dirigisme system, 232 distances in social network analysis, 23–24 distribution patterns, 211 Dorian Hexapolis, 138 Doubly Constrained Gravity Model (DCGM), 28 Early Bronze Age (EBA) Cyclades, 7, 69, 77 Early Iron Age, 10, 103 Early Modern era, 9 easiest networks, 44–45 Eastern Sigillata, 187, 200, 206, 209–210, 214–217, 246 ecology in Mediterranean prehistory, 1–2 economic networks. see religious and economic networks in the Greco-Roman Mediterranean economic organization, 84 edges in social network analysis, 23–24 effective distance, 42 ego networks, 145, 147f Egyptian New Kingdom, 63 Eigenvalue centrality, 33 entropy, 41–42, 46 environmental constraints of the sea, 3, 4, 12 epistemic approach of entropy, 46 epiteichismata, 115 ethnoarchaeology of maritime coastal community assumptions and hypotheses, 83–84 introduction to, 83 maritime habitus, 84–86, 85f overview of, 83–88, 85f peripheral status and identity formation, 87–88 summary of, 253 Euboian Gulf, 102–103, 104–105, 106–107 eupolia, 106 expert knowledge, 107, 250 exploratory network analysis, 187, 209–210, 212 Exponential Random Graph models, 30 face-to-face basis of ancient societies, 1, 2, 3, 73 fishermen, 94 Flotilla Fresco, 63 ForceAtlas2 graph, 151, 151f, 153–154f Foucault, Michel, 137, 249 fraction of effort, 71 friendship bias, 178 Gela shipwreck, 151 Geographic Information Systems (GIS), 22–23, 186 Geometric Greece, 39 Gephi software, 151, 152–153, 223 global network measures, 112–113, 200, 201t, 202t, 203t, 204f, 205f, 206f

globalization, 83, 153, 155, 247 Goldilocks scenario, 54 goods to market model, 138 “grammar” (and stylistic change model), 172 Grand Ribaud F shipwreck, 151 gravitational pull, 71, 97 gravity models, 26–28, 27f Greco-Roman Mediterranean. see religious and economic networks in the Greco-Roman Mediterranean Greek colonization, 12 Greek myth, 96 Greek polytheism, 96 hegemon, 114 Hellenistic Delos, 102 Hellenistic Rhodian amphoras, 181 Hellenistic world, 10 Hierarchical Clustering Methods, 33 Hinduism, 88 Hittite New Kingdom, 63 human cargo, 239 hypothetical “trade network,” 132–133 ICRATES (Inventory of Crafts and Trade in the Roman East), 188, 193, 209 identity formation, 87–88 imperfect optimization model, 71–72 inside-out geography, 5 interaction distance, 73 interactions in social network analysis, 23–25 intermarriage ties, 249 international spirit, 78 interregional/intercultural maritime interaction sphere, 9, 48, 51, 73, 76, 82, 154 Intervening Opportunity Model, 45, 50 intervisibility, 73, 76, 87 intraregional/intracultural maritime interaction sphere, 73, 76, 146, 147, 154, 248 Ionian cups, 135, 142, 150 Iphigenia in Tauris (Euripides), 103, 109, 111 Iron Age Mediterranean, 95 Kalamianos, 79–82, 80–81f Kekova Adası shipwreck, 139, 142, 143f Kepçe Burnu shipwreck, 139 Kerala (India) fishing communities, 84–85 Kolonna, 77–79, 78f krateriskoi, 104 Kristiansen, Kristian, 242f, 242, 243 La Méditerranée et le monde méditerranéen à l’époque de Philippe II (Braudel), 5 landscape learning, 240 Language Change model, 172–175, 173f, 174f, 175f, 177 Late Bronze Age (LBA), 7, 39, 103, 118. see also Aegean Bronze Age Late Minoan IA (LM IA) period, 48, 55, 56

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:05:26, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.012

IN D EX

Late Minoan IB (LM IB) period, 48, 55 Latenian Northern Europe, 39 Liburna project, 231 linear cost model, 26 logarithmic cost model, 26 long distance trips, 27 longue durée, 114 The Making of the Middle Sea (Broodbank), 5, 94 maritime habitus, 84–86, 85f, 240 maritime interaction, 1–2, 6–7, 66 maritime networks. see also archaeological social networks; networks/network models; religious and economic networks in the Greco-Roman Mediterranean Byzantine maritime trade, 219–232, 228f, 231f communication, 9, 15 creating connections, 1–3 local-scale maritime networks, 61–62 Mediterranean maritime interaction, 1–2, 6–7, 9–16 mobility considerations, 3–9 multi-scalar framework, 67–73, 74f past maritime connectivity, 213 religious networks, 250–251 terrestrial networks vs., 239–241 maritime small world, 73–75 Massaliot amphoras, 150 material culture, 83, 222 material diasporas, 2 maximum distance network (MDN), 28, 29f maximum entropy principle, 26 Mediterranean maritime interaction, 1–2, 6–7, 9–16 memory work, 241, 250–251 Mesoscopic approaches, 22 micro-ecologies, 5 microcanonical ensembles, 46 Middle Bronze Age (MBA), 7, 39, 42, 71 middle ground, 136–137 Middle Helladic (MH) period, 78 mimicry model of amphoras, 168–172, 169f, 171f, 175–177 miniature continents, 5, 12 mobility considerations, 1, 3–9 model discrepancy, 39 modeling archaeological and historical data, 15–16 cultural interaction, 13 environmental variables, 3–4, 138 general approach to, 22–24 interactions, 24–25 networks of interaction, 142 ontic approach, 46 prehistoric periods, 69 relational thinking and, 242 simulation modeling, 244 social variables, 3–4 technological parameters, 138

259

variability in, 243–244 modern transport systems, 26 Monte Carlo approaches, 22, 30 most likely networks, 41–44, 43f movement of goods (dirigisme), 229 multi-scalar framework, 67–73, 74f multiethnic middle ground, 137 Muslims, 88 mutation-rate settings, 169, 174 Mycenaean coastal worlds, 67–72 Mycenaean economy, 88 Mycenaean Greeks, 63–67 natural experiments in history, 40 Neolithic age, 7, 9, 69 Netlogo Mimicry model, 169f network metaphors, 8 network science, defined, 185f, 186 networks/network models. see also maritime networks; religious and economic networks in the Greco-Roman Mediterranean; robust social network analysis; social networks affiliation networks, 222, 225–227f, 231f, 232, 245 Archaic through Hellenistic amphora record, 175–177 archeological setting and, 164–167, 166–167f bipartite network, 223f, 224 Byzantine maritime trade, 219–232, 228f, 231f complete networks, 194 cost-benefit models, 51 description in robust social network analysis, 25, 26f easiest networks, 44–45 ego network for analysis, 145, 147f exploratory network analysis, 187, 209–210, 212 global network measures, 112–113, 200, 201t, 202t, 203t, 204f, 205f, 206f hypothetical “trade network,” 132–133 introduction to, 163–164 Language Change model, 172, 173f, 174f, 175f, 177 local-scale maritime networks, 61–62 mimicry model, 168–172, 169f, 171f, 175–177 one-mode networks, 152, 222, 246 pre-eruption networks, 48–54, 49f, 52f preferential attachment model, 110–111, 169, 170, 174, 177–181, 178–179f, 180f Retail Gravity Model, 43, 50, 51 summary of, 181–182 node rankings, 200, 207–208, 208f nomima, 114 Odyssey (Homer), 76 one-mode networks, 152, 222, 246 onward distribution patterns, 165 Pabuç Burnu shipwreck, 139, 140f, 147, 149 Pactio Sicardi, 230 paleocoastal reconstruction, 79

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:05:26, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.012

260

I N DEX

Pearson correlation coefficients, 33 Peloponnesian War, 109, 111 peripheral status of maritime coastal community, 87–88 personal possessions from shipwrecks, 13 Pointe Lequin 1A shipwreck, 133, 142, 151 political power, 84, 239 ports and harbors guardian of harbors, 104 Mycenaean anchorages and harbors, 67–68 pottery, 12, 79, 94 Eastern Sigillata, 187, 200, 206, 209–210, 214–217, 246 Mycenaean-style, 79 preferential attachment model, 110–111, 169, 170, 174, 177–181, 178–179f, 180f premodern sea travel, 62 presence/absence technique, 191 Principal Component Analysis (PCA), 33, 36 Principle of Insufficient Reason, 41, 45 Principle of Maximum Ignorance, 41 progradational (advancing seaward) shorelines, 67 proletarians at sea, 109 proteleia, defined, 105–106 Proximal Point Analysis (PPA), 8, 28, 29f, 44, 69, 70, 238 pyrrhichai, 104 Radiation Model, 45, 50, 52f, 55, 56 recessive (eroding landward) shorelines, 67 refugee plight, 5 regional capital, 25 regional/intra-cultural maritime interaction sphere, 75–76 religious and economic networks in the GrecoRoman Mediterranean, 250–251 cabotage connectivity, 97, 113f, 116 cultic practices, 96, 101–106, 102f, 251 cults and cabotage, 101–106, 102f global networkedness, 112–113 introduction to, 5, 93–98 maritime knowledge, 106–107 maximum connectivity, 110–111 qualitative approach, 98–101, 100f weak ties, 108–110 Rihll and Wilson Gravity Models (RWGMs), 25, 35f, 36 Roman Empire, 10 Roman Mediterranean, 7 Royal Tombs at Salamis, 155 Saronic Gulf, 73, 76–82, 77f, 78f, 80–81f, 84 Saronic Harbors Archaeological Research Project (SHARP), 79, 83 Scandinavian archaeology, 6 seaborne routes, 7, 9 sensitivity analysis, 194

Shaft Grave period, 71, 79 shipwrecks. see also individually named shipwrecks amphoras and, 12 analysis of, 245, 246 introduction to, 6, 132–136, 134f middle grounds, 136–137 as mobile nodes, 12 social network graphing, 145–149, 147f, 148f social relationships and, 154–157, 156f visualization of, 149–154, 151f, 153–154f zones of interaction, 138, 140f, 141f, 143f, 144f, 155, 156f similarity networks for tableware, 192–193 Simple Gravity Model (SGM), 27, 51, 55 small coastal settlements, 77 A Small Greek World (Malkin), 97 small world model, 73–75, 247 smallest-scale entities, 22 social bonds, 3 social integration, 109 Social Network Analysis (SNA) analysis of, 242f, 247 comparison of networks, 31–36, 32t, 33t, 34f, 35f culture, heterogeneity and, 248–249 distances, 23–24 edges, 23–24 graphing of shipwrecks, 145–149, 147f, 148f gravity models, 26–28, 27f growth of, 96 interactions, 23–25 introduction to, 8, 22–24, 238–239 maritime vs. terrestrial networks, 239–241 maximum distance network, 28, 29f memory work, 241, 250–251 modeling scales, 22–23 network description, 25, 26f proximal point analysis, 28, 29f religion and, 250–251 site-centric social network analysis, 23 statistical measurements, 226 stochastic model, 28–31, 30f summary of, 36, 252–253 technological innovation, 247–248 vertex-centric social network analysis, 23 social organization, 84 social potential, 45 social relationships and shipwrecks, 154–157, 156f social thermodynamics, 46 socioeconomic considerations, 12 spatial distribution of artifacts, 8 spatial network analysis, 238 state-supplied food, 228 stochastic model, 28–31, 30f strength of weak ties, 108–110 structuring logic to coastal life, 83–84 symbolic artifacts, 2

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:05:26, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.012

IN D EX

tableware distribution in Roman Eastern Mediterranean Brainerd-Robinson coefficient, 190f, 192f, 194, 212 Brainerd-Robinson coefficient, distribution of, 194–197, 195–196f, 198–199t clustering coefficients, 200, 207–208, 208f data, 188–189, 189f, 189t exploratory network analysis, 187, 209–210, 212 global network measures, 200, 201t, 202t, 203t, 204f, 205f, 206f node rankings, 200, 207–208, 208f per 25-year period, 208–209, 214–217 similarity networks, 192–193 summary of, 211–213 Tan Fabric Southern Ionian amphoras, 146 technological innovation and network dynamics, 247–248 terrestrial networks, 3, 239–241 Theran eruption agency and, 41 best networks, 45, 47f easiest networks, 44–45 immediate post-eruption networks, 54–55 later post-eruption networks, 55–56 Minoan culture, 47, 48t most likely networks, 41–44, 43f outlook, 57

261

overview, 39–40 pre-eruption networks, 48–54, 49f, 52f Thiessen polygons (Voronoi Diagrams), 25 Third Science Revolution, 242 topological sensitivity, 39 trade and commerce benefits of, 71 Byzantine maritime trade, 219–232, 228f, 231f hypothetical “trade network,” 132–133 Roman commerce, 12 Traditional Social Network Analysis (SNA), 23 Trojan War cycle, 103 two-mode network, 246 Uluburun shipwreck, 248 uninformative results, 211 Viking Age Scandinavia, 230 visualization of shipwrecks, 149, 153–154f Vita Ansgarii (Sindbæk), 23 Works and Days (Hesiod), 75–76 xoanon, 114 Xtent model, 25 zones of control models, 25, 31 zones of interaction, 138, 140f, 141f, 143f, 144f, 155, 156f

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:05:26, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.012

Downloaded from https://www.cambridge.org/core. Stockholm University Library, on 18 Nov 2018 at 08:05:26, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108555685.012