This is the first installment in a series of blog posts focusing on craft in archaeology. Here’s a link to the call for submissions. The posts will explore craft in archaeology from the perspective of field practices, analytical and interpretative frameworks, and social impacts on the discipline. The posts will appear every Thursday for as long as we get contributions and compiled into a e-book by The Digital Press at the University of North Dakota.
Shawn Graham, The Electric Archaeologist
Assistant Professor, Department of History, Carleton University
What is digital archaeology? Is it craft?
In “Going Postal“, the Royal Mail is kaput. A critical mass of letters warp space and time, and this being Discworld, magic too, affecting events as they try to effect their own delivery. Knowledge = power = energy = mass, Pratchett tells us in the context of the Library (another concentration of magical words). Words – data – have curious power in the Discworld. Golems, creatures of clay, obey the words in their heads, enslaving them. When in “Feet of Clay“, someone puts the receipt for the purchase of a golem into that golem’s head, the golem is set free. Pratchett of course is not known as a philosopher of science or of epistemology, but the illustrations are useful as we think about the consequences of our own tombs of dead words and self aware word-based robot slaves. Ultimately, when you have these things, you are not as in control as you would like. All things strive. The craft of digital archaeology is about this tension. Our digital methods in archaeology in the field record and organize the contingencies, serendipitous encounters, chance observations, joys and sorrows such that we can excise all that is human from our story about the past.
Slabs like the squared off clots Of a blue cream. Sunk for centuries under grass [Seamus Heany, Door into the Dark 1969] Until I found Bann clay. Like wet daylight or viscous satin under the felt and frieze Of humus layers. The true diatomite Discovered in a little sucky hole, Grey-blue, dull-shining, scentless, touchable – Like the earthâ€™s old ointment box, sticky and cool. [Seamus Heany, To a Dutch Potter in Ireland, 1996]
Very compact, Blue-ish grey to white, 10YR/8/1, pliable, clay 90% silt 10%, 35-17 cm, probably natural.
(originally quoted on Electric Archaeology in a post on 2007/12/10 What do we do then? We might then put that “data” on the web, where nothing much else happens subsequently. From graves to tombs of words. There is no craft here (witchcraft, practical necromancy, or other). There are only our golems, standing quietly by, awaiting further orders.
There is a genre of after-action reporting in video gaming called the ‘playthrough’. A subgenre of this is the ‘speedrun‘, a video of a player beating a game in the fastest time possible, by finding the equivalent of a least-cost path through the branches of possible experiences afforded by the game. Speedrunners are not craftsmen, but analog algorithms optimized for a particular path. Excavation is theatre, Tilley tells us. I’d argue that it’s more akin to a game. The site report is but one path through the excavation. If the excavation in potentia could play out in many possible ways, the site report is the expert playthrough. And like most playthroughs it is lifeless. Contingency and chance and serenditpity and insight are excised. The playthrough is not the game. It is but one path (an efficient path) through our tomb of dead words. Other paths might produce other, equally enlightening truths; possibly more enlightening ones. This is not Discworld; our texts do not warp space and time except in only figurative ways (although consider a citation network). Our golems are not yet set free. But. The day is coming. We are beginning to use our knowledge of the structures and relationships and semantic meanings in our data to structure the internet itself: to traverse these linkages is to perform the structure of knowledge. Archaeological data is not on the web, but is of the web. We have also started to create sentient word-driven creatures to do our bidding, without us necessarily telling them how or what to find. These unsupervised learning algorithms hold much promise for us as archaeologists. Both of these ideas, of linked open data and the use of unsupervised learningalgorithmns are premised on a kind of idea-space. Linked Open Data links data that is semantically related, that has meaning, thus are ‘closer’ in some sense. The links themselves express something about how the knowledge was produced and what it represents. It’s not a network, but a meshwork in an Ingoldian sense. Not a landscape, but a taskscape. Natural language unsupervised learning looks for patterns in these taskscapes – language does work; it is built out of the careful juxtaposition and grammatical linkages of words and ideas – making decisions about what goes with what. Thus there is no reason why alternative playthroughs could be generated and examined. We could tune our golem to emphasize the patterns containing these elements or relationships rather than those ones. In short, digital archaeology is not the use of digital information management tools in the field. Rather, it is the careful crafting and contemplation of algorithmic representations of the past. One could call this “visualization”, if visualization means ‘to communicate meaning’ (and see the tremendous work done at the recent HeritageJam in this regard). However, it is the step before communication that I am most interested in here: data mining. Topic modeling for instance carries no rule-book for deciding how many topics to generate, or which of the many signals created by the algorithm to pay most attention to. Indeed, what topic models produce are meshworks woven through the dead words of the corpus, all the dead cells, loci, contexts and other echoes of the excavation, and through these paths different voices can be heard again. What voice? Not the voice of the past, certainly. Here is the excavation at Prescott Street. In these models of meshworks I can clearly see the “official” interpretation of the site. But I can see other voices too. I can even put names to them:
Figure 1: refuse pr greg crees pit rubbish kind determine paula representing previously rounded bs discovered full gradual probing based enclosing struck
Figure 2: pit david unspecified ross edge roman brenna lowest shallow expect final basal presume dimensions marcus pebbles angular appeared covering diffuse
Greg and David record aspects of the excavation that are, for the most part, unique to and uniquely expressed by, them. We won’t find that in the official report, though. Understanding the output of data mining requires craft. Were it art presumably my interpretation would be unique from yours; both might reveal great truths, but they’d hardly be reproducible or of the level of ‘great’ art that transcends time and place. Science? Were it science then I could give you the method and workflow and you’d be able to exactly replicate what I have done. There would be one and one result only. No, there is randomness and probability involved in data mining, natural language processing, and unsupervised learning. Just as there is randomness and probability involved in excavation, in site formation processes, in the stigmergy of humans in a given culture interacting with each other and with their environment. What we are dealing with is craft in the same way every shaker chair is unique yet recognizably of a type, working with the grain of its materials and all of the improbabilities that came to bear on this one knot in space and time. Craft involves tacit knowledge and induction into secrets. Rob Nelson revealed to me the secrets of getting MALLET to work. It took me several months to figure it all out. In digital work though the bar is always moving. What was once hard becomes easier- Ian Milligan, Scott Weingart and I pulled back the curtain and revealed to all of us poor humanists how to make it work. And yet, even as we write morehandbooks to digital craft (‘hand book’ – hand – digits – digital) we find that what works on my machine does not work on Ian’s. Why? Craft again. Our digital workshops are responsive to the tasks (data) we most often do. Retooling reveals the tacit embodied knowledges we didn’t even see. Craft is alive and well in digital archaeology.
Recognizing/validating/extracting/creating (surely the Germans have an appropriate compound verb) all the meshwork of paths/playthroughs latent in this digital soil, is craft. Craft means that there will be masters, guilds, secrets, and initiations. It can’t be avoided when there is tacit knowledge involved. I leave to the reader to decide whether or not this is desirable. I do know that digital archaeology, as I conceive it, is every bit as ‘slow’, as ‘punk’, as other archaeologies. There are other issues at stake. Chris Godsen asked, ‘What do objects want?’ My digital golems: I did not create them. Who knows their wants and desires? Digital archaeology, more than perhaps other kinds of archaeology, shares authority in the co-creation of archaeological knowledge with quasi-independent algorithms. There is literally something unhuman about knowledge created/uncovered this way. Stephen Ramsay once said, ‘algorithms are thoughts; chainsaws are tools‘. The tacit knowledge, the unspoken bits about working with these digital golems, these thoughts-encoded: that’s the part of the craft that needs the most work right now.