Simulating pots in ancient Jerash

An explorable explanation of computational modelling in archaeology

One of the main goals of archaeology is to understand past societies. This not an easy task: imagine trying to tell the story of a city based on a few ceramic fragments. Here, we will show you how modelling and computer simulations can help reconstruct how people lived in the past – or at least make sense out of a bunch of broken pots.

This is Jerash:

Jerash was a city in the ancient world which flourished for several centuries from late Hellenistic times through the Roman, Byzantine and early Islamic periods. Today, what’s left of Jerash makes it a great case study for understanding how an ordinary ancient city in the region worked. The archaeological dig at Jerash unearthed many fragments of pots that fell broadly in two categories: local pots and imported pots. However, there were a lot more local pots than imported pots. In fact, more than 99% of the fragments came from local pots. Why would that be? Perhaps pots imported from distant lands were simply not available in as high quantities in Jerash’s market stalls? In the simulation below move the slider to change the availability of pots and press ‘PLAY’:

This is a simple simulation of the broken pots that this buyer would leave behind over their lifetime. What happens if you set the availability in the middle, equal for both local and imported pots? You might notice that sometimes you will get 49%/51% or 51%/49%. This is because in simulations like this, the availability slider only represents a probability that a certain pot will be bought, so over many simulations, the average should be 50%/50%. What happens when the availability is around 75:25? Does the simulation behave as you expected? Or perhaps people just preferred buying local pots over imported pots. Imagine we could go back in time and look at a the number of pots being sold. In the simulation below move the slider to change the preference of pots and press ‘PLAY’:

Let’s suppose that Jerash potters had been making pots for a long time before imported pots were introduced in the market. We can easily add this new factor to our previous simulation using a new slider. This slider will determine how long it took for the foreign pottery to be introduced to the Jerash market. Note that if we set the time ‘gap’ to zero, then our simulation behaves exactly like the one above. Can you get a big discrepancy in the amount of pottery even without a strong preference towards either type of pottery?

You might have started to wonder: is this just play, or whether we learn something from it? One thing that we can now ask is which combinations of purchase preference and temporal gap would explain our archaeological data.

Before we analyse the results for our simulations above, let’s talk through a very simple example. Imagine each of the two sliders only had three options. That would give us a total of nine (3×3 = 9) combinations that we could simulate, and we can visualize this as a 3×3 grid as in the figure below. At the end of each simulation we could compare our results to our data. If the simulation looks like our data, we can colour it green, and if the simulation result is very different from the data we can colour it pink.

In this example, knowing the value of one of our sliders is not always enough to predict whether our simulation will reproduce our data. It is the combination of slider values that really matters. How important is each slider for explaining the data? Does one slider have a bigger effect on changes in the data than another?

Now let’s see what this really looks like for our simulations when we consider the full spectrum of slider settings. On one axis you will see the values of the preference, and on the other you will see the values for the temporal gap. The colour of each point in the graph will tell you the percentage of local pottery that the model predicts: the dark green zone of slider settings predicts more than 99.8% of local pottery, whereas the dark purple zone explains less than 20%. Try hovering your mouse above one of the colours! If you are using Firefox, the animation might not work. Try another browser (e.g. Chrome, Safari, Edge).

We are now going to address one last hypothesis. This hypothesis suggests that local pottery was used more frequently than imported pottery. This would imply a higher probability of breaking local pottery than imported pottery, which would be reflected in the overall numbers of pots in each household.

We could hypothesise that local Jerash pottery is overrepresented in the archaeological record because it is was used daily and was replaced more often, whereas imported pottery was used less and broke less often. To consider this hypothesis, we need to change our model to differentiate between buying a pot and breaking the pot, and we will only count broken pots.

We assume that the number of pots in a household is going to be constant, so when a pot breaks it will be replaced by a new pot. The type of pot that is bought will depend on the preferences of pot type: i.e. the new pot bought to replace a broken pot is not necessarily of the same type as the pot that broke. The more frequently a type of pot is used, the more likely this type of pot will end up on the pile of discarded broken pots. In this model you can see on the left the number of pots in the household, and on the right you see the broken pots that are discarded and excavated by archaeologists. Try out this model by clicking the buttons to use, break and buy one pot at a time.

Breaking pots one by one might be fun, but it’s a bit slow so we speed things up a bit in the next simulation. Your preference for using either local or imported pottery is now represented by a slider. Each time you press ‘PLAY’ the model simulates the use, breaking, and buying of pots 100 times according to the slider settings (you will notice the number of pots in the house will change very quickly as the simulation breaks pots at super speed). Notice how there needs to be a really strong preference for both buying the local product and the frequency of using the local product in order to get simulation results where only 1% of the broken deposited pots is imported pottery.

That’s it: you have just applied computer simulation to archaeological research. This explorable explanation obviously just offered a very abstract and simplified example of archaeological research, the real stuff involves much more archaeological data analysis and computer coding work. But there’s a few things that will have become clear from this explanation:

• Complex archaeological theories can be broken into little parts and formally represented.
• By using computer modelling we can simulate the archaeological data we would expect to see as the result of our theory.
• We can determine the plausibility of our archaeological theories by comparing simulated data with actual archaeological data.
• Computer modelling allows us to specify our theories, narrowing down the influence of our theories on the archaeological record (think about the graph with the parameter space: only a tiny portion of slider settings could reproduce the archaeological data even though half of the graph represented our archaeological theory).
• Abstracting and simplifying our archaeological theories is necessary for computer simulation to work, and it’s provocative: which is a good thing! It forces us to rethink our theories and specify or reject them.
• Some archaeological theories which sound simple are intuitive can turn out to have surprising counter-intuitive results. The human brain is particularly bad at forecasting the interplay between multiple theoretical scenarios in more complex models: that’s why we use computers.
• These simple models obviously don’t capture the full complexity of the ancient inhabitants of Jerash and their relationship with pottery. They should be seen as thought experiments to be merged with traditional quantitative and qualitative archaeological approaches.

Want to play around a little bit more with simulated pots and Jerash people? Here’s a sandbox model that includes all variables introduced in this explorable explanation. Together they show a pretty complex theory of preference that can be explored in a formal way: