Figure 1: The structure of BCal
The authors of this paper were part of the team who originally developed the Bayesian radiocarbon calibration framework and, as such, have for some time had access to the software and hardware needed to undertake the necessary calculations. The software is extremely specialised and the computer power needed quite large. Initially, we were able to work with a number of archaeologists who wished to use our methods. Very quickly, however, demand from archaeologists outstripped what we could offer and so we sought other ways to provide researchers with what they need to make interpretations of their radiocarbon and archaeological information. Our first step towards this was to write specialist computer software (known as mexcal).
mexcal is object oriented general purpose Bayesian radiocarbon calibration software written (mainly by Christen) in C++. It is designed to be able to handle a wide range of archaeological situations and has features for including many types of relative and absolute prior information. mexcal uses simulation techniques known as Markov Chain Monte Carlo (MCMC) methods to perform its calibrations and consequently requires high performance computing resources. We do not discuss the use of MCMC methods here, but for a good introduction to MCMC methods, their use in Bayesian inference and their application to radiocarbon calibration see Gilks et al. (1996).
mexcal takes as input a textual representation of the calibration problem and from this creates C++ objects that are used to undertake the calibration. mexcal produces, as output, an enormous sample of likely calendar dates for each parameter of the radiocarbon problem under study (these parameters represent components such as the calendar dates associated with each of the radiocarbon samples, and the dates of archaeological events of interest such as phase and sequence boundaries).
Although mexcal is a very powerful tool in the hands of an experienced user, it was never intended to be user-friendly. In fact, it was designed to be the number crunching core of our own radiocarbon calibration research tools. As a result, it requires access to fairly considerable high performance computing resources and at least some knowledge of structured computer programming. This meant that a number of other issues had to be tackled before we could make our implementation of the Bayesian radiocarbon calibration framework available for use by others.
The BCal solution is an on-line Bayesian radiocarbon calibration tool, which allows users with a standard World-wide Web browser access to easy to use, graphical user interface tools. These tools act as a mediator between the user and the statistical parts of the software (as shown in Figure 1).
BCal builds up calibration definitions on the basis of answers the user gives in response to project-specific questions generated by the software one step at a time. Once the calibration definition is complete, mexcal code is automatically generated and submitted for calibration. When calibration is complete, the user is given access to a range of tools for creating graphical and statistical summaries. These are based on the raw MCMC samples from the posterior distributions of the dates of interest output by mexcal.
Individual users of BCal do not require powerful computing resources on their own desktop, even if they want to calibrate large numbers of radiocarbon determinations from a complex and highly stratified archaeological site. This is because the actual mexcal calibration process takes place on a dedicated, UNIX machine funded by a Cardiff University research grant. Jobs are queued to the server as and when it has space to handle them and the majority of projects calibrate in an hour or so at most.
Figure 1: The structure of BCal
© Internet Archaeology
URL: http://intarch.ac.uk/journal/issue7/buck/buck2.html
Last updated: Mon Sept 6 1999