The Pictish symbol stones

The stones were classified in 1903 by J. Romilly Allen and Joseph Anderson into three groups that are still used: Class I, undressed boulders or slabs incised with symbols only; Class II, dressed slabs carved in relief with a cross and symbols; and Class III, cross-slabs without the symbols⁠[10]. The national catalogue, The Pictish Symbol Stones of Scotland, published by the Royal Commission on the Ancient and Historical Monuments of Scotland in 2008, lists every known stone with photographs and locations⁠[1]. Each stone also has a record in Canmore, the national record of the historic environment, maintained by Historic Environment Scotland⁠[11].

New stones still turn up. In 2022 a team led by Gordon Noble of the University of Aberdeen excavated a symbol stone at Aberlemno in Angus, carved with triple ovals, a comb and mirror, a crescent and V-rod and double discs, some cut over others; it had been reused as paving in an eleventh- or twelfth-century building⁠[2]⁠[12].

The symbol inventory is small and stable. Lee and colleagues built a pictorial corpus of every symbol on stones, metalwork and caves, defined a catalogue of symbol types from it, and found the corpus to be short (one to three symbols per stone) and small, under a thousand symbols in all⁠[4]⁠[3].

The dating changed in 2018. Earlier art-historical work placed the start of the system in the sixth or seventh century AD. Noble, Goldberg and Hamilton dated symbol-bearing contexts directly: the carvings from the promontory fort of Dunnicaer near Stonehaven came from a settlement at its height in the third to fourth centuries AD; bone objects and settlement layers in the Northern Isles showed the symbols in use in the fifth century; and Bayesian modelling dated the fort at Rhynie, with its symbol stones, to the late fourth to early sixth century⁠[6]⁠[13]. The authors argued that the symbols formed a naming system that developed at the same time as Irish ogham and Scandinavian runes, beyond the Roman frontier⁠[6]. Excavation at Rhynie had already linked symbol stones to a high-status settlement⁠[14].

The main computational test is the 2010 entropy study. Lee, Jonathan and Ziman built a two-parameter decision tree, based in part on bigram conditional entropy, that sorts small symbol corpora by the kind of communication system they come from, calibrated on known writing systems and non-linguistic sets. It placed the Pictish symbols with writing and not with random or heraldic systems⁠[4]⁠[9]. A longer companion paper concluded that the symbols are most likely words and mostly names⁠[3].

The method was challenged at once. A linguist's comment argued that turning carvings into linear symbol strings and the features of archaic writing systems could bias the result, that the line between writing and drawing is less sharp than the method assumes, and that the conclusion remained pending⁠[15]; the authors replied⁠[16]. Sproat questioned the method and the review practices of general science journals that published such claims⁠[8]; Lee and colleagues responded⁠[17]. In 2014 Sproat tested the published measures on a larger set of written and non-linguistic corpora and found none of them useful as published. One of Lee's measures with a different cut-off, and a new measure based on repetition, did separate the two groups, and both classified the Pictish symbols as non-linguistic⁠[9].

Three readings are live. The first is that the symbols are a writing system for names. Ross Samson argued in 1992 that the paired symbols record personal names; the 2010 statistics and the 2018 dating were read in the same direction⁠[5]⁠[3]⁠[6]. The second is that they are emblems, standing for rank, kin group, office or territory without encoding words, which is consistent with Sproat's 2014 classification⁠[7]⁠[9]. The third, that they are decoration or religious images without a fixed meaning, is hard to square with the strict conventions and pairing⁠[4]. Readings of individual symbols, for example the crescent and V-rod as a name element or a rank, have not been tested in a way that could fail.

A bilingual would settle it: a stone with symbols and a Latin or ogham inscription naming the same person. Short of that, two tests can be run. One is whether symbol pairs pattern like name elements, with constraints on which symbol takes first and second place similar to those in recorded Pictish and Celtic names. The other is a corrected version of the statistical test, run with a null model built from non-linguistic emblem systems of similar size and pairing structure, which would show whether the 2010 result survives Sproat's objection. Systematic radiocarbon dating of more symbol-bearing contexts would show how the inventory changed over time.