Detecting writing: Towards a measure for distinguishing linguistic from non-linguistic sign systems
Building on the work of Sproat (2013, 2014) this paper proposes a new measure for discriminating linguistic from non-linguistic sign systems. We hypothesise that non-linguistic sign systems will exhibit significantly higher conditional entropy when the adjacency of duplicate symbols is controlled for, among other measures (?). We test this prediction against a specific subset of semasiographic codes that are designed to be used in conjunction with spoken or sung language. These sign systems have been selected because they most closely approximate full writing without purporting to model linguistic structure. As such they provide the best chance of isolating the properties of glottographic writing that are not shared by other non-linguistic sign systems. The two non-linguistic codes that we analyse are Blissymbols and Naxi Dongba. For these we have produced parallel corpora whereby signs can be aligned with the relevant elements of an associated utterance. If our measure is successful we will apply it to two undeciphered and controversial sign systems: the Glozel script of France and the Rongorongo script of Rapa Nui.