Agent-based modelling of accent retraction in Balto-Slavic

Refreshment will be provided. The talk is also available online (see booking details for the link)

This presentation concerns a method for simulating a type of drift attested in historical linguistics. Over time, the phonology of a language can change without more drastic outside influences (such as language contact, or analogical levelling), because in spoken communication both speaker and listener are responsible for phonetic variation (see e.g. Ohala 1981). This means that there is always a distribution of pronunciation variants in flux, irrespective of language users’ commitment to conserve traditional pronunciations.

An aspect of Indo-European philology that has long attracted attention are Balto-Slavic words with accent position that does not match the position that is reconstructed for the ancestor language Proto-Indo-European (PIE) on the basis of Greek and Vedic correspondences. For example, Lithuanian dùkterį (‘daughter’, accusative singular) deviates from PIE *dhugh2térm̥ with accent on the second syllable. Retraction of accent is more generally considered to be a morphological change, since the accent may be thought to ‘belong’ to specific morphemes. However, it has been suggested (Jasanoff 2017) that retraction in pre-Balto-Slavic could in some cases be phonetically motivated. The possibility of such right-to-left accent shift on phonetic (and not simply analogical) grounds warrants experimental investigation.

I use this problem as a test case for modelling a prosodic aspect of language change using Multi-Agent Simulation (MAS). MAS is a technique widely used to model physical processes and in more recent years has been applied to certain linguistic phenomena such as diachronic schwa deletion in Hindi (Choudhury et al. 2006) and the formation of a common spoken accent among scientists living and working in Antarctica (Harrington et al. 2019). In my simulation, linguistic interactions are represented by a simple communication game which is repeated thousands of times. Among themselves agents share numerical vectors which stand for polysyllabic strings with prominence (e.g. higher pitch or longer duration) on certain syllables, but in every exchange the vectors pass through noise which represents the vicissitudes of articulation, sound propagation, and perception. Agents selected at random interpret this input according to their ‘mental lexicon’ which is constantly being updated. I demonstrate how categorical change of accent position can emerge from many interactions of agents making small, incremental adjustments to their pronunciations.


Choudhury, M., A. Basu, and S. Sarkar. 2006. Multi-agent simulation of emergence of schwa deletion pattern in Hindi. Journal of Artificial Societies and Social Simulation 9 (2). Accessed 23rd July 2022 at

Harrington, J., M. Gubian, M. Stevens, and F. Schiel. 2019. Phonetic change in an Antarctic winter. Journal of the Acoustical Society of America 146, 3327-3332.

Jasanoff, J. 2017. The Prehistory of the Balto-Slavic Accent. Leiden: Brill.

Ohala, J. 1981. The listener as a source of sound change. In C. S. Masek, R. A. Hendrik, & M. F. Miller (eds), Papers from the parasession on language and behavior. Chicago: Chicago Linguistics Society. 178-203.