Letters From the Past: Modeling Historical Sound Change Through Diachronic Character Embeddings

Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 6713 - 6722 May 22-27, 2022 c⃝2022 Association for Computational Linguistics Letters From the Past: Modeling Historical Sound Change Through Diachronic Character Embeddings Sidsel Boldsen1 and Patrizia Paggio1,2 1Department of Nordic Studies and Linguistics, University of Copenhagen 2Institute of Linguistics and Language Technology, University of Malta {sbol,paggio}@hum.ku.dk Abstract While a great deal of work has been done on NLP approaches to lexical semantic change de- tection, other aspects of language change have received less attention from the NLP commu- nity. In this paper, we address the detection of sound change through historical spelling. We propose that a sound change can be captured by comparing the relative distance through time between the distributions of the charac- ters involved before and after the change has taken place. We model these distributions us- ing PPMI character embeddings. We verify this hypothesis in synthetic data and then test the method’s ability to trace the well-known historical change of lenition of plosives in Dan- ish historical sources. We show that the mod- els are able to identify several of the changes under consideration and to uncover meaning- ful contexts in which they appeared. The methodology has the potential to contribute to the study of open questions such as the relative chronology of sound shifts and their geograph- ical distribution. 1 Introduction The study of sound change goes back to the be- ginnings of modern linguistics in early nineteenth century, when scholars such as Rask and Grimm started making hypotheses about the way sound changes over time, which in turn lead to the dis- covery of regular sound correspondences between ancient languages and the identification of cognates in modern ones (Murray, 2015). Since spoken language from the past is not avail- able, sound change in ancient languages must be deduced from written records by considering devel- opment in spelling through time. In addition, while we may be able to see from the written records that a change did occur, less is known on the specific dynamics of the change. Details of these dynamics include knowledge of when the change started to appear, how long it took for it to be complete, what was the relative chronology of individual sounds in a larger shift, what was the geographical distribu- tion of a change and so forth. Due to the sparsity of linguistic evidence, de- tailed empirical studies of chronological sound change are difficult to conduct. This is especially the case for older stages of languages, where little written text was produced, and much of what did exist has been lost in transmission. However, as we move forward in history to the rise of bureaucracy, for example in medieval Europe, we see that an ex- tensive amount of written records were made. Text from this period of time is interesting in the context of a study of sound change because it shows great variability in spelling patterns. With the digitaliza- tion …

Content truncated to 500 words