Detecting and modeling syntactic change

Abstract

All languages change over time, but the mechanisms of such change are not well understood. In the age of big data, one approach to studying language change is to use automated methods to analyze corpora that contain millions of words spanning many decades. While existing computational work has focused on semantic change, automated analysis of syntactic change is the focus of this thesis. Our data consist primarily of parliamentary debates, a genre that has been consistent over the past century, and we use dependency parses of the sentences spoken in the British and Canadian Parliaments to investigate how the syntax of English has changed over time. First, we identify verbs that initially appeared in transitive constructions in the British parliamentary debates after the mid-nineteenth century, and we use the list of identified verbs to test hypotheses about the process by which a verb becomes transitive. Second, we investigate passivization and examine which factors are predictive of whether a verb is used in the passive voice so that we can identify possible causes for the observed decline in the rate of passivization. Third, we study how syntactic changes related to colloquialization have propagated over the past century in the Canadian parliament; due to the availability of individual-level sociolinguistic data, we can connect inferences at the aggregate level to those at the individual level while considering variables such as age, gender, cohort, and party, and we find political power to have an important effect. Finally, we probe Transformer-based language models fine-tuned on different time periods for evidence of syntactic change.--Author's abstract