A Computational Evaluation of Two Laws of Semantic Change

For more than a century scholars have proposed laws of semantic change that characterize how words change in meaning over time. Two such laws are the law of differentiation, which proposes that near-synonyms tend to differentiate in meaning over time, and the law of parallel change, which proposes that related words tend to undergo parallel changes in meaning. Researchers have identified a handful of changes that are consistent with each proposed law, but there are no systematic evaluations that assess the validity and generality of these competing laws. Here we evaluate these laws by using a large corpus to assess how thousands of related words changed in meaning over the twentieth century. Our analyses show that the law of parallel change applies more broadly than the law of differentiation, and thereby illustrate how large-scale computational analyses can place laws of semantic change on a more secure footing.