ABSTRACT
Detecting diachronic semantic shifts in words has always been a laborious task for linguists, demanding months and years of observations and dictionary mining. The recent emergence of large and representative national corpora has somewhat simplified this task without undermining the academic value and importance of linguistic expertise. However, corpora can only ease the efforts of linguists to search for examples and date them. In order to come up with a list of changed words for further testing, a researcher has nothing to rely on but personal intuition and knowledge. The first corpus-based approaches to detecting language change mostly employed simple frequency metrics. Vector space models have been studied and used for decades. The idea of employing changes in distributional semantic models to track semantic shifts is not in itself new. Detecting the semantic shifts that words undergo over time demands the ability somehow to compare the reference with the updated models.
