Detecting Semantic Change Using LDA in Historical Texts: a Case Study on Dutch

Activity: Talk or presentation typesInvited talk

Simon Hengchen - Speaker

This talk was part of the Language Technology Lab Seminars series.

Abstract:
Semantic change detection is relevant to many, including historians who want to better understand their sources, or lexicographers who wish to compile dictionaries. While the traditional way of detecting semantic change is to “read a lot” (Cavallin 2012), the availability of large diachronic corpora in digital form and computing power allow for a more automatic and efficient way to tackle this task. This talk is in two parts: first, an LDA -based method to detect semantic change in historical, dirty text will be presented, and then a case study will illustrate the approach. In our case study, we demonstrate a language-agnostic method on a corpus of badly-OCRed Belgian socialist newspapers in Dutch from the 19th and 20th centuries. This case study thus hints at the reproducibility of the method on other, less-resourced languages.

Cavallin, K. (2012). Automatic extraction of potential examples of semantic change using lexical sets. In KONVENS, pages 370–377
10 Oct 2017

ID: 101804308