Rizun, Nina and Revina, Aleksandra (2019): Business Sentiment Analysis. Concept and Method for Perceived Anticipated Effort Identification. CORE https://core.ac.uk/download/pdf/301385454.pdf [01.11.2023]
Lei Lei and Dilin Liu (2021): Conducting Sentiment Analysis. Cambridge University Press https://doi.org/10.1017/9781108909679 [01.11.2023]
Sweta Saraff et al. (2020): Application of Sentiment Analysis in Understanding Human Emotions and Behaviour. EAI https://doi.org/10.4108/eai.6-10-2020.166547 [01.11.2023]

This project explores the use of sentiment analysis in historical sociolinguistics to detect language variation caused by emotions. We apply sentiment analytic techniques to examine features of speech-like language in the Salem Witchcraft Papers corpus.

Sentiment analytic techniques collect numerical language users’ ratings of emotional, psychological, cognitive, and semantic features of words. These ratings are then applied to words in a document to estimate overall emotional characteristics of language in the document. Sentiment analysis is commonly used in commercial applications (cf. Rizun and Revina 2019), in corpus linguistics (cf. Lei and Liu 2021), and behavioral psychology (cf. Saraff et al. 2020), but its potential for (historical) sociolinguistic research has yet to be fully explored.

Historical texts pose particular challenges for sentiment analysis, because—in addition to better-studied patterns of language variation and change including the emergence or disappearance of words or changes in lexical semantics—emotional features associated with words may change over time. This creates an affective layer to Labov’s “historical paradox,” since we cannot know the extent to which people in the past “felt” differently about words.

In this project, we demonstrate possibilities for addressing these challenges to use sentiment analysis in historical sociolinguistics. We explore the Salem Witchcraft Papers (SWP)—a corpus of protocols recorded during the Salem Witch Hunt in 1692 and 1693. During that period, 200 individuals were accused of witchcraft and stood trial—with 30 being found guilty and 19 of these being executed. 140 protocols from these trials have so far been transcribed and made digitally available (https://salem.lib.virginia.edu/home.html).

To address problems of diachronic change in emotional features of language, we leverage the technical infrastructure provided by NLTK (https://www.nltk.org) and Python. Our preliminary results demonstrate that it is (a) possible to quantify and visualize emotions within the SWP corpus. Additionally, we have discovered that (b) a computed emotional value for a specific witch paper does not necessarily correlate with the verdict. Moreover, we suggest that (c) in cases where sentiment and verdict do not align, the detected higher diversity and quantity of collocations possibly point towards ongoing semantic change.

Thus, our approach allows us to quantify sentiment ratings as predictors of semantic change. In doing so, we demonstrate the potential of sentiment analysis as a tool for sociolinguistics generally and in historical sociolinguistics specifically. We also suggest a natural language processing solution to Labov’s “historical paradox” in the context of lexical emotional and affective features. As such, our approach more broadly demonstrates the value of leveraging new computational technologies to develop novel solutions to the challenges of studying diachronic sociolinguistic change.