TL;DR
This paper investigates the long-standing linguistic claim that women use euphemisms more than men. Using a large-scale quantitative analysis of four diachronic English text corpora, the authors found no evidence to support this claim. Their findings suggest that this widely held belief, originating from anecdotal observations, may not hold true across diverse language use.
What Problem Does This Paper Address?
For decades, linguists like Robin Lakoff and Otto Jespersen have proposed that women tend to use euphemisms more frequently than men. The proposed reasons for this include a desire to speak more tactfully and to avoid directly mentioning taboo or "unladylike" topics. This difference has been linked to broader theories about "women's language" and its role in language change, with the idea that women might even lead the formation of new euphemisms. However, this claim has largely been based on anecdotal evidence and has not been systematically tested across historical language use. This paper aims to fill that gap by quantitatively evaluating the gender asymmetry in euphemism usage over time.
Methods
The core of this study is a quantitative analysis of euphemism usage across different genders and time periods. The authors defined euphemism as a less distasteful word or phrase used as a substitute for something harsher or more offensive. They assembled a list of 106 euphemism-taboo pairs to analyze. For each pair, they calculated a "euphemism usage proportion" for each gender. This proportion is defined as the frequency of the euphemism divided by the sum of the frequencies of the euphemism and its corresponding taboo expression. This metric aims to capture whether a gender tends to use the euphemistic form over the taboo form.
To handle variations in language and to ensure robust results, the authors employed several techniques:
- Lemmatization: They used the NLTK WordNetLemmatizer and part-of-speech tagger to reduce words to their base forms, ensuring that different inflected forms of euphemisms and taboos were captured.
- Gender Classification: For corpora where gender was not explicitly labeled, they used titles (Mr./Ms.) and a gender-guessing R package based on first names and birth years. Texts where gender could not be determined were discarded.
- Sampling and Binning: To address gender imbalance in the datasets and mitigate data sparsity, they used random sub-sampling and created time bins (2-year bins for Reddit/NYT, 10-year bins for Canadian Parl./US Congr.).
- Frequency Thresholding: Pairs that did not meet a certain frequency threshold (appearing in at least 10% of 25 samples) were excluded to avoid unreliable results from very sparse data.
- Statistical Testing: A one-tailed Welch's unequal variances t-test was used to compare euphemism proportions between genders for each pair.
Datasets
The study utilized four large diachronic text corpora:
- Reddit: A social media platform, representing an informal register.
- New York Times Annotated Corpus (NYT): A collection of newspaper articles, representing a formal register.
- Canadian Parliamentary dataset (Canadian Parl.): Transcripts of parliamentary debates, representing a formal register.
- United States Congressional dataset (US Congr.): Transcripts of congressional debates, representing a formal register.
These corpora were chosen for their size, diachronic nature, and the ability to approximate author/speaker gender. The NYT, Canadian Parl., and US Congr. corpora are considered to embody the "middle class politeness criterion" discussed in prior work.
Key Insights and Ideas
- Challenging a Linguistic Orthodoxy: The most significant insight is the direct refutation of a long-held linguistic belief about gendered language use, based on robust quantitative evidence rather than anecdote.
- Euphemism Treadmill and Innovation: The paper touches upon the "euphemism treadmill" (Pinker, 1994), where euphemisms become taboo and are replaced by new ones. While Lakoff viewed euphemism use as linguistic conservatism, the authors suggest it could also be a sign of innovation. Their findings, however, do not support the idea that women are leading this process.
- The Nuance of "Politeness": The study highlights that not all forms of "polite speech" are used equally by genders. While previous studies found women use polite forms more, this doesn't automatically extend to euphemisms.
- Topic-Specific Differences: The authors acknowledge that while overall gender differences in euphemism use might be minimal, specific topics or euphemism-taboo pairs might show gendered preferences (e.g., "chest" vs. "breast").
Key Results
- No Gender Asymmetry in Euphemism Use: Across all four corpora and throughout the analyzed time periods (spanning 1951-2018), the study found no consistent evidence that women use euphemisms with a higher proportion than men.
- Robustness of Findings: The results remained consistent even when analyzing subsets of the data, such as multi-word phrases or pairs that met frequency thresholds across all corpora.
- Majority of Pairs Show No Difference: For most euphemism-taboo pairs, there was no significant difference in euphemism proportion between men and women.
- Specific Pair Examples:
- For the pair "lady" vs. "woman," women did not use the euphemism "lady" with a higher proportion than men. In fact, men used it with a higher proportion in three of the four corpora.
- For pairs like "chest" vs. "breast," men consistently preferred the euphemism more than women across all corpora.
- Data Discarded: A notable percentage of data was discarded due to the inability to determine speaker gender (e.g., 46% of NYT data).
Limitations and Future Work
- Corpus Limitations: The authors acknowledge limitations of their chosen corpora. Political and newspaper language might be regulated, potentially minimizing gender differences. Reddit data relied on self-reported gender, and natural conversation data, which Lakoff and Jespersen might have been more concerned with, was not included.
- Euphemism List: The list of 106 euphemism-taboo pairs was gathered manually and is relatively small, not covering all taboo topics or euphemism types.
- Ambiguity: Some words in the list have multiple senses, which could affect analysis, though multi-word phrases were analyzed separately to mitigate this.
- Future Work:
- Automatic Euphemism Detection: Developing automatic methods could help create more comprehensive euphemism lists for future studies.
- Leading Euphemism Innovation: Future research could directly investigate who leads the formation of new euphemisms and the euphemism treadmill, acknowledging that this might be context- and topic-dependent.
- Historical Data: Investigating older corpora (pre-1950s) could provide further diachronic insights, though data sparsity and gender identification challenges are significant hurdles.
Summary updated