Euphemistic Phrase Detection by Masked Language Model

Abstract

It is a well-known approach for fringe groups and organizations to use euphemismsordinary-sounding and innocent-looking words with a secret meaning-to conceal what they are discussing. For instance, drug dealers often use "pot" for marijuana and "avocado" for heroin. From a social media content moderation perspective, though recent advances in NLP have enabled the automatic detection of such single-word euphemisms, no existing work is capable of automatically detecting multi-word euphemisms, such as "blue dream" (marijuana) and "black tar" (heroin). Our paper tackles the problem of euphemistic phrase detection without human effort for the first time, as far as we are aware. We first perform phrase mining on a raw text corpus (e.g., social media posts) to extract quality phrases. Then, we utilize word embedding similarities to select a set of euphemistic phrase candidates. Finally, we rank those candidates by a masked language model-SpanBERT. Compared to strong baselines, we report 20-50% higher detection accuracies using our algorithm for detecting euphemistic phrases.

Proceedings of the Second Workshop on Understanding Implicit and Underspecified Language

“…Computational approaches to processing euphemisms (Felt and Riloff, 2020;Zhu et al, 2021;Zhu and Bhat, 2021;Magu and Luo, 2018;Kapron-King and Xu, 2021;Gavidia et al, 2022) have shown much promise, but the dynamic nature of euphemisms remains an obstacle. A euphemism annotation task conducted by Gavidia et al (2022) shows that the inherent ambiguity of euphemisms leads to low agreement in what qualifies as a euphemism.…”

“…Their approach is interesting, as it is the first of its kind and their use of sentiment analysis to identify euphemisms has inspired our work. Zhu et al (2021) approach the task of discovering euphemisms from the lens of content moderation. Their goal was the detection of euphemisms used for formal drug names on social media.…”

“…This work is similar to Magu and Luo (2018), who also explore euphemisms as code words in hate speech. Zhu and Bhat (2021) and Zhu et al (2021) both treat detection and identification as a masked language problem where they use a masked language model (MLM) as a filter to get rid of sentences that are not related to their seedlist of euphemisms and then again to find euphemistic candidates. Like Felt and Riloff (2020), Zhu et al (2021) and Zhu and Bhat (2021) show promise, though their narrow topic focus limit the kinds of euphemisms that can be found.…”

Searching for PETs: Using Distributional and Sentiment-Based Methods to Find Potentially Euphemistic Terms

“…Computational approaches to processing euphemisms (Felt and Riloff, 2020;Zhu et al, 2021;Zhu and Bhat, 2021;Magu and Luo, 2018;Kapron-King and Xu, 2021;Gavidia et al, 2022) have shown much promise, but the dynamic nature of euphemisms remains an obstacle. A euphemism annotation task conducted by Gavidia et al (2022) shows that the inherent ambiguity of euphemisms leads to low agreement in what qualifies as a euphemism.…”

“…Their approach is interesting, as it is the first of its kind and their use of sentiment analysis to identify euphemisms has inspired our work. Zhu et al (2021) approach the task of discovering euphemisms from the lens of content moderation. Their goal was the detection of euphemisms used for formal drug names on social media.…”