class: inverse, center, middle background-image: url(data:image/png;base64,#https://cc.oulu.fi/~scoats/oululogoRedTransparent.png); background-repeat: no-repeat; background-size: 80px 57px; background-position:right top; exclude: true --- class: title-slide <br><br><br><br><br> .pull-right[ <span style="font-family:Rubik;font-size:24pt;font-weight: 700;font-style: normal;float:right;text-align: right;color:white;-webkit-text-fill-color: black;-webkit-text-stroke: 0.8px;">What the heck? Euphemisms and emotionality in Anglophone government meetings</span> ] <br><br><br><br> <p style="float:right;text-align: right;color:white;font-weight: 700;font-style: normal;-webkit-text-fill-color: black;-webkit-text-stroke: 0.5px;"> Steven Coats<br> University of Oulu, Finland<br> <a href="mailto:steven.coats@oulu.fi">steven.coats@oulu.fi</a><br> 12<sup>th</sup> SwiSca 9, Helsinki<br> January 22nd, 2025<br> </p> --- layout: true <div class="my-header"><img border="0" alt="oululogo" src="https://cc.oulu.fi/~scoats/oululogonewEng.png" width="80" height="80"></div> <div class="my-footer"><span>Coats                                  What the heck? | SwiSca 9</span></div> --- exclude: true <div class="my-header"><img border="0" alt="oululogo" src="https://cc.oulu.fi/~scoats/oululogonewEng.png" width="80" height="80"></div> <div class="my-footer"><span>Coats                                  What the heck? | SwiSca 9</span></div> --- ## Outline 1. Introduction and Background 2. Data: CoNASE, CoBISE, CoANZSE 3. Methods: Text search, audio analysis 4. Results: Relative frequencies and anger intensity 5. Caveats, summary and outlook .footnote[Slides for the presentation are on my homepage at https://cc.oulu.fi/~scoats] <div class="my-header"><img border="0" alt="oululogo" src="https://cc.oulu.fi/~scoats/oululogonewEng.png" width="80" height="80"></div> <div class="my-footer"><span>Coats                                  What the heck? | SwiSca 9</span></div> --- ### Introduction and Background The use of *hell* and **what the hell** is considered to be mildly offensive in most Anglophone contexts - This study considers use of *what the hell* and euphemistic substitutes (*what the heck*) in local government meetings - Impolite and potentially face-threatening language is usually limited in such meetings (communicative maxims such as courtesy and professionalism: no swearing) - We use corpus-linguistic methods to analyze transcripts and **speech emotion recognition** (SER) to analyze utterance emotionality - Premise of SER: Affective information, independent of the semantic content of lexical items, is encoded in the **_acoustic features of conversational speech utterances_** - This are reliably perceived <span class="small">(Campbell, 2004; Swain et al., 2018)</span> and are consistent across languages and cultures <span class="small">(Cowen et al., 2019)</span> 1. In which Anglophone cultures is *what the hell* more/less common in these contexts? 2. In which cultures are euphemistic substitutes more common? 3. What collocates are most frequent with these utterances? 4. What can we infer about the emotionality of such utterances, based on the underlying audio? --- ### Data: CoNASE, CoBISE, CoANZSE <span class="small">(Coats 2024, 2023)</span> US, Canada, England, Scotland, Wales, Northern Ireland, the Republic of Ireland, Australia, and New Zealand - [CoNASE](https://cc.oulu.fi/~scoats/CoNASE.html): 1.25b tokens, 2,572 channels/locations, 301,846 word-timed, part-of-speech-tagged Automatic Speech Recognition (ASR) transcripts - [CoBISE](https://cc.oulu.fi/~scoats/CoBISE.html): 112m tokens, 452 locations, 38,680 transcripts - [CoANZSE](https://cc.oulu.fi/~scoats/CoANZSE.html): 195m tokens, 482 locations, 57k transcripts Freely available for research use; download from the Harvard Dataverse ([CoNASE](https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/X8QJJV), [CoBISE](https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/UGIIWD), [CoANZSE](https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/GW35AK)) .small[.pull-left[ **CoNASE** Country | Channels|Videos|Tokens |Length (h) --------------|---------|------|-----------|----------- US |2,189 |270,931 |1,149,030,824 | 141,455.11 Canada | 383 |30,916 |103,035,369 |12,586.77 **CoANZSE** (also [CoANZSE Audio](https://coanzse.org): searchable online CoAZNSE data, including audio and forced alignment files) Country | Channels|Videos|Tokens |Length (h) --------------|---------|------|-----------|----------- Australia |408 |38,786 |111,470,235 | 13,885.1 New Zealand | 74 |18,029 |84,058,661 |1,083.75 ]] <div style="top:-40px"> .small[.pull-right[ **CoBISE** Country | Channels|Videos|Tokens |Length (h) -------------------|---------|------|-----------|----------- England |324 |23,657|72,879,173 |8,518.39 Northern Ireland | 10 |1,898 |6,508,505 |774.17 Republic of Ireland| 26 |2,525 |6,264,276 |680.81 Scotland |75 |8,135 |17,111,396 |1,845.35 Wales |18 |2,465 |8,800,264 |982.66 ]] </div> --- ### Focus on regional and local council channels Many recordings of meetings of elected councillors: advantages in terms of representativeness and comparability - Speaker place of residence (cf. videos collected based on place-name search alone) - Topical contents and communicative contexts comparable - Content either in the public domain (US) or can be used under "fair use" or "fair dealings" provisions of copyright law (e.g. Australian Copyright Act of 1968) --- ### Data collection and processing - Identification of relevant channels (lists of councils with web pages -> scrape pages for links to YouTube) - Inspection of returned channels to remove false positives - Retrieval of ASR transcripts using [YT-DLP](https://github.com/yt-dlp/yt-dlp) - Geocoding: String containing council name + address + country location to Google's geocoding service - PoS tagging with SpaCy --- ### Example video <iframe width="560" height="315" src="https://www.youtube.com/embed/cn8vWlUae7Y?rel=0&&showinfo=0&cc_load_policy=1&cc_lang_pref=en" frameborder="0" allow="accelerometer; autoplay; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe> --- ### WebVTT file  --- ### Methods: Text search, audio analysis Searches conducted with Apache-Lucene-based **Blacklab** search engine <span class="small">(De Does et al., 2017)</span> - Basic search (in Corpus Query Language): <span class="small">**[lemma="wh.\*"] [lemma="the|in"] [pos="JJ|RB"]? [lemma="hell|heck.\*|devil|fuck.\*|deuce|sam|fudge|feck|eff"]**</span> Emotionality/anger of targeted segments: **emotion2vec** <span class="small">(Ma et al., 2023)</span> For each search hit: - Retrieve the corresponding 6-second audio segment from the video, using **yt-dlp** - Send the audio segment to emotion2vec inference pipeline - Record the emotionality values .small[ Utterance | anger|happiness|neutrality|sadness|unknown --------------|---------|------|-----------|----------- *What the hell did they mean with that* |0.5 |0.2 |0.2 | 0.05|0.05 *what the heck are you doing negotiating for us* |0.33 |0.1 |0.27 | 0.2|0.1 (etc.) ||||| ] --- ### Results: Most frequent types and collocates, CoNASE .pull-left[  ] .pull-right[  ] --- ### Results: CoNASE  --- ### Results: Most frequent types and collocates, CoBISE .pull-left[  ] .pull-right[  ] --- ### Results: CoBISE  --- ### Results: Most frequent types and collocates, CoANZSE .pull-left[  ] .pull-right[  ] --- ### Results: CoANZSE  --- ### Results: Geographical comparison | **country** | **total** | **_hell_** | **_heck_** | **_devil_** | **_fuck_** | **_sam hill_** | **_fudge_** | **_eff_** | | ---------------- | ---------- | ----------- | ----------- | ------------ | ----------- | --------------- | ------------ | ---------- | | Australia | 1.489 | 1.175 | 0.224 | 0.09 | | | | | | Canada | 2.253 | 0.488 | **1.605** | 0.141 | 0.009 | | | | | England | 0.947 | 0.576 | 0.233 | 0.082 | | | 0.027 | | | Ireland | **2.394** | **2.074** | 0.16 | 0.16 | | | | | | New Zealand | 1.903 | 1.308 | 0.523 | 0.048 | | | | 0.012 | | Northern Ireland | 0.765 | 0.459 | | **0.306** | | | | | | Scotland | 1.227 | 0.993 | 0.175 | | | | | 0.058 | | USA | 2.07 | 0.792 | 1.195 | 0.05 | 0.006 | 0.003 | 0.003 | 0.006 | | Wales | 1.363 | 0.682 | 0.568 | 0.114 | | | | | --- ### Results: Most frequent types and collocates - North American speakers have a tendency to use *heck* more than do speakers in the UK, Ireland, Australia, or New Zealand - Second-person pronouns are mostly indirect speech (e.g. *he’s gonna say what the hell are you doing*); avoidance of face-threatening utterances - Many hits are from the “public comment” period at the end of a meeting --- ### Results: Anger intensity | **country** | **_hell_** | **_heck_** | **_fuck_** | **_devil_** | **_sam hill_** | **_eff_** | | ---------------- | ---------- | ---------- | ---------- | ----------- | -------------- | --------- | | AUS | 0.18 | 0.005 | 0 | 0.001 | 0 | 0 | | CAN | 0.208 | **0.076** | 0.004 | **0.165** | 0 | 0 | | England | 0.16 | 0.01 | 0 | 0.003 | 0 | 0 | | Ireland | 0.099 | 0.001 | 0 | 0 | 0 | 0 | | New Zealand | 0.147 | 0.002 | 0 | 0.015 | 0 | 0 | | Northern Ireland | **0.252** | 0 | 0 | 0.006 | 0 | 0 | | Scotland | 0.167 | 0.002 | 0 | 0 | 0 | 0 | | USA | 0.176 | 0.032 | 0.144 | 0.056 | 0 | 0.008 | | Wales | 0.01 | 0.031 | 0 | 0 | 0 | 0 | - **_hell_** is stronger than euphemisms - **_heck_** and other euphemisms are strongest in Canada --- ### Results: Anger intensity and euphemism - Confirmation in naturalistic data that **_what the hell_** has significantly higher anger ratings than do euphemistic equivalents<span class="small"> (Bowers & Pleydell-Pearce, 2011)</span> - First empirical evidence in naturalistic data for the **X-phemism cycle** <span class="small">(Allan and Burridge 1991, 2006)</span>, also referred to as the **euphemism treadmill** <span class="small">(Pinker 1994)</span> - A dysphemism (**_what the hell_**) has become too taboo for Canadians? - It is being displaced by the euphemistic **_what the heck/what the devil_** - These are, in turn, now showing higher emotional valence --- ### Caveats - ASR is not perfect - No demographic data, but some can be semi-automatically annotated <span class="small">(cf. Bredin 2023; Plaquet & Bredin 2023; Ferreira 2024)</span> - Labelled emotion datasets used to train emotion2vec, such as IEMOCAP, may not reflect the nuances of emotional states <span class="small">(Cowen et al., 2019; Öhman, 2021)</span>; some contain acted emotions --- ### Summary - Relative frequencies for the use of *what the hell* in these contexts are comparable in English-speaking cultures - Ireland, Australia, and New Zealand show higher usage of *hell* - US and Canada higher usage of *heck* and other euphemisms - Differences in anger intensity for *what the hell* and its euphemistic substitutes across national varieties of English: *hell* is much stronger than *heck*; *what the heck* is "angrier" in Canada (X-phemism cycle) - Approach represents a foundation for future research into national and regional variation in the use of swearing and emotional speech in general --- #### References .hangingindent[.small[ Allan, K. and Burridge, K. (1991). *Euphemism & dysphemism: Language used as shield and weapon*. Oxford University Press. Allan, K. and Burridge, K. (2006). *Forbidden words: Taboo and the censoring of language*. Cambridge University Press. Bowers, J. S. and Pleydell-Pearce, C. W. (2011). [Swearing, euphemisms, and linguistic relativity](https://doi.org/10.1371/journal.pone.0022341). *PloS One* 6(7), e22341. Bredin, H. (2023). [pyannote.audio 2.1 speaker diarization pipeline: Principle, benchmark, and recipe](https://www.isca-archive.org/interspeech_2023/bredin23_interspeech.html).In *INTERSPEECH 2023*, 1983-1987. Campbell, N. (2004). Perception of affect in speech: Towards an automatic processing of paralinguistic information in spoken conversation. In *Proceedings of Interspeech 2004*, 881–884. Coats, S. (2024). [CoANZSE Audio: Creation of an online corpus for linguistic and phonetic snalysis of Australian and New Zealand Englishes](https://aclanthology.org/2024.lrec-main.302). In *Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024)*, 3407-3412. Coats, S. (2023). [Dialect corpora from YouTube](https://doi.org/10.1515/9783111017433-005). In B. Busse, N. Dumrukcic, and I. Kleiber (eds.), *Language and linguistics in a complex world*, 79–102. De Gruyter. Cowen, A. S., Laukka, P., Elfenbein, H. A., Liu, R., Keltner, D. (2019). The primacy of categories in the recognition of 12 emotions in speech prosody across two cultures. *Nature Human Behavior* 3(4), 369–382. https://doi.org/10.1038/s41562-019-0533-6 De Does, J., Niestadt, J., and Depuydt, K (2017). Creating research environments with BlackLab. In J. Odijk and A. van Hessen (eds.), *CLARIN in the Low Countries*, 245–257. Ubiquity Press. Ferreira, A. I. S. (2024). [wav2vec2-large-xlsr-68353-gender-recognition-librispeech](https://huggingface.co/alefiury/wav2vec2-large-xlsr-53-gender-recognition-librispeech). Ma, Z., Zheng, Z., Ye, J., Li, J., Gao, Z., Zhang, S., Chen, X. (2023). [emotion2vec: Self-supervised pre-training for speech emotion representation](https://doi.org/10.48550/arXiv.2312.15185 ). *arXiv*:2312.15185. Öhman, E. (2021). Emotion annotation: Rethinking emotion categorization. In Reinsone, S., Skadiņa, I., Baklāne, A., Daugavieti, J. (eds.), *Post-Proceedings of the 5th Digital Humanities in the Nordic Countries Conference, Riga, Latvia, October 21–23, 2020*, 134–144. Pinker, S. (1994, 3 April). The game of the name. *The New York Times*. Plaquet, A. and Bredin, H. (2023). [Powerset multi-class cross entropy loss for neural speaker diarization](https://doi.org/10.21437/Interspeech.2023-205). In *Proceedings of Interspeech 2023*, 3222-3226. Swain, M., Routray, A., and Kabisatpathy. P. (2018). Databases, features and classifiers for speech emotion recognition: A review. *International Journal of Speech Technology* 21, 93–120. ] ]