Elamur Research⌕
☰
home / papers / 2020 / I Feel Offended, Don't Be Abusive! Implicit/Explicit Messages in Offensive and Abusive Language

I Feel Offended, Don't Be Abusive! Implicit/Explicit Messages in Offensive and Abusive Language

Summary

Automatic detection of abusive language remains difficult, partly because existing datasets rarely record how overtly a message is abusive. The authors build on the typology of Waseem et al. (2017), which classifies abuse by two factors: whether it has a target and how explicit it is. They apply this lens to OLID/OffensEval, an English Twitter corpus from SemEval 2019 Task 6 that annotates targets but ignores explicitness.

They first profile the corpus using class sizes, message length, an "offensive prior" score derived from the extended Wiegand et al. (2018a) lexicon, and tf-idf keywords. They also re-implement a simple lexicon-lookup classifier in the style of the Duluth system, and it performs very competitively. Together, these findings suggest that the keyword-driven data collection favoured explicitly offensive content.

The authors then add an explicit/implicit annotation layer to OLID and write new guidelines for abusive language, producing AbuseEval v1.0. This resource supports research on implicit abuse, targets, the need for context, and the interplay between offensive and abusive language.

Key points

  • OLID is annotated hierarchically through Figure Eight crowdsourcing:
    • sub-task A: offensive or not;
    • sub-task B: targeted or not;
    • sub-task C: target is an individual, a group, or other.
  • The analysis focuses on sub-task A, where class sizes are:
    • training: 4,400 OFF vs. 8,840 NOT;
    • test: 240 OFF vs. 620 NOT;
    • overall, about 67% of messages are non-offensive.
  • 58% of the offensive material originally came from tweets that Twitter's "safe" filter had already flagged.
  • The average offensive prior separates the classes clearly, and the difference is significant in both splits (Mann-Whitney, p<0.05):
    • training: 0.2547 (OFF) vs. 0.0614 (NOT);
    • test: 0.2306 (OFF) vs. 0.0370 (NOT).
  • Message length differs significantly between classes only in the training split, not in the test split.
  • The dictionary baseline uses 861 stemmed terms from Wiegand et al.'s extended lexicon with an offensiveness score of at least 0.75. It is compared with:
    • NULI (fine-tuned BERT, the top OffensEval system);
    • Duluth (a 563-word blacklist);
    • an SVM baseline.
  • The enriched OLID and AbuseEval v1.0 are released at github.com/tommasoc80/AbuseEval.

Interesting ideas

  • I1. Lexicon-based bias score. Averaging weighted-lexicon scores per class gives a cheap check of whether a dataset's positive class is dominated by overt vocabulary. This is worth running on any new toxicity corpus before training on it.
  • I2. Trivial classifier as a diagnostic. If a single rule ("flag the message if it contains any high-scoring word") rivals strong models, the dataset probably rewards surface cues rather than understanding. This sanity check transfers to many classification benchmarks.
  • I3. New layers on existing data. Adding an explicit/implicit layer and separate abuse guidelines to an established corpus, rather than collecting from scratch, keeps comparability with prior results. It also exposes what the original labels missed.
  • I4. Separate dimensions instead of one label. Treating offensiveness, abusiveness, explicitness, and targets as distinct annotation axes avoids blending different phenomena into a single binary tag. This is useful when designing guidelines for other languages.
  • I5. Statistical checks on split consistency. Comparing per-class properties such as length with significance tests separately in training and test reveals distribution mismatches between splits. Such mismatches could otherwise silently affect evaluation.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

Fields: Linguistics & languages

Cited by

73 papers of our library cite it (of the 140673 whose bibliographies we read); 4 of them have a page here, 69 not yet.

Cite

The paper itself, not this page. BibTeX from the library's record of the work.

@inproceedings{caselli2020i,
  title = {{I Feel Offended, Don't Be Abusive! Implicit/Explicit Messages in Offensive and Abusive Language}},
  author = {Caselli, Tommaso and Basile, Valerio and Mitrović, Jelena and Kartoziya, Inga and Granitzer, Michael},
  booktitle = {Proceedings of the Twelfth Language Resources and Evaluation Conference},
  year = {2020},
  url = {https://aclanthology.org/2020.lrec-1.760/}
}

References

42 references, 7 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Basile, V., Bosco, C., Fersini, E., Nozza, D., Patti, V., Rangel Pardo, F. M., Rosso, P., and Sanguinetti, M. (2019). SemEval-2019 task 5: Multilingual detection of hate speech against immigrants and women in twitter. In Proceedings of the 13th International Workshop on Semantic Evaluation, pages 54… find
  2. Bosco, C., Dell’Orletta, Felice, F. P., Sanuguinetti, M., and Tesconi, M. (2018). Overview of the EVALITA Hate Speech Detection (HaSpeeDe) Task. In Tommaso Caselli, et al., editors, Proceedings of the 6th evaluation campaign of Natural Language Processing and Speech tools for Italian (EVALITA’18), T… find
  3. Chen, Y., Zhou, Y., Zhu, S., and Xu, H. (2012). Detecting offensive language in social media to protect adolescent online safety. In Proceedings of the 2012 ASE/IEEE International Conference on Social Computing and 2012 ASE/IEEE International Conference on Privacy, Security, Risk and Trust, SOCIALCO… find
  4. Davidson, T., Bhattacharya, D., and Weber, I. (2019). Racial bias in hate speech and abusive language detection datasets. In Proceedings of the Third Workshop on Abusive Language Online, pages 25–35, Florence, Italy, August. Association for Computational Linguistics. find our card
  5. Devlin, J., Chang, M.-W., Lee, K., and Toutanova, K. (2019). BERT: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (… find
  6. Fleiss, Joseph., L. (1971). Measuring nominal scale agreement among many raters. Psychological Bulletin, 76(5):378–382. find
  7. Fortuna, P. and Nunes, S. (2018). A survey on automatic detection of hate speech in text. ACM Computing Surveys (CSUR), 51(4):85. find
  8. Founta, A. M., Djouvas, C., Chatzakou, D., Leontiadis, I., Blackburn, J., Stringhini, G., Vakali, A., Sirivianos, M., and Kourtellis, N. (2018). Large scale crowdsourcing and characterization of twitter abusive behavior. In Twelfth International AAAI Conference on Web and Social Media. find
  9. Grice, H. P., Cole, P., and Morgan, J. (1975). Speech acts: Syntax and semantics. find
  10. Ibrohim, M. O. and Budi, I. (2019). Multi-label hate speech and abusive language detection in indonesian twitter. In Proceedings of the Third Workshop on Abusive Language Online, pages 46–57. find
  11. Jurgens, D., Hemphill, L., and Chandrasekharan, E. (2019). A just and comprehensive strategy for using NLP to address online abuse. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 3658–3666, Florence, Italy, July. Association for Computational Lingui… find
  12. Kennedy, G., McCollough, A., Dixon, E., Bastidas, A., Ryan, J., Loo, C., and Sahay, S. (2017). Technology solutions to combat online harassment. In Proceedings of the First Workshop on Abusive Language Online, pages 73–77. find
  13. Kumar, R., Ojha, A. K., Malmasi, S., and Zampieri, M. (2018). Benchmarking aggression identification in social media. In Proceedings of the First Workshop on Trolling, Aggression and Cyberbullying (TRAC-2018), pages 1–11, Santa Fe, New Mexico, USA, August. Association for Computational Linguistics. find
  14. Liu, P., Li, W., and Zou, L. (2019). NULI at SemEval-2019 task 6: Transfer learning for offensive language detection using bidirectional transformers. In Proceedings of the 13th International Workshop on Semantic Evaluation, pages 87–91, Minneapolis, Minnesota, USA, June. Association for Computation… find
  15. Merriam-Webster Online. (2009). Merriam-Webster Online Dictionary. find
  16. Mishra, P., Del Tredici, M., Yannakoudakis, H., and Shutova, E. (2018). Author profiling for abuse detection. In Proceedings of the 27th International Conference on Computational Linguistics, pages 1088–1098. find
  17. Mitrović, J., O’Reilly, C., Mladenović, M., and Handschuh, S. (2017). Ontological representations of rhetorical figures for argument mining. Argument & Computation, 8:267–287. find
  18. Mladenović, M., Krstev, C., Mitrović, J., and Stanković, R. (2017). Using lexical resources for irony and sarcasm classification. In Proceedings of the 8th Balkan Conference in Informatics, BCI 2017, Skopje, Macedonia, September 20 - 23, 2017, pages 13:1–13:8. ACM. find
  19. Nobata, C., Tetreault, J., Thomas, A., Mehdad, Y., and Chang, Y. (2016). Abusive language detection in online user content. In Proceedings of the 25th International Conference on World Wide Web, WWW ’16, pages 145– 153, Republic and Canton of Geneva, Switzerland. International World Wide Web Confere… find
  20. Oxford English Dictionary. (2019). Oxford English Dictionary Online. find
  21. Pamungkas, E. W., Basile, V., and Patti, V. (2019). Stance classification for rumour analysis in twitter: Exploiting affective information and conversation structure. CoRR, abs/1901.01911. — [arXiv:1901.01911](https://arxiv.org/abs/1901.01911) find
  22. Papegnies, E., Labatut, V., Dufour, R., and Linarès, G. (2017). Detection of abusive messages in an on-line community. In CORIA. find
  23. Park, J. H. and Fung, P. (2017). One-step and two-step classification for abusive language detection on twitter. CoRR, abs/1706.01206. find
  24. Pedersen, T. (2019). Duluth at SemEval-2019 task 6: Lexical approaches to identify and categorize offensive tweets. In Proceedings of the 13th International Workshop on Semantic Evaluation, pages 593–599, Minneapolis, Minnesota, USA, June. Association for Computational Linguistics. find
  25. Poletto, F., Stranisci, M., Sanguinetti, M., Patti, V., and Bosco, C. (2017). Hate speech annotation: Analysis of an italian twitter corpus. In CEUR WORKSHOP PRO-CEEDINGS, volume 2006, pages 1–6. CEUR-WS. find
  26. Razavi, A. H., Inkpen, D., Uritsky, S., and Matwin, S. (2010). Offensive language detection using multi-level classification. In Proceedings of the 23rd Canadian Conference on Advances in Artificial Intelligence, AI’10, pages 16–27, Berlin, Heidelberg. Springer-Verlag. find
  27. Ribeiro, M. H., Calais, P. H., Santos, Y. A., Almeida, V. A., and Meira Jr, W. (2018). Characterizing and detecting hateful users on twitter. In Twelfth International AAAI Conference on Web and Social Media. find
  28. Ross, B., Rist, M., Carbonell, G., Cabrera, B., Kurowsky, N., and Wojatzki, M. (2017). Measuring the reliability of hate speech annotations: The case of the european refugee crisis. arXiv preprint arXiv:1701.08118. — [arXiv:1701.08118](https://arxiv.org/abs/1701.08118) find
  29. Sap, M., Card, D., Gabriel, S., Choi, Y., and Smith, N. A. (2019). The risk of racial bias in hate speech detection. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 1668–1678, Florence, Italy, July. Association for Computational Linguistics. find our card
  30. Schmidt, A. and Wiegand, M. (2017). A survey on hate speech detection using natural language processing. In Proceedings of the Fifth International Workshop on Natural Language Processing for Social Media. Association for Computational Linguistics, Valencia, Spain, pages 1– 10. find our card
  31. Strawson, P. F. (1964). Intention and convention in speech acts. The philosophical review, 73(4):439–460. find
  32. Vidgen, B., Harris, A., Nguyen, D., Tromble, R., Hale, S., and Margetts, H. (2019). Challenges and frontiers in abusive content detection. In Proceedings of the Third Workshop on Abusive Language Online, pages 80–93, Florence, Italy, August. Association for Computational Linguistics. find
  33. Waseem, Z. and Hovy, D. (2016a). Hateful symbols or hateful people? predictive features for hate speech detection on twitter. In Proceedings of the NAACL Student Research Workshop, pages 88–93, San Diego, California, June. Association for Computational Linguistics. find
  34. Waseem, Z. and Hovy, D. (2016b). Hateful symbols or hateful people? predictive features for hate speech detection on twitter. In Proceedings of the NAACL student research workshop, pages 88–93. find
  35. Waseem, Z., Davidson, T., Warmsley, D., and Weber, I. (2017). Understanding abuse: A typology of abusive language detection subtasks. In Proceedings of the First Workshop on Abusive Language Online, pages 78–84. find our card
  36. Wiegand, M., Ruppenhofer, J., Schmidt, A., and Greenberg, C. (2018a). Inducing a lexicon of abusive words– a feature-based approach. In Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Paper… find
  37. Wiegand, M., Siegel, M., and Ruppenhofer, J. (2018b). Overview of the germeval 2018 shared task on the identification of offensive language. Austrian Academy of Sciences, Vienna September 21, 2018. find
  38. Wiegand, M., Ruppenhofer, J., and Kleinbauer, T. (2019). Detection of Abusive Language: the Problem of Biased Datasets. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers),… find our card
  39. Zampieri, M., Malmasi, S., Nakov, P., Rosenthal, S., Farra, N., and Kumar, R. (2019a). Predicting the Type and Target of Offensive Posts in Social Media. In Proceedings of NAACL. find our card
  40. Zampieri, M., Malmasi, S., Nakov, P., Rosenthal, S., Farra, N., and Kumar, R. (2019b). SemEval-2019 task 6: Identifying and categorizing offensive language in social media (OffensEval). In Proceedings of the 13th International Workshop on Semantic Evaluation, pages 75–86, Minneapolis, Minnesota, USA… find our card
  41. Language Resource References Basile, Valerio and Bosco, Cristina and Fersini, Elisabetta and Nozza, Debora and Patti, Viviana and Rangel Pardo, Francisco Manuel and Rosso, Paolo and Sanguinetti, Manuela. (2019). HatEval - Multilingual Detection of Hate Speech Against Immigrants and Women in Twitter. find
  42. Zampieri, Marcos and Malmasi, Shervin and Nakov, Preslav and Rosenthal, Sara and Farra, Noura and Kumar, Ritesh. (2019). Offensive Language Identification Dataset - OLID. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/papers/caselli-2020-i-feel-offended-don-t/