---
title: "LT3 at SemEval-2022 Task 6 — paper by Kaminska et al. (2022)"
description: "The paper describes the LT3 team's entry for English subtask A of SemEval-2022 Task 6 (iSarcasmEval). 15 references."
authors: ["Olha Kaminska", "Chris Cornelis", "Veronique Hoste"]
year: 2022
venue: "Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval-2022)"
source: "https://aclanthology.org/2022.semeval-1.138/"
doi: "10.18653/v1/2022.semeval-1.138"
licence: "CC BY 4.0 (ACL Anthology)"
licence_url: "https://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/kaminska-2022-lt3-semeval-2022-task-6/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/kaminska-2022-lt3-semeval-2022-task-6/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2022](https://md.research.elamur.ai/free/en/papers/2022/page/5/) / LT3 at SemEval-2022 Task 6: Fuzzy-Rough Nearest Neighbor Classification for Sarcasm Detection

# LT3 at SemEval-2022 Task 6: Fuzzy-Rough Nearest Neighbor Classification for Sarcasm Detection

**Authors:** Olha Kaminska, Chris Cornelis and [Veronique Hoste](https://md.research.elamur.ai/free/en/authors/veronique-hoste/) · **Year:** 2022 · **Venue:** [Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval-2022)](https://md.research.elamur.ai/free/en/venues/semeval/) ([SemEval 2022](https://md.research.elamur.ai/free/en/venues/semeval/2022/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2022.semeval-1.138/) · [PDF](https://aclanthology.org/2022.semeval-1.138.pdf) · [DOI 10.18653/v1/2022.semeval-1.138](https://doi.org/10.18653/v1/2022.semeval-1.138) · [OpenAlex](https://openalex.org/W4287890943)

**Citations:** 6 citations (OpenAlex, 2026-10-14)

## Summary
The paper describes the LT3 team's entry for English subtask A of SemEval-2022 Task 6 (iSarcasmEval). The task is to decide whether a short social-media text is sarcastic. The labels were given by the authors of the texts themselves, and non-sarcastic examples outnumber sarcastic ones about three to one. Instead of a fully neural classifier, the team used an instance-based and more interpretable method. Each text is turned into a vector, and then the Fuzzy-Rough Nearest Neighbor classifier with Ordered Weighted Average operators (OWA-FRNN) is applied. The authors compared six embedding methods, three preprocessing options, several neighbour counts and mean-vote ensembles, all with 5-fold cross-validation. The winning setup was raw text, Twitter-roBERTa irony embeddings and k = 5. It reached 9th place on the leaderboard. Because each prediction comes from a few concrete training neighbours, the authors could inspect errors directly. This showed that shared topic drives similarity more than sarcasm does. Later experiments showed that hyperparameters tuned by cross-validation transferred poorly to the test set.

## Key points
- Data: iSarcasmEval English subtask A, with 2,601 non-sarcastic and 867 sarcastic training texts; the metric is F1 on the sarcastic class.
- Embeddings compared: Word2Vec, DeepMoji, USE, BERT, SBERT and cardiffnlp Twitter-roBERTa-base for irony. The roBERTa vectors, taken from an inner layer, clearly won, with a cross-validation F1 of 0.3722 against 0.3157 for the next best.
- The classifier uses additive OWA weights over fuzzy lower and upper approximations, with cosine similarity rescaled to the range [0, 1]. Mean-vote ensembles of embeddings all scored below roBERTa alone.
- Official test results were sarcastic F1 0.4242 (9th place), macro-F 0.6552 and precision 0.6422 (both 7th), and accuracy 0.8100 (8th).
- After the competition, k = 17 scored lower in cross-validation (0.3408) but reached 0.5 on test, which would have meant 4th place. Plain wkNN gave very similar numbers: 0.4299 on test at k = 5 and 0.4969 at k = 17.

## Interesting ideas
- The classifier's explanation is simply the handful of labelled training examples that decided the class, so every verdict can be audited by a human.
- Class membership is scored in two directions: how similar an item is to its own class, and how dissimilar it is to the other classes. Both scores are combined with soft rank-weighted averaging rather than taken from a single nearest point.
- Error analysis found that neighbours were mostly chosen because they shared a topic, not because they were sarcastic, so the embedding confuses topic with the target property.
- The value of k chosen by cross-validation was far from the best value on the test set, which shows how unstable model selection is when the target class is small.
- Averaging the votes of weaker models with one strong model reduced quality, and pruning the weak members helped.

## For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

- **I1** (frames): the irony or implicature detector in prag could attach its nearest labelled precedents to each decision, which keeps the step explainable without needing a separate explanation model.
- **I2** (world): when assigning an entity to an ontology class from WordNet or Wikidata, this two-sided soft scoring could make the choice less brittle than a single best match.
- **I3** (learn): add negative controls that pair same-topic items with opposite labels, to check whether a component has learned the phenomenon or just the subject matter.
- **I4** (learn): report the spread of scores across folds and settings rather than trusting one tuned optimum, especially for rare categories such as irony.
- **I5** (learn): when several witness parsers agree on silver data, weight each witness by its measured reliability instead of giving all of them an equal vote.

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

## Institutions

- [Ghent University](https://md.research.elamur.ai/free/en/universities/ghent-university/) — University, Ghent, Belgium · 18 works · 2,888 citations

📚 Shelves (4): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 9,086 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 8,889 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 9,296 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 9,206 works

🧭 Topics (2): [English](https://md.research.elamur.ai/free/en/topics/english/) · 1,373 works · [😊 Sentiment & emotion](https://md.research.elamur.ai/free/en/topics/sentiment-emotion/) · 608 works

## Cite

Cite as: Olha Kaminska, Chris Cornelis and Veronique Hoste. 2022. LT3 at SemEval-2022 Task 6: Fuzzy-Rough Nearest Neighbor Classification for Sarcasm Detection. Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval-2022).

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

```bibtex
@inproceedings{kaminska2022lt3,
  title = {{LT3 at SemEval-2022 Task 6: Fuzzy-Rough Nearest Neighbor Classification for Sarcasm Detection}},
  author = {Olha Kaminska, Chris Cornelis and Veronique Hoste},
  booktitle = {Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval-2022)},
  year = {2022},
  doi = {10.18653/v1/2022.semeval-1.138},
  url = {https://doi.org/10.18653/v1/2022.semeval-1.138}
}
```

- [← previous IISERB Brains at SemEval-2022 Task 6: A Deep-learning Framework to Identify Intended Sarcasm in English](https://md.research.elamur.ai/free/en/papers/shekhawat-2022-iiserb-brains-semeval-2022-task/)
- [next → Making Italian Parliamentary Records Machine-Actionable: the Construction of the ParlaMint-IT corpus](https://md.research.elamur.ai/free/en/papers/agnoloni-2022-making-italian-parliamentary-records-machine/)
- [↑ back to papers of 2022](https://md.research.elamur.ai/free/en/papers/2022/page/5/)

## References

15 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Ibrahim Abu Farha, Silviu Oprea, Steve Wilson, and Walid Magdy. 2022. Semeval 2022: isarcasmeval -intended sarcasm detection in english and arabic. In Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval-2022). [find](https://scholar.google.com/scholar?q=Ibrahim+Abu+Farha%2C+Silviu+Oprea%2C+Steve+Wilson%2C+and+Walid+Magdy%2E+2022%2E+Semeval+2022%3A+isarcasmeval+%2Dintended+sarcasm+detection+in+english+and+arabic%2E+In+Proceedings+of+the+16th+International+Workshop+on)
2. Francesco Barbieri, Jose Camacho-Collados, Luis Espinosa Anke, and Leonardo Neves. 2020. [TweetEval: Unified benchmark and comparative evaluation for tweet classification](https://md.research.elamur.ai/free/en/papers/barbieri-2020-tweeteval-unified-benchmark-comparative-evaluation/). In Findings of the Association for Computational Linguistics: EMNLP 2020, pages 1644–1650, Online. Association for Computational Linguistics.
3. Daniel Chandler and Rod Munday. 2011. A dictionary of media and communication. OUP Oxford. [↗](https://doi.org/10.1093/acref/9780199568758.001.0001)
4. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. [Bert: Pre-training of deep bidirectional transformers for language understanding](https://md.research.elamur.ai/free/en/papers/devlin-2019-bert-pre-training-deep-bidirectional/). 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), pages 4171– 4186. machine learning with fuzzy rough sets. In IJCRS 2020: Proceedings of the International Joint Conference on Rough Sets, volume 12179 of Lecture Notes in Artificial Intelligence, pages 491–499. Springer.
5. Saif M. Mohammad, Felipe Bravo-Marquez, Mohammad Salameh, and Svetlana Kiritchenko. 2018. [Semeval-2018 Task 1: Affect in tweets](https://md.research.elamur.ai/free/en/papers/mohammad-2018-semeval-2018-task-1-affect/). In Proceedings of International Workshop on Semantic Evaluation (SemEval-2018), New Orleans, LA, USA.
6. Nils Reimers and Iryna Gurevych. 2019. [Sentence-BERT: Sentence embeddings using Siamese BERT-networks](https://md.research.elamur.ai/free/en/papers/reimers-2019-sentence-bert-sentence-embeddings-siamese/). In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 3982–3992, Hong Kong, China. Association for Computational Linguistics. [↗](https://doi.org/10.18653/v1/d19-1410)
7. Cynthia Van Hee, Els Lefever, and Véronique Hoste. 2018. [Semeval-2018 task 3: Irony detection in english tweets](https://md.research.elamur.ai/free/en/papers/hee-2018-semeval-2018-task-3-irony/). In Proceedings of The 12th International Workshop on Semantic Evaluation, pages 39– 50.
8. Sarah Vluymans, Neil Mac Parthaláin, Chris Cornelis, and Yvan Saeys. 2019. Weight selection strategies for ordered weighted average based fuzzy rough sets. Information Sciences, 501:155–171. [↗](https://doi.org/10.1016/j.ins.2019.05.085)
9. Sahibsingh A Dudani. 1976. The distance-weighted k-nearest-neighbor rule. IEEE Transactions on Systems, Man, and Cybernetics, (4):325–327. [↗](https://doi.org/10.1109/tsmc.1976.5408784)
10. Bilal Ghanem, Jihen Karoui, Farah Benamara, Paolo Rosso, and Véronique Moriceau. 2020. Irony detection in a multilingual context. Advances in Information Retrieval, 12036:141. [find](https://scholar.google.com/scholar?q=Bilal+Ghanem%2C+Jihen+Karoui%2C+Farah+Benamara%2C+Paolo+Rosso%2C+and+V%C3%A9ronique+Moriceau%2E+2020%2E+Irony+detection+in+a+multilingual+context%2E+Advances+in+Information+Retrieval%2C+12036%3A141%2E)
11. Richard Jensen and Chris Cornelis. 2011. Fuzzy-rough nearest neighbour classification and prediction. Theoretical Computer Science, 412(42):5871–5884. [↗](https://doi.org/10.1016/j.tcs.2011.05.040)
12. Olha Kaminska, Chris Cornelis, and Veronique Hoste. 2021a. Fuzzy-rough nearest neighbour approaches for emotion detection in tweets. [find](https://scholar.google.com/scholar?q=Olha+Kaminska%2C+Chris+Cornelis%2C+and+Veronique+Hoste%2E+2021a%2E+Fuzzy%2Drough+nearest+neighbour+approaches+for+emotion+detection+in+tweets%2E)
13. Olha Kaminska, Chris Cornelis, and Veronique Hoste. 2021b. Nearest neighbour approaches for emotion detection in tweets. In Proceedings of the Eleventh Workshop on Computational Approaches to Subjectivity, Sentiment and Social Media Analysis, pages 203–212. [↗](https://doi.org/10.48550/arxiv.2107.05394)
14. Oliver Urs Lenz, Daniel Peralta, and Chris Cornelis. 2019. Scalable approximate frnn-owa classification. IEEE Transactions on Fuzzy Systems, 28(5):929–938. [↗](https://doi.org/10.1109/tfuzz.2019.2949769)
15. Oliver Urs Lenz, Daniel Peralta, and Chris Cornelis. 2020. fuzzy-rough-learn 0.1: a Python library for [find](https://scholar.google.com/scholar?q=Oliver+Urs+Lenz%2C+Daniel+Peralta%2C+and+Chris+Cornelis%2E+2020%2E+fuzzy%2Drough%2Dlearn+0%2E1%3A+a+Python+library+for)

*Updated 24 October 2026*