---
title: "I2C at SemEval-2022 Task 6 — paper by Diaz et al. (2022)"
description: "The authors, a team from Universidad de Huelva, took part in SubTask A of SemEval-2022 Task 6 (iSarcasmEval)… 0.2451 F1. 13 references."
authors: ["Pablo Gonzalez Diaz", "Pablo Cordon", "Jacinto Mata", "Victoria Pachón"]
year: 2022
venue: "Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval-2022)"
source: "https://aclanthology.org/2022.semeval-1.122/"
doi: "10.18653/v1/2022.semeval-1.122"
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/diaz-2022-i2c-semeval-2022-task-6/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/diaz-2022-i2c-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/8/) / I2C at SemEval-2022 Task 6: Intended Sarcasm Detection on Social Networks with Deep Learning

# I2C at SemEval-2022 Task 6: Intended Sarcasm Detection on Social Networks with Deep Learning

**Authors:** Pablo Gonzalez Diaz, Pablo Cordon, [Jacinto Mata](https://md.research.elamur.ai/free/en/authors/jacinto-mata/) and Victoria Pachón · **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-1.122/) · [PDF](https://aclanthology.org/2022.semeval-1.122.pdf) · [DOI 10.18653/v1/2022.semeval-1.122](https://doi.org/10.18653/v1/2022.semeval-1.122) · [OpenAlex](https://openalex.org/W4287887076)

**Citations:** 1 citations (OpenAlex, 2026-10-14)

## Summary
The authors, a team from Universidad de Huelva, took part in SubTask A of SemEval-2022 Task 6 (iSarcasmEval), a binary task: decide whether an English tweet is meant sarcastically. They built two training sets from the task data. One is the original, imbalanced tweet collection. The other pairs each sarcastic tweet with its author-written literal rephrase, so that a model sees the same message with and without sarcasm. Text was left untouched (casing, emojis and punctuation kept), on the view that sarcasm depends on surface form. About 30 configurations were compared: classical classifiers over CountVectorizer/TFIDF features, an LSTM with GloVe embeddings, and cased and uncased BERT. BERT-base-cased trained on the Tweet–Rephrase pairs scored best on their internal split (0.86 accuracy and F1). On the official test set, however, it reached only 0.2451 F1 for the sarcastic class, placing 30th. The large gap shows how hard intended sarcasm is and how misleading in-house validation can be.

## Key points
- **Data:** The original dataset has 3467 tweets (2600 non-sarcastic, 867 sarcastic). The Tweet–Rephrase set is balanced at 1734 rows: 867 sarcastic tweets, each paired with its literal rephrase labelled 0.
- **Models compared:**
  - LinearSVC, DecisionTree and RandomForest over CountVectorizer and TFIDF features;
  - an LSTM with 200d GloVe vectors (batch 32, 10 epochs, early stopping);
  - BERT-base-cased and BERT-base-uncased (batch 32, 5 epochs).
- **Evaluation setup:** All experiments used a stratified 80/20 split. The final model was retrained on the full training data once the test set was released.
- **Results on the original data:** BERT-base-cased reached 0.72 accuracy and 0.43 F1 (sarcastic class), ahead of uncased BERT (0.39 F1) and the LSTM (0.38 F1).
- **Results on Tweet–Rephrase and test:** BERT-base-cased reached 0.86, uncased BERT 0.84 and the LSTM 0.62 F1. The official test score was 0.2451 F1, ranking 30th.

## Interesting ideas
- **Contrastive minimal pairs:** Training on pairs of a sarcastic tweet and its literal restatement isolates the signal of sarcasm from the topic, since both members share content and differ only in stance.
- **Keep surface form:** Leaving case, emojis and punctuation in place, and preferring a cased model, follows from the view that sarcasm cues live in form; the cased model did beat the uncased one in both settings.
- **Validation that does not match the test task:** Scores near 0.86 on the paired data fell to 0.2451 on the real test. Balanced pairs make an easier task than spotting sarcasm in natural, skewed tweet streams.
- **The rephrase as a meaning anchor:** Each literal rephrase spells out what the sarcastic tweet actually intends, which gives a gold paraphrase of the implied meaning.

## For Mova

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

- **I1** (frames): The irony module in crate prag could be tested on such pairs, checking that the frame and speaker-attitude output flips between the two versions while the who-did-what-to-whom roles stay the same.
- **I2** (tokens): The en tokenizer should keep casing, emoji and repeated punctuation as features on tokens, so that later irony detection can still see them.
- **I3** (learn): Evaluation should always include a natural-distribution held-out set and negative controls, and not rely on validation data built the same way as the training data.
- **I4** (back): Such pairs could check whether generating text from the meaning graph of a sarcastic input produces something close to the literal rephrase.

*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

- [Universidad de Huelva](https://md.research.elamur.ai/free/en/universities/universidad-de-huelva/) — University, Huelva, Spain · 3 works · 8 citations

📚 Shelves (4): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 16,194 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 16,039 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 15,891 works

🧭 Topics (2): [English](https://md.research.elamur.ai/free/en/topics/english/) · 2,263 works · [😊 Sentiment & emotion](https://md.research.elamur.ai/free/en/topics/sentiment-emotion/) · 972 works

## Cite

Cite as: Pablo Gonzalez Diaz, Pablo Cordon, Jacinto Mata and Victoria Pachón. 2022. I2C at SemEval-2022 Task 6: Intended Sarcasm Detection on Social Networks with Deep Learning. 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{diaz2022i2c,
  title = {{I2C at SemEval-2022 Task 6: Intended Sarcasm Detection on Social Networks with Deep Learning}},
  author = {Pablo Gonzalez Diaz, Pablo Cordon, Jacinto Mata and Victoria Pachón},
  booktitle = {Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval-2022)},
  year = {2022},
  doi = {10.18653/v1/2022.semeval-1.122},
  url = {https://doi.org/10.18653/v1/2022.semeval-1.122}
}
```

- [← previous High aspect in the English <em>be going to</em> construction: Syntactic evidence](https://md.research.elamur.ai/free/en/papers/duncan-2022-high-aspect-english-em-be/)
- [next → JCT at SemEval-2022 Task 6-A: Sarcasm Detection in Tweets Written in English and Arabic using Preprocessing Methods and Word N-grams](https://md.research.elamur.ai/free/en/papers/kerner-2022-jct-semeval-2022-task-6/)
- [↑ back to papers of 2022](https://md.research.elamur.ai/free/en/papers/2022/page/8/)

## References

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

1. Abu Farha, Ibrahim, Silviu Oprea, Steven Wilson, and Walid Magdy. 2022. ["SemEval-2022 Task 6: iSarcasmEval Intended Sarcasm Detection in English and Arabic."Association for Computational Linguistics,](https://md.research.elamur.ai/free/en/papers/farha-2022-semeval-2022-task-6-isarcasmeval/).
2. Chollet, Francois and others. 2015. "Keras." . [https://github.com/fchollet/keras](https://github.com/fchollet/keras). [↗](https://github.com/fchollet/keras)
3. Devlin, Jacob, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2018. "[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/)." . https://arxiv.org/abs/1810.04805. [↗](https://arxiv.org/abs/1810.04805)
4. Harris, Charles R., K. Jarrod Millman, van der Walt, Stéfan J., Ralf Gommers, Pauli Virtanen, David Cournapeau, Eric Wieser, et al. 2020. "[Array Programming with NumPy](https://md.research.elamur.ai/free/en/papers/harris-2020-array-programming-numpy/)." Nature 585: 357–362. doi:10.1038/s41586-020-2649-2. [↗](https://doi.org/10.1038/s41586-020-2649-2)
5. Hochreiter, Sepp and Jürgen Schmidhuber. 1997. "Long Short-Term Memory." Neural Computation 9 (8): 1735-1780. doi:10.1162/neco.1997.9.8.1735. https://doi.org/10.1162/neco.1997.9.8.1735. Embeddings, Convolutional Neural Networks and Incremental Parsing.". [↗](https://doi.org/10.1162/neco.1997.9.8.1735)
6. Loper, Edward and Steven Bird. 2002. "NLTK: The Natural Language Toolkit." Philadelphia, Pennsylvania, Association for Computational Linguistics, . doi:10.3115/1118108.1118117. https://doi.org/10.3115/1118108.1118117 . [↗](https://doi.org/10.3115/1118108.1118117)
7. McKinney, Wes and others. 2010. "Data Structures for Statistical Computing in Python."Austin, TX, . [find](https://scholar.google.com/scholar?q=McKinney%2C+Wes+and+others%2E+2010%2E+%22Data+Structures+for+Statistical+Computing+in+Python%2E%22Austin%2C+TX%2C+%2E)
8. Pedregosa, Fabian and Varoquaux, Ga\el and Gramfort, Alexandre and Michel, Vincent and Thirion, Bertrand and Grisel, Olivier and Blondel, Mathieu and Prettenhofer, Peter and Weiss, Ron and Dubourg, Vincent and others. 2011. "Scikit-Learn: Machine Learning in Python." Journal of Machine Learning Research 12 (Oct): 2825-2830. [↗](https://doi.org/10.48550/arxiv.1201.0490)
9. Pennington, Jeffrey, Richard Socher, and Christopher Manning. 2014. "[Glove: Global Vectors for Word Representation](https://md.research.elamur.ai/free/en/papers/pennington-2014-glove-global-vectors-word-representation/)."01. doi:10.3115/v1/D14-1162. [↗](https://doi.org/10.3115/v1/d14-1162)
10. Shah, Bhumi and Margil Shah. 2021. "A Survey on Machine Learning and Deep Learning Based Approaches for Sarcasm Identification in Social Media."Springer Singapore, . [find](https://scholar.google.com/scholar?q=Shah%2C+Bhumi+and+Margil+Shah%2E+2021%2E+%22A+Survey+on+Machine+Learning+and+Deep+Learning+Based+Approaches+for+Sarcasm+Identification+in+Social+Media%2E%22Springer+Singapore%2C+%2E)
11. Suzuki, Shota, Ryohei Orihara, Yuichi Sei, Yasuyuki Tahara, and Akihiko Ohsuga. 2017. "Sarcasm Detection Method to Improve Review Analysis.". [find](https://scholar.google.com/scholar?q=Suzuki%2C+Shota%2C+Ryohei+Orihara%2C+Yuichi+Sei%2C+Yasuyuki+Tahara%2C+and+Akihiko+Ohsuga%2E+2017%2E+%22Sarcasm+Detection+Method+to+Improve+Review+Analysis%2E%22%2E)
12. Vaswani, Ashish, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. 2017. [Attention is all You Need](https://md.research.elamur.ai/free/en/papers/vaswani-2017-attention-all-need/).
13. Verma, Palak, Neha Shukla, and A. P. Shukla. 2021. "Techniques of Sarcasm Detection: A Review.". doi:10.1109/ICACITE51222.2021.9404585. Honnibal, Matthew and Ines Montani. "spaCy 2: Natural Language Understanding with Bloom [↗](https://doi.org/10.1109/icacite51222.2021.9404585)

*Updated 1 November 2026*