---
title: "NULL at SemEval-2022 Task 6 — paper by Rahgouy et al. (2022)"
description: "The authors argue that a contextual encoder alone poorly captures intent, so they add class-level style… 29 references, cited by 2 papers here."
authors: ["Mostafa Rahgouy", "Hamed Babaei Giglou", "Taher Rahgooy", "Cheryl Seals"]
year: 2022
venue: "Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval-2022)"
source: "https://aclanthology.org/2022.semeval-1.120/"
doi: "10.18653/v1/2022.semeval-1.120"
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/rahgouy-2022-null-semeval-2022-task-6/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/rahgouy-2022-null-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/) / NULL at SemEval-2022 Task 6: Intended Sarcasm Detection Using Stylistically Fused Contextualized Representation and Deep Learning

# NULL at SemEval-2022 Task 6: Intended Sarcasm Detection Using Stylistically Fused Contextualized Representation and Deep Learning

**Authors:** Mostafa Rahgouy, Hamed Babaei Giglou, Taher Rahgooy and Cheryl Seals · **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-1.120/) · [PDF](https://aclanthology.org/2022.semeval-1.120.pdf) · [DOI 10.18653/v1/2022.semeval-1.120](https://doi.org/10.18653/v1/2022.semeval-1.120) · [OpenAlex](https://openalex.org/W4287887240)

**Citations:** 4 citations (OpenAlex, 2026-10-14) · 2 in our library, of the 141,840 papers whose bibliographies we read

## Summary
The paper addresses SubTask A of SemEval-2022 Task 6 (iSarcasmEval): deciding whether a short social-media text is sarcastic, in English and Arabic. The labels come from the writers themselves. The authors argue that a contextual encoder alone poorly captures intent, so they add class-level style signals.

Their Char-LDSE representation converts TF-IDF-weighted character n-grams (plus word unigrams for English) into per-class term-probability statistics. These are summarized as a 104-dimensional vector per text for each class. The sarcastic-class vector is concatenated with XLM-RoBERTa base embeddings and fed to a CNN. The non-sarcastic vector, joined the same way, goes to a BiLSTM. The two outputs are merged and passed to a small feed-forward classifier.

Dev-set experiments compare linear baselines, LDSE-only, XLM-RoBERTa-only and combined models. Weighted cross-entropy and TF-IDF-based augmentation add gains, especially for Arabic. Error analysis shows that false positives dominate, often in replies that lack thread context. The work is a practical example of fusing cheap stylistic statistics with pretrained multilingual representations.

## Key points
- Char-LDSE uses character n-grams of range (2, 3) for both languages, plus word unigrams for English only. Each text gets max, min, std, average and 100 quantiles of its class-weighted term scores (104 features).
- Each 104-dim LDSE vector is joined with 768-dim XLM-RoBERTa base embeddings, giving 872-dim inputs. Two Conv2D layers (kernels (4, 872) and (3, 872), 100 channels each) and a BiLSTM each produce 200 features. The resulting 400-dim vector goes through a 400→200→2 head with GELU.
- A 10% split of the training data served as the dev set for tuning and model selection. The official data carries author-provided sarcasm labels.
- On dev, weighted cross-entropy added about 1% (English) and 2% (Arabic). TF-IDF augmentation added about 1% and 4%. Together they added about 2% and 6%.
- Analysis of 436 English false positives found three recurring problems:
  - missing conversational context;
  - apparent label noise and preprocessing damage, such as a price/ticker line reduced to fragments;
  - emojis, present in roughly 60% of these cases.

## Interesting ideas
- Turn class-conditional term weights into a fixed-size profile per text (extremes, spread, quantiles) so a small labeled set yields a compact style signal regardless of text length.
- Feed each class-specific view to its own encoder (one view to a CNN for local patterns, the other to a BiLSTM for longer dependencies), then fuse the results. This lets complementary evidence stay separate until the decision.
- The error analysis shows that many "sarcastic" false alarms are replies stripped of the thread they answer. This suggests the task is partly undecidable without context.
- Rendering emojis as plain words, and aggressively cleaning numbers and tickers, discarded cues and produced meaningless fragments.
- An ablation ladder (n-gram baselines → each representation alone → combination → loss weighting → augmentation) isolates what each component contributes.

## For Mova

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

- **I1** (frames): the irony component could attach such a profile as a secondary cue next to literal-vs-intended incongruity, and list which n-grams pushed the score.
- **I2** (learn): when training a classifier over parser features, keeping distinct evidence channels separate until a late merge would make per-channel ablations and explanations easier.
- **I3** (frames): the irony detector should be able to output "needs discourse context" rather than forcing a sarcastic/literal label when the antecedent post is missing.
- **I4** (tokens): the tokenizer should keep emojis, currency amounts and tickers as typed tokens with their own features, so later stages can reason about them instead of losing them.
- **I5** (learn): evaluation reports could follow the same ladder so each added module shows its own measured gain against simpler controls.

*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

- [Auburn University](https://md.research.elamur.ai/free/en/universities/auburn-university/) — University, Auburn, United States · 2 works · 151 citations

Also on the byline: [University of Tabriz](https://ror.org/01papkj44) · [University of West Florida](https://ror.org/002w4zy91).

📚 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 (3): [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 · [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/) · 1,484 works

## Cite

Cite as: Mostafa Rahgouy, Hamed Babaei Giglou, Taher Rahgooy and Cheryl Seals. 2022. NULL at SemEval-2022 Task 6: Intended Sarcasm Detection Using Stylistically Fused Contextualized Representation and 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{rahgouy2022null,
  title = {{NULL at SemEval-2022 Task 6: Intended Sarcasm Detection Using Stylistically Fused Contextualized Representation and Deep Learning}},
  author = {Mostafa Rahgouy, Hamed Babaei Giglou, Taher Rahgooy and Cheryl Seals},
  booktitle = {Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval-2022)},
  year = {2022},
  doi = {10.18653/v1/2022.semeval-1.120},
  url = {https://doi.org/10.18653/v1/2022.semeval-1.120}
}
```

- [← previous 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/)
- [next → SSR: Utilizing Simplified Stance Reasoning Process for Robust Stance Detection](https://md.research.elamur.ai/free/en/papers/yuan-2022-ssr-utilizing-simplified-stance-reasoning/)
- [↑ back to papers of 2022](https://md.research.elamur.ai/free/en/papers/2022/page/5/)

## References

29 references, 9 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, Steven Wilson, and Walid Magdy. 2022. [SemEval-2022 Task 6: iSarcasmEval, Intended Sarcasm Detection in English and Arabic](https://md.research.elamur.ai/free/en/papers/farha-2022-semeval-2022-task-6-isarcasmeval/). In Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval-2022). Association for Computational Linguistics.
2. Abien Fred Agarap. 2019. Deep learning using rectified linear units (relu). [↗](https://doi.org/10.48550/arxiv.1803.08375)
3. Ameeta Agrawal, Aijun An, and Manos Papagelis. 2020. Leveraging transitions of emotions for sarcasm detection. In Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval, pages 1505–1508. [find](https://scholar.google.com/scholar?q=Ameeta+Agrawal%2C+Aijun+An%2C+and+Manos+Papagelis%2E+2020%2E+Leveraging+transitions+of+emotions+for+sarcasm+detection%2E+In+Proceedings+of+the+43rd+International+ACM+SIGIR+Conference+on+Research+and+Development)
4. Ravinder Ahuja and SC Sharma. 2021. Transformer-based word embedding with cnn model to detect sarcasm and irony. Arabian Journal for Science and Engineering, pages 1–14. [↗](https://doi.org/10.1007/s13369-021-06193-3)
5. Nastaran Babanejad, Heidar Davoudi, Aijun An, and Manos Papagelis. 2020. [Affective and contextual embedding for sarcasm detection](https://md.research.elamur.ai/free/en/papers/babanejad-2020-affective-contextual-embedding-sarcasm-detection/). In Proceedings of the 28th International Conference on Computational Linguistics, pages 225–243.
6. Yann LeCun, Patrick Haffner, Léon Bottou, and Yoshua Bengio. 1999. Object recognition with gradient-based learning. In Shape, Contour and Grouping in Computer Vision, page 319, Berlin, Heidelberg. Springer-Verlag. [find](https://scholar.google.com/scholar?q=Yann+LeCun%2C+Patrick+Haffner%2C+Le%CC%81on+Bottou%2C+and+Yoshua+Bengio%2E+1999%2E+Object+recognition+with+gradient%2Dbased+learning%2E+In+Shape%2C+Contour+and+Grouping+in+Computer+Vision%2C+page+319%2C+Berlin%2C+Heidelberg%2E+Sp)
7. Meimei Li, Chen Lang, Min Yu, Yue Lu, Chao Liu, Jianguo Jiang, and Weiqing Huang. 2020. Scx-sd: Semi-supervised method for contextual sarcasm detection. In International Conference on Knowledge Science, Engineering and Management, pages 288– 299. Springer. [find](https://scholar.google.com/scholar?q=Meimei+Li%2C+Chen+Lang%2C+Min+Yu%2C+Yue+Lu%2C+Chao+Liu%2C+Jianguo+Jiang%2C+and+Weiqing+Huang%2E+2020%2E+Scx%2Dsd%3A+Semi%2Dsupervised+method+for+contextual+sarcasm+detection%2E+In+International+Conference+on+Knowledge+Scienc)
8. Francesco Barbieri, Horacio Saggion, and Francesco Ronzano. 2014. [Modelling sarcasm in twitter, a novel approach](https://md.research.elamur.ai/free/en/papers/barbieri-2014-modelling-sarcasm-twitter-novel-approach/). In proceedings of the 5th workshop on computational approaches to subjectivity, sentiment and social media analysis, pages 50–58.
9. Santosh Kumar Bharti, Korra Sathya Babu, and Sanjay Kumar Jena. 2015. Parsing-based sarcasm sentiment recognition in twitter data. In 2015 IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining (ASONAM), pages 1373– 1380. IEEE. [find](https://scholar.google.com/scholar?q=Santosh+Kumar+Bharti%2C+Korra+Sathya+Babu%2C+and+Sanjay+Kumar+Jena%2E+2015%2E+Parsing%2Dbased+sarcasm+sentiment+recognition+in+twitter+data%2E+In+2015+IEEE%2FACM+International+Conference+on+Advances+in+Social+Netwo)
10. Alexis Conneau, Kartikay Khandelwal, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer, and Veselin Stoyanov. 2020. Unsupervised cross-lingual representation learning at scale. [find](https://scholar.google.com/scholar?q=Alexis+Conneau%2C+Kartikay+Khandelwal%2C+Naman+Goyal%2C+Vishrav+Chaudhary%2C+Guillaume+Wenzek%2C+Francisco+Guzma%CC%81n%2C+Edouard+Grave%2C+Myle+Ott%2C+Luke+Zettlemoyer%2C+and+Veselin+Stoyanov%2E+2020%2E+Unsupervised+cross%2Dling)
11. Hamed Babaei Giglou, Taher Rahgooy, Jafar Raz-mara Mostafa Rahgouy, and Zahra Rahgooy. 2021. Profiling Haters on Twitter using Statistical and Contextualized Embeddings—Notebook for PAN at CLEF 2021. In CLEF 2021 Labs and Workshops, Notebook Papers. CEUR-WS.org. [find](https://scholar.google.com/scholar?q=Hamed+Babaei+Giglou%2C+Taher+Rahgooy%2C+Jafar+Raz%2Dmara+Mostafa+Rahgouy%2C+and+Zahra+Rahgooy%2E+2021%2E+Profiling+Haters+on+Twitter+using+Statistical+and+Contextualized+Embeddings%E2%80%94Notebook+for+PAN+at+CLEF+2021%2E+)
12. José Ángel González, Lluı́s-F Hurtado, and Ferran Pla. 2020. Transformer based contextualization of pre-trained word embeddings for irony detection in twitter. Information Processing & Management, 57(4):102262. [↗](https://doi.org/10.1016/j.ipm.2020.102262)
13. Hunter Gregory, Steven Li, Pouya Mohammadi, Natalie Tarn, Rachel Draelos, and Cynthia Rudin. 2020. [A transformer approach to contextual sarcasm detection in twitter](https://md.research.elamur.ai/free/en/papers/gregory-2020-transformer-approach-contextual-sarcasm-detection/). In Proceedings of the Second Workshop on Figurative Language Processing, pages 270–275.
14. Hacker. 2011. Merriam-webster.com (8 may 2011). [find](https://scholar.google.com/scholar?q=Hacker%2E+2011%2E+Merriam%2Dwebster%2Ecom+%288+may+2011%29%2E)
15. Devamanyu Hazarika, Soujanya Poria, Sruthi Gorantla, Erik Cambria, Roger Zimmermann, and Rada Mihalcea. 2018. [Cascade: Contextual sarcasm detection in online discussion forums](https://md.research.elamur.ai/free/en/papers/hazarika-2018-cascade-contextual-sarcasm-detection-online/). arXiv preprint arXiv:1805.06413. [↗](https://arxiv.org/abs/1805.06413)
16. Dan Hendrycks and Kevin Gimpel. 2020. Gaussian error linear units (gelus). [find](https://scholar.google.com/scholar?q=Dan+Hendrycks+and+Kevin+Gimpel%2E+2020%2E+Gaussian+error+linear+units+%28gelus%29%2E)
17. Avinash Kumar, Vishnu Teja Narapareddy, Pranjal Gupta, Veerubhotla Aditya Srikanth, Lalita Bhanu Murthy Neti, and Aruna Malapati. 2021. Adversarial and auxiliary features-aware bert for sarcasm detection. In 8th ACM IKDD CODS and 26th COMAD, pages 163–170. [find](https://scholar.google.com/scholar?q=Avinash+Kumar%2C+Vishnu+Teja+Narapareddy%2C+Pranjal+Gupta%2C+Veerubhotla+Aditya+Srikanth%2C+Lalita+Bhanu+Murthy+Neti%2C+and+Aruna+Malapati%2E+2021%2E+Adversarial+and+auxiliary+features%2Daware+bert+for+sarcasm+detect)
18. Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, and Veselin Stoyanov. 2019. Roberta: A robustly optimized bert pretraining approach. [find](https://scholar.google.com/scholar?q=Yinhan+Liu%2C+Myle+Ott%2C+Naman+Goyal%2C+Jingfei+Du%2C+Mandar+Joshi%2C+Danqi+Chen%2C+Omer+Levy%2C+Mike+Lewis%2C+Luke+Zettlemoyer%2C+and+Veselin+Stoyanov%2E+2019%2E+Roberta%3A+A+robustly+optimized+bert+pretraining+approach%2E)
19. Edward Ma. 2019. Nlp augmentation. [https://github.com/makcedward/nlpaug](https://github.com/makcedward/nlpaug). [↗](https://github.com/makcedward/nlpaug)
20. Silviu Oprea and Walid Magdy. 2020. [isarcasm: A dataset of intended sarcasm](https://md.research.elamur.ai/free/en/papers/oprea-2020-isarcasm-dataset-intended-sarcasm/).
21. Rolandos Alexandros Potamias, Georgios Siolas, and Andreas-Georgios Stafylopatis. 2020. [A transformer-based approach to irony and sarcasm detection](https://md.research.elamur.ai/free/en/papers/potamias-2019-transformer-based-approach-irony-sarcasm/). Neural Computing and Applications, 32(23):17309– 17320. [↗](https://doi.org/10.1007/s00521-020-05102-3)
22. Francisco Rangel, Marc Franco-Salvador, and Paolo Rosso. 2017. A low dimensionality representation for language variety identification. [↗](https://doi.org/10.48550/arxiv.1705.10754)
23. Ellen Riloff, Ashequl Qadir, Prafulla Surve, Lalindra De Silva, Nathan Gilbert, and Ruihong Huang. 2013. [Sarcasm as contrast between a positive sentiment and negative situation](https://md.research.elamur.ai/free/en/papers/riloff-2013-sarcasm-as-contrast-between-positive/). In Proceedings of the 2013 conference on empirical methods in natural language processing, pages 704–714.
24. M. Schuster and K.K. Paliwal. 1997. Bidirectional recurrent neural networks. IEEE Transactions on Signal Processing, 45(11):2673–2681. [↗](https://doi.org/10.1109/78.650093)
25. Tony Veale and Yanfen Hao. 2010. Detecting ironic intent in creative comparisons. In ECAI 2010, pages 765–770. IOS Press. [↗](https://doi.org/10.3233/978-1-60750-606-5-765)
26. Byron C Wallace, Laura Kertz, Eugene Charniak, et al. 2014. [Humans require context to infer ironic intent (so computers probably do, too)](https://md.research.elamur.ai/free/en/papers/wallace-2014-humans-require-context-infer-ironic/). In Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 512–516.
27. Janet CE Watson et al. 2011. Word stress in arabic. [↗](https://doi.org/10.1002/9781444335262.wbctp0124)
28. Deirdre Wilson. 2006. The pragmatics of verbal irony: Echo or pretence? Lingua, 116(10):1722–1743. Language in Mind: A Tribute to Neil Smith on the Occasion of his Retirement. [↗](https://doi.org/10.1016/j.lingua.2006.05.001)
29. Hamed Yaghoobian, Hamid R. Arabnia, and Khaled Rasheed. 2021. Sarcasm detection: A comparative study. Dataset English Train Dev Test Arabic Train Dev Test Sarcastic Non-Sarcastic Overall 772 95 200 2349 252 1200 3121 347 1400 678 67 200 2113 244 1200 2791 311 1400 Parameter Batch Size Learning Rate Epoch Loss Function Weighted Loss Optimizer Weight Decay Dropout [↗](https://doi.org/10.48550/arxiv.2107.02276)

*Updated 24 October 2026*