---
title: "IronyMagnet at SemEval-2018 Task 3: A Siamese network for"
description: "The authors treat irony as a clash between parts of a message, so they build IronyMagnet, a Siamese… 9 references, cited by 6 papers here."
authors: ["Aniruddha Ghosh", "Tony Veale"]
year: 2018
venue: "Proceedings of the 12th International Workshop on Semantic Evaluation"
source: "https://aclanthology.org/S18-1093/"
doi: "10.18653/v1/s18-1093"
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/ghosh-2018-ironymagnet-semeval-2018-task-3/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/ghosh-2018-ironymagnet-semeval-2018-task-3/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2018](https://md.research.elamur.ai/free/en/papers/2018/page/5/) / IronyMagnet at SemEval-2018 Task 3: A Siamese network for Irony detection in Social media

# IronyMagnet at SemEval-2018 Task 3: A Siamese network for Irony detection in Social media

**Authors:** [Aniruddha Ghosh](https://md.research.elamur.ai/free/en/authors/aniruddha-ghosh/) and [Tony Veale](https://md.research.elamur.ai/free/en/authors/tony-veale/) · **Year:** 2018 · **Venue:** [Proceedings of the 12th International Workshop on Semantic Evaluation](https://md.research.elamur.ai/free/en/venues/semeval/) ([SemEval 2018](https://md.research.elamur.ai/free/en/venues/semeval/2018/)) · **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/S18-1093/) · [PDF](https://aclanthology.org/S18-1093.pdf) · [DOI 10.18653/v1/s18-1093](https://doi.org/10.18653/v1/s18-1093) · [OpenAlex](https://openalex.org/W2805862862)

**Citations:** 22 citations (OpenAlex, 2026-10-14) · [6](https://md.research.elamur.ai/free/en/papers/ghosh-2018-ironymagnet-semeval-2018-task-3/cited-by/) in our library, of the 141,840 papers whose bibliographies we read

## Summary
The paper tackles irony recognition in English tweets for SemEval-2018 Task 3. The task had a binary subtask and, for the first time in a shared task, a fine-grained one with four labels: irony by polarity contrast, situational irony, other verbal irony, and non-irony. The authors treat irony as a clash between parts of a message, so they build IronyMagnet, a Siamese model. Each tweet is cut into two fragments, and each fragment goes through its own branch made of frozen GloVe embeddings and an LSTM. The element-wise difference of the two branch outputs is then classified by a dense layer with softmax. Because every admissible cut point becomes a separate training pair, the small training set of 3834 tweets is effectively multiplied. Text normalization and lexical abstraction through SentiWordNet and WordNet add robustness. The system reaches an F1 of roughly 0.72 on the binary subtask and 0.51 on the fine-grained one, clearly ahead of the authors' earlier CNN+LSTM composite model. The result suggests that modelling internal contrast explicitly helps figurative-language detection when data is scarce.

## Key points
- Data: 3834 training tweets and 784 test tweets. The binary split is balanced (1923 non-ironic vs 1911 ironic). In the fine-grained split, polarity contrast dominates (1390), with situational irony at 316 and other verbal irony at 205.
- Splitting scheme: each fragment must contain at least r = 3 words, so a tweet of n words yields n − 2r + 1 fragment pairs, each carrying the tweet's label.
- Preprocessing steps:
  - character elongations are collapsed and matched to the closest WordNet word by edit distance;
  - glued repeats are separated;
  - URLs are dropped;
  - hashtags are segmented into words;
  - emojis are replaced by their Unicode names;
  - rare non-WordNet tokens are removed.
- Augmentation: strongly polar words are replaced with the tokens "positive"/"negative" using SentiWordNet, and words are replaced by their WordNet category.
- Setup and results:
  - Model: LSTM with 32 units and recurrent dropout 0.5, a 16-unit ReLU dense layer, Xavier init, Adam (lr 0.001), built in Keras.
  - Binary subtask: F1 0.7234 vs 0.5257 for the Ghosh and Veale (2016) model.
  - Fine-grained subtask: F1 0.5074 vs 0.3988.

## Interesting ideas
- Instead of classifying a sentence as a whole, compare two of its halves through shared-weight encoders and classify the gap, which turns incongruity into an explicit signal.
- Enumerating all cut points of an example, subject to a minimum fragment length, is a cheap way to multiply scarce labelled data without new annotation.
- Abstracting surface words into sentiment tokens or ontology categories lets a model learn contrast patterns that are independent of specific vocabulary.
- Social-media noise is normalized into dictionary words before modelling: elongations are snapped to the nearest WordNet entry, hashtags are segmented, and emojis are spelled out by name.
- The error analysis shows that a contrast between two fragments fails when the incongruous elements sit very close together or depend on world knowledge, as with weather in Minnesota.

## For Mova

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

- **I1** (frames): The irony module in prag could compare the evaluation frame of one clause against the situation frame of another and flag a polarity clash, giving an explainable reason rather than a bare label.
- **I2** (learn): When gold irony or implicature examples are few, clause-boundary splits of each sentence could generate many silver contrast pairs for training and testing.
- **I3** (world): Mapping entities and predicates to WordNet categories and polarity values in the compiled knowledge layer would let clash detection generalize across topics.
- **I4** (tokens): The en tokenizer could include such reversible normalizations so that tweets and informal text produce cleaner lemmas and tags for the parser.
- **I5** (answer): Evaluation of irony-sensitive readers should include adjacency and world-knowledge cases as separate test slices, so that failures are attributed to the right stage.

*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

- [University College Dublin](https://md.research.elamur.ai/free/en/universities/university-college-dublin/) — University, Dublin, Ireland · 10 works · 1,520 citations

📚 Shelves (4): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 9,085 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 8,888 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/) · 609 works

## Cite

Cite as: Aniruddha Ghosh and Tony Veale. 2018. IronyMagnet at SemEval-2018 Task 3: A Siamese network for Irony detection in Social media. Proceedings of the 12th International Workshop on Semantic Evaluation.

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

```bibtex
@inproceedings{ghosh2018ironymagnet,
  title = {{IronyMagnet at SemEval-2018 Task 3: A Siamese network for Irony detection in Social media}},
  author = {Aniruddha Ghosh and Tony Veale},
  booktitle = {Proceedings of the 12th International Workshop on Semantic Evaluation},
  year = {2018},
  doi = {10.18653/v1/s18-1093},
  url = {https://doi.org/10.18653/v1/s18-1093}
}
```

- [← previous A Systematic Review of Automated Grammar Checking in English Language](https://md.research.elamur.ai/free/en/papers/soni-2018-systematic-review-automated-grammar-checking/)
- [next → NLPRL-IITBHU at SemEval-2018 Task 3: Combining Linguistic Features and Emoji pre-trained CNN for Irony Detection in Tweets](https://md.research.elamur.ai/free/en/papers/rangwani-2018-nlprl-iitbhu-semeval-2018-task/)
- [↑ back to papers of 2018](https://md.research.elamur.ai/free/en/papers/2018/page/5/)

## References

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

1. Jane Bromley, Isabelle Guyon, Yann LeCun, Eduard Säckinger, and Roopak Shah. 1994. Signature verification using a” siamese” time delay neural network. In Advances in Neural Information Processing Systems, pages 737–744. [↗](https://doi.org/10.1142/9789812797926_0003)
2. George Lakoff. 1993. metaphor. The contemporary theory of Stephanie Lukin and Marilyn Walker. 2013. [Really? well. apparently bootstrapping improves the performance of sarcasm and nastiness classifiers for online dialogue](https://md.research.elamur.ai/free/en/papers/lukin-2013-really-well-apparently-bootstrapping-improves/). In Proceedings of the Workshop on Language Analysis in Social Media, pages 30–40.
3. Paula Carvalho, Luı́s Sarmento, Mário J Silva, and Eugénio De Oliveira. 2009. Clues for detecting irony in user-generated contents: oh...!! it’s so easy;-. In Proceedings of the 1st international CIKM workshop on Topic-sentiment analysis for mass opinion, pages 53–56. ACM. [find](https://scholar.google.com/scholar?q=Paula+Carvalho%2C+Lu%C4%B1%CC%81s+Sarmento%2C+Ma%CC%81rio+J+Silva%2C+and+Euge%CC%81nio+De+Oliveira%2E+2009%2E+Clues+for+detecting+irony+in+user%2Dgenerated+contents%3A+oh%2E%2E%2E%21%21+it%E2%80%99s+so+easy%3B%2D%2E+In+Proceedings+of+the+1st+international+CI)
4. Abhijit Mishra, Kuntal Dey, and Pushpak Bhattacharyya. 2017. [Learning cognitive features from gaze data for sentiment and sarcasm classification using convolutional neural network](https://md.research.elamur.ai/free/en/papers/mishra-2017-learning-cognitive-features-gaze-data/). In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), volume 1, pages 377–387.
5. Ellen Riloff, Ashequl Qadir, Prafulla Surve, Lalin-dra 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 EMNLP, volume 13, pages 704–714.
6. Oren Tsur, Dmitry Davidov, and Ari Rappoport. 2010. Icwsm-a great catchy name: Semi-supervised recognition of sarcastic sentences in online product reviews. In ICWSM. [find](https://scholar.google.com/scholar?q=Oren+Tsur%2C+Dmitry+Davidov%2C+and+Ari+Rappoport%2E+2010%2E+Icwsm%2Da+great+catchy+name%3A+Semi%2Dsupervised+recognition+of+sarcastic+sentences+in+online+product+reviews%2E+In+ICWSM%2E)
7. A. Utsumi. 2000. Verbal irony as implicit display of ironic environment: Distinguishing ironic utterances from nonirony. Journal of Pragmatics. [↗](https://doi.org/10.1016/s0378-2166(99)00116-2)
8. Cynthia Van Hee, Els Lefever, and Véronique Hoste. 2016. [Monday mornings are my fave:)# not exploring the automatic recognition of irony in english tweets](https://md.research.elamur.ai/free/en/papers/hee-2016-monday-mornings-are-my-fave/). In Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 2730–2739.
9. Meishan Zhang, Yue Zhang, and Guohong Fu. 2016. [Tweet sarcasm detection using deep neural network](https://md.research.elamur.ai/free/en/papers/zhang-2016-tweet-sarcasm-detection-deep-neural/). In Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 2449–2460.

*Updated 24 October 2026*