---
title: "NTUA-SLP at SemEval-2018 Task 3: Tracking Ironic Tweets"
description: "The paper describes the NTUA-SLP entry to SemEval-2018 Task 3, which asks systems to find irony… 29 references, cited by 20 papers here."
authors: ["Christos Baziotis", "Athanasiou Nikolaos", "Pinelopi Papalampidi", "Athanasia Kolovou", "Georgios Paraskevopoulos", "Nikolaos Ellinas", "Alexandros Potamianos"]
year: 2018
venue: "Proceedings of the 12th International Workshop on Semantic Evaluation"
source: "https://aclanthology.org/S18-1100/"
doi: "10.18653/v1/s18-1100"
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/baziotis-2018-ntua-slp-semeval-2018-task-s18-1100/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/baziotis-2018-ntua-slp-semeval-2018-task-s18-1100/)

[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/4/) / NTUA-SLP at SemEval-2018 Task 3: Tracking Ironic Tweets using Ensembles of Word and Character Level Attentive RNNs

# NTUA-SLP at SemEval-2018 Task 3: Tracking Ironic Tweets using Ensembles of Word and Character Level Attentive RNNs

**Authors:** [Christos Baziotis](https://md.research.elamur.ai/free/en/authors/christos-baziotis/), Athanasiou Nikolaos, Pinelopi Papalampidi, [Athanasia Kolovou](https://md.research.elamur.ai/free/en/authors/athanasia-kolovou/), Georgios Paraskevopoulos, Nikolaos Ellinas and Alexandros Potamianos · **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/S18-1100/) · [PDF](https://aclanthology.org/S18-1100.pdf) · [DOI 10.18653/v1/s18-1100](https://doi.org/10.18653/v1/s18-1100) · [OpenAlex](https://openalex.org/W2798264625)

**Citations:** 54 citations (OpenAlex, 2026-10-14) · [20](https://md.research.elamur.ai/free/en/papers/baziotis-2018-ntua-slp-semeval-2018-task-s18-1100/cited-by/) in our library, of the 141,840 papers whose bibliographies we read

## Summary
The paper describes the NTUA-SLP entry to SemEval-2018 Task 3, which asks systems to find irony in English tweets. Subtask A is a binary decision (ironic or not). Subtask B separates three irony types (polarity clash, other verbal irony, situational irony) from non-ironic tweets. The team trained two independent two-layer BiLSTM classifiers, one over words and one over characters, each with a self-attention layer. They then combined the two by averaging their predictions (late fusion). The word model starts from word2vec vectors trained on 550 million unlabeled tweets. Text passes through the ekphrasis preprocessor. No lexicons, handcrafted features or extra labeled data were used. The official submissions placed second in both subtasks. After the deadline, the authors stopped normalizing the embedding vectors, and the improved models would have ranked first in both. The work shows that two cheap views of the same text, combined late, can match feature-engineered irony detectors. Attention heat-maps give some insight into which tokens drive each decision.

## Key points
- The unlabeled corpus holds 550M English tweets from Apr 2014 to Jun 2017. Embeddings were trained with word2vec skip-gram in Gensim, using negative sampling 5 and minimum count 20, giving an 800,000-word vocabulary.
- ekphrasis does Twitter-aware tokenization, spell correction and hashtag segmentation. Both spell correction and segmentation use the Viterbi algorithm, with uni/bigram priors taken from the unlabeled corpus. It also adds annotation tags such as `<allcaps>`, `<repeated>` and `<hashtag>`.
- Official F1 scores were 0.6719 for Subtask A and 0.4959 for Subtask B, both second place. The abstract reports 42 and 31 teams, while the result tables list 43 and 29.
- After the competition, the averaged word+char ensemble reached F1 0.7856 (A) and 0.5358 (B). These numbers are averaged over 10 runs. They beat the TF-IDF BOW and N-BOW baselines, which used a linear SVM with C = 0.6.
- On its own, the word-level model was the strongest. The char-level model was weaker than the baselines in Subtask B but still added some value to the ensemble.

## Interesting ideas
- The preprocessor marks features like shouting, elongation, emphasis and hashtags with explicit tags instead of deleting them, so cues that carry irony stay in the input.
- Hashtags are split into words with a Viterbi search, using word statistics from a large in-domain corpus as priors, so compounds like "#TwinPeaks" become readable words.
- Irony is split into polarity clash, other verbal irony and situational irony (an expectation that fails), which is a richer target than a single "ironic" label.
- A word-level and a character-level model, each weaker alone, are averaged after training, and their attention maps show they focus on different tokens.
- The authors found that normalizing embedding vectors hurt performance, because vector length carries meaning as well as direction. They also reported results averaged over 10 training runs.

## For Mova

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

- **I1** (tokens): The en tokenizer could attach similar markup flags (caps, repeated punctuation, emphasis) as token features, so later stages can read them as pragmatic signals.
- **I2** (tokens): A corpus-prior Viterbi segmenter could split run-together tokens and hashtags before tagging, yielding real lemmas instead of unknown words.
- **I3** (frames): The prag irony module could use this three-way scheme, checking for a clash between literal and intended evaluation, or for a violated expectation in the world model.
- **I4** (learn): Witness parsers could be chosen on purpose to have different input granularities, and the points where they disagree could flag silver data for teacher review.
- **I5** (learn): Evaluations should average several seeds and treat preprocessing choices like vector normalization as tested variables, not defaults.

*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

- [Athens University of Economics and Business](https://md.research.elamur.ai/free/en/universities/athens-university-of-economics-and-business/) — University, Athens, Greece · 15 works · 9,302 citations
- [National Technical University of Athens](https://md.research.elamur.ai/free/en/universities/national-technical-university-of-athens/) — University, Athens, Greece · 2 works · 386 citations

Also on the byline: [National and Kapodistrian University of Athens](https://ror.org/04gnjpq42).

📚 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 (1): [English](https://md.research.elamur.ai/free/en/topics/english/) · 1,373 works

## Cite

Cite as: Christos Baziotis, Athanasiou Nikolaos, Pinelopi Papalampidi, Athanasia Kolovou, Georgios Paraskevopoulos, Nikolaos Ellinas and Alexandros Potamianos. 2018. NTUA-SLP at SemEval-2018 Task 3: Tracking Ironic Tweets using Ensembles of Word and Character Level Attentive RNNs. 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{baziotis2018ntua,
  title = {{NTUA-SLP at SemEval-2018 Task 3: Tracking Ironic Tweets using Ensembles of Word and Character Level Attentive RNNs}},
  author = {Christos Baziotis, Athanasiou Nikolaos, Pinelopi Papalampidi, Athanasia Kolovou, Georgios Paraskevopoulos, Nikolaos Ellinas and Alexandros Potamianos},
  booktitle = {Proceedings of the 12th International Workshop on Semantic Evaluation},
  year = {2018},
  doi = {10.18653/v1/s18-1100},
  url = {https://doi.org/10.18653/v1/s18-1100}
}
```

- [← previous RNNs as psycholinguistic subjects: Syntactic state and grammatical dependency](https://md.research.elamur.ai/free/en/papers/futrell-2018-rnns-psycholinguistic-subjects-syntactic-state/)
- [next → Stance Detection in Fake News A Combined Feature Representation](https://md.research.elamur.ai/free/en/papers/ghanem-2018-stance-detection-fake-news-combined/)
- [↑ back to papers of 2018](https://md.research.elamur.ai/free/en/papers/2018/page/4/)

## References

29 references, 11 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Michael Heilman, Dani Yogatama, Jeffrey Flanigan, and Noah A. Smith. 2011. [Part-of-speech tagging for twitter: Annotation, features, and experiments](https://md.research.elamur.ai/free/en/papers/gimpel-2011-part-speech-tagging-twitter-annotation/). In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies: Short Papers-Volume 2, pages 42–47. Association for Computational Linguistics.
2. Dzmitry Bahdanau, Kyunghyun Cho, and Yoshua Bengio. 2014. [Neural machine translation by jointly learning to align and translate](https://md.research.elamur.ai/free/en/papers/bahdanau-2014-neural-machine-translation-jointly-learning/). CoRR, abs/1409.0473. [↗](https://arxiv.org/abs/1409.0473)
3. Francesco Barbieri and Horacio Saggion. 2014. Automatic detection of irony and humour in twitter. In ICCC, pages 155–162. [find](https://scholar.google.com/scholar?q=Francesco+Barbieri+and+Horacio+Saggion%2E+2014%2E+Automatic+detection+of+irony+and+humour+in+twitter%2E+In+ICCC%2C+pages+155%E2%80%93162%2E)
4. Sepp Hochreiter, Yoshua Bengio, Paolo Frasconi, and Jürgen Schmidhuber. 2001. Gradient Flow in Recurrent Nets: The Difficulty of Learning Long-Term Dependencies. A field guide to dynamical recurrent neural networks. IEEE Press. [find](https://scholar.google.com/scholar?q=Sepp+Hochreiter%2C+Yoshua+Bengio%2C+Paolo+Frasconi%2C+and+J%C3%BCrgen+Schmidhuber%2E+2001%2E+Gradient+Flow+in+Recurrent+Nets%3A+The+Difficulty+of+Learning+Long%2DTerm+Dependencies%2E+A+field+guide+to+dynamical+recurrent+n)
5. Christos Baziotis, Nikos Pelekis, and Christos Doulk-eridis. 2017. [Datastories at semeval-2017 task 4: Deep lstm with attention for message-level and topic-based sentiment analysis](https://md.research.elamur.ai/free/en/papers/baziotis-2017-datastories-semeval-2017-task-4/). In Proceedings of the 11th International Workshop on Semantic Evaluation (SemEval-2017), pages 747–754.
6. Sepp Hochreiter and Jürgen Schmidhuber. 1997. Long short-term memory. Neural computation, 9(8):1735–1780. [↗](https://doi.org/10.1162/neco.1997.9.8.1735)
7. Yoshua Bengio, Patrice Simard, and Paolo Frasconi. 1994. Learning long-term dependencies with gradient descent is difficult. IEEE transactions on neural networks, 5(2):157–166. [↗](https://doi.org/10.1109/72.279181)
8. Aditya Joshi, Pushpak Bhattacharyya, and Mark J Carman. 2017. [Automatic sarcasm detection: A survey](https://md.research.elamur.ai/free/en/papers/joshi-2016-automatic-sarcasm-detection-survey/). ACM Computing Surveys (CSUR), 50(5):73. [↗](https://doi.org/10.48550/arxiv.1602.03426)
9. James Bergstra, Daniel Yamins, and David D. Cox. 2013. Making a Science of Model Search: Hyperparameter Optimization in Hundreds of Dimensions for Vision Architectures. ICML (1), 28:115–123. [find](https://scholar.google.com/scholar?q=James+Bergstra%2C+Daniel+Yamins%2C+and+David+D%2E+Cox%2E+2013%2E+Making+a+Science+of+Model+Search%3A+Hyperparameter+Optimization+in+Hundreds+of+Dimensions+for+Vision+Architectures%2E+ICML+%281%29%2C+28%3A115%E2%80%93123%2E)
10. Aditya Joshi, Vaibhav Tripathi, Kevin Patel, Pushpak Bhattacharyya, and Mark Carman. 2016. [Are word embedding-based features useful for sarcasm detection?](https://md.research.elamur.ai/free/en/papers/joshi-2016-are-word-embedding-based-features/) arXiv preprint arXiv:1610.00883. [↗](https://arxiv.org/abs/1610.00883)
11. Kyunghyun Cho, Bart Van Merriënboer, Caglar Gulcehre, Dzmitry Bahdanau, Fethi Bougares, Holger Schwenk, and Yoshua Bengio. 2014. Learning phrase representations using RNN encoder-decoder for statistical machine translation. arXiv preprint arXiv:1406.1078. [↗](https://arxiv.org/abs/1406.1078)
12. Daniel Jurafsky and H. James. 2000. Speech and language processing an introduction to natural language processing, computational linguistics, and speech. [find](https://scholar.google.com/scholar?q=Daniel+Jurafsky+and+H%2E+James%2E+2000%2E+Speech+and+language+processing+an+introduction+to+natural+language+processing%2C+computational+linguistics%2C+and+speech%2E)
13. Jihen Karoui, Farah Benamara, Véronique Moriceau, Nathalie Aussenac-Gilles, and Lamia Hadrich Belguith. 2015. [Towards a contextual pragmatic model to detect irony in tweets](https://md.research.elamur.ai/free/en/papers/karoui-2015-contextual-pragmatic-model-detect-irony/). In 53rd Annual Meeting of the Association for Computational Linguistics (ACL 2015), pages PP–644.
14. Ronan Collobert and Jason Weston. 2008. A unified architecture for natural language processing: Deep neural networks with multitask learning. In Proceedings of the 25th International Conference on Machine Learning, pages 160–167. ACM. [find](https://scholar.google.com/scholar?q=Ronan+Collobert+and+Jason+Weston%2E+2008%2E+A+unified+architecture+for+natural+language+processing%3A+Deep+neural+networks+with+multitask+learning%2E+In+Proceedings+of+the+25th+International+Conference+on+Mac)
15. Diederik Kingma and Jimmy Ba. 2014. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980. [↗](https://arxiv.org/abs/1412.6980)
16. Bhuwan Dhingra, Zhong Zhou, Dylan Fitzpatrick, Michael Muehl, and William W Cohen. 2016. Tweet2vec: Character-based distributed representations for social media. arXiv preprint arXiv:1605.03481. [↗](https://arxiv.org/abs/1605.03481)
17. CC Liebrecht, FA Kunneman, and APJ van Den Bosch. 2013. [The perfect solution for detecting sarcasm in tweets# not](https://md.research.elamur.ai/free/en/papers/liebrecht-2013-perfect-solution-detecting-sarcasm-tweets/).
18. Delia Irazú Hernańdez Farías, Viviana Patti, and Paolo Rosso. 2016. Irony detection in twitter: The role of affective content. ACM Transactions on Internet Technology (TOIT), 16(3):19. [↗](https://doi.org/10.1145/2930663)
19. Tomas Mikolov, Ilya Sutskever, Kai Chen, Greg S. Corrado, and Jeff Dean. 2013. [Distributed representations of words and phrases and their compositionality](https://md.research.elamur.ai/free/en/papers/mikolov-2013-distributed-representations-words-phrases-their/). In Advances in Neural Information Processing Systems, pages 3111–3119. [↗](https://doi.org/10.48550/arxiv.1310.4546)
20. Aniruddha Ghosh and Tony Veale. 2016. [Fracking sarcasm using neural network](https://md.research.elamur.ai/free/en/papers/ghosh-2016-fracking-sarcasm-neural-network/). In Proceedings of the 7th Workshop on Computational Approaches to Subjectivity, Sentiment and Social Media Analysis, pages 161–169. [↗](https://doi.org/10.13140/rg.2.2.16560.15363)
21. Razvan Pascanu, Tomas Mikolov, and Yoshua Bengio. 2013. On the difficulty of training recurrent neural networks. ICML (3), 28:1310–1318. [↗](https://doi.org/10.48550/arxiv.1211.5063)
22. Adam Paszke, Sam Gross, Soumith Chintala, Gregory Chanan, Edward Yang, Zachary DeVito, Zeming Lin, Alban Desmaison, Luca Antiga, and Adam Lerer. 2017. Automatic differentiation in pytorch. [find](https://scholar.google.com/scholar?q=Adam+Paszke%2C+Sam+Gross%2C+Soumith+Chintala%2C+Gregory+Chanan%2C+Edward+Yang%2C+Zachary+DeVito%2C+Zeming+Lin%2C+Alban+Desmaison%2C+Luca+Antiga%2C+and+Adam+Lerer%2E+2017%2E+Automatic+differentiation+in+pytorch%2E)
23. Aniruddha Ghosh and Tony Veale. 2017. [Magnets for sarcasm: Making sarcasm detection timely, contextual and very personal](https://md.research.elamur.ai/free/en/papers/ghosh-2017-magnets-sarcasm-making-sarcasm-detection/). In Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pages 482–491. Fabian Pedregosa, Gaël Varoquaux, Alexandre Gramfort, Vincent Michel, Bertrand Thirion, Olivier Grisel, Mathieu Blondel, Peter Prettenhofer, Ron Kevin Gimpel, Nathan Schneider, Brendan O’Connor, Dipanjan Das, Daniel Mills, Jacob Eisenstein, Weiss, Vincent Dubourg, and others. 2011. Scikit-learn: Machine learning in Python. Journal of Machine Learning Research, 12(Oct):2825–2830.
24. Radim Řehůřek and Petr Sojka. 2010. Software Framework for Topic Modelling with Large Corpora. In Proceedings of the LREC 2010 Workshop on New Challenges for NLP Frameworks, pages 45–50, Valletta, Malta. ELRA. [http://is.muni.cz/](http://is.muni.cz/) publication/884893/en. [↗](http://is.muni.cz/)
25. Toby Segaran and Jeff Hammerbacher. 2009. Beautiful Data: The Stories Behind Elegant Data Solutions. "O’Reilly Media, Inc.". [find](https://scholar.google.com/scholar?q=Toby+Segaran+and+Jeff+Hammerbacher%2E+2009%2E+Beautiful+Data%3A+The+Stories+Behind+Elegant+Data+Solutions%2E+%22O%E2%80%99Reilly+Media%2C+Inc%2E%22%2E)
26. Nitish Srivastava, Geoffrey E. Hinton, Alex Krizhevsky, Ilya Sutskever, and Ruslan Salakhutdinov. 2014. Dropout: A simple way to prevent neural networks from overfitting. Journal of Machine Learning Research, 15(1):1929–1958. [find](https://scholar.google.com/scholar?q=Nitish+Srivastava%2C+Geoffrey+E%2E+Hinton%2C+Alex+Krizhevsky%2C+Ilya+Sutskever%2C+and+Ruslan+Salakhutdinov%2E+2014%2E+Dropout%3A+A+simple+way+to+prevent+neural+networks+from+overfitting%2E+Journal+of+Machine+Learning+R)
27. 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, pages 162–169. [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%2C+pages+162%E2%80%93169%2E)
28. 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, SemEval-2018, New Orleans, LA, USA. Association for Computational Linguistics.
29. Benjamin J Wilson and Adriaan MJ Schakel. 2015. Controlled experiments for word embeddings. arXiv preprint arXiv:1510.02675. [↗](https://arxiv.org/abs/1510.02675)

*Updated 24 October 2026*