---
title: "Parallel Deep Learning-Driven Sarcasm Detection from Pop"
description: "The paper addresses sarcasm detection from text alone, without audio or video cues. 13 references."
authors: ["Sourav Das", "Anup Kumar Kolya"]
year: 2020
venue: "arXiv:2106.05752 · In RAAI 2020. Advances in Intelligent Systems and Computing, vol 1355 (2021)"
source: "https://arxiv.org/abs/2106.05752"
doi: "10.1007/978-981-16-1543-6_6"
arxiv: "2106.05752"
licence: "CC BY-NC-SA 4.0 (arXiv)"
licence_url: "http://creativecommons.org/licenses/by-nc-sa/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/das-2021-parallel-deep-learning-driven-sarcasm/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/das-2021-parallel-deep-learning-driven-sarcasm/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2020](https://md.research.elamur.ai/free/en/papers/2020/page/8/) / Parallel Deep Learning-Driven Sarcasm Detection from Pop Culture Text and English Humor Literature

# Parallel Deep Learning-Driven Sarcasm Detection from Pop Culture Text and English Humor Literature

**Authors:** Sourav Das and Anup Kumar Kolya · **Year:** 2020 · **Venue:** arXiv:2106.05752 · In RAAI 2020. Advances in Intelligent Systems and Computing, vol 1355 (2021) · **Publisher:** [Springer Nature](https://md.research.elamur.ai/free/en/publishers/springer-nature/) · **Licence:** [CC BY-NC-SA 4.0 (arXiv)](http://creativecommons.org/licenses/by-nc-sa/4.0/)

**Links:** [Publisher's page](https://doi.org/10.1007/978-981-16-1543-6_6) · [PDF](https://arxiv.org/pdf/2106.05752) · [DOI 10.1007/978-981-16-1543-6\_6](https://doi.org/10.1007/978-981-16-1543-6_6) · [arXiv 2106.05752](https://arxiv.org/abs/2106.05752) · [OpenAlex](https://openalex.org/W3168784625)

**Citations:** 4 citations (OpenAlex, 2026-10-14)

## Summary
The paper addresses sarcasm detection from text alone, without audio or video cues. The authors use the transcript part of MUStARD, which contains sitcom dialogues: 690 indexed lines, each 7 to 65 words long. They count the most frequent words and turn token indices into weighted embedding vectors. These vectors feed four LSTM branches that share the same size, inputs and hyperparameters. Each branch is bidirectional with dense layers, and the branches differ only in their final activation: softmax, sigmoid, ReLU or tanh. Training uses Adam at a learning rate of 0.01, 400-dimensional embeddings and 500 epochs.

The softmax branch reaches 98.95% training accuracy and the sigmoid branch 96.88%. The ReLU and tanh branches never reduce their loss. On held-out material (two Project Gutenberg humor books plus SIGN, Sarcasm V2 and SARC), the best test accuracy is 98.31%. The authors report large gains over earlier results on Sarcasm V2 and SARC. They also admit the setup ignores conversational context, since the training lines are isolated punchlines.

## Key points
- Training data is the text modality of MUStARD: 690 dialogue lines from TV comedy shows, with roughly balanced sarcastic and non-sarcastic labels in the source set.
- The most frequent corpus words are mostly discourse markers: 'oh' (74), 'know' (49), 'like' (46), 'yeah' (43), 'well' (39), 'right' (34) and 'really' (30).
- The architecture, called pLSTM, has four isolated parallel BiLSTM stacks with dropout of 0.6 on embeddings and 0.4 on bidirectional layers. It is trained with categorical cross-entropy.
- Test accuracy for pLSTM+softmax:
  - 98.31% on The Comedy of Errors and 96.93% on Three Men in a Boat;
  - 95.89% on SIGN;
  - 94.06% on Sarcasm V2, against a previous 76.00;
  - 93.00% on SARC, against a previous 77.00.
- F-scores on The Comedy of Errors are 97.03 for softmax and 94.58 for sigmoid. Evaluation used 5-fold shuffled splits at a 3:2 ratio.

## Interesting ideas
- **Controlled single-component comparison.** Identical branches share inputs and tuning and vary only one component, so any performance gap can be attributed to that component; this pattern works for any ablation.
- **Discourse markers as sarcasm cues.** The top words in sitcom sarcasm are interjections and hedges like 'oh', 'yeah', 'well', 'right' and 'really', which suggests these function words carry ironic tone.
- **Out-of-domain evaluation.** The authors test a model trained on short TV lines against long literary humor (Shakespeare, Jerome K. Jerome) to see whether it transfers across genres.
- **Context-free sarcasm data has a built-in weakness.** The authors note that isolated sarcastic lines can read as neutral without the surrounding conversation, so labels on single sentences may be unreliable.
- **Tracking each branch separately.** Because the branches are isolated, a failing variant (ReLU, tanh) shows up immediately instead of being averaged away inside an ensemble.

## For Mova

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

- **I1** (learn): Run several parser or classifier variants side by side on the same silver batch, changing exactly one setting, so each difference in agreement has a clear, explainable cause.
- **I2** (frames): The irony module in prag could use these markers, tied to their position in the UD tree (e.g. a sentence-initial discourse dependent), as explainable triggers for an "ironic reading" hypothesis.
- **I3** (answer): Test irony and implicature handling not only on curated examples but also on classic humorous prose, checking whether readers of fables and tales hold up across styles.
- **I4** (world): Irony judgments should consult the world state built from earlier sentences, such as an expectation contradicted by events, rather than deciding from one sentence alone.
- **I5** (learn): Log each witness parser's agreement separately before merging, so that a consistently diverging witness can be detected and down-weighted.

*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

Also on the byline: [Maulana Abul Kalam Azad University of Technology, West Bengal](https://ror.org/030tcae29) · [RCC Institute of Information Technology](https://ror.org/007mfn583).

📚 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: Sourav Das and Anup Kumar Kolya. 2020. Parallel Deep Learning-Driven Sarcasm Detection from Pop Culture Text and English Humor Literature. arXiv:2106.05752 · In RAAI 2020. Advances in Intelligent Systems and Computing, vol 1355 (2021).

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

```bibtex
@misc{das2020parallel,
  title = {{Parallel Deep Learning-Driven Sarcasm Detection from Pop Culture Text and English Humor Literature}},
  author = {Sourav Das and Anup Kumar Kolya},
  year = {2020},
  doi = {10.1007/978-981-16-1543-6\_6},
  eprint = {2106.05752},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.1007/978-981-16-1543-6_6}
}
```

- [← previous On the Specificity of Event Actualization in an English Verb Phrase Headed by a Verb of Movement](https://md.research.elamur.ai/free/en/papers/axelrood-2020-specificity-event-actualization-english-verb/)
- [next → Punctuation Practice in Early Modern English Scientific Writing: The Case of MS 3009 at the Wellcome Library, London](https://md.research.elamur.ai/free/en/papers/criado-pena-2020-punctuation-practice-early-modern-english/)
- [↑ back to papers of 2020](https://md.research.elamur.ai/free/en/papers/2020/page/8/)

## References

13 references, 8 of them with a card of ours. ↗ leads to the work.

1. Mukherjee, S., Bala, P.K.: Sarcasm detection in microblogs using Naïve Bayes and fuzzy clustering. Technology in Society. 48, 19–27 (2017). https://doi.org/10.1016/j.tech-soc.2016.10.003. [↗](https://doi.org/10.1016/j.tech-soc.2016.10.003)
2. Avvaru, A., Vobilisetty, S., Mamidi, R.: [Detecting Sarcasm in Conversation Context Using Transformer-Based Models](https://md.research.elamur.ai/free/en/papers/avvaru-2020-detecting-sarcasm-conversation-context-transformer/). In: Proceedings of the Second Workshop on Figurative Language Processing. pp. 98–103. Association for Computational Linguistics, Online (2020).
3. Gregory, H., Li, S., Mohammadi, P., Tarn, N., Draelos, R., Rudin, C.: [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. pp. 270–275. Association for Computational Linguistics, Online (2020).
4. Majumder, N., Poria, S., Peng, H., Chhaya, N., Cambria, E., Gelbukh, A.: Sentiment and Sarcasm Classification With Multitask Learning. IEEE Intelligent Systems. 34, 38–43 (2019). https://doi.org/10.1109/MIS.2019.2904691. [↗](https://doi.org/10.1109/mis.2019.2904691)
5. Son, L.H., Kumar, A., Sangwan, S.R., Arora, A., Nayyar, A., Abdel-Basset, M.: Sarcasm Detection Using Soft Attention-Based Bidirectional Long Short-Term Memory Model With Convolution Network. IEEE Access. 7, 23319–23328 (2019). https://doi.org/10.1109/ACCESS.2019.2899260. [↗](https://doi.org/10.1109/access.2019.2899260)
6. Sundararajan, K., Palanisamy, A.: Multi-Rule Based Ensemble Feature Selection Model for Sarcasm Type Detection in Twitter, [https://www.hindawi.com/journals/cin/2020/2860479/](https://www.hindawi.com/journals/cin/2020/2860479/). [↗](https://www.hindawi.com/journals/cin/2020/2860479/)
7. Castro, S., Hazarika, D., Pérez-Rosas, V., Zimmermann, R., Mihalcea, R., Poria, S.: [Towards Multimodal Sarcasm Detection (An _Obviously_ Perfect Paper)](https://md.research.elamur.ai/free/en/papers/castro-2019-towards-multimodal-sarcasm-detection-obviously/). In: Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics. pp. 4619–4629. Association for Computational Linguistics, Florence, Italy (2019).
8. Devlin, J., Chang, M.-W., Lee, K., Toutanova, K.: [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). pp. 4171–4186. Association for Computational Linguistics, Minneapolis, Minnesota (2019).
9. Peled, L., Reichart, R.: [Sarcasm SIGN: Interpreting Sarcasm with Sentiment Based Monolingual Machine Translation](https://md.research.elamur.ai/free/en/papers/peled-2017-sarcasm-sign-interpreting-sarcasm-sentiment/). In: Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). pp. 1690–1700. Association for Computational Linguistics, Vancouver, Canada (2017).
10. Oraby, S., Harrison, V., Reed, L., Hernandez, E., Riloff, E., Walker, M.: [Creating and Characterizing a Diverse Corpus of Sarcasm in Dialogue](https://md.research.elamur.ai/free/en/papers/oraby-2016-creating-characterizing-diverse-corpus-sarcasm/). In: Proceedings of the 17th Annual Meeting of the Special Interest Group on Discourse and Dialogue. pp. 31–41. Association for Computational Linguistics, Los Angeles (2016).
11. Khodak, M., Saunshi, N., Vodrahalli, K.: [A Large Self-Annotated Corpus for Sarcasm](https://md.research.elamur.ai/free/en/papers/khodak-2018-large-self-annotated-corpus-sarcasm/). In: Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018). European Language Resources Association (ELRA), Miyazaki, Japan (2018).
12. Ilić, S., Marrese-Taylor, E., Balazs, J., Matsuo, Y.: [Deep contextualized word representations for detecting sarcasm and irony](https://md.research.elamur.ai/free/en/papers/ilic-2018-deep-contextualized-word-representations-detecting/). In: Proceedings of the 9th Workshop on Computational Approaches to Subjectivity, Sentiment and Social Media Analysis. pp. 2–7. Association for Computational Linguistics, Brussels, Belgium (2018).
13. Pelser, D., Murrell, H.: Deep and Dense Sarcasm Detection. arXiv:1911.07474 [cs]. (2019). [↗](https://arxiv.org/abs/1911.07474)

*Updated 24 October 2026*