Elamur Research⌕
☰
home / papers / 2026 / Pixel Phantoms at SemEval-2026 Task 3: Language-Specific Transformer Regression for Dimensional Aspect-Based Sentiment Analysis

Pixel Phantoms at SemEval-2026 Task 3: Language-Specific Transformer Regression for Dimensional Aspect-Based Sentiment Analysis

Paper · 2026 · Jithu Morrison S and Abisha Rose S

Summary

The paper describes the Pixel Phantoms entry for Subtask 1 of SemEval-2026 Task 3 (DimABSA). In this subtask, systems assign continuous Valence and Arousal scores on a 1–9 scale to given aspect terms, replacing discrete polarity labels. The team submitted to every language–domain pair in Track A and Track B, 15 combinations in total. Instead of using one multilingual encoder everywhere, they picked a pretrained backbone per language: deberta-v3-base, cl-tohoku/bert-base-japanese-v3, DeepPavlov/rubert-base-cased, bert-base-chinese, and Davlan's Swahili mBERT. For Ukrainian, Tatar and Nigerian Pidgin they fell back to xlm-roberta-base. Every backbone uses the same regression head:

  • the aspect is placed in front of the review as a prompt;
  • the CLS and mean-pooled vectors are concatenated;
  • a sigmoid output is rescaled with learnable scale and shift parameters;
  • training uses a 0.7·MSE + 0.3·MAE loss.

The best results were Japanese Hotel (RMSE 0.7297, rank 13/21) and Chinese restaurant (0.9823, against 1.8959 for the Kimi-K2 baseline). The work suggests that dedicated encoders help in well-resourced languages. Low-resource languages and unfamiliar domains remain weak spots.

Key points

  • The architecture went through four versions (V1–V4). The final V4 uses deberta-v3-base, max length 256, LR 1e-5, early-stopping patience 3, gradient clipping and FP16. Learnable scale and shift start at 8.0 and 1.0.
  • Ukrainian (restaurant, xlm-roberta-base): RMSE 1.5937, better than the Kimi-K2 baseline at 1.7805. Tatar (2.0729 vs 1.9380) and Russian with rubert (1.7686 vs 1.7768) were worse than or roughly equal to the baseline.
  • Track B, compared with Mistral-3 14B:
    • German politics: 1.5509 vs 1.5910, rank 8/12.
    • English environment: clearly worse, 2.0893 vs 1.6430.
  • For Japanese, the system used fugashi and ipadic tokenization plus an aspect prompt written in Japanese. In V1, feeding text and aspect as a plain sentence pair worked worse than the prompt format.
  • The authors identify three weaknesses: Arousal is harder to predict than Valence, the model cannot tell apart repeated mentions of the same aspect, and low-resource languages remain hard. Their future work includes transferring Russian models to Ukrainian.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Fields, topics and concepts

📚 Shelves 6The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 worksPsychology · 1,649 worksExperimental and Cognitive Psychology · 930 works

🧭 Topics 1Research subjects and languages the work is about.Sentiment & emotion · 1,035 works

Cite

Cite as Jithu Morrison S and Abisha Rose S. 2026. Pixel Phantoms at SemEval-2026 Task 3: Language-Specific Transformer Regression for Dimensional Aspect-Based Sentiment Analysis. Proceedings of the 20th International Workshop on Semantic Evaluation (2026).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.semeval-1.112

@inproceedings{s2026pixel,
  title = {{Pixel Phantoms at SemEval-2026 Task 3: Language-Specific Transformer Regression for Dimensional Aspect-Based Sentiment Analysis}},
  author = {Jithu Morrison S and Abisha Rose S},
  booktitle = {Proceedings of the 20th International Workshop on Semantic Evaluation (2026)},
  year = {2026},
  doi = {10.18653/v1/2026.semeval-1.112},
  url = {https://doi.org/10.18653/v1/2026.semeval-1.112}
}

References

12 references, 6 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Jonas Becker, Liang-Chih Yu, Shamsuddeen Hassan Muhammad, Jan Philip Wahle, Terry Ruas, Idris Abdulmumin, Lung-Hao Lee, Nelson Odhiambo, Lilian Wanzare, Wen-Ni Liu, Tzu-Mi Lin, Zhe-Yu Xu, Ying-Lung Lin, Jin Wang, Maryam Ibrahim Mukhtar, Bela Gipp, and Saif M. Mohammad. 2026. Dimstance: Multilingual… find
  2. Alexis Conneau, Kartik 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. In Proceedings of ACL.
  3. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. Bert: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of NAACL.
  4. Yue Gao and 1 others. 2022. Aspect-prompted encoding for aspect-based sentiment analysis. In Proceedings of ACL. find
  5. Pengcheng He, Xiaodong Liu, Jianfeng Gao, and Weizhu Chen. 2021. Deberta: Decoding-enhanced bert with disentangled attention. In Proceedings of ICLR.
  6. Lung-Hao Lee, Liang-Chih Yu, Natalia Loukashe-vich, Ilseyar Alimova, Alexander Panchenko, Tzu-Mi Lin, Zhe-Yu Xu, Jian-Yu Zhou, Guangmin Zheng, Jin Wang, Sharanya Awasthi, Jonas Becker, Jan Philip Wahle, Terry Ruas, Shamsuddeen Hassan Muhammad, and Saif M. Mohammad. 2026. Dimabsa: Building multilingu… find
  7. 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. arXiv. ArXiv:1907.11692.
  8. Maria Pontiki, Dimitris Galanis, Harris Papageorgiou, Ion Androutsopoulos, and Suresh Manandhar. 2014. Semeval-2014 task 4: Aspect based sentiment analysis. In Proceedings of SemEval.
  9. Maria Pontiki, Dimitris Galanis, Harris Papageorgiou, Ion Androutsopoulos, and Suresh Manandhar. 2016. Semeval-2016 task 5: Aspect based sentiment analysis. In Proceedings of SemEval.
  10. James A. Russell. 1980. A circumplex model of affect. Journal of Personality and Social Psychology, 39(6):1161–1178. find
  11. Amy B. Warriner, Victor Kuperman, and Marc Brysbaert. 2013. Norms of valence, arousal, and dominance for 13,915 english lemmas. Behavior Research Methods. find
  12. Liang-Chih Yu, Jonas Becker, Shamsuddeen Hassan Muhammad, Idris Abdulmumin, Lung-Hao Lee, Ying-Lung Lin, Jin Wang, Jan Philip Wahle, Terry Ruas, Alexander Panchenko, Ilseyar Alimova, Kai-Wei Chang, Lilian Wanzare, Nelson Odhiambo, Bela Gipp, and Saif M. Mohammad. 2026. SemEval-2026 task 3: Dimension… find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/s-2026-pixel-phantoms-semeval-2026-task/