Elamur Research⌕
☰
home / papers / 2026 / Modeling Perspectives in NLP: Parameter-Efficient Perspective Conditioning for Span Extraction and Summarization

Modeling Perspectives in NLP: Parameter-Efficient Perspective Conditioning for Span Extraction and Summarization

Paper · 2026 · Harikrishnan Gurushankar Saisudha and Sabine Bergler

Summary

Healthcare community question-answering threads mix several viewpoints: personal experience, advice, causes, follow-up questions and factual information. This paper tackles both subtasks of the PerAnsSumm 2025 shared task, which builds on the PUMA corpus. Task A finds and labels perspective spans in answers. Task B writes a summary for each perspective. The authors did not compete in the task but reuse its data and evaluation.

Their pipeline starts with a multi-label classifier, a fine-tuned BART encoder with an LLM few-shot alternative, which predicts the perspectives present in each question–answer pair. These predictions then steer the downstream models through one of two lightweight mechanisms:

  • prefix vectors, inspired by PLASMA
  • a perspective-aware attention layer, drawing on AWAN and label attention

Span detection becomes binary BIO tagging, run separately for each perspective. Compared with a joint multi-label tagging baseline, explicit conditioning strongly improves span recall. Summaries reach ROUGE scores similar to or slightly above full fine-tuning. The attention variant does this while training only a small share of the parameters. This suggests structured viewpoint control can be both efficient and interpretable.

Key points

  • Data splits: 2,533 training, 959 validation and 640 test-seen instances (test-seen is drawn from validation), plus a 50-item official test set. The five perspectives are cause, suggestion, experience, question and information. PUMA itself contains 3,167 Yahoo! Answers threads.
  • On span proportional-match recall, Perspective-Attention with key length 1 scores 56.24, against 18.01 for the joint BIO baseline. The baseline keeps the highest proportional precision (74.25).
  • Strict span matching remains hard for every system, with precision and recall of roughly 3–12.
  • For summarization, Prefix Tuning with prefix length 5 achieves the best ROUGE-L (34.84) and ROUGE-2 (18.90). The baseline scores 32.28 and 18.12. Perspective-Attention with key length 16 reaches 34.01 / 18.84.
  • The classifier uses the final EOS hidden state of the BART encoder, followed by a sigmoid output layer trained with class-weighted binary cross-entropy. BART was chosen because it accepts longer inputs than BERT or RoBERTa.

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

🎓Concordia UniversityMontreal · Canada9 works · 673 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 works

🧭 Topics 1Research subjects and languages the work is about.Summarization · 232 works

Cite

Cite as Harikrishnan Gurushankar Saisudha and Sabine Bergler. 2026. Modeling Perspectives in NLP: Parameter-Efficient Perspective Conditioning for Span Extraction and Summarization. Proceedings of the the fifth edition of NLPerspectives.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.63317/5oecmktyh8bf

@inproceedings{saisudha2026modeling,
  title = {{Modeling Perspectives in NLP: Parameter-Efficient Perspective Conditioning for Span Extraction and Summarization}},
  author = {Harikrishnan Gurushankar Saisudha and Sabine Bergler},
  booktitle = {Proceedings of the the fifth edition of NLPerspectives},
  year = {2026},
  doi = {10.63317/5oecmktyh8bf},
  url = {https://doi.org/10.63317/5oecmktyh8bf}
}

References

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

  1. Siddhant Agarwal, Md. Shad Akhtar, and Shweta Yadav. 2025. Overview of the PerAnsSumm 2025 Shared Task on Perspective-aware Healthcare Answer Summarization. In Proceedings of the Second Workshop on Patient-Oriented Language Processing (CL4Health), pages 445–455, Albuquerque, New Mexico. Association… find
  2. Satanjeev Banerjee and Alon Lavie. 2005. ME-TEOR: An Automatic Metric for MT Evaluation with Improved Correlation with Human Judgments. In Proceedings of the ACL workshop on Intrinsic and Extrinsic Evaluation Measures for Machine Translation and/or Summarization, pages 65–72. find
  3. Samuel Bowman, Gabor Angeli, Christopher Potts, and Christopher D Manning. 2015. A Large Annotated Corpus for Learning Natural Language Inference. In Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pages 632–642.
  4. Federico Cabitza, Andrea Campagner, and Valerio Basile. 2023. Toward a Perspectivist Turn in Ground Truthing for Predictive Computing. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 37, pages 6860–6868.
  5. Alessandra Teresa Cignarella, Cristina Marco, and Davide Bernardi. 2025. Perspectivist Approaches to Natural Language Processing: a survey. Language Resources and Evaluation (LREC), 59(2):1719–1746. find
  6. Zhiheng Huang, Wei Xu, and Kai Yu. 2015. Bidirectional lstm-crf models for sequence tagging. ArXiv, abs/1508.01991.
  7. Dongsuk Jang, Haoxin Li, and Arman Cohan. 2025. YaleNLP @ PerAnsSumm 2025: Multi-Perspective Integration via Mixture-of-Agents for Enhanced Healthcare QA Summarization. In Proceedings of the Second Workshop on Patient-Oriented Language Processing (CL4Health), pages 415–427. find
  8. John Lafferty, Andrew McCallum, and Fernando Pereira. 2001. Conditional Random Fields: Probabilistic Models for Segmenting and Labeling Sequence Data. ICML. find
  9. Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Veselin Stoyanov, and Luke Zettlemoyer. 2020. BART: Denoising Sequence-to-Sequence pre-training for Natural Language Generation, Translation, and Comprehension. In Proceedings of the 58th annual meeting of the…
  10. Xiang Lisa Li and Percy Liang. 2021. Prefix-Tuning: Optimizing Continuous Prompts for Generation. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers), pages 4582–… find
  11. Chin-Yew Lin. 2004. ROUGE: A Package for Automatic Evaluation of Summaries. In Proceedings of Text Summarization Branches Out Workshop at ACL 2004, pages 74–81.
  12. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. In Proceedings of the 2019 conference of the North American chapter of the association for Computational Linguistics: Human Language Technologies,…
  13. Alexander R Fabbri, Xiaojian Wu, Srini Iyer, and Mona Diab. 2021. Multi-Perspective Abstractive Answer Summarization. arXiv preprint arXiv:2104.08536. — [arXiv:2104.08536](https://arxiv.org/abs/2104.08536) find
  14. Simona Frenda, Gavin Abercrombie, Valerio Basile, Alessandro Pedrani, Raffaella Panizzon, Ilya Loshchilov and Frank Hutter. 2017. Sgdr: Stochastic gradient descent with warm restarts. In 5th International Conference on Learning Representations, ICLR 2017, Toulon, France, April 24-26, 2017, Conferenc… find
  15. Ilya Loshchilov and Frank Hutter. 2019. Decoupled weight decay regularization. In 7th International Conference on Learning Representations, ICLR 2019, New Orleans, LA, USA, May 6-9, 2019. find
  16. Gauri Naik, Sharad Chandakacherla, Shweta Yadav, and Md Shad Akhtar. 2024. No perspective, no perception!! Perspective-aware Healthcare Answer Summarization. In Findings of the Association for Computational Linguistics: ACL 2024, pages 15919–15932, Bangkok, Thailand. Association for Computational Li… find
  17. Tabea Pakull, Hendrik Damm, Henning Schäfer, Peter Horn, and Christoph Friedrich. 2025. Wis-PerMed @ PerAnsSumm 2025: Strong Reasoning Through Structured Prompting and Careful Answer Selection Enhances Perspective Extraction and Summarization of Healthcare Forum Threads. In Proceedings of the Second… find
  18. Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. 2002. BLEU: a Method for Automatic Evaluation of Machine Translation. In Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics, pages 311–318.
  19. Nils Reimers and Iryna Gurevych. 2019. Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 3982–3992.
  20. Kaitao Song, Xu Tan, Tao Qin, Jianfeng Lu, and Tie-Yan Liu. 2020. MPNet: Masked and Permuted Pre-training for Language Understanding. Advances in neural information processing systems (NEURIPS), pages 16857–16867.
  21. Narjes Tahaei and Sabine Bergler. 2025. Beyond Consensus: Use of Demographics for Datasets that Reflect Annotator Disagreement. In First Workshop on Bridging NLP and Public Opinion Research at COLM2025. find
  22. Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. Advances in neural information processing systems (NeurIPS).
  23. Thanh Vu, Dat Quoc Nguyen, and Anthony Nguyen. 2020. A label attention model for icd coding from clinical text. In Proceedings of the Twenty-Ninth International Joint Conference on Artificial Intelligence, IJCAI-20, pages 3335–3341. International Joint Conferences on Artificial Intelligence Organiza… find
  24. An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, et al. 2025. Qwen3 Technical Report. arXiv preprint arXiv:2505.09388.
  25. Tianyi Zhang, Varsha Kishore, Felix Wu, Kilian Q. Weinberger, and Yoav Artzi. 2020. BERTScore: Evaluating Text Generation with BERT. In International Conference on Learning Representations, ICLR 2020. 13. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/saisudha-2026-modeling-perspectives-nlp-parameter-efficient/