Until recently, language resources supporting many tasks in Natural Language Processing (NLP) and other areas of Artificial Intelligence (AI) have been based on the assumption of a single ‘ground truth’ label sought via aggregation, adjudication, or statistical means. However, the field is increasingly focused on subjective and controversial tasks, such as quality estimation or abuse detection, in which multiple points of view may be equally valid; subjectivity, indeed, is considered one of the main causes of Human Label Variation (Plank, 2022).
Data Perspectivism is a proposed solution to deal with subjectivity (Cabitza et al., 2023) and is the main topic of this workshop. Perspectivist approaches to NLP aim at leveraging the variation in human-made language data and modeling different human points of view, in order to develop more informative and fairer models of language and communication.
In the previous editions of the workshop, different aspects of perspectivist NLP were discussed, including ties to personalisation, computer vision and multimedia research. The fourth edition of the workshop will widen the discussed methodology to include not only current and ongoing work on collecting and labelling non-aggregated datasets, but also approaches to mining, modelling and inclusion of diverse perspectives in data, evaluation, and applications of multi-perspective Machine Learning models. In addition, it will involve techniques from social science and Human-Computer Interaction, such as participatory approaches and how they can be implemented at all stages of the supervised learning pipeline.
Authors are also invited to share their language resources (data, tools, services, etc.) and provide essential information (i.e., also technologies, standards, evaluation kits, etc.) that have been used for the work or are a result of their research. In addition, authors will be required to adhere to ethical research policies on AI and may include an ethics statement in their papers.
For the first time, the workshop will also host the new edition of Learning with Disagreement (Le-Wi-Di) shared task, which will explore new approaches to modelling and evaluation of Perspectivist benchmarks.
The NLPerspectives workshop will be co-located with the 14th edition of EMNLP 2025 in Suzou, China, in November 5-9, 2025 and online.
Submissions
The papers should be submitted as a PDF document, conforming to the formatting guidelines provided in the call for papers of the EMNLP conference. Templates are provided here.
We accept three types of submissions:
- Regular research papers;
- Non-archival submissions: like research papers, but will not be included in the proceedings;
- (Non-archival) research communications: 4-page abstracts summarising relevant research published elsewhere.
NLPerspectives will also accept submissions that have been rejected from ACL rolling review, provided they are accompanied by their reviews and they fit the topic of the workshop.
Research papers (archival or non-archival) may consist of up to 8 pages of content. Research communications may consist of up to 4 pages of content. On acceptance, authors may add one additional page to accommodate changes suggested by the reviewers.
Topics
We invite original research papers from a wide range of topics, including but not limited to:
- Non-aggregated data collection and annotation frameworks
- Descriptions of corpora collected under the perspectivist paradigm
- Multi-perspective Modelling and Machine Learning
- Evaluation of multi-perspective models/ models of disagreement
- Multi-perspective disagreement as applied to NLP evaluation
- Fairness and inclusive modelling
- Perspectivist approaches for social good
- Applications of multi-perspective modelling
- Computing with (dis)agreement
- Perspectivist Natural Language Generation
- Perspectivism in multimodal AI
- Foundational aspects of perspectivism
- Participatory approaches and human label variation
- Opinion pieces and reviews on perspectivist approaches to NLP
Submissions are open to all, and are to be submitted anonymously (and must conform to the instructions for double-blind review). All papers will be refereed through a double-blind peer review process by at least three reviewers, with final acceptance decisions made by the workshop organisers. Scientific papers will be evaluated based on relevance, significance of contribution, impact, technical quality, scholarship, and quality of presentation.
Shared Task
The third edition of the shared task Learning with Disagreement (LeWiDi) will be co-located for the first time with NLPerspectives. Positioned within the growing body of research that critically examines label harmonization practices and the reliance on a single ground truth, this year’s shared task challenges participants to leverage both instance-level disagreement and annotator-level information in classification. The proposed tasks feature datasets that address disagreement in both interpretation and labeling—with a dataset for Natural Language Inference (NLI) and another for paraphrase detection—as well as subjective tasks, including irony and sarcasm detection.
More information on the tasks and evaluation will be be available at https://le-wi-di.github.io/
Important Dates
- June 27: Paper submission
- July 25: Notification of acceptance
- August 29: Camera-ready papers due
- November 8, 2025: NLPerspectives workshop at EMNLP