Language models And RePresentations (LARP) brings together researchers that explore how information is structured, encoded and used in computational language systems. We encourage submissions on both neural (sub-symbolic) and discrete (symbolic) representations from the fields of computational linguistics and artificial intelligence or their intersection.

The conference is organised by the Centre for Linguistic Theory and Studies in Probability (CLASP, https://gu-clasp.github.io/), University of Gothenburg. The conference will be held between September 8 and 9 in Gothenburg, Sweden (on-site and hybrid).

Important dates

• Submission deadline (archival): April 28, 2025, anywhere on Earth
• Notification of acceptance (archival): June 20, 2025, anywhere on Earth
• Submission deadline (non-archival): August 1, 2025, anywhere on Earth
• Notification of acceptance (non-archival): August 8, 2025, anywhere on Earth
• Camera ready: August 8, 2025, anywhere on Earth
• Registration deadline: TBA
• Conference: September 8–9, 2025, University of Gothenburg, Sweden

Invited speakers

TBA

Topics of interest

We hope to see innovative work that considers neural and symbolic learning and processing in terms of different modelling perspectives. Papers are invited on the following topics as they relate to natural language:

• Neuro-symbolic integration: novel hybrid frameworks combining symbolic representations with neural network learning for enhanced reasoning and natural language processing
• Explainable machine learning: techniques that allow for better interpretability, transparency, and explainability of neural, symbolic and neuro-symbolic architectures
  
• Logical constraints in neural networks: methods that use logical structures (e.g., knowledge bases, ontologies) for post-hoc or inherent explainability
  
• Automated reasoning systems providing human-interpretable rationales for decisions
  
• Symbolic planning and control in neural workflows
  
• Application-driven scenarios (robots, autonomous systems) showcasing benefits of symbolic approaches
  
• Techniques that integrate symbolic representations into text or multimodal generation
  
• Approaches that enforce domain knowledge, consistency, or adherence to constraints in text and/or multimodal generation
  
• Fine-tuning and in-context learning strategies that incorporate logical or rule-based knowledge

This list is illustrative but is not intended to be exhaustive.

Submission Requirements

We accept only direct submissions at this link.

Archival track
Archival track will feature the following types of submissions to appear in conference proceedings: we accept long papers (max 8 pages) and short papers (max 4 pages). Long and short papers must describe substantial, original, and unpublished research. Supplementary materials, appendices, a section on limitations and ethical concerns do not count towards the page limit. Archival accepted papers will be published in the 2025 ACL Anthology as a CLASP Conference Proceedings. Papers should be electronically submitted via the OpenReview system at this link. Submissions should be .pdf files and use the LaTeX or Word templates provided for ACL submissions (https://github.com/acl-org/acl-style-files). Archival submissions must be anonymous. Please make sure that you select the right track when submitting your paper. Contact the organisers if you have questions.

Non-archival track
At the time of submission, authors may indicate that their paper should be considered for the non-archival track. The format for non-archival submissions is the same for both long and short papers as it is for the archival submissions. Non-archival papers will not undergo the peer review process. They will be evaluated by the programme committee for clarity and content relevance before the decision by the PC is made. Non-archival papers do not need to be anonymous. If accepted, they are to be published on the conference website and presented as posters.

Poster abstracts
We invite researchers to submit abstracts in the above areas of interest. Abstract submissions are non-archival. This is a great opportunity to get feedback on work in progress or to present previously published work to a new audience. The deadline for abstract submission is the same as for non-archival papers. Notifications of acceptance will be sent out by August 8, 2025. Abstract submissions should be .pdf files and use the LaTeX or Word templates provided for ACL submissions (https://github.com/acl-org/acl-style-files). Abstracts should not exceed 2 pages (supplementary materials, appendices, a section on limitations and ethical concerns are not included) and be submitted via OpenReview system at this link. The acceptance decision on abstracts will go through the same procedure as papers for the non-archival track. Accepted abstracts will be presented as posters.

Concurrent Submissions

Papers that have been or will be submitted to other conferences or publications must indicate this at submission time using a footnote on the title page of the submissions. We will not accept publications or presentation papers that overlap significantly in content or results with papers that will be (or have been) published elsewhere.

Authors of papers accepted for presentation at LARP must notify the program chairs by the camera-ready deadline as to whether the paper will be presented. All accepted papers must be presented at the conference to appear in the Proceedings.

Camera Ready Versions

Camera ready versions must be deanonymised. Archival submissions get 1 more page to address comments from reviewers: long papers can be maximum up to 9 pages, short papers can be maximum up to 5 pages.

About CLASP

LARP is organised by the Centre for Linguistic Theory and Studies in Probability (CLASP, https://gu-clasp.github.io/) at the Department of Philosophy, Linguistics and Theory of Science (FLoV), University of Gothenburg. CLASP focuses its research on the application of probabilistic and information theoretic methods to the analysis of natural language. CLASP is concerned both with understanding the cognitive foundations of language and developing efficient language technology. We work at the interface of computational linguistics/natural language processing, theoretical linguistics, and cognitive science.