Call for Papers
The 1st Workshop on Discourse Structure in Neural NLG invites papers of a theoretical and experimental nature on discourse structure and controllability of neural models for natural language generation (NNLG). The event will be co-located with INLG in Tokyo, Japan. Relevant topics include (but are not limited to):
- Limits of current end-to-end NNLG, for example with respect to
- Sentence planning and discourse structure
- Control and interpretability
- Methods for improving discourse coherence and cohesion in NNLG
- Handling discourse connectives
- Lexical choice, aggregation and generating referring expressions
- Avoiding unnecessary repetition or enhancing contextually appropriate diversity
- Representations and model structure in NNLG
- Use of more explicit guidance or structure in the input
- Use of intermediate structure
- End-to-end vs. modular systems
- Evaluation of discourse coherence and cohesion in NNLG
We welcome three categories of regular papers:
- Long papers are most appropriate for presenting substantial research results and must not exceed eight (8) pages of content (with any number of additional pages for references).
- Short papers are more appropriate for presenting an ongoing research effort and must not exceed four (4) pages of content (with any number of additional pages for references).
- Demo papers should be no more than two (2) pages in length, including references, and should describe implemented systems. Authors of demo papers should be willing to present a demo of their system during the workshop.
Submissions should follow the new ACL Author Guidelines and policies for submission, review and citation, and be anonymized for double blind reviewing. Dual submission to other venues (conferences/workshops/journals) is not permitted to avoid overtaxing limited reviewer resources, though cross-submissions are allowed (see below).
Regular papers will be included in the workshop proceedings and appear in the ACL Anthology.
We also welcome two categories of non-archival submissions:
- Abstracts of no more than 350 words are appropriate for preliminary but interesting unpublished ideas that would benefit from additional exposure and discussion.
- Cross-submissions are appropriate for papers on relevant topics which have appeared in or been submitted to alternative venues (such as other NLP or ML conferences). Authors should indicate the alternative venue in the Acknowledgments section.
Submissions should be emailed to email@example.com. Please indicate the desired submission category in the email.
Selection of non-archival submissions will be determined solely by the organizing committee based on their relevance to the workshop aims. Review of non-archival submissions will not be anonymous. Selected submissions will be presented as posters at the workshop and will not appear in the proceedings.