5. Publication Rules¶
All publications must cite the three scientific archives that make all of this work possible.
How to cite each archive¶
| Archive | Citation format | Where to find it |
|---|---|---|
| HomeBank | (1) Cite the specific corpus using its HomeBank documentation page. (2) Cite the archive itself: VanDam M, et al. (2016). HomeBank. Seminars in Speech and Language, 37, 128–141. DOI: 10.1055/s-0036-1580745 | Full citation rules |
| Databrary | Author(s) (year). Title of dataset. Databrary. https://doi.org/10.17910/XXXXX | Volume page on Databrary |
| The Language Archive | Author(s) (year). Title of collection. The Language Archive. https://hdl.handle.net/XXXXX (Accessed YYYY-MM-DD) | Collection page on TLA. See also: archive.mpi.nl/tla |
5.1 Model Development and Benchmarking Publications¶
When custodians publish work specifically concerned with the development, training, evaluation, or benchmarking of automated models, they are entitled to do so without including corpus collectors as co-authors. The custodians' obligation is to cite each dataset used, following the citation guidelines outlined above for the relevant archive.
[!NOTE] In rare cases, citing everyone may not be possible. See Scenario C.
Example: Okko submits a paper at Interspeech presenting a neural network trained using data from several corpora. Each corpus is cited in the references. No corpus collector co-authorship is required because they made no contributions to the paper's intellectual content.
5.2 Research Beyond Model Development: Consultation Requirement¶
When custodians intend to use consortium data for research that addresses substantive scientific questions beyond model performance, they must notify each affected corpus collector in advance. The notification must:
- Describe the research question and the planned use of the data.
- Offer the corpus collector the option to withdraw their data from the specific project. We suggest giving at least two weeks for the reply to arrive, and more over holiday times (which vary across countries, with some concentration around July-August in the Northern hemisphere, January-February in the Southern hemisphere).
- Offer the corpus collector the option to participate as a co-author, subject to their meeting standard co-authorship criteria. Make sure to spell out what precise contributions are expected and the ideal timeline for them.
Example: Sho proposes a paper examining phonological variation in child-directed speech across multiple language communities using consortium corpora. Each corpus collector whose data is involved must be notified and offered the option to withdraw their data or join as co-author.
5.3 Bilateral Agreements¶
The rules in Sections 5.1 and 5.2 represent the minimum baseline applicable to all ExELang Consortium members. Corpus collectors are free to negotiate separate bilateral agreements with custodians that supplement, modify, or extend these baseline terms.