Mobilizing archival collections: The Open Text Collections project

Presented by Christian Doehler, Sebastian Nordhoff and Mandana Seyfeddinipur at the Language Documentation and Archiving Conference, Berlin & Online, 4-6 Sept, 2024 Franz Boas established the “Boasian Trilogy” in language documentation and description (Himmelmann 1998), consisting of a grammatical description, a dictionary, and a text collection. All three are necessary to get a comprehensive overview of a language, a language portrait, or snapshot in time. More im- portantly, the three components complement each other. While we have good publishing outlets for grammars (e.g. Comprehensive Grammar Library) and dictionaries (e.g. Dictionaria), such is not the case for text collections. This means that only few of them are published, and even fewer follow the FAIR principles of findability, accessibility, interoperability, and reusability (Wilkin- son 2016). The project Open Text Collections (http://opentextcollections.org) will remedy this by making high quality text collections from endangered lan- guages available in an open interoperable format. Next to providing pdfs or printed books to the communities themselves, this setup will also provide the data in Cross-Linguistic Data Format (Forkel et al. 2018) for downstream use in NLP applications. Most archive collections are the result of a language documentation project, often part of the collectors’ dissertation projects. While countless hours are in- vested into the structuring and glossing of texts, in many cases, however, these texts are not made available in a reusable way. Linguists tend to have them somewhere on their hard drive, or uploaded to an archive but there is no gen- erally established way of publishing them, at least not in a format which would feed further research downstream (e.g. linguistic typology, corpus- based lan- guage description, or NLP). This means that these valuable results of language documentation often fail to be discovered. Open Text Collections will provide a quality venue for publishing text collec- tions, following the setup established by Language Science Press. The platform is community-driven and aims at being attractive to both data producers (i.e. language documenters) as well as data users (i.e. language communities, ty- pologists, NLP practitioners). For data producers, the platform will provide rigorous peer review, quality control, and top-notch publishing (pdf and print- on-demand), making sure that the time invested in a text collection will not harm job prospects. For data consumers, different outlets will be available to suit different needs: printed books will be available for communities; a search interface (prototype available at https://imtvault.org) will be available for typologists, and all data will be available as CLDF dump for NLP practitioners. By making reuse easy, the research will spread more widely, which in turn is very attractive for the data producers. As of today, there are 5 regional boards and 40+ proposed text collections. This presentation will showcase the platform, its motivations, and its benefits for data producers and consumers.