Multi-lingual documentation: the ArviZ project website case

Multi-lingual documentation The ArviZ project website as an example Tomás Capretto Oriol Abril Pla About Exploratory Analysis of Bayesian Models All parts of the Bayesian modeling workflow outside of inference and sampling ●Uncertainty visualization ●Prior elicitation ●Inference diagnostics ●Model comparison ●Model criticism ●Prior and likelihood sensitivity analysis ●Simulation based calibration 1 million weekly downloads from PyPI Community and volunteer driven Publicly available governance Aim for lazy consensus and do-ocracy with mechanisms for the elected steering council to step in and accelerate decisions 2-3 active maintainers plus ~5 active recurrent contributors (out of 19 core contributors) One of the maintainers works at Aalto university and can work on ArviZ as part of their job → Aalto university is an institutional partner ~1 GSoC participants every year since 2019 Why go into translations at all? 1 - To attract Argentina based research organization act as institutional partner 2 - I am a Catalan The language barrier can’t be overcome individually: Participation On average, non-native English researchers need: ●Up to 91% more time to read English papers. ●Up to 51% more time to write papers and their papers will be rejected 2.6 times more often. If accepted, they’ll have to revise it 12.5 more times. ●When they present in English, need up to 94% more time to prepare their presentations. They will often avoid English speaking conferences and oral presentations. Amano T, Ramírez-Castañeda V, Berdejo-Espinola V, Borokini I, Chowdhury S, Golivets M, et al. (2023) The manifold costs of being a non-native English speaker in science. PLoS Biol 21(7): e3002184. https://doi.org/10.1371/journal.pbio.3002184 The language barrier can’t be overcome individually: Belonging ●[...] language presents a barrier to participation or belonging in open source communities. ●[...] a sense of belonging involves more than the ability to contribute code in English. Those who prefer languages other than English were less likely to feel welcome or believe that different backgrounds have equal opportunities to participate in open source projects. ●those [contributors] preferring German, Spanish, Portuguese, and French were all more likely to disagree with OSS having clear paths to leadership Hilary Carter and Jessica Groopman, “Diversity, Equity, and Inclusion in Open Source: Exploring the Challenges and Opportunities to Create Equity and Agency Across Open Source Ecosystems”, foreword by Jim Zemlin, The Linux Foundation, December, 2021, https://www.linuxfoundation.org/research/the-2021-linux-foundation -report-on-diversity-equity-and-inclusion-in-open-source Overview of translation and localization Translation, localization, et al ●Translating content is the process of adapting the content to another language ●Localization extends translation to also adapt significant, non-textual components such as cultural references or currencies ●Internationalization refers to technological solutions which allow software to adapt to different regions without requiring changes to the source code Still work to do Perks and quirks: Fuzzy matching It’s off by default (thankfully) Perks and quirks: No fuzzy matching Old message New message Old translation No new translation Perks and quirks: Inconsistent spacing Translation workflow Social aspects Collaboration between translators Collaborative translations are a constant tradeoff between consistency+adequacy or speed+generality ●Division of labour ●Style guide ●Glossary Conclusions and Learnings