Vavanagi: a Community-run Platform for Documentation of the Hula Language in Papua New Guinea

Bri, Raphael, Ekaterina · 2026

Read Vavanagi: a Community-run Platform for Documentation of the Hula Language in Papua New Guinea by Bri, Raphael, Ekaterina (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-14210.txt.

Introduction

Olewale, Merx, and Vylomova present Vavanagi, a community-run platform that documents the Hula language of Papua New Guinea. Hula has about 10,000 speakers. The platform gathers English-Hula translation and voice records from the crowd, with elder-led review. So far, 77 translators and 4 reviewers have made over 12,000 sentence pairs across 9,000 Hula words. The authors rate community control on a ladder and place Vavanagi at the top rung. Data choices and design all sit with the Hula community.

Catalog record

CREATORBri, Raphael, Ekaterina
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE19476
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.14210

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Vavanagi: a Community-run Platform for Documentation of the Hula Language in Papua New Guinea Bri, Raphael, Ekaterina Abstract We present Vavanagi, a community-run platform for Hula (Vula’a), an Austronesian language of Papua New Guinea with approximately 10,000 speakers. Vavanagi supports crowdsourced English-Hula text translation and voice recording, with elder-led review and community-governed data infrastructure. To date, 77 translators and 4 reviewers have produced over 12k parallel sentence pairs covering 9k unique Hula words. We also propose a multi-level framework for measuring community involvement, from consultation to fully community-initiated and governed projects. We position Vavanagi at Level 5: initiative, design, implementation, and data governance all sit within the Hula community, making it, to our knowledge, the first community-led language technology initiative for a language of this size. Vavanagi shows how language technology can bridge village-based and urban members, connect generations, and support cultural heritage on the community’s own terms. Olewale Merx Vylomova keywords community-led language technology, language documentation, Pacific languages † † address: 1 Vula’a Kunenai Community, Central Province, Papua New Guinea 2 The University of Melbourne, Melbourne, Australia † † email: [email protected] 1 Introduction Papua New Guinea (PNG) is widely recognized for its high linguistic diversity [ 1 , 2 , 3 ] . Hula (Vula’a), spoken in Central Province, belongs to the Austronesian language family [ 4 , 5 ] , and has approximately 10,000 speakers [ 6 ] . As in many PNG settings, language shifts linked to urban mobility and the dominance of Tok Pisin create pressure on Hula’s position, and demand for practical, community-owned language documentation [ 7 ] . We present Vavanagi (Figure 1 ), a community-run platform for collecting English-Hula parallel text and voice data through crowdsourced translation and elder-led review. The resulting corpus is intended to serve as the foundation for downstream speech technologies, including machine translation (MT) and automated speech recognition (ASR). We document how Vavanagi is the first community-led language technology initiative for a language of this size, with project initiative, design, implementation, and data governance all within the community, demonstrating data sovereignty in practice. To situate this contribution within the landscape of community-based language technology, we also propose a five-level community involvement spectrum, from community consultation to community-governed work (Figure 2 ). This spectrum is intended as a reusable conceptual tool for the field, providing shared vocabulary for comparing the degree of community agency across documentation and technology projects. Figure 1: Overview of the Vavanagi platform and Hula language 2 Background: Community involvement in language technology 2.1 Community involvement in language documentation In linguistic research. Collaborative fieldwork has long been debated in linguistics. Bowern and Warner [ 8 ] argue that community agency and research goals are mutually beneficial and that ethical documentation requires respecting community sovereignty. More recent frameworks go further, distinguishing gradations from community-based to community-led research in which design authority progressively shifts to the community [ 9 ] . Evidence from Pacific contexts shows that community-led documentation can yield larger and higher-quality datasets than visiting-linguist approaches, while supporting long-term sustainability of the work [ 10 ] . In language technology. Participatory approaches in text and speech technology have grown considerably. At scale, the Masakhane initiative demonstrated that grassroots collaboration can produce MT datasets and benchmarks for 30+ African languages [ 11 , 12 ] . Tool design work increasingly centres linguistic sovereignty [ 13 ] , and position papers argue that serving communities requires prioritising their needs rather than treating them as data sources [ 14 ] . Process-focused work extends this to speech technology, advocating participatory design at every stage [ 15 , 16 ] , while platforms such as the Mozilla Data Collective [ 17 ] position contributors as data stewards who govern how their datasets are licensed and used. Figure 2: Community involvement spectrum, from consultation to community governance. See mapping of prior work onto the spectrum in Table 1 . 2.2 Community involvement spectrum Prior work emphasizes participatory practice, but community participation takes meaningfully different forms [ 18 , 19 , 15 , 13 ] . Without explicit distinctions, projects ranging from occasional consultation to full community governance may all be described as “community-based”, obscuring the extent of community agency. A shared vocabulary is needed to compare projects and identify where authority truly lies. Four dimensions capture where a project sits on this continuum: (i) initiative origin : who starts the project; (ii) design authority : who shapes its goals and methods; (iii) operational leadership : who leads day-to-day activity; and (iv) data governance : who owns and controls the resulting data. These dimensions tend to stack: externally initiated projects are more likely to retain external design authority, which in turn correlates with external operational control. This ordering yields a five-level spectrum (Figure 2 ). • Level 1: Community consulted. Community members are consulted (e.g., before project launch), but day-to-day project activity remains external. • Level 2: Community engaged, externally led. Community members contribute data or feedback, while planning and operational control remain with external teams. • Level 3: Community-led operations, externally designed. Community members lead daily implementation, but core project design originates outside the community. • Level 4: Community-led with co-shaped design. Project initiative starts externally, but community priorities substantially reshape goals, methods, and outputs. • Level 5: Fully community-initiated and governed. Initiative, decision-making, implementation, and data governance all sit within the language community. Table 1 maps a selection of prior work (corpus gathering for very low-resource languages) onto this spectrum. Community-led initiatives for low-resource languages have become more common in recent years, including in Africa [ 11 ] , Central Asia [ 20 ] , and the Middle East [ 21 ] , but these efforts target languages with speaker populations in the millions, at least two orders of magnitude larger than Hula. To our knowledge, Vavanagi is the first community-led language documentation project for a language of this size. Table 1: Mapping of prior work onto the community involvement spectrum. Columns: I nitiative origin, D esign authority, O perational leadership, data G overnance. Codes: E = externally led, S = shared/co-shaped, C = community-led. Work I D O G Level Petrariu et al. [ 22 ] – Aromanian E E E E 2 Li et al. [ 23 ] – Khinalug E E E E 2 Prabowo et al. [ 24 ] – Meher/Woirata E E C S 3 Cox et al. [ 19 ] – Tsuut’ina E S S S 4 Frontull et al. [ 25 ] – Ladin S S C S 4 Krajinovic et al. [ 10 ] – Nafsan S S C S 4 Vavanagi (this work) – Hula C C C C 5 3 Methodology Admin Import English data Translators Enter text and voice translations Reviewers Review, flag issues, comment Admin Export approved translations feedback and revisions Figure 3: Workflow on the Vavanagi platform, with one admin user, 77 translators, and 4 reviewers. 3.1 Project initiative and governance Inception. The Hula community is organised online around a WhatsApp community, within which there are interest groups (e.g., canoe racing, church, history, language). In the language interest group, members discussed long-running Bible translation efforts and whether AI could help accelerate translation work. This raised the need for

Read the complete text