Breaking the Silence: Challenges in Deploying Language Technologies for Low-Resource Agricultural Contexts
Speech interfaces are rapidly emerging as a cornerstone of digital agriculture, promising to make advisory systems more inclusive for low-literate and linguistically diverse farming communities. Yet, the technologies enabling this transformation—Automatic Speech Recognition (ASR) and Natural Language Processing—remain underdeveloped for most African languages. Our study presents both a systematic review of ASR across development domains in Africa and recent progress in designing, training, and evaluating ASR models for agricultural communication.
Our session seeks to inform stakeholders on three major contributions from our study:
- how transfer learning and multilingual modeling can effectively overcome the data limitations associated with low-resource languages,
- how field-level data can be used to demonstrate the practical limits and opportunities of ASR for agriculture, and
- how these methods can be directed toward practical, field-ready solutions that support agricultural practice and policy rather than remaining confined to controlled experimental settings.
Our research highlights the need to pair technical innovation with contextual grounding. Through open, community-driven workflows for data creation and model deployment, it supports local institutions such Farm Radio International in building inclusive, scalable speech technologies for African agriculture—ensuring local voices are central to digital transformation.
Our session cuts across several of this year's themes but is closely aligned with the data and impact track. Our study examines the entire ASR development pipeline, from collection, curation, and management of speech datasets to the development of ASR models and their implementation and deployment in agricultural contexts aimed at strengthening food systems and agricultural productivity.
Speakers
Nelson Mganga is a Data Scientist with a strong foundation in Statistics and Development Economics. His work with the International Food Policy Research Institute focuses on designing and deploying Automatic Speech Recognition and Natural Language Processing tools for low-resource languages.
Nelson's recent work, a collaboration with Farm Radio International, has been centered on building speech and language technologies tailored to agricultural contexts, with the goal of expanding access to advisory services and empowering rural communities.
Nelson’s work bridges advanced AI research with practical, scalable solutions that drive inclusive development.
Eliot Jones-Garcia is a Senior Research Analyst with the Natural Resources and Resilience Unit based in Washington, DC. His research focuses on human-AI interaction, user-centered design, and the ethical and responsible development of AI.
At IFPRI, he explores the conceptual and methodological advancements of AI, the skills and governance structures needed to responsibly integrate AI into agricultural research, and the socio-technical barriers that shape its effectiveness in extension and advisory services.
Eliot is finalizing a PhD on the digitalization of agricultural advisory services at Wageningen University & Research. He holds a master’s in Rural Development and Innovation from Wageningen University and a bachelor’s in International Agriculture from the University of Greenwich.
Breaking the Silence: Challenges in Deploying Language Technologies for Low-Resource Agricultural Contexts
Session Type
Breakout Sessions
Description
Theme: Data Theme: Charting the Future of Data in Development and Humanitarian Response. Explore the rapidly shifting data landscape – from AI’s potential and pitfalls to challenges associated with responsible data sharing, interoperability, and the power dynamics in data collection and use.Primary Tag: Agriculture and ICT
Back to Session Gallery