Low-Resource, High-Impact: Building Corpora for Inclusive Language Technologies

This tutorial 1 is designed for NLP practitioners, researchers, and developers working with multilingual and low-resource languages who seek to create more equitable and socially impactful language technologies. Participants will walk away with a practical toolkit for building end-to-end NLP pipelines for underrepresented languages-from data collection and web crawling to parallel sentence mining, machine translation, and downstream applications such as text classification and multimodal reasoning. The tutorial presents strategies for tackling the challenges of data scarcity and cultural variance, offering hands-on methods and modeling frameworks. We will focus on fair, reproducible, and community-informed development approaches, grounded in realworld scenarios. We will showcase a diverse set of use cases covering over 10 languages from different language families and geopolitical contexts, including both digitally resource-rich and severely underrepresented languages.

💡 Research Summary

This tutorial paper, “Low-Resource, High-Impact: Building Corpora for Inclusive Language Technologies,” presents a comprehensive guide for NLP practitioners and researchers aiming to develop equitable language technologies for multilingual and low-resource contexts. It offers a practical toolkit for constructing end-to-end NLP pipelines, covering the full spectrum from initial data collection and web crawling to parallel sentence mining, machine translation, and downstream applications like text classification and multimodal reasoning.

The content is structured into three core parts. Part 1 lays the foundation on data annotation basics, detailing principles, workflows, and best practices with a focus on quality, scalability, and ethics. It explores strategies for low-resource settings with limited LLM support, including hybrid human-in-the-loop approaches. Part 2 delves into concrete case studies, providing real-world examples of pipeline construction. These include community annotation for language identification of web data, parallel sentence mining for endangered languages like Upper and Lower Sorbian, data collection and evaluation for low-resource machine translation shared tasks (using the ESA framework), and cross-lingual knowledge transfer for building text classification systems. A significant case study also examines the creation of JEEM, a culturally-grounded benchmark for image captioning and VQA across four low-resource Arabic dialects, highlighting challenges in annotator recruitment and dialect-specific quality control. Part 3 synthesizes insights from interviews with 10-15 NLP experts involved in benchmark creation for socially impactful applications, shedding light on often-overlooked practical hurdles, trade-offs, and community engagement strategies.

Throughout, the tutorial emphasizes fair, reproducible, and community-informed development approaches. It addresses critical ethical considerations, such as the responsible employment and fair compensation of human annotators, while also discussing the limitations of LLMs in low-resource scenarios. The tutorial presenters themselves reflect diversity in gender, language background, career stage, and affiliation, aligning with the core mission of promoting true inclusivity in NLP. Attendees are expected to gain hands-on methodologies and modeling frameworks to tackle data scarcity and cultural variance, grounded in a diverse set of use cases spanning over 10 languages from different families and geopolitical contexts.