Bridging the Data Provenance Gap Across Text, Speech and Video

2024-12-19

arXiv.org

AI Summary

This study addresses systemic deficiencies in multimodal AI training data concerning source transparency, licensing restrictions, and geographic/linguistic representativeness. Method: We conduct the first longitudinal provenance audit spanning text, speech, and video modalities (1990–2024), covering nearly 4,000 publicly available datasets. Our approach integrates fine-grained human annotation, cross-modal metadata standardization, provenance chain tracing, and quantitative analysis of geographic and linguistic coverage. Results: We find that over 80% of mainstream datasets impose non-commercial use restrictions; multilingual and multi-regional representation shows no substantive improvement post-2013; and web crawling and synthetic data dominate sourcing, with licensing constraints exhibiting implicit proliferation. We publicly release the complete audit dataset, providing an empirical foundation and methodological framework to advance data traceability, de-Westernization, and responsible AI development.

Technology Category

• Machine Learning: Multimodal Learning
• Natural Language Processing: Language Grounding & Multi-modal NLP
• Data Mining & Knowledge Management: Mining of Visual, Multimedia & Multimodal Data

Application Category

• Security and Privacy: Data transparency and provenance
• Web Mining and Content Analysis: Mining multimedia, multimodal, multilingual, cross-lingual Web data
• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search

Abstract

Progress in AI is driven largely by the scale and quality of training data. Despite this, there is a deficit of empirical analysis examining the attributes of well-established datasets beyond text. In this work we conduct the largest and first-of-its-kind longitudinal audit across modalities--popular text, speech, and video datasets--from their detailed sourcing trends and use restrictions to their geographical and linguistic representation. Our manual analysis covers nearly 4000 public datasets between 1990-2024, spanning 608 languages, 798 sources, 659 organizations, and 67 countries. We find that multimodal machine learning applications have overwhelmingly turned to web-crawled, synthetic, and social media platforms, such as YouTube, for their training sets, eclipsing all other sources since 2019. Secondly, tracing the chain of dataset derivations we find that while less than 33% of datasets are restrictively licensed, over 80% of the source content in widely-used text, speech, and video datasets, carry non-commercial restrictions. Finally, counter to the rising number of languages and geographies represented in public AI training datasets, our audit demonstrates measures of relative geographical and multilingual representation have failed to significantly improve their coverage since 2013. We believe the breadth of our audit enables us to empirically examine trends in data sourcing, restrictions, and Western-centricity at an ecosystem-level, and that visibility into these questions are essential to progress in responsible AI. As a contribution to ongoing improvements in dataset transparency and responsible use, we release our entire multimodal audit, allowing practitioners to trace data provenance across text, speech, and video.

Problem

• Analyzes data sourcing trends across text, speech, video.
• Examines licensing restrictions in multimodal AI datasets.
• Assesses geographical and linguistic representation in datasets.

Innovation

• Largest multimodal dataset audit
• Tracing data provenance chains
• Analyzing dataset sourcing restrictions