Toward a Coarse-Labeled Spoken Language Identification Dataset for Central Alaskan Yup’ik and Samoan from US Broadcast Archives

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Intermediate, quick

Summary

An ongoing effort is developing a coarse-labeled Spoken Language Identification (LID) dataset for Central Alaskan Yup'ik and Samoan, sourced from US public broadcast archives. This initiative addresses the sparse coverage of these indigenous languages by existing public LID systems like HuggingFace, Meta's MMS-LID, Whisper, and VoxLingua107. While Meta's largest MMS-LID variant covers Yup'ik and three variants cover Samoan, Whisper and VoxLingua107 lack both. The project benchmarks current LID systems, trains a prototype MLP classifier using frozen wav2vec 2.0 representations, and reports preliminary corpus statistics, off-the-shelf model performance, and prototype results. Future work aims for phonologically-informed fine-tuning.

Key takeaway

For NLP Engineers expanding Spoken Language Identification (LID) coverage for under-resourced languages, recognize that current public systems like Whisper and VoxLingua107 often lack support for languages such as Central Alaskan Yup'ik and Samoan. You should consider leveraging US public broadcast archives to build custom, coarse-labeled datasets. Additionally, explore training simple classifiers on frozen wav2vec 2.0 representations and investigate phonologically-informed fine-tuning to improve model performance.

Key insights

Public LID systems often lack coverage for indigenous languages, necessitating custom dataset creation.

Principles

• Public LID systems show sparse indigenous language coverage.
• US broadcast archives offer valuable language data.
• Phonological typology can guide LID fine-tuning.

Method

A coarse-labeled LID dataset is built from broadcast archives, existing systems are benchmarked, and a prototype MLP classifier is trained on frozen wav2vec 2.0 representations.

In practice

• Benchmark existing LID systems on new datasets.
• Utilize frozen wav2vec 2.0 for classifier training.
• Explore phonologically-informed fine-tuning directions.

Topics

• Spoken Language Identification
• Indigenous Languages
• Central Alaskan Yup'ik
• Samoan
• Wav2vec 2.0
• Broadcast Archives
• Low-Resource Languages

Best for: Research Scientist, AI Scientist, NLP Engineer, Machine Learning Engineer

Related on AIssential

• Robust Language Identification for Romansh Varieties — The paper titled "Robust Language Identification for Romansh Varieties," authored by Charlotte Model, Sina Ahmadi, and Jannis Vamvas, was presented at the 11th Edition of the Swiss Text Analytics Conf…
• Māori Text-to-Speech Model Spurns Big Tech’s Values — Indigenous communities can develop high-fidelity AI language models prioritizing data sovereignty and community ownership, offering an alternative to big tech.
• Curation of a Palaeohispanic Dataset for Machine Learning — A new structured dataset enables machine learning analysis of under-resourced Palaeohispanic languages.
• Low-resource Language Discrimination Towards Chinese Dialects with Transfer learning and Data Augmentation — Transfer learning and data augmentation effectively overcome scarce annotation resources for low-resource Chinese dialect discrimination.
• The Data Problem in Low-Resource Languages — Data availability, quality, and standardization are critical for NLP model performance across diverse languages.
• "Sorry, I Didn't Catch That": How Speech Models Miss What Matters Most — Speech models fail on high-stakes, short utterances despite low benchmark word error rates, especially for non-English speakers.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.