Low-Resource Speech Technology
ASR, cross-lingual transfer, data-efficient adaptation, naturalistic speech, style mismatch, and robust evaluation for languages with limited training data.
ASRWhisperXLS-RCross-lingual transfer
Linguistics · Speech Technology · African Languages
Lateef Adeleke
PhD Researcher in Linguistics & Low-Resource Speech & Language Technology
I build datasets, linguistic analyses, and computational tools for underdescribed African languages. My work connects primary fieldwork with corpus development, grammatical analysis, low-resource speech modeling, and linguistically informed AI evaluation.
ELDP Principal Investigator
Multi-year Uneme documentation project
ELAR →
AfricaNLP 2026
Linguistically informed ASR evaluation
ACL Anthology →
Indigenous NLP Excellence Award
NLDSTC Awards · 2026
Award announcement →
Donald M. & Janet C. Barnard Fellow
University of Rochester · 2025–2026
Fellowship announcement →
Research program
From primary language data to better language technology
I work across a research pipeline that is often split across different disciplines: collecting and documenting language data, turning it into structured corpora, analyzing linguistic systems and variation, training computational models, and using linguistic knowledge to understand model behavior.
Documentation, Corpora & Resources
Community-based fieldwork, audiovisual documentation, annotation, corpus building, lexical resources, multilingual dictionaries, and archival workflows designed for both linguistic and computational reuse.
ELANFLExELARCorpus design
Grammar, Variation & AI Evaluation
Tense, aspect, negation, grammatical tone, clause structure, variation, and model-error analysis grounded in language-specific phonology and grammar.
ToneTAMVariationError analysis
Featured publication
Linguistically Informed Evaluation of Multilingual ASR for African Languages
Standard aggregate metrics can obscure what speech systems actually learn about low-resource tonal languages. This work evaluates multilingual ASR on Yorùbá and Uneme with character-, phonological-feature-, and tone-sensitive measures, revealing systematic model behavior hidden by word-level accuracy alone.
Featured project
Documenting Uneme language, variation, and indigenous knowledge
I lead an ELDP-supported, multi-year documentation project on Uneme, an Edoid language of Nigeria. The project combines naturalistic audiovisual documentation with corpus development, grammatical analysis, lexical resources, archival outputs, and speech technology.
Document
Naturalistic discourse, oral histories, cultural practices, traditional knowledge, material culture, and targeted linguistic elicitation.
Structure
Transcription, annotation, grammatical analysis, variation across communities, and an Uneme–English multimedia lexical resource.
Model
Speech datasets and low-resource ASR experiments that test how modern models handle naturalistic, tonal, and highly constrained-data conditions.
Recent highlights
News & recognition
2026
Awarded the 2026 Indigenous NLP Excellence Award — Nigerian Languages Data and Sci-Tech Conference.
2026
EACL — Attended EACL (19th Conference of the European Chapter of the ACL), Rabat.
2025
Training — Completed ELDP Language Documentation Training, Berlin.
2025
Awarded the Donald M. and Janet C. Barnard Fellowship — University of Rochester.
2025
Awarded ELDP Individual Graduate Scholarship — began the documentation of Uneme language and indigenous iron technology.
Scholarly Engagement
Selected Talks & Presentations
Recent presentations spanning low-resource speech technology, African language structure, grammatical theory, and language documentation.
2026
Linguistically Informed Evaluation of Multilingual ASR in African Languages
7th Workshop on African Natural Language Processing (AfricaNLP), Rabat, Morocco
7th Workshop on African Natural Language Processing (AfricaNLP), Rabat, Morocco
2026
Reflexes of Movement in Uneme
55th Colloquium on African Languages and Linguistics, Leiden
55th Colloquium on African Languages and Linguistics, Leiden
2026
Polarity-Conditioned Asymmetries in Uneme
57th Annual Conference on African Linguistics (ACAL 57), Buffalo, New York
57th Annual Conference on African Linguistics (ACAL 57), Buffalo, New York
2026
Reflexes of Movement in Uhami
57th Annual Conference on African Linguistics (ACAL 57), Buffalo, New York
57th Annual Conference on African Linguistics (ACAL 57), Buffalo, New York
2025
Research Presentation
Department of Linguistics Omnibus Meeting, University of Rochester
Department of Linguistics Omnibus Meeting, University of Rochester
2025
Morphosyntax of TAM and Negation in Uhami
56th Annual Conference on African Linguistics (ACAL 56), University of Minnesota
56th Annual Conference on African Linguistics (ACAL 56), University of Minnesota
Broader vision
African languages shouldn't remain passive "low-resource benchmarks". They should actively contribute to both linguistic theory and the design of next-generation language technology.
As a long-term goal, I want to develop research approaches in which language documentation, linguistic theory, corpus development, and machine learning reinforce one another. I am especially interested in technologies that work with the realities of underdescribed languages— limited data, naturalistic speech, tonal and morphological complexity, and speaker and dialect variation.
Collaboration
Interested in collaborating?
I welcome conversations with researchers, language communities, research labs, and organizations. I am especially interested in collaborations that connect data, language structure, and computational modeling in ways that expand both scientific understanding and technological access.