Author

V. Olufemi

View on Semantic Scholar

2 papers indexed here

We haven’t gathered this author’s papers yet. Follow them and we’ll fetch their work.

Not the right person? Other researchers publish under this name.

Book Open access Jul 2026

The Alignment Gap: A Benchmark Demonstrating the Lack of Cross-Lingual Mapping in Dialect-Specialized Language Models - The Case of Ehugbo

Cross-lingual information retrieval (CLIR) for low-resource dialects remains underexplored, despite millions of speakers worldwide. This work addresses a critical gap by introducing the first information retrieval (IR) benchmark resource for Ehugbo (the Afikpo dialect of Igbo with \textasciitilde 150,000 speakers in Nigeria), constructed from a high-quality parallel multimodal corpus: 1 hour of transcribed Ehugbo Bible audio aligned with standard English translations. This parallel design enables rigorous evaluation of retrieval models across language pairs. Our benchmark reveals a surprising and counterintuitive finding: the ''Alignment Gap'', where African-centric foundation models (Serengeti, Afro-XLMR, AfriBERTA) that excel at linguistic familiarity with Igbo achieve <5% retrieval accuracy, while global models like LaBSE achieve 85% despite less exposure to the language family. Through diagnostic analysis (t-SNE visualizations, tokenization studies, error patterns), we show that regional models lack cross-lingual alignment bridges despite deep language understanding, while global models achieve language invariance through explicit translation supervision. This finding has immediate implications for the design of multilingual systems: pre-training diversity alone is insufficient for dialectal IR. We release our Ehugbo corpus, results, and evaluation splits on GitHub to enable future work on dialect-specific fine-tuning and alignment strategies for African languages.

Ukachi Agnes Eze-Mbey, V. Olufemi, A. Bahizire et al. · Annual International ACM SIG... · 0 citations

Preprint Aug 2026

TransSLR: A Lightweight Transformer for Sign Language Recognition

Automated Sign Language Recognition for under-represented languages remains a largely unsolved problem. Central African Sign Language (CASL) exemplifies this gap: the only available bench-mark, CASL-W60, has a best reported accuracy of 69.93%, and we show that the common heuristic of fine-tuning high-resource models fails to close it. This failure stems from two compounding factors: the limited scale of available CASL data and the significant lexical and visual domain gap between CASL and large-scale corpora such as WLASL, which renders pre-trained representations largely uninformative. To address this, we propose TransSLR, a lightweight Temporal Transformer Encoder trained from scratch on 64-frame normalized pose sequences, with average pooling and a classification head. By operating on geometric keypoint representations rather than raw RGB, TransSLR achieves signer-independent generalization without relying on visual appearance. On the CASL-W60 benchmark, TransSLR establishes a new state-of-the-art accuracy of 80.39%, surpassing the prior best by +10.46%. Beyond accuracy, our encoder-only design significantly reduces computational overhead, making deployment feasible in resource-constrained environments. We conduct extensive experiments on the CASL-W60 benchmark, comparing against RGB-based and multimodal baselines, and demonstrate that TransSLR achieves state-of-the-art performance.

Lucia Yen Wanchi, S. Johnny, V. Olufemi et al. · 0 citations