Publications & patents

Systems and Methods for Image Privacy and De-identification

U.S. Patent No. 12,613,996 · Issued April 28, 2026

Salomon Kabongo · Co-inventor

Co-invented a computer-vision system for detecting and obscuring people and personally identifiable text in images.

Computer Vision · Image Privacy · De-identification · AI/ML

INJONGO: A Multicultural Intent Detection and Slot-filling Dataset for 16 African Languages

ACL 2025 · Long Papers

Co-authored a culturally grounded, open benchmark for evaluating intent detection and slot filling across 16 African languages.

African Languages · Intent Detection · Slot Filling · LLM Evaluation

IrokoBench: A New Benchmark for African Languages in the Age of Large Language Models

NAACL 2025 · Long Papers

Co-authored a human-translated benchmark for evaluating natural-language inference, mathematical reasoning, and knowledge-based question answering across 17 African languages.

African Languages · LLM Evaluation · Reasoning · Multilingual NLP

Effective Context Selection in LLM-based Leaderboard Generation: An Empirical Study

NLDB 2024

First-author study of how document representations and context selection affect extraction quality, reliability, and computational efficiency in LLM-based leaderboard generation.

Large Language Models · Context Selection · Information Extraction · Evaluation

Systems and Methods for Advanced Duplicate Image Search and Analysis

Pending U.S. Published Patent Application · App. 18/652,500 · Publication No. US20240411724A1

Salomon Kabongo · Co-inventor

Co-invented a system combining hash-based exact matching with visual-similarity analysis to identify duplicate and near-duplicate images.

Computer Vision · Vector Embeddings · Similarity Hashing · AI/ML

Zero-shot Entailment of Leaderboards for Empirical AI Research

ACM/IEEE Joint Conference on Digital Libraries (JCDL 2023)

First-author evaluation of whether supervised leaderboard-extraction models generalize to previously unseen labels, accompanied by a zero-shot evaluation dataset.

Zero-shot Evaluation · Natural Language Inference · Information Extraction

Bibletts & LiSTra: African Speech Corpora

Interspeech 2022 · NeurIPS 2022 Black in AI Workshop

Co-authored BibleTTS, a multilingual speech corpus, and developed LiSTra, an English-to-Lingala speech-translation dataset and baseline using cascade ASR and machine translation alongside an end-to-end architecture.

ASR · Speech Translation · Lingala · Low-Resource NLP

Automated Mining of Leaderboards for Empirical AI Research

ICADL 2021 · International Journal on Digital Libraries

Award: ICADL 2021 Best Paper Award

Introduces an end-to-end approach for extracting empirical AI results from scientific text and representing them as searchable, knowledge-graph-based leaderboards.

Knowledge Graphs · Information Extraction · NLP · Scholarly IE

LiSTra Automatic Speech Translation: English to Lingala Case Study

NeurIPS 2021 · Black in AI Workshop (Spotlight)

Presents the Lingala Speech Translation (LiSTra) dataset and a reusable pipeline for constructing similar resources in other low-resource languages. Reports cascade ASR-to-MT and transformer-based end-to-end baselines.

ASR · Machine Translation · Lingala · Transformers

Participatory Research for Low-Resourced Machine Translation

EMNLP Findings 2020 · AfricaNLP Workshop ICLR 2020

Contributed to a community-led study of participatory methods for building machine-translation research, datasets, and models for African languages.

Machine Translation · African Languages · Low-Resource NLP · Community

An Empirical Investigation into the Properties of Standard Word Embeddings

MSc Thesis · University of the Western Cape / AIMS South Africa

Reviews mechanisms for computing word embeddings, investigates popular toolkits and embedding matrices, and experiments with selected implementations to better understand their characteristics and properties.

Word Embeddings · NLP · Deep Learning