Language is central to how people communicate, reason, and express knowledge, yet language technologies remain unevenly developed across the world's linguistic diversity. Many African languages face limited representation in language models, evaluation benchmarks, and training data. The Language Intelligence team researches multilingual language understanding, language generation, and model evaluation to advance language technologies that better represent African languages.
Research teams: Speech Intelligence · Language Intelligence · Language Resources · Model Intelligence · Responsible AI
Investigating the capabilities, limitations, and possibilities of language technologies across African languages.
Understanding language requires more than recognising words. Research in this area examines how models capture meaning, interpret context, and represent linguistic relationships across African languages, with particular attention to languages that remain underrepresented in digital data and modern language models.
Generating language that feels natural and preserves meaning across linguistic and cultural contexts presents a distinct set of challenges. This research explores multilingual text generation, cross-lingual transfer, and model adaptation to advance the generation of coherent, contextually appropriate text across African languages.
Reliable language technology demands more than strong performance on widely used benchmarks. Research here investigates evaluation methodologies, language-specific benchmarks, and comparative assessments that expose model limitations, measure linguistic coverage, and reveal gaps in how language technologies serve Africa's diverse linguistic communities.