Abstract

Idioms and proverbs (òwe) are central to Yorùbá discourse, yet computational Yorùbá NLP has concentrated on compositional tasks, leaving non-compositional expressions largely unaddressed. This study presents YorùbáÒwe, a publicly deployed web application for bidirectional Yorùbá↔English idiom and proverb translation. Its engine combines input normalization, sliding-window proverb detection, exact lexicon matching, near-match retrieval, rule-based context disambiguation, and word-level fallback. The engine's production source code was evaluated directly through a multi-layer evaluation. An initial evaluation against a 13-entry seed lexicon traced every Yorùbá-side miss (69.23% exact-match recall) to a hard-coded 12-token segmentation ceiling. The lexicon was then expanded to approximately 578 entries from a 568-pair Yorùbá–English proverb corpus, which was divided deterministically (80/20, random seed 42) into a 454-record Development/Lexicon Set and a 114-record Held-Out Test Set. All 13-unit tests passed, and the production build succeeded. On the Development/Lexicon Set the system retrieved the matching entry for 451/454 records (99.34% exact-match rate) and handled 806/818 controlled variants (98.53%); these figures describe lexicon-supported behavior, not general translation accuracy. On the Held-Out Test Set, evaluated with the held-out entries removed from the lexicon, reference-overlap scores were very low (BLEU 0.26, chrF 6.64, ROUGE-L F1 6.41; 0/114 exact textual matches), whereas a ChatGPT baseline on the same 114 records scored BLEU 38.42, chrF 58.77, and ROUGE-L F1 66.00 (4/114 exact textual matches). These metrics measure similarity to a single reference translation and do not establish translation quality. A human evaluation of 20 selected proverbs by three evaluators (60 judgments) gave mean ratings of 4.72/5 for meaning preservation, 4.72/5 for English naturalness, and 4.83/5 for cultural appropriateness, with 59/60 (98.33%) judged acceptable; an earlier 77-respondent evaluation of the 13 seed entries is also reported. YorùbáÒwe is therefore characterised as a lexicon-based, retrieval-oriented system whose coverage depends on its stored proverb knowledge.

References

• Jimoh, T. A., De Wille, T., & Nikolov, N. S. (2025). Bridging gaps in natural language processing for Yorùbá: A systematic review of a decade of progress and prospects. arXiv:2502.17364.
• Liu, L., Ghosh, S., & Jiang, T. (2025). Evaluating the impact of verbal multiword expressions on machine translation. arXiv:2508.17458.
• Ademowo, A. J., & Balogun, N. (2014). Proverbs and conflict management in Africa: A study of selected Yoruba proverbs and proverbial expressions. International Journal of Literature, Language and Linguistics, 1(1), 39–44.
• Owomoyela, O. (2005). Yorùbá proverbs. University of Nebraska Press.
• Ehineni, T. O. (2016). A discourse structural analysis of Yorùbá proverbs in interaction. Colombian Applied Linguistics Journal, 18(1), 71–83.
• Baldwin, T., & Kim, S. N. (2010). Multiword expressions. In N. Indurkhya & F. J. Damerau (Eds.), Handbook of natural language processing (2nd ed., pp. 267–292). CRC Press, Taylor and Francis Group.
• Carpuat, M., & Diab, M. (2010). Task based evaluation of multiword expressions: A pilot study in statistical machine translation. In Human Language Technologies: The 2010 Annual Conference of the North American Chapter of the Association for Computational Linguistics (pp. 242–245). Association for Computational Linguistics.
• Eludiora, S., & Ayoade, B. A. (2021). Design and implementation of English to Yorùbá verb phrase machine translation system. arXiv:2104.04125.
• Olusanya, J., Masakhane NLP, & Abolade, D. (2025). OWE YOR: Leveraging transformer based models for Yorùbá proverb classification. Conference paper, OpenReview.
• Inuwa Dutse, I. (2025). NaijaNLP: A survey of Nigerian low resource languages. arXiv:2502.19784.
• Odeniyi, O. A., Akinwumi, A. R., & Ogunsuyi, O. O. (2026). Development of a web-based number-to-Yorùbá name translation system: A rule-based computational linguistics approach to numeral text normalisation for a low-resource Nigerian language. International Journal of Applied Information Systems (IJAIS), 13(4). Foundation of Computer Science (FCS), New York, USA. www.ijais.org
• Al Saqqa, S., Sawalha, S., & Abdel Nabi, H. (2020). Agile software development: Methodologies and trends. International Journal of Interactive Mobile Technologies, 14(11), 246–270.
• Molina Ríos, J., & Pedreira Souto, N. (2019). Approach of agile methodologies in the development of web based software. Information, 10(10), Article 305.
• Nielsen, J. (2000). Why you only need to test with 5 users. Nielsen Norman Group. www.nngroup.com/articles/why-you-only-need-to-test-with-5-users/