{"id":5137,"date":"2026-05-13T15:08:00","date_gmt":"2026-05-13T19:08:00","guid":{"rendered":"https:\/\/oge.mit.edu\/msrp\/?post_type=profiles&#038;p=5137"},"modified":"2026-08-11T17:03:14","modified_gmt":"2026-08-11T21:03:14","slug":"michael-owusu","status":"publish","type":"profiles","link":"https:\/\/oge.mit.edu\/msrp\/profiles\/michael-owusu\/","title":{"rendered":"Michael Owusu"},"content":{"rendered":"<div class=\"wp-block-image\">\n<figure class=\"alignleft size-full is-resized\"><img loading=\"lazy\" decoding=\"async\" width=\"400\" height=\"599\" src=\"https:\/\/oge.mit.edu\/msrp\/wp-content\/uploads\/sites\/2\/2026\/05\/Owusu-Michael.jpg\" alt=\"by Corban Swain\" class=\"wp-image-5623\" style=\"aspect-ratio:1;object-fit:cover;width:200px;height:auto\" srcset=\"https:\/\/oge.mit.edu\/msrp\/wp-content\/uploads\/sites\/2\/2026\/05\/Owusu-Michael.jpg 400w, https:\/\/oge.mit.edu\/msrp\/wp-content\/uploads\/sites\/2\/2026\/05\/Owusu-Michael-200x300.jpg 200w\" sizes=\"auto, (max-width: 400px) 100vw, 400px\" \/><\/figure>\n<\/div>\n\n\n<div class=\"wp-block-group\"><div class=\"wp-block-group__inner-container is-layout-constrained wp-block-group-is-layout-constrained\">\n<p class=\"wp-block-paragraph\"><strong>Morehouse College<\/strong><br>Faculty Advisor: Prof. Marzyeh Ghassemi<br>Research Supervisors: Kumail Hamoud, Hara Moraitaki<br>Department: Electrical Engineering and Computer Science<\/p>\n<\/div><\/div>\n\n\n\n<div style=\"height:0px\" aria-hidden=\"true\" class=\"wp-block-spacer\"><\/div>\n\n\n\n<h4 class=\"wp-block-heading\"><strong>Biography<\/strong><\/h4>\n\n\n\n<p class=\"wp-block-paragraph\">Michael Owusu is a Computer Science major at Morehouse College from Kumasi, Ghana. He is<br>interested in artificial intelligence and machine learning. This summer, he is conducting research<br>at MIT under the mentorship of Dr. Marzyeh Ghassemi, where he studies multilingual safety in large<br>language models. He enjoys building software, attending research conferences, and learning about<br>the problems other researchers are working on. When he gets the chance, he also enjoys sharing his<br>own work.<\/p>\n\n\n\n<p class=\"has-text-align-center wp-block-paragraph\"><strong>Negation Understanding in Vision-Language Models: A Twi (Akan) Extension of NegBench<br>Michael Owusu\u00b9, Kumail Alhamoud\u00b2, Hara Moraitaki\u00b2 and Marzyeh Ghassemi\u00b2<\/strong><br>\u00b9Department of Computer Science, Morehouse College<br>\u00b2Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology<\/p>\n\n\n\n<p class=\"has-text-align-center wp-block-paragraph\">Vision-language models power image search, but they ignore negation: asked for \u201ca photo with no<br>cars,\u201d they return cars. Such errors reverse meaning, which matters where absence is the message,<br>e.g., a scan showing \u201cno evidence of pneumonia.\u201d NegBench (Alhamoud et al., 2025) documented this<br>in English and showed that finetuning on synthetic negated captions recovers 28 points; whether the<br>failure or the fix holds elsewhere was untested. Within an ongoing HealthyML project extending<br>NegBench across languages, we built the first negation benchmark for Twi, an unevaluated Ghanaian<br>language: 5,914 human-verified multiple-choice questions and 500 captions translated by a native<br>speaker (first author). Across ten models and five languages, reading a language and understanding<br>negation prove separate abilities. NLLB-CLIP matches 14.0% of Twi captions to their images, against<br>61.0% in English, yet answers 1.2% of Twi negation questions correctly, below 25% chance. The fix<br>does not transfer: one finetuned model drops from 54.5% in English to 17.5% in Twi, below its<br>untuned baseline. Measuring image-text similarity, the object a caption names matters 8\u201339x more<br>than whether it is negated: a matching shortcut, not missing data. Negation must be addressed in<br>training, and without benchmarks this failure stays invisible.<\/p>\n","protected":false},"featured_media":5435,"template":"","profile_category":[25],"class_list":["post-5137","profiles","type-profiles","status-publish","has-post-thumbnail","hentry","profile_category-2026-interns"],"acf":[],"_links":{"self":[{"href":"https:\/\/oge.mit.edu\/msrp\/wp-json\/wp\/v2\/profiles\/5137","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/oge.mit.edu\/msrp\/wp-json\/wp\/v2\/profiles"}],"about":[{"href":"https:\/\/oge.mit.edu\/msrp\/wp-json\/wp\/v2\/types\/profiles"}],"version-history":[{"count":2,"href":"https:\/\/oge.mit.edu\/msrp\/wp-json\/wp\/v2\/profiles\/5137\/revisions"}],"predecessor-version":[{"id":5739,"href":"https:\/\/oge.mit.edu\/msrp\/wp-json\/wp\/v2\/profiles\/5137\/revisions\/5739"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/oge.mit.edu\/msrp\/wp-json\/wp\/v2\/media\/5435"}],"wp:attachment":[{"href":"https:\/\/oge.mit.edu\/msrp\/wp-json\/wp\/v2\/media?parent=5137"}],"wp:term":[{"taxonomy":"profile_category","embeddable":true,"href":"https:\/\/oge.mit.edu\/msrp\/wp-json\/wp\/v2\/profile_category?post=5137"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}