Panel 15 : NLP and Intercultural Ethics
27 April | 2.00 pm | Chair: Eleanor Drage | Venue: Bawden Room
Presentation 1: Building NLP models to teach local languages in Africa
Presenter: Aderonke Busayo Sakpere and Makuochi Samuel Nkwo
Abstract: Language is key for communicating the values, norms and beliefs of a group of people. Research has shown that a lot of indigenous languages in Africa face the risk of extinction. It has also been noted that kids are best taught their mother tongue in their formative years. It’s evident from the various research and a preliminary exploratory study carried out that there is indeed a wide gap that exists when it comes to designing technology that can support kids to learn indigenous language. As a result, in this study we attempted to bridge the existing gaps by designing a platform that can support kids in Nigeria to learn a native language. As a first step, the proposal demonstrated how a prototype developed on the concept of gamification and constructivism pedagogy gave positive results such as fun in learning language, ability to improve critical thinking. However, on further analysis we realized that the NLP tool that supports the application doesn’t produce the native language using local accent. As a result, it negatively affects the learning of our kids to correctly pronounce Yoruba language. We therefore propose a native NLP tool that can help kids correctly learn indigenous or local accents.
We propose using the Naive Bayes, which is an NLP model that gives the best accuracy, when compared with others such as Decision Tree and Random Forest Algorithm. We will incorporate the Yoruba language dataset with which we use to train the NLP model. Building the NLP pipeline step-by-step, we will integrate the app with tools that will help kids to correctly learn indigenous or local accents of the Yoruba language.
Author bios: Aderonke Busayo Sakpere holds a PhD in Computer Science from the University of Cape Town, South Africa. She is a faculty member at the University of Ibadan with about 12 years of experience in lecturing, research and mentoring. As a lecturer, she teaches both undergraduate and postgraduate students, with research interests spanning Human Computer Interaction (HCI), Data Privacy, ICT for Development. Dr Sakpere’s continued academic and research excellence has earned her various honors and awards. They include MIT Empowering The Teachers Fellowship, Selected Young Scientist by Lindau Nobel Laureate Meetings, Top 200 young researcher by Heidelberg Laureate Foundation (HLF), Hasso Plattner Institute (HPI) Fellowship, among others. She is the founder of Tech Girls Club which was set up to foster networking and interaction among females in technology. She recently served as a Consultant/Research Assistant with a digital innovation company named Qhala, on a Wikimedia Foundation Project.
Makuochi Samuel Nkwo is a Postdoctoral Research Fellow, Research Innovation and Development Unit, University of Namibia, Windhoek. He is also a lecturer in Computer Science with over 8 years of teaching and research experience. He has done high-quality and impactful single and collaborative research works in health & wellness, teaching & learning, eCommerce & workplaces, as well as in sustainable environments. Also, he has authored and co-authored, published, and presented over 26 peer-reviewed papers in reputable journals and conferences around the world. He is currently working in the field of Responsible and Human-Centered Artificial Intelligence and Ethics.
Presentation 2: Critical Evaluation of AI-Powered Digital Dialogue Tools in Peacemaking: Insights from United Nations’ Digital Dialogues in Libya and Yemen
Presenter: Ali Altiok
Abstract: This paper critically examines the use of AI-powered digital dialogue tools by United Nations’ peacemakers in Libya and Yemen. AI-powered digital dialogues claim to transform the peacemaking field as peace mediators deploy AI-powered digital dialogues to facilitate real-time conversations between thousands of individuals simultaneously (Alavi et al. 2022). Peace mediation teams inform and receive instant feedback from large numbers of civil society actors. As such, proponents of these innovations claim that AI-powered digital dialogue tools have the potential to speed up the signing of peace agreements without sacrificing principles of inclusivity (Bilich et al. 2019). Since June 2020, mediators of the United Nations have been using AI-powered digital dialogue tools in the Libyan and Yemeni peace processes. However, the evidence supporting such positive contributions of technological innovations to peacemaking efforts is thin (Richmond and Visoka 2021; Richmond and Tellidis 2020). Through qualitative interviews with individuals who participated in digital dialogue sessions, this study provides insights into the benefits, shortcomings, and risks of using digital dialogue tools in peacemaking processes. Drawing on the cultural turn in Peace Studies, the study critiques the techno-solutionist promise of AI-powered digital dialogue tools. Preliminary findings suggest that using AI-powered digital dialogues exacerbates the impact of digital divides, produces new privacy concerns, and violates conflict sensitivity principles and do-no-harm approaches. The application of AI-based digital dialogue tools in Libya and Yemen also signals that the penetration of techno-solutionism in the peacemaking field reproduces widely criticized colonial power dynamics between the global North and South, and between the global and local. Finally, the study shares critical reflections on whether and how AI-powered digital dialogue should be deployed to make peace processes more inclusive.
Author bio: Ali Altiok is a doctoral student in the joint Ph.D. program on Peace Studies and Political Science at the University of Notre Dame. His research focuses on political inclusion and the securitization of young people in the context of peacebuilding processes. Previously, he has worked as a researcher, policy officer, and coordinator for several international peacebuilding organizations in New York City. As an in-house research consultant at the United Nations Peacebuilding Support Office, he supported data analysis and narrative development of the Missing Peace: Independent Progress Study on Youth, Peace, and Security, mandated by the United Nations Security Council Resolution 2250. He co-authored the policy paper, We Are Here: An Integrated Approach to Youth Inclusive Peace Processes for the UN Secretary General’s Envoy on Youth. His work highlights what is transformative to youth agency in peacebuilding and social change. He is a former member of the United Networks of Young Peacebuilders. He holds an M.A. in Political, Legal, and Economic Philosophy from the University of Bern (Switzerland) and an M.A. in Peace and Security Studies from the University of Hamburg (Germany). He is originally from Turkey.
Presentation 3: Towards Building a Gender-Inclusive Khaya AI for English⇔Twi Text Translator
Presenter: Abigail Oppong
Abstract: We present a work in progress on how NLP Ghana is working on examining gender bias in the Khaya AI text translator. Khaya AI Text translator is the first text translator app for African languages. Currently, five languages are represented on the app; we focus on examining gender bias in the Twi→English and English→Twi machine translation to serve as the benchmark for the other African languages on the translator. NLP Ghana publicly released an open-sourced English-Twi Parallel corpus in 2021, which set a baseline for building the text translator to enable Twi Speakers and those in the Diaspora to communicate effectively from Twi to English or English to Twi. Upon testing some hand-crafted test set examples, we examine some representational and allocational gender biases depicting the stereotypical concept in the community. Building from efforts already done, like how Google Translate has worked on examining gender bias in the translation for specific languages, English, Spanish, French, Indonesia, etc., we present the first project to investigate gender bias in an African Language translator, Khaya AI. We set out guidelines to help with a participatory data collection approach from the data team and explore our models to help provide text translators free from such biases to serve the community.
Author bio: Abigail Oppong is the lead/Co-Investigator for AI ethics at NLP Ghana, part of the 2023 list for 100 Women in AI Ethics from the Women AI Ethics due to her work in examining gender in AI Applications for African AI systems. She is an early career researcher focused on AI ethics in the sector of health and communication and an enthusiast about fairness and ethics in Natural Language Processing. She is a young Ambassador for Ariel Foundation International under the United Nations, focusing on issues affecting youths and women in Africa and making the voices of African Youths heard. Changemaker, Young Activist, and a Social Entrepreneur. As a member of the Ghana Natural Language Processing group. Her research project is based on mitigating gender bias in NLP applications. Recipients of the "Young Leaders Creating a Better World for All award" at the Women Economic Forum, 2019. She has made several AFI publications that seek to speak, influence, and impact life. She is part of the Board Members of Women Media and Changes working hard to ensure young women's voices are represented well in digital media. She is also a Co-founder of Photo4her, an initiative that seeks to empower teenage mothers with photography skills. Currently, she is running a project called "The Level Her Up project" that aims to help female student entrepreneurs with the needed digital skills to thrive in managing their businesses while in their studies.