Language Technologies for Crisis Preparedness and Response - Professor William D. Lewis

Language Technologies for Crisis Preparedness and Response - Prof. William D. Lewis, University of Washington, USA 

Abstract

The advent of the digital era has made Language Technologies (LT) both promising and important in disaster response. LTs can, for instance, be used to triage the need for immediate assistance based on text messages or social media, translate vast amounts of data related to ongoing pandemics, or engage in conversations to guide natural disaster victims. However, these advances are only limited to a few dozen of the more than 6500 languages spoken or signed in the world today, neglecting millions of people and widening the “digital divide”, effectively reducing the technologies’ real-world utility.

Language Technologies for Crisis Preparedness and Response (LT4CPR), a recently funded National Science Foundation initiative, seeks to bolster the work in the LT disciplines (e.g., machine translation, speech recognition, natural language processing, etc.) on under-served and under-resourced languages spoken in communities that have been affected by disasters or crises, or communities that might be so affected in the future. Our goal is to build resources to encourage work on these languages across the broader LT disciplines. However, LT is of little value if it is not used by community members or by those providing support. We have been collaborating with Translators without Borders (CLEAR Global) for many years, and see the grant as a means to deepen that relationship and broaden the language and technology reach at their disposal. Likewise, we have sought out connections to crisis response researchers, crisis informatics researchers, aid agencies, related NGOs, etc. as part of this initiative.

In this talk, I will review some of the historic use of LTs in the crisis response space that led to LT4CPR, and then review the work we have been doing as part of the LT4CPR grant. Much of our work thus far has focused on reviewing the current state of the art, surveying recent literature published on LT in the crisis space, evaluating the quality and utility of existing systems, identifying the needs of emergency managers, and planning the path forward that will have the most utility to the crisis community. Recently developed technologies, such as multilingual neural machine translation (MNMT) and large language models (LLMs), show great promise in the context of providing aid, however, they should not be used without a thorough evaluation of their strengths, weaknesses and limits.

About the Speaker

William D. Lewis is a researcher with extensive experience in machine translation (MT), low-resource languages, and language technologies for crisis response. For nearly 14 years, until December 2020, he worked on the Machine Translation Incubation team at Microsoft, most of that time embedded at Microsoft Research and later Microsoft Azure. Prior to that, he helped found and teach in the Computational Linguistics M.S. Program at the University of Washington with Emily Bender and Fei Xia. Earlier, he served as faculty at California State University, Fresno, where he helped establish the university’s Computational Linguistics and Cognitive Science programs and served as the inaugural chair of both.

His research focuses on MT for low-resource and endangered languages, data curation, and the use of language technologies in crisis settings. At Microsoft Translator, he led efforts to develop and deploy MT for under-resourced languages including Haitian Kreyòl, White Hmong, Querétaro Otomí, Icelandic, Canadian French, and Inuktitut among others. Following the 2010 Haiti earthquake, he helped lead the rapid development of the world’s first commercial Haitian Kreyòl MT system in fewer than five days, an achievement that helped shape his long-term research agenda on crisis response. This work later informed broader efforts such as TICO-19 and the NSF-funded Language Technologies for Crisis Preparedness and Response (LT4CPR) project, in collaboration with colleagues at George Mason University and Translators without Borders (now part of CLEAR Global).

All are welcome, please register using this form: CenTraS Research Talk 4 June, 17:00-18:00

This Photo by Unknown Author is licensed under CC BY-SA-NC