Anders Søgaard: 3 Nov. 2022, 12:00 (IST)
Language Technology and Inequality
Language tech isn’t the Great Leveller (yet)
Language: English
Before Internet, information was not readily available to everyone. While some people could go to libraries and share photocopies, others didn’t live even within driving distance of a library or other form of information centre. Internet ushered in wider, though unequal, access to information. If you enter English terms in a Google search, you get more and much better search results than if you search in, say, Greenlandic. Machine translation holds out the promise of universal access, but the quality of machine translations varies greatly depending on the language. This is the upshot of trends and dynamics in AI research and development that this paper seeks to explain. I also outline some potential remedies for these inequalities, which go beyond merely improving translation quality, since translation systems are based on assumptions about literacy that are not always borne out. This point is illustrated by a brief look at initial efforts to support communication in refugee camps.
Anders Søgaard is a professor of computer science and philosophy at the University of Copenhagen and the holder of various awards, including an ERC Starting Grant, a Google Focused Research Award, and Best Paper awards in NLP and AI. Søgaard has previously worked at the University of Potsdam, Amazon and Google Research.
Before Internet, information was not readily available to everyone. While some people could go to libraries and share photocopies, others didn’t live even within driving distance of a library or other form of information centre. Internet ushered in wider, though unequal, access to information. If you enter English terms in a Google search, you get more and much better search results than if you search in, say, Greenlandic. Machine translation holds out the promise of universal access, but the quality of machine translations varies greatly depending on the language. This is the upshot of trends and dynamics in AI research and development that this paper seeks to explain. I also outline some potential remedies for these inequalities, which go beyond merely improving translation quality, since translation systems are based on assumptions about literacy that are not always borne out. This point is illustrated by a brief look at initial efforts to support communication in refugee camps.
Anders Søgaard is a professor of computer science and philosophy at the University of Copenhagen and the holder of various awards, including an ERC Starting Grant, a Google Focused Research Award, and Best Paper awards in NLP and AI. Søgaard has previously worked at the University of Potsdam, Amazon and Google Research.
