Study shows languages with more speakers tend to be harder for machines to learn

Just a few months ago, many people would have found it unimaginable how well artificial intelligence-based "language models" could imitate human speech. What ChatGPT writes is often indistinguishable from human-generated text.

A research team at the Leibniz Institute for the German Language (IDS) in Mannheim, Germany have now used text material in 1,293 different languages to investigate how quickly different computer language models learn to "write." The surprising result: languages that are spoken by a large number of people tend to be more difficult for algorithms to learn than languages with a smaller linguistic community. The study is ...