Hebrew University's Inbal Arnon finds baby finches learn songs by language's rules

An international team including cognitive scientist Prof. Inbal Arnon of the Hebrew University of Jerusalem reported in Science Advances, announced on 17 August 2026, that the songs of young Bengalese finches share some of the deepest statistical features of human language. Her co-authors were Prof. Simon Kirby of the University of Edinburgh, whale-song expert Dr. Ellen Garland of the University of St Andrews, Kazuo Okanoya of Teikyo University and Miki Takahasi of Japan's RIKEN Center for Brain Science.

In every human language, a handful of words are used extremely often while most are rare, in a precise mathematical pattern known as Zipf's law. Children seem to use such regularities to break the continuous stream of speech into words. Arnon and her colleagues had previously applied a speech-segmentation method developed for infant language learning to more than 30 hours of humpback whale recordings and found that whale song follows the same pattern.

In the new study they analysed recordings of six Bengalese finches between 60 and 120 days old, the period in which the birds learn and polish their songs. The finches' songs could also be split into recurring note sequences that behave like words, and those units followed a Zipf-like frequency distribution even in the birds' earliest attempts. Much as a child picks up words from several adults, young finches assemble their songs from fragments learned from different tutors.

The authors argue that the pattern appears wherever communication is passed down by learning, because it makes a signal easier to learn, rather than because it carries meaning. Finch song has no words in the human sense and is used mainly in courtship, yet it converges on the same mathematical structure. The researchers wrote that some of language's basic building blocks can emerge wherever communication is learned.