A Babel of voices: making the Desert Island Discs corpus

This article presents a new corpus resource with wide-ranging potential in corpus-assisted discourse studies. We compiled a corpus containing over 2400 episodes (close to 12 million words) of Desert Island Discs transcripts, between 1951 and 2025, and enriched it with fine-grained extra-linguistic information. In 80 years (the first broadcast was in 1942) the format of the iconic BBC radio program has remained unchanged: every week the host casts away on an imaginary desert island a well-known guest, with a choice of eight music tracks which guide the interview about the castaway’s memories, personal life, and career. This remarkable consistency makes it an interesting barometer of cultural British life over the past eight decades. But, as well as being a formidable dataset for studies of diachronic variation, Desert Island Discs is also – in the words of one of the castaways – a “library of Babel of human stories, experience and language” (Desert Island Discs, David Mitchell 13 November 2020) and could be a versatile tool to explore a number of questions about identity, memory, emotions, storytelling, and other typical interests of corpus-assisted discourse studies.