Linguistic diversity in machine learning training data: Content analysis of English and Finnish soundscape descriptions in audio captioning

Laura Hekanaho, Emilia Tuuri, Maija Surakka, Maija Hirvonen · 2026

Read Linguistic diversity in machine learning training data: Content analysis of English and Finnish soundscape descriptions in audio captioning by Laura Hekanaho, Emilia Tuuri, Maija Surakka, Maija Hirvonen (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/plos-pone-0350043.txt.

Description

Machines that caption sounds learn from human-written descriptions, and Laura Hekanaho and colleagues compared those descriptions in English and Finnish. Their 2026 PLOS One study is among the first to inspect this training data as language, including descriptions written by blind and sighted Finns.

The Finnish collections varied far more than the English one, which the authors trace partly to English writers working in a second language. A close reading of one hundred fifty descriptions showed what people choose to mention about a soundscape and what they leave out. The findings matter for accessibility tools, since captions trained on thin descriptions will describe the world thinly.

Catalog record

CREATORLaura Hekanaho, Emilia Tuuri, Maija Surakka, Maija Hirvonen
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE66713
LICENSECC_BY
RIGHTSCC BY
SOURCEPLOS

Source: https://doi.org/10.1371/journal.pone.0350043

Browse the full library · About the Tower of Babel