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/PMC13257990.txt.

Description

Laura Hekanaho and colleagues wrote a 2026 article on the language used in soundscape descriptions that train machine learning systems. Audio captioning means writing text that describes what can be heard. The team compared caption datasets in English and Finnish, two very different languages. The Finnish data came from both sighted and non-sighted writers, while the English data came from sighted writers, many using English as a second language.

A broad corpus analysis found more variation in the Finnish datasets than in the English one, which the authors link to second-language writers in the English data. A closer reading of 150 descriptions found that non-sighted writers used slightly richer wording, above all for describing spatial layout, meaning where sounds sit in space. The authors warn that dataset language quality shapes what machines learn.

Catalog record

CREATORLaura Hekanaho, Emilia Tuuri, Maija Surakka, Maija Hirvonen
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE66258
LICENSEhttps://creativecommons.org/licenses/by/4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/article/MED/PMC13257990

Browse the full library · About the Tower of Babel