Summary
"To what extent does experience affect child language acquisition? Answering this is crucial for our understanding of human cognition, but current evidence is insufficient because causal links between experience and outcomes cannot be established with typical approaches. We will study how the speech infants hear affects their learning of the sounds and structures of their native language, as represented in the speech the same infants produce, using an innovative and ecological technique: Audio-recordings collected with a recorder worn by the child throughout a normal day. Fifteen large-scale corpora will be combined, totaling over 25,000h of audio, collected from over 1,400 infants, living in rural and urban sites in every peopled continent, and learning a wide variety of mostly unrelated languages.
We will develop cutting-edge machine learning algorithms to automatically analyze these long, messy, and multilingual recordings. This will yield measures representing infants' experiences (the speech addressed to them, and the speech they overhear) as well as how advanced their own vocalizations are.
A subset of data come from Randomized Control Trials (RCTs) where treatment groups have received behavioral interventions aimed at increasing the quantity of speech infants are addressed. We aim to establish to what extent speech addressed to the infants and overheard speech affect their development by triangulating analyses of variation across individuals, cultures, and treatment-control contrasts.
In parallel, we will use a reverse-engineering approach to assess the theoretically-predicted effects of experience. A variety of learning mechanisms will be implemented in unsupervised learning systems which take as input audio like that heard by infants and produce as output ""vocalizations"". If a system yields the same experience-outcome curves found in the infants’ data, then those learning mechanisms plausibly resemble those used by infants."
We will develop cutting-edge machine learning algorithms to automatically analyze these long, messy, and multilingual recordings. This will yield measures representing infants' experiences (the speech addressed to them, and the speech they overhear) as well as how advanced their own vocalizations are.
A subset of data come from Randomized Control Trials (RCTs) where treatment groups have received behavioral interventions aimed at increasing the quantity of speech infants are addressed. We aim to establish to what extent speech addressed to the infants and overheard speech affect their development by triangulating analyses of variation across individuals, cultures, and treatment-control contrasts.
In parallel, we will use a reverse-engineering approach to assess the theoretically-predicted effects of experience. A variety of learning mechanisms will be implemented in unsupervised learning systems which take as input audio like that heard by infants and produce as output ""vocalizations"". If a system yields the same experience-outcome curves found in the infants’ data, then those learning mechanisms plausibly resemble those used by infants."
Unfold all
/
Fold all
More information & hyperlinks
| Web resources: | https://cordis.europa.eu/project/id/101001095 |
| Start date: | 01-11-2021 |
| End date: | 31-10-2026 |
| Total budget - Public funding: | 2 007 023,00 Euro - 2 007 023,00 Euro |
Cordis data
Original description
"To what extent does experience affect child language acquisition? Answering this is crucial for our understanding of human cognition, but current evidence is insufficient because causal links between experience and outcomes cannot be established with typical approaches. We will study how the speech infants hear affects their learning of the sounds and structures of their native language, as represented in the speech the same infants produce, using an innovative and ecological technique: Audio-recordings collected with a recorder worn by the child throughout a normal day. Fifteen large-scale corpora will be combined, totaling over 25,000h of audio, collected from over 1,400 infants, living in rural and urban sites in every peopled continent, and learning a wide variety of mostly unrelated languages.We will develop cutting-edge machine learning algorithms to automatically analyze these long, messy, and multilingual recordings. This will yield measures representing infants' experiences (the speech addressed to them, and the speech they overhear) as well as how advanced their own vocalizations are.
A subset of data come from Randomized Control Trials (RCTs) where treatment groups have received behavioral interventions aimed at increasing the quantity of speech infants are addressed. We aim to establish to what extent speech addressed to the infants and overheard speech affect their development by triangulating analyses of variation across individuals, cultures, and treatment-control contrasts.
In parallel, we will use a reverse-engineering approach to assess the theoretically-predicted effects of experience. A variety of learning mechanisms will be implemented in unsupervised learning systems which take as input audio like that heard by infants and produce as output ""vocalizations"". If a system yields the same experience-outcome curves found in the infants’ data, then those learning mechanisms plausibly resemble those used by infants."
Status
SIGNEDCall topic
ERC-2020-COGUpdate Date
27-04-2024
Images
No images available.
Geographical location(s)
Search
