Change search
ReferencesLink to record
Permanent link

Direct link
Modelling speech imitation
Stockholm University, Faculty of Humanities, Department of Linguistics, Phonetics.ORCID iD: 0000-0003-3981-2551
Stockholm University, Faculty of Humanities, Department of Linguistics, Phonetics. (Jollergruppen)ORCID iD: 0000-0002-7980-3601
2008 (English)In: IROS 2008: From motor to interaction learning in robots, 2008Conference paper (Other academic)
Abstract [en]

The concept of imitation is often pointed out as one of the cornerstones in infants' early language acquisition. Still there are few studies concerning vocal adult-child imitations reported in the literature and results from those are often inconsistent. One reason for the inconsistencies is the lack of a stringent model for what should be classified as imitations. This is not only a problem when trying to learn something about adult-child interactions, but also when trying to make robots that can learn to interact naturally with humans. In order to interact vocally a robot is typically equipped with artificial models of the ear and the vocal tract connected by an artificial neural network. This model is inspired by the motor theory of speech perception [1] and the more recent discovery of mirror neurons [2]. While the robot can use babbling to create an initial map between the acoustic signal and the corresponding vocal tract positions, it needs to overcome interspeaker differences and to acquire key positions of the vocal tract to be able to communicate with humans or other robots. Imitation games are therefore used to train the networks [3, 4, 5]. As we have shown in our previous work [6] these imitation games should preferably go both ways. Having the robot imitating the caregiver is useful for directing the robot towards keypoints, while having the caregiver imitating the robot is more important for learning the map and overcome interspeaker differences. As we will show in this work, both types of imitations can also be found in adult-child interactions. However, while robots usually follow very strict imitation games with predefined turn-taking behaviors, adult-child interactions tend to be much more complex. For the robot to be able to learn its maps under such natural conditions it has to be able to separate imitations from non-imitations. The question we want to answer in this work is therefore the following. How can the robot decide when a pair of utterances should be considered as vocal imitations of each other?

Place, publisher, year, edition, pages
Keyword [en]
speech; imitation; motor; interaction; Learning; Infants; Infant; Language; Language acquisition; Ear; Vocal tract; motor theory of speech perception; Speech Perception
National Category
General Language Studies and Linguistics
URN: urn:nbn:se:su:diva-14789OAI: diva2:181309
IROS 2008
Available from: 2008-10-31 Created: 2008-10-31 Last updated: 2014-04-29Bibliographically approved

Open Access in DiVA

No full text

Other links

Search in DiVA

By author/editor
Gustavsson, LisaLacerda, Francisco
By organisation
General Language Studies and Linguistics

Search outside of DiVA

GoogleGoogle Scholar
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

Total: 38 hits
ReferencesLink to record
Permanent link

Direct link