Word segmentation abilities in infants under acoustically degraded conditions

• In Event: Word segmentation and statistical learning

Abstract

The current study adopts a psychoacoustic approach to explore the effect of acoustic degradation on wordform perception in 6-month-old infants with normal hearing. This approach models the decomposition of the spectral and temporal components of speech by the auditory system. Signal-processing algorithms called “vocoders” have been used to selectively alter the spectral and temporal variations of speech signals. Moreover, some experimental conditions have been designed to simulate the reception of the speech signal transmitted by current cochlear-implant processors for normal-hearing listeners.

Four vocoder experiments were conducted using a passage-wordform segmentation task via the Headturn Preference Procedure (HPP) on French-learning 6-month-olds (Experiments 1, 2 and 3) and 10-month-olds (Experiment 4). In all experiments, infants with normal hearing (N=24) first accumulated one minute of looking time to target word(s) from a 6-sentence passage in the familiarization phase. Following, in the test phase, infants’ looking time preference for isolated familiar v. novel words was measured.

Original stimuli from Nishibayashi & Nazzi (2016) were either processed by an 8-band vocoder (Experiments 1 & 2) to reduce drastically the fast temporal information and the fine spectral details of original stimuli as current cochlear implant processor, or by a 16-band vocoder reducing to a less extent the spectral details of the stimuli (Experiment 3 & 4).
In Experiment 1, 6-month-old infants were presented with two targets in familiarization and with two novel words and the two target words at test. Likewise, Experiment 2 used an 8-band vocoder, but presented only one target and one novel word at test to 6-month-olds. Experiments 3 and 4 used a less-degraded 16-band vocoder and presented one target and one novel word at test to 6- and 10-month-olds. If infants are able to rely on spectrally reduced speech signals, it is expected they will show preferential orientation toward familiar stimuli, as shown by previous studies using natural stimuli.

At 6 months of age, no difference in preferential orientation time was found between familiar v. novel stimuli in the two 8-band vocoder conditions nor in the 16-band vocoder condition (Experiments 1, 2, 3). At 10 months of age, a significant preference for novel words was observed in the 16-band vocoder condition (Experiment 4).

The null results at 6 months suggest that fine spectro-temporal information may be necessary for lexical processing in wordform segmentation tasks at this age. Four months later, infants appear to be able to use reduced spectro-temporal information to detect word-forms. To better understand the role of spectral information for word segmentation abilities during early development, a new group of 10-month-olds will be tested in a more degraded condition (8-band vocoder).

Authors

• Monica Hegde, Université de Paris (Integrative Neuroscience and Cognitive Center-INCC)
  Presenting Author

• Irene de la Cruz-Pavía, University of the Basque Country UPV/EHU & Ikerbasque

• Laurianne Cabrera, Centre National de la Recherche Scientifique (CNRS)

• Thierry Nazzi, University of Paris