LETTERS TO NATURE Speed and cerebral correlates of syllable discrimination in infants

THE remarkable linguistic abilities of human neonates are well documented. Young infants can discriminate phonemes even if they are not used in their native language, an ability which regresses during the first year of life. This ability to discriminate is often studied by repeating a stimulus for several minutes until some behavioural response of the infant habituates, and later examining whether the response recovers when the stimulus is changed. This method, however, does not reveal how fast infants can detect phonetic changes, nor what brain mechanisms are involved. We describe here high-density recordings of event-related potentials in three-month-old infants listening to syllables whose first consonants differed in place of articulation. Two processing stages, corresponding to an increasingly refined analysis of the auditory input, were identified and localised to the temporal lobes. A late frontal response to novelty was also observed. The infant brain recognizes a phonetic change in less than 400 ms. Sixteen healthy twoto three-month-old infants (average 81 days old, range 63-91), born to monolingual American-English families, were tested. Infants were seated in a carrier affixed to the parent, and their heads were covered with a geodesic sensor net, a very light mesh made of 58 electrodes encased in sponges soaked with a saline solution. Infants faced a loudspeaker placed on top of a TV monitor in a sound-attenuated room. To avoid eye and head movement, a silent video showing attentiongrabbing coloured objects was played continuously. The video was not synchronised with the auditory stimuli, thus preventing any visually evoked potentials from appearing after averaging in synchrony with the auditory stream. On each trial, a sequence of five syllables (/ba/ or /ga/) was presented. In half the trials, one syllable, designated as the standard, was repeated five times (standard trials). In the other half (deviant trials), the standard was repeated only four times, followed by one instance of the other syllable, designated as the deviant. Because repeated and deviant trials were randomly mixed, infants could not predict the nature of the fifth stimulus. Thus any significant difference in event-related potentials (ERPs) to repeated and deviant trials indicated that the two syllables had been discriminated. The onset and topography of the observed differences were used to infer the speed and brain mechanisms of syllable discrimination. The evolution of ERPs with successive repetitions of the standard was also studied. In human adults, reduced ERPs to repetitive sounds as well as novelty-specific responses have been described . Each syllable presentation elicited a distinctive waveform characterized by two peaks (Fig. 1). The maturation of these peaks from birth to six months has been previously described", but their functional significance has not been elucidated. The first identifiable event, peak 1, reached its maximum about 220 ms after stimulus onset. Peak 1 amplitude was highest in response to the first syllable of each trial. As soon as the standard was repeated, a significant decrease was observed. Further repetitions of the standard did not lead to further decrease, and there was no recovery to the deviant syllable (Fig. 2). Thus, by 200 ms, the acoustical analysis had not proceeded far enough to separate two syllables that were quite similar in pitch, duration and intensity. The generators of peak 1 appeared insensitive to the subtle acoustical differences that encoded phonetic information. The scalp topography of peak 1 showed synchronous anterior positivities and posterior negativities with a temporal-central inversion plane (Fig. 3), suggesting bilateral generators inside the temporal lobes. We speculate that peak 1 reflected the activation of primary and secondary auditory areas. The next identifiable event, peak 2, reached its maximum about 390 ms after stimulus onset. Again, the amplitude of peak 2 decreased significantly between the first and the second presentation of the standard, with no further decrease to subsequent