Proceedings Fonetik 2009 - Institutionen för lingvistik

Proceedings, FOETIK 2009, Dept. of Linguistics, Stockholm Universityand between low and low rising tone in the contrast[ka1]-[ka4]. Although the two tonal contrastsmust be considered nonnative to Swedishinfants, non-Thai listeners in general seem torely on complex acoustic variables when tryingto discriminate tone (Burnham et al., 1996),which therefore makes it difficult to predict thediscrimination of tonal contrast change.After recording, all speech stimuli were presentedto an expert panel consisting of two Thainative speakers and one trained phonetician inorder to select the three best exemplars perstimulus type out of ten (see Table 1 for variationof utterance duration between the selectedexemplars).Table 1. The table shows utterance duration in msfor all selected exemplars per stimulus type andaverage duration in ms per stimulus type (Thai toneis demarcated by number).b̬ a0 ba0 p h a0 da0 d̬ a0 ta h 0 ka1 ka2 ka41 606 576 667 632 533 607 613 550 5352 626 525 646 634 484 528 558 502 5383 629 534 629 599 508 585 593 425 502M 620 545 647 622 508 573 588 492 525Within each trial, the first contrast of each pairwas repeated two to five times until the secondcontrast was presented twice after a brief interstimulusinterval of 300 ms. Each stimulus type(native consonantal, nonnative consonantal,nonnative tonal) was presented twelve timeswithin a block. Within each block, there were36 change trials and nine no-change trials. Achange trial repeated identical exemplars for thefirst contrast and then presented the identicalexemplar of the second contrast twice. A nochangetrial had identical first and secondsound exemplars, presented randomly betweenfour and seven times. A completed experimentconsisted of three blocks à 145 trials.EquipmentThe EEG recordings took place in a radiationinsulatednear-soundproof test chamber at thePhonetics Lab at Stockholm University.Infant brain activation was measured byEGI Geodesic Hydrocel GSN Sensor nets with124 electrodes on the infant net sizes. These nettypes permit EEG measurement without requiringgel application which makes them particularlycompatible with infant research; potassiumchloride and generic baby shampoo serveas conductive lubricants instead. All electrodeimpedances were kept below 50 kΩ at measurementonset. All EEG channel data was amplifiedwith an EGI NetAmps 300 amplifier andrecorded with a sampling rate of one sampleevery 4 ms. The program Netstation 4.2.1 wasused to record and analyse the ERPs.The stimuli were presented with KOSSloudspeakers, mounted at a distance of about100 cm in front of the child. The volume wasset to 55 dB at the source. The experiment wasprogrammed and controlled by the e-prime 1.2software.ProcedureAll infant participants were seated in their parent’slap, facing a TV screen on which silencedshort cartoon movie clips played during the experimentto entertain the infants and keep themas motionless as possible. The infants werepermitted to eat, breastfeed, sleep, as well assuck on dummies or other objects during stimulusexposure.Dependent on the randomisation of the firstcontrast between two and five repetitions, theduration of the entire experiment varied between10 and 13 min. Infant and parent behaviourwas monitored through an observer windowand the experiment was aborted in the caseof increasing infant fussiness - this happened inone case after 8 min of stimulus exposure.Data treatmentThe EEG recordings were filtered with a bandpassfilter of 0.3 to 50 Hz and clipped into 1000ms windows starting at the onset of the secondcontrast. These windows were then cleanedfrom all 10 ms segments during which the ERPcurve changed faster than 200 µV to removemeasurement artefacts caused by body movementand eye blinks. If more than 80% of thesegments of one single electrode were markedas artefacts, the entire data from that electrodewas not included in the average.ResultsIn accordance with other infant speech perceptionERP studies (e.g., Friederici et al., 2007),the international 10-20 electrode system wasselected to structure the EEG data. Within thissystem, the analysis focused on electrode T3,situated at the temporal lobe in the left hemisphere,as MMN in 8-month-old infants haspreviously been found to be largest in T3 (Panget al., 1998).132