F0 peaks are a necessary condition for German infants’ perception of stress in metrical segmentation Katharina Zahner & Bettina Braun Department of Linguistics, University of Konstanz {katharina.zahner|bettina.braun}@uni-konstanz.de Abstract Infants exposed to stress-timed languages have been demonstrated to use the stressed syllable to localize word beginnings. More recently, intonation has been shown to interfere with stress-based segmentation [1, 2]: For instance, stress-based segmentation was limited to accented words with bell-shaped accents that had the f0 peak aligned with the stressed syllable (“medial-peak” accents); segmentation failed when the f0 peak preceded or followed the stressed syllable in the target word’s contour. Here, we test whether metrical segmentation is caused by the f0 peak on the stressed syllable or by the tonal alternation (LHL, bell-shaped contour in [2]). This allows us to probe whether f0 peaks are necessary cues for metrical segmentation. To this end, we replicated Zahner, et al. [2] but used cup-shaped intonation contours on the targets resulting in a tonal alternation in the opposite direction (HLH). Looking times obtained in a head-turn preference experiment showed no evidence of segmentation for the cup- shaped contours. This suggests that an f0 peak is a necessary condition for the stressed syllables to be used in stress-based segmentation, at least for German infants. Index Terms: stress, pitch accent type, infant, German 1. Introduction In order to extract units from speech, infants use both general strategies, e.g., transitional probabilities between syllables, and language-specific cues ([3, 4] for overviews). Regarding language-specific cues, prosodic properties of the ambient language shape segmentation behavior. For example, infants from stress-timed languages develop a stress-based strategy and interpret stressed syllables as word onsets, for Dutch [5-8], English [9-13], and German [14-16]. Two recent studies on German showed that intonation affects infants’ segmentation behavior [1, 2]. Specifically, using electrophysiological measures, Männel and Friederici [1] demonstrated that word-form recognition is influenced by accentuation, i.e., whether or not a trochee (Sirup [ɑziə.rʊp] ‘syrup’) receives a pitch accent. Infants’ event-related potentials (ERPs) showed that the recognition of trochees differs with age and is modulated by accentuation: 6-month- olds only recognized trochees that were accentuated in familiarization, surfacing in a positive ERP response 500ms after word onset, i.e., before the end of the trochee. At 9 months, recognition was independent of accentuation and manifested in a (mature) negative response 400ms after word onset; this effect was followed by a late negativity only for accentuated words. At 12 months, infants recognized words independent of accentuation during familiarization (negative response 350ms after word onset). Hence, from that study, we may infer that accentuation generally facilitates segmentation.  Yet, a behavioural study by Zahner, et al. [2] suggests pitch accent type and the resulting consequences of (mis)alignment between the f0 peak and the stressed syllable (rather than accentedness in general) to modulate the segmentation success for German infants. Acoustically, pitch accent types differ in the alignment of the f0 peak in regard to the stressed syllable, making the position of f0 peak an unreliable cue to stress [17], see Fig. 1. In medial-peak accents (H*), used to introduce new information to the discourse [18], the f0 peak and the stressed syllable coincide. In early-peak accents (H+L*), signaling accessible information [19], the f0 peak precedes the stressed syllables, while it follows the stressed syllable in late-peak contours (L*+H / L* H-^H%), commonly employed for sentence-initial topics [20] or in (polar) questions [21].  ba na na ba na na  ba na na Medial-peak  Early-peak   Late-peak  Figure 1: Different pitch accent types on a trisyllabic word with stress on the second syllable (e.g., banana). Specifically, Zahner, et al. [2] familiarized German 9- months-olds with trisyllabic words (WSW stress pattern, e.g., Lagune [la.ɑDZuə.nə] ‘lagoon’) in sentences; recognition of the embedded SW units (e.g., gune [ɑDZuə.nə]) was tested. The WSW words were presented in one of 3 naturally occurring intonation conditions (between-subjects): medial-peak condition (peak-stress alignment), early- or late-peak contours (peak-stress misalignment), Fig. 1. Infants recognized the SW units only when f0 peak and the stressed syllable coincided. The authors concluded that a high-pitched accent (bell-shaped with the f0 peak on the stressed syllable, i.e., including a rising-falling movement) is important for stress perception and consequently segmentation. Here, we test whether an f0 peak on the stressed syllable is a necessary cue for German infants to use a stressed syllable for metrical segmentation or whether it is the tonal alternation (LHL) that rendered the syllable with alternating pitch (the one with the high tonal target) particularly salient. In other words, infants may be particularly sensitive to the stressed syllable when the neighboring syllables differ in f0, e.g., LHL or HLH, but less sensitive when there is little change, LLH or HLL, as in the two misalignment conditions in [2]. This is compatible with an early view on stress perception on the utterance level by Bolinger [22], arguing that it is a “wide departure from a contour” (in any direction) that makes a syllable stand out [22: 112]. It would also be compatible with PREPRESS PROOF FILE CAUSAL PRODUCTIONS 1