A cross-language comparison of the use of stress in word segmentation

Abstract

In spite of our illusions to the contrary, there are few acoustic cues to word boundaries in spoken language. While statistical probabilities between adjacent speech units provide language-general information for speech segmentation, this study shows that language-specific information may also play an important role. An f0 excursion, a correlate of stress, was provided on the first or last syllable of each word in an artificial language. As expected, French listeners succeeded with final-syllable stress but not initial-syllable stress. However, Dutch listeners appeared to benefit from both initial and final stress, rather than initial stress only. When the results were split according to the language background of the talker (French or Dutch), it became clear that the Dutch listeners only benefited from initial stress when the talker was Dutch, but not when the talker was French. Given that the majority of Dutch listeners spoke French, it is suggested that they readily adapted their word segmentation strategies when faced with French-sounding input (i.e., French talker or final stress). In support of this, Australian English listeners with no knowledge of French clearly benefited from initial stress only, consistent with English stress patterning. [Work supported by Conseil Regional de Bourgogne, NWO, CNRS, NIHDC00403.]