The perceptual nature of stress shifts

Abstract

According to the rhythm rule in metrical phonology, a word's stress alternates from the second syllable to the first when followed by a word with first syllable stress, a situation also known as a stress clash. For example, the second-syllable stressed word sixTEEN will be produced as SIXteen in the phrase SIXteen CANdles. Using pseudo-words in different rhythmic contexts, we demonstrate that stress clash resolution has a strong perceptual component and can be an auditory illusion. In Experiment 1, participants were more likely to misattribute stress in a clash condition (sixTEEN heard as SIXteen in sixTEEN CANdles) than in isolation (sixTEEN) or in a non-clash condition (sixTEEN caNALS). In Experiment 2, we failed to find support for the hypothesis that nuclear stress drove these ratings. We discuss these findings in light of various theories and mechanisms for parsing spontaneous speech.

Keywords:

• beat clash
• rhythm
• auditory illusion
• nuclear stress
• speech perception

Notes

1. It should be noted that in the limited context of the rhythm rule discussion, we chose to illustrate this point in classic metrical phonology terms (Lieberman & Prince, 1977) for simplicity as well as in keeping with previous literature. Because of a lack of isomorphism between the syntactic and prosodic domains, a more complex phrasal environment can lead to other grid eurhythmy problems. Thus, a similar but more complete and faithful analysis should be carried out in the framework of nested boundaries as prescribed by prosodic phonology.

2. An anonymous reviewer suggested that the items containing real word syllabic components might bias listeners to report stress on the real word syllables, or to avoid reporting stress the real word syllables. While possible, we note that any such bias was balanced across conditions and would have affected each condition equally.

3. See Sluijter and van Heuven (1996) for conflicting evidence in Dutch regarding the relative importance of duration in the production of stress.

4. “B” here refers to the estimates for each level in the model relative to the baseline level. These represent the unstandardized regression coefficients on the logit scale.

5. Temperley (2009) is the only work that we are aware of that has examined contextual-specific stress variants of citation forms in a spontaneous speech corpus. Temperley (2009) found that contextual stress patterns do not increase (they actually decrease) the regularity of SW alternations, providing further evidence that speakers do not produce the stress shift.