Abstract
The paper argues that in order to locate stress in languages such as Macedonian, Latin and Cairene Arabic, where words have one stressed syllable, it is necessary to assign metrical structure to the entire word even though most of it is subsequently erased. In a discussion of Latin enclitic stress it is shown that this erasure of metrical structure must be combined with stages in the derivation where previously assigned metrical structure is scrupulously respected. This leads to a digression concerning the similar enclitic stress in the Austronesian language Manam. Attention is then focussed on the fact that in some languages — e.g., in Winnebago — foot boundaries may occur inside syllables that have more than one stress-bearing element, whereas in other languages — e.g., in Cairene Arabic and Yupik Eskimo — syllable-internal foot bound-aries are not allowed. To deal with this type of variability it is proposed that in addition to idiosyncratic stresses the theoretical framework must admit also idiosyncratic constituent boundaries. The effects of these theoretical innovations are illustrated by an examination of stress assignment in different Yupik dialects.