The present paper provides a survey of the influence exerted by rhythm on various domains of linguistic constituency. It fosters a layered conception of the structure of linguistic utterances (similar to the one present in neural network models) and argues that the intensity of rhythmic effects decreases with an increasing distance between the linguistic subsystem concerned and the rhythmic system. The latter is part of the phonological layer, which, by hypothesis, borders on the morphological layer (comprising affixes as well as free grammatical words), which, in turn, connects with the syntactic layer, which is itself framed by the vast system of semantic concepts. Thus, in the production of an utterance, the linguistic system seeks to solve a problem of rhythmic well-formedness by means of minimal modifications as close as possible to the place where the problem arises. For this reason, most rhythmic problems are solved on the rhythm-internal level, many are solved on the morphological level, some on the syntactic level and only a few on the semantic level.
The argument is based on a large number of representative case studies, surveying grammatical variation phenomena that have been shown to be sensitive to the rhythmic preference for alternating stressed and unstressed syllables (Schlüter 2005). Among the phenomena studied are numerous instances in which the application of rhythm rules suffices to render the construction grammatically acceptable. In many cases, however, the presence or absence of a grammatical morpheme is determined by the need for a buffer syllable. There are some variation phenomena concerning optional morphemes that still are or used to be free syntactic forms, which occupy the borderline between morphology and syntax. A few other phenomena clearly belong to the domain of syntax, but hardly any cases have been detected where a semantic alteration can be ascribed to a rhythmic problem. In addition to the raw numbers of relevant phenomena testifying to rhythmic influences, the empirical, corpus-based methodology employed in the case studies allows us to quantify the magnitude of the effects obtained.
The qualitative and quantitative gradation of rhythmic influences on other levels of linguistic structure can be taken to derive from the layered structure of a neural network. In this network, the layers (corresponding to the modules in traditional grammatical theory) interpenetrate each other, but only to a limited extent. Morphological choices are very strongly determined by the resultant rhythmic constellations, syntax is much less under the sway of rhythm, and semantics tolerates at most minimal adjustments in favour of a better rhythm.