Published online by Cambridge University Press: 30 July 2012
There have been claims in the literature that the variability of compound stress assignment in English can be explained with reference to the informativeness of the constituents (e.g. Bolinger 1972, Ladd 1984). Until now, however, large-scale empirical evidence for this idea has been lacking. This paper addresses this deficit by investigating a large number of compounds taken from the British National Corpus. It is the first study of compound stress variability in English to show that measures of informativeness (the morphological family sizes of the constituents and the constituents' degree of semantic specificity) are indeed highly predictive of prominence placement. Using these variables as predictors, in conjunction with other factors believed to be relevant (see Plag et al. 2008), we build a probabilistic model that can successfully assign prominence to a given construction. Our finding, that the more informative constituent of a compound tends to be most prominent, fits with the general propensity of speakers to accentuate important information, and can therefore be interpreted as evidence for an accentual theory of compound stress.
The authors wish to thank Sabine Arndt-Lappe, Kristina Kösling, Gero Kunter and three anonymous Journal of Linguistics referees for their feedback on earlier versions. Special thanks also to Harald Baayen for discussion and support. This work was made possible by an AHRC postgraduate award (114 200) and a major studentship from Newnham College, Cambridge, to the first author as well as two grants from the Deutsche Forschungsgemeinschaft (PL151/5-1, PL 151/5-3) to the second author, all of which are gratefully acknowledged.