Semantics and phonology constrain compound formation (original) (raw)

Berent and Pinker (2007) presented five experiments concerning the formation of compounds, especially the apparent restriction on the occurrence of "regular" plurals as modifiers (as in *RATS-EATER). Their data were said to support a "words and rules" approach to inflectional morphology, and to contradict the approach developed by Haskell, MacDonald, and Seidenberg (2003) in which multiple probabilistic constraints, mainly involving semantic and phonological properties of words, determine degree of acceptability. We examine Berent and Pinker's studies and show that a) their experiments tested hypotheses that are incorrectly ascribed to our theory, and b) their data are actually compatible with our account. Contrary to the words and rules approach, there are phonological effects on modifier acceptability that cannot be subsumed by a grammatical rule. Berent and Pinker (2007) have offered a response to our article (Haskell, MacDonald, & Seidenberg, 2003), which provided evidence against the theory that Pinker (1994, 1999) has proposed as an explanation of the apparent bias against regular plurals as modifiers in compounds (e.g., MICE-EATER is acceptable to many English speakers, whereas *RATS-EATER is not). 1 This analysis of compound formation has been repeatedly presented as supporting broader conclusions about the nature of grammatical knowledge, the distinction between words and rules, and the need for innate constraints to explain how such knowledge can be acquired (e.g., Gordon, 1985; Pinker, 1994, 1999). This account is based on level-ordering, a conceptual framework that arose within phonological and morphological theory some years ago (Kiparsky, 1982; Siegel, 1974) and which drew distinctions between regular and irregular inflectional processes. Since its introduction, the level-ordering account has been subject to exten