The Roles of Phonotactics and Frequency in the Learning of Alternations

Several recent studies have examined the acquisition of morphophonological alternations, with apparently conflicting findings. Production and comprehension studies with both real and nonce words suggest that 31⁄2 year old Dutch-learning children do not yet have productive knowledge of the morphophonological voicing alternation (Zamuner, Kerkhoff, Fikkert, and Westrek 2005; Zamuner, Kerkhoff, and Fikkert 2006; Zamuner, Kerkhoff, and Fikkert 2007). Although five year olds perform well on real alternating words (Kerkhoff 2007), even seven year olds are reluctant to extend alternations to novel forms (Kerkhoff 2004). At the same time, in an artificial language learning task, White, Peperkamp, Kirk, and Morgan (2008) find evidence that one year olds are already learning morpho-phonological alternations. This paper presents the results of computational simulations that may help explain these divergent findings. Given data representative of the Dutch voicing distribution, the computational model predicts learning curves consistent with the Dutch acquisition findings, with a long delay for alternating forms. The focus of this paper is on analyzing the behavior of the model to determine what properties of the input or assumptions about the learning process underlie the observed effects. The analyses reveal that two independent properties of the input distribution in Dutch conspire against the alternating segments in Dutch. A major focus of the analyses is on the interaction of prior phonotactic learning with the learning of alternations. A number of researchers have observed that phonotactic learning may be helpful in the learning of alternations (Hayes 2004; Prince and Tesar 2004; Jarosz 2006; Tesar and Prince 2003 / 2007). Despite the delay in the learning of voicing alternations in Dutch, the analyses suggest the observed effects are consistent with early phonotactic learning that aids in subsequent learning of alternations. The rest of the paper is organized as follows. Section 2 presents the learning model. Section 3 reviews experimental findings on the acquisition of the Dutch