Pretonic schwa reduction in dutch: frequency effects interact with phonotactics

Abstract Word frequency plays a role in reduction processes: high-frequency words undergo more reduction than low-frequency words. Pretonic schwa reduction in Dutch is susceptible to frequency effects in this way, but the well-formedness of the cluster that remains after schwa deletion also plays a role. An experiment shows that cluster well-formedness and frequency effects in fact interact with each other. This suggests that phonological models should incorporate both lexical as well a grammatical information. Key words: Frequency effects, phonotactics, pretonic schwa, reduction. Introduction It has been observed for Dutch, that schwa deletion is more likely to be applied, when the resulting consonant cluster consists of an obstruent and subsequent liquid (Booij 1995). Thus all other things being equal, gelijk >$%–O- $%–QLS@µPHDQ¶, because in Dutch $O is a well-formed cluster but $n is not. However, pretonic vowels in Dutch are also known to be more reduced in high-frequency (HF) words than in low-frequency (LF) words (van Oostendorp 1999). This raises the question of whether frequency effects interact with the well-formedness of the cluster that remains after deletion. Is it the case that frequency effects are only found in potential well-formed onset clusters? Or is it rather the case that schwa reduction in consonant-liquid clusters is dependent on word frequency? How does frequency interact with phonotactics exactly? To investigate the effect of cluster well-formedness (CWF) and frequency, a word reading task was carried out. HF and LF words were extracted from Corpus Gesproken Nederlands (Spoken Dutch Corpus) available on www.tst.inl.nl and from the CELEX database on www.celex.mpi. Each HF word with WKHFRQWH[W&%&- was matched with an LF word following a number of criteria: the consonants preceding the schwa and following the schwa were identical, the number of syllables was identical, the number of phonemes was matched as much as possible and stress placement was identical. The corpus frequency counts may deviate from the frequency of the individual lexicon of the subjects. It is well-known in psycholinguistics that word frequency is related to naming latency, in the sense that HF words have shorter latencies than LF words (Oldfield & Wingfield 1965, Jescheniak &