Where do syllables come from?

Abstract

Where do syllables come from? Evelyn Martens and Walter Daelemans and Steven Gillis and Helena Taelman Universitaire Instelling Antwerpen Universiteitsplein 1 2610 Wilrijk Belgium Abstract Young children are able to segment words into sylla- bles, even though there are no perceptual or acous- tic cues that indicate syllable boundaries in the pri- mary linguistic data. We show that information about word boundaries can be used to predict syl- lable boundaries by replicating the results of exper- iments done by Gillis and De Schutter (1996) with children who syllabified Dutch disyllabic monomor- phemes with a single intervocalic consonant. Word boundary probabilities were statistically computed in child language corpora and used to predict sylla- ble boundaries with a simple statistical model. The children’s syllabification behavior could be simu- lated using word-boundary probabilities estimated from child language corpora. Similar results were obtained for three di↵erent corpora. In our sim- ulations, we also investigate the question whether children acquire their knowledge of word boundaries from words from the input, from the intake, or from their own output. Introduction The syllable is an important construct in phono- logical descriptions of languages (Van der Hulst & Ritter, 1999) as well as in models of language ac- quisition (Jusczyk, 1997) and language processing (Levelt, 1989). In most contemporary phonologi- cal theories the syllable plays an important role at the segmental level (e.g., in consonant harmony) as well as at the supra-segmental level (e.g., in stress assignment). Across languages syllables adhere to a number of universal principles (Venneman, 1988) and Clements (1990) proposes a universally valid al- gorithm for syllabifying words. One of its operat- ing principles is ’sonority sequencing’: a syllable has rising sonority from the left edge to the vocalic nu- cleus and falling sonority from the vowel to the right edge. Irrespective of the theoretical framework in which the universals of syllabification are cast, it is accepted that the language universals, such as those incorporated in Clements’ algorithm, can be over- ruled by language-specific constraints. For instance, at the end of a syllable long vowels are universally accepted, but languages di↵er as to whether there can be a short vowel at the end of a syllable (Kager, In sharp contrast to the relatively clear phonolog- ical picture stands the phonetic reality: what are the acoustic correlates of the syllable in the speech stream? For instance, acoustic correlates of the ’sonority sequencing principle’ are very difficult to determine, which led phoneticians to define the syl- lable from a phonetic point of view as that entity of which the word syllable has three. The syllabic nucleus (the vowel) is fairly easy to detect, but the syllable boundaries are not straightforward. For in- stance, the /I/ in bitter is the nucleus of the first syllable, but where is the boundary of that syllable: immediately after the vowel /bI.t@r/ or after the first consonant /bIt.@r/ or in the middle of the first consonant /bIt.t@r/, a case of ambisyllabicity? This brings us to the core issue addressed in the present paper: if from a structural perspective syllables are easy to describe, but if it is very difficult to depict the acoustic correlates of the syllable and its bound- aries, it is an outstanding question how children ar- rive at detecting syllables and their boundaries. Nevertheless, in early speech perception (Jusczyk, 1997) as well as in speech production (Wijnen, 1988) children appear to use syllables as organizing enti- ties. The question is: how does a child acquire the knowledge of the structure of syllables? In the acquisition literature there are basically two approaches: in a nativist approach, the universals of syllable structure are thought to be innately given: they are described as inborn parameters (Fikkert, 1998), or as inborn constraints (Kager, 1999; Lev- elt, Schiller, & Levelt, 2000). Acquiring the struc- ture of syllables requires a child to figure out the language-specific parameter setting or the language- specific constraint ranking. Thus, the broad outlines are genetically given, so that only on the basis of the ambient language the child has to determine where precisely her mother tongue fits into these outlines. Appealing as this may sound, it is unclear on what basis parameters are set or constraints are ranked. The cues for parameter setting or constraint ranking can only be found in the input. However, the acous- tic correlates of the syllable are not clear in the input (see second paragraph). The alternative approach is that children do not start from a preset body of knowledge, but instead