The proto-lexicon: Segmenting word-like units from the speech stream

There is growing evidence from experimental studies that infants start acquiring a vocabulary already when they are between six to nine months old (Bergelson & Swingley, 2012; Parise & Csibra, 2012; Tincoff & Jusczyk, 1999). Clearly, infants must have learned what these words mean prior to their lab visit, that is, learned them in their natural environment. However, as infant speech corpora reveal, there are surprisingly few one-word utterances directed to 6-12-month-olds in maternal speech (excluding exclamations, fillers and social expressions that typically do not combine into phrases): ranging from 2% (Morgan, 1996) to 9% (Brent & Siskind, 2001). Even when parents are explicitly instructed to teach their child certain words, they present these words predominantly in multi-word utterances (Aslin et al., 1996; Johnson et al., 2013). It appears that, for most children, most of the words they hear are presented in continuous speech. As Figure 2.1 demonstrates for a typical infant-directed example from Dutch, words are generally glued together without clear pauses between them to signal word onset and offset (unlike written text, in which words are conveniently separated by spaces). While it is debated whether or not isolated words alone can function as a starting point for vocabulary acquisition (Brent & Siskind, 2001; Depaolis, Vihman & Keren-Portnoy, 2014; Lew‐Williams, Pelucchi & Saffran, 2011; Yang, 2004), it is undeniably true that in order to learn from the input, infants are required to segment the speech stream into separate word-like units from very early on. From adult research, we know that the speech signal contains several cues that assist successful word extraction from speech, but these cues are probabilistic rather than deterministic: no single cue appears sufficient to signal word boundaries (Cutler, 2012). Fortunately, the past two decades have provided us with two experimental methods that allow us to examine when and under what conditions infants can segment words from speech. In what follows next, we will first discuss the main methods (Section 2.1), before we focus on the cues that infants can use to detect words from speech (Section 2.2). Section 2.3 underscores how vital early speech segmentation skill is for subsequent language development, while Section 2.4 ends with future directions.