Words

When children start to utter their first words, an enormous segmentation work lies behind them: they had to find out what exactly in the speech stream constitutes a word. It was suggested that the supposed sensitivity to prosodic boundaries would help infants to segment words from fluent speech, as prosodic boundaries always coincide with word boundaries. Recent research has provided evidence that 8-month-old infants more readily segment words that occur at boundary positions as opposed to words that occur in utterance-medial positions. Seidl and Johnson (2006) found that American English- learning infants recognised monosyllabic words and non-words, like cash and geff, that had been previously presented in utterance-initial or utterance-final position easier than monosyllabic words that had been presented in utterance-medial position.

18 _{This classification follows Hirsh-Pasek and Golinkoff (1996, p. 17ff). The rationale behind the use of this}

terminology is that both inside-out theories, like nativism, and outside-in theories, like cognitivism, assume innate capacities which allow the acquisition of the language. The difference is that inside-out theories claim these innate capacities to be domain-specific, that is, specifically and uniquely for language acquisition, whereas outside-in theories believe them to be domain-general.

In a second study, Seidl and Johnson (2008) provided further evidence for their Edge

Hypothesis. In a series of experiments testing 11-month-olds on the recognition of

monosyllabic vowel-initial words and non-words, like ash and eeb, they found differences in the facilitation effect of the positions: the segmentation and recognition of words presented in utterance-final position seemed to be easiest, followed by words presented in utterance-initial position and words that were presented in utterance-medial position seemed to be more difficult to segment. Thus, the occurrence of a word boundary that coincided with the prosodically marked right boundary of an utterance seemed to facilitate the segmentation of whole words in fluent speech.

The facilitation effect for segmenting words from the utterance-final position was also found for bisyllabic words in German-learning 9-month-olds. Schmitz et al. (submitted) found that infants more easily segmented and recognised bisyllabic words, like Balken ‘beam’ and Pinsel ‘paint-brush’, that had been presented in utterance-final position as opposed to words that had been presented in utterance-medial position. Thus, the sensitivity to prosodic boundaries not only helped infants to segment monosyllabic words which are identical with the prosodically marked final syllable of an utterance, but also allowed the segmentation of bisyllabic words from the utterance-boundary. That the sensitivity to a phrase boundary helped infants in the recognition of words was also found in the study by Gout and colleagues (Gout, Christophe & Morgan, 2004) which will be discussed in detail in the next section of the present paper (2.2.3.2).

The segmentation of monosyllabic words from fluent speech, in which the position of the target word was variable, was also found in earlier studies. Jusczyk and Aslin (1995) provided evidence that 7.5-month-old, but not 6-month-old American English- learning infants were able to recognise monosyllabic content words, like cup and feet, in fluent speech. Infants of this age were not only able to recognise stressed content words from fluent speech, but also unstressed closed-class elements, like the determiner das ‘the’ in German 7.5-month-olds (Höhle & Weissenborn, 2003). Shi and colleagues (Shi, Werker & Cutler, 2003) found that 13-month-olds, but not 8-month-olds, recognised real function words, like its, as opposed to nonsense function words, like ots, in an NP19 (see also Shi, Cutler, Werker & Cruickshank, 2006). Furthermore, Shi and colleagues (Shi, Marquis &

19

That 14-month-old infants used high frequent function words, like determiners, to categorise the subsequent word as a noun could be shown in a study by Höhle and colleagues (Höhle et al., 2004).

Gauthier, 2006) found that French-learning infants at the age of 6 months also recognised and segmented a determiner, like la ‘the’(fem.) within an NP and are able to discriminate

between the determiner and a nonsense function word like ta, at the age of 8 months. In all these studies, it was argued that the high frequency of occurrence of function words in the input may have allowed infants to recognise these words, despite the fact that they are prosodically less salient than the stressed content words. In addition, function words like determiners usually occur at the left boundary of an utterance in languages like German, English or French and therefore might also be easier to segment for the infants.

The sensitivity to the rhythm of their native language also seems to influence infants’ ability to segment bisyllabic words from fluent speech. Jusczyk and colleagues (Jusczyk, Houston & Newsome, 1999) found that 7.5-month-old American English- learning infants were able to recognise trochaic words, like kingdom or doctor, from fluent speech, but not iambic words, like guitar or surprise. That is, the infants seemed to rely on the metrical segmentation strategy appropriate for stress-timed languages (cf., Cutler, 1994, 1996). The iambs were missegmented, that is, the infants seemed to perceive the strong syllable of the iamb as a word onset and integrate the next following syllable to form another trochee, that is, infants seemed to perceive a word combination like ‘guitar is’ as ‘taris’ (Jusczyk et al., 1999, Experiment 11). The experiments also suggested that the infants represented the whole trochee as a coherent unit rather than only the strong initial syllable, that is, kingdom rather than king. From 10.5 months on, the infants were also able to recognise iambic words in fluent speech and also represented these iambic words as coherent units. This was interpreted in a way that by the end of the first year the infants no longer rely only on the rhythmic segmentation strategy to acquire the language, but also use other information to integrate knowledge, like allophonic cues.

Recent studies on German infants’ discrimination of trochaic vs. iambic lists of bisyllabic sequences, like 'gaba20 vs. ga'ba, provided evidence that 6-months-olds, but not 4-month-olds, preferred to listen to the trochaic over the iambic sequences (Höhle, Bijeljac-Babic, Herold, Weissenborn & Nazzi, submitted). However, Friederici and colleagues (Friederici, Friedrich & Christophe, 2007), who used a non-behavioural experimental technique, namely the Event-Related Potentials (ERP) paradigm, provided

evidence that infants as young as 4 months are able to discriminate trochaic and iambic stimuli. This was reflected in a mismatch response (MMR) of the brain, in which a positive wave in the EEG signal was found for the stimulus type that is less frequent in their language, that is, to the iambic stress pattern in German infants. Thus, it seems that infants are able to discriminate trochaic from iambic stimuli already by 4 to 6 months of age. But solving the more demanding task of recognising trochees in fluent speech can first be observed at about 7.5 months of age. Only by the age of 10.5 months, the infants seem to integrate other speech cues to allow them to discriminate iambic words from fluent speech.

But how do infants learn about the differences between a multi-syllabic word and a sequence of words, which is a prerequisite for identifying phrases and grammatical relations within phrases in the speech signal? Johnson and colleagues (Johnson, van de Weijer & Jusczyk, 2001) tested whether 7.5-month-old infants perceive a 3-word phrase as a coherent unit, that is, as a single word or not. Based on the finding by Santelmann and colleagues (Santelmann, Houston & Jusczyk, 1997) that 7.5-month-old infants were able to identify a three-syllabic word with a strong-weak-strong stress pattern like parachute as a coherent unit, Johnson et al. (2001) familiarised infants with sentence passages which contained a 3-word-phrase like pair of mugs, which also has a strong-weak-strong stress pattern. During the test phase, however, the 7.5-month-olds did not listen longer to the word lists containing the familiarised combination of words than to the lists containing previously unfamiliarised combinations of words. Johnson et al. (2001) concluded that the infants did not perceive and represent the 3-word-phrase as a coherent unit, that is, not as a single trisyllabic word, and therefore did not recognise the phrase in the test phase. They suggest that infants integrate speech cues like coarticulation cues as a means to perceive word boundaries and distinguish between multisyllabic words and multi-word phrases.

To summarise, infants are able to use a range of cues in their perception of words in fluent speech, including prosodic cues which occur at the boundaries of utterances. The studies cited above show that already by 7.5 months infants are able to discriminate monosyllabic, bisyllabic and even trisyllabic words from fluent speech and that they use their knowledge to further segment the speech stream and, during the second year of life,

start to categorise words from the input. Furthermore, at the age of 7.5 months, the infants are able to perceive a 3-word-phrase as a non-coherent unit.

The next two sections will present studies on infants’ perception of phrases and clauses in fluent speech. Many of these studies used the insertion of artificial pauses at boundaries or within utterances to investigate infants’ recognition of phrases and clauses. That is, these studies used the insertion of a pause as a gap inserted between phrases or within phrases, without addressing the question of infants’ sensitivity to pauses and the pause hierarchy. Therefore, studies on the perception of pauses will be presented in a separate section (2.2.3.4) in the present paper.