'The (Non)Realization in Children's Utterances: Evidence for a Rhythmic Constraint'

[pagina 59]

The (non)realization of unstressed elements in children's utterances: evidence for a rhythmic constraintGa naar voetnoot*

Frank Wijnen

University of Utrecht

Evelien Krikhaar

University of Groningen

and

Els den Os

University of Amsterdam

Abstract

In this study it is argued that the omission of closed class morphemes and of unstressed syllables within words is related to their common characteristic, viz. that they are unstressed, rhythmically weak parts of utterances. Several strands of evidence indicate that it is unlikely that children are unable to perceive these elements in the input speech. The pattern of (non)realization of unstressed syllables within content words and the class of determiners, was analysed in two Dutch children from 1;6 to 2;11. It appeared that polysyllabic words were quite generally truncated in such a way that they fitted a trochaic (strong-weak) pattern, particularly in the early samples. Some observations with respect to the (non)realization of determiners are suggestive of an influence of an SW-constraint on the realization of noun phrases. These findings support the hypothesis that in the course of utterance preparation, words and phrases are mapped onto S(W) templates. Anecdotal evidence suggests that the dissolution of the SW-constraint coincides with the acquisition of specific aspects of stress assignment in Dutch, such as quantity sensitivity.

[pagina 60]

Introduction

The omission of unstressed syllables from polysyllabic words is widely documented in children acquiring various, divergent languages. It appears to be a virtually universal characteristic of early child language. Several explanations for this phenomenon have been advanced, which represent the classic oppositions in developmental psycholinguistic theorizing: perception vs. production, and performance vs. competence (Clark & Clark, 1977). Not many of these explanations, however, have related syllable omissions to another, fairly wide-spread phenomenon in early child language, viz. the absence of so-called grammatical morphemes (Brown, 1973). In the present study an attempt is made to relate both phenomena, starting from the simple observation that most grammatical morphemes and unstressed syllables within words share the phonological property of being unstressed, or rhythmically weak.

The absence of grammatical morphemes from early (Stage I, or telegraphic) child utterances has been explained by Brown (1973), and others after him, by referring to their semantic complexity. Grammatical morphemes are ‘modulators’ of meaning. The concepts they express (e.g. determinacy, number) are more abstract than those expressed by the content words, and could, therefore, be more difficult to acquire. In support of this idea, Brown provides evidence showing that the order of appearance of the famous first 14 grammatical morphemes can be explained by their cumulative semantic complexity.

More recently, attempts have been made to relate the absence of these morphemes to the acquisition of syntax. Many of the grammatical morphemes are specifically linked to what syntacticians call functional projections. For instance, articles and other determiners, as well as pronouns, are heads of DET projections, and auxiliaries are heads of INFL projections. It has been argued that functional projections are not yet present in the syntactic representations of Stage I children, so that, even if they knew the morphemes that are linked to them, they would be unable to express them, as they cannot be syntactically licensed (Radford, 1990). It is held that the presence and character of functional categories varies parametrically across languages. Eventually, on the basis of positive evidence, children who are confronted with languages such as English or Dutch will alter their phrase structure representations in order to include the appropriate functional categories. This will be reflected by the appearance of the associated morphemes in language production.

The semantic and syntactic explanations of the absence of grammatical morphemes are challenged by Pye's (1983) observations on the acquisition of Quiché. Pye noted that two-year-old children, in their attempts to use the morphologically complex verb forms of Quiché, often did not produce the

[pagina 61]

verb's root morpheme (i.e. a semantically ‘simple’ content word), but a syllable consisting of the last consonant of the root morpheme plus a VC-shaped termination suffix, expressing various highly abstract semantic and syntactic features (Pye, 1983: 585ff.). This can be related to the phonological form of the verbs. Morpheme and syllable boundaries do not coincide in Quiché verbs, due to resyllabification. The part that children produce corresponds to the syllable bearing primary stress. Pye argues that stress patterns in the input language are a major determinant of the acquisition and use of words by children. The grammatical morphemes in languages such as Dutch and English are generally unstressed, even unstressable. Thus, although semantic and syntactic factors undoubtedly contribute to the acquisition of the grammatical morphemes (it is hard to see, for instance, how the order of acquisition can be explained by reference to prosodic differences alone), their initial absence from the utterances of children who acquire these languages might be explained by their prosodic characteristics. This explanation suggests a connection to the omission of unstressed syllables within words. The question then is what kind of process or mechanism related to stress might account for the omission of both unstressed syllables and grammatical morphemes.

As a first option, we might look at the perception of speech. In comparison to stressed syllables, unstressed syllables have a shorter duration and a lower amplitude. Moreover, they do not involve the pitch modulations that are associated with intonational accents. Consequently, unstressed syllables may be masked by stressed syllables, and the resulting reduction in perceptibility may be the reason why children omit them in their own speech. There are several arguments against this solution. First, Pye (1983) observed that children acquiring Quiché do also produce verb roots, particularly when the verb occurs sentence-medially. Pye connects this to the variation in stress placement as a function of the verb's sentence position in the input language: the final syllable is stressed when the verb occurs in sentence-final position, whereas the verb root is stressed in sentence-medial position. The children's behaviour indicates that they perceive and store both verb root and suffix. Consequently, the fact that they systematically avoid producing both parts within the same utterance cannot be ascribed to a perceptual problem. It is on the basis of data such as these that Gleitman, Gleitman, Landau & Wanner (1988) suggest that for young speakers, ‘word’ and ‘stressed syllable’ are equivalent.

A second argument against the perceptual explanation can be derived from experimental studies of children's sentence comprehension. Shipley, Smith & Gleitman (1969) and Petretic & Tweney (1977) found that children who omit function words from their speech nevertheless are more likely to respond appropriately to utterances containing function words than to utterances without them. Moreover, Gerken, Landau & Remez (1990), using

[pagina 62]

an imitation paradigm with existing and nonsense function words, showed that although stress plays a major role in the tendency to exclude function words from imitative responses, existing function words were responded to differently from nonsense words, which suggests that, despite their relative inconspicuousness, these words were perceived and processed.

Finally, it appears that parts of the syllables that are omitted from multisyllabic words (particularly consonants or consonant features) are retained, and may be substituted for, or added to the segments of one of the realized syllables (Allen & Hawkins, 1980; Fikkert, 1991). Examples from Fikkert's data base of Dutch child phonology are given in (1):

(1a)	telefoon	/‘te:lə'fon/	‘telephone’:	[,tɔ:]	(Tom 1;7.11)
(b)	ballon	/bα'lɔn/	‘balloon’:	['plɔm]	(Tom 1;7.9)
(c)	konijn	/ko'nειn/	‘rabbit’:	['ηει]	(Tom 1;5.28)

In (1 a) the vowel of the realized syllable roughly corresponds to that in the adult model, but it is preceded by the onset of the word's initial (unstressed) syllable. In (1 b) the initial /b/, pronounced as [p] is added to the second (stressed) syllable onset. In (1 c) finally, the initial nasal in the child's realization is velarized, presumably as a result of the velar stop /k/ in the adult form. Examples such as these, which are typical for most children, suggest that the segmental information contained in the unstressed syllables is perceived and stored. Thus, these examples are also hard to reconcile with the hypothesis that listeners tend to consider stressed syllables as word onsets (Cutler & Norris, 1988), which predicts that information preceding the syllable bearing primary stress would be excluded from language learners' initial lexical representations.

If perceptual explanations of children's omissions of unstressed syllables and morphemes fail, as the above overview suggests, the logical next step would be to look for an explanation in the production domain. Allen & Hawkins (1978, 1980) have proposed that the omission of unstressed elements is the result of a condition on the rhythmical structure of utterances. After an initial stage in which only monosyllabic utterances are allowed, children's utterances will tend to meet two requirements: (1) weak (unstressed) and strong (stressed) syllables should alternate; and (2) initial syllables should be strong. Together, these requirements constitute a trochaic bias, which, as Allen & Hawkins surmise, reflects a universal of human perceptual and motor performance.Ga naar voetnoot1 Although in adult speech the universal tendency to alternate between strong and weak syllables in a regular fashion is still discernible (see Kelly & Bock, 1988, and their references), stress and rhythm are largely determined by language-specific stress rules. However, in the speech of children who have not yet acquired language-

[pagina 63]

specific rules of stress and rhythm, the trochaic bias is argued to be directly reflected by the widely reported tendency to omit unstressed syllables that either are word-initial (W₁S(W₂) → S(W₂)), or that precede other unstressed syllables (SW₁W₂ → SW₂). Allen & Hawkins' claim has been challenged by Hochberg (1988) on the basis of imitative and naturalistic data from children acquiring Spanish. Hochberg states that a rhythmic bias cannot be discerned in her subjects' speech. It should be noted, however, that she eliminated polysyllabic realizations of monosyllabic target words and truncated realizations of polysyllabic targets - in other words, the data on which Allen & Hawkins' claim is based - from her analyses. Even so, it appears that in Hochberg's data set, children's realizations with final stress are less frequent than those with penultimate stress. Thus, the contrast between Allen & Hawkins and Hochberg seems to be largely superficial.

Allen & Hawkins' trochaic bias may be interpreted as a constraint on phonological (prosodic) representations that is specific to an early stage in the development of language. At least two different conceptions of the level of processing and representation at which this constraint is implemented are possible. First, it might be part of the system by which the stored phonemic or featural information (which is based on auditory input) is transformed into output representations. Traditionally, child phonologists have described this ‘transduction system’ in terms of rewrite rules (Menyuk, Menn & Silber, 1986). To attain compatibility with modern non-linear phonology, we might assume that the transduction system involves the mapping of ‘melodic’ (i.e. segmental or featural) information onto phonological representations at various - syllabic and prosodic - levels of description. This seems to provide a natural fit with current models of language production, which assume that planning involves the insertion of information stored in memory (words, segments, features) into planning frames that are derived from the relevant sets of grammatical rules (Dell, 1986; Levelt, 1989; Wijnen, 1990, 1992). Supposedly, the only type of frame that is allowed at the level of prosodic planning during an early stage of language acquisition is one that captures the Strong-Weak structure. We will refer to this hypothetical representation as the ‘SW-template’. A question that may be raised is whether it is derived from universal grammar (Dresher & Kaye, 1990), or whether it reflects the child's initial inductive hypothesis concerning the metrical structure of the input language. In the latter case, we would expect to find the trochaic constraint only in children acquiring languages such as English or Dutch, in which 75% or more of the words are trochaic (Cutler & Carter, 1987; Quené, 1992).

Alternatively, the trochaic constraint might take effect within the lexicon. This would imply that immediately upon acquisition, a new word would be adapted to the trochaic template. This hypothesis leads to predictions that differ from those related to the previous one in two respects. First, we would

[pagina 64]

not expect to find free variation in the realization of word forms. For instance, a word like banana [bə'nænə] should always be realized as, e.g. ['nænə], for this would represent the stored form. Secondly, an effect on the realization of phrases and sentences is ruled out. In contrast, the assumption that a trochaic constraint takes effect during transduction (i.e. utterance planning) makes variation in realized forms conceivable. So, for instance ['nænə], ['bænə] and [bə'nænə] might occur as realizations of banana. Moreover, not only words, but also phrases and sentences might be adapted to fit the template, so that a particular realization may depend on the rhythmic structure of the surrounding utterance. The complete utterance, however, should agree with the trochaic pattern.

Clearly, only the ‘planning’ version of the rhythmic constraint hypothesis can account for the omission of both unstressed syllables and grammatical morphemes by utilizing their common phonological characteristic. Some evidence in support of this hypothesis was gathered by Gerken (1990, 1991) in a sentence imitation task with two-year-old children. She found that unstressed closed class words, such as determiners and pronouns, which constituted weak parts of iambic feet were systematically omitted. In a later study (Gerken, 1992), a positive correlation was found between two-year-olds' MLUs and their tendency to preserve weak (unstressed) syllables in the imitation of multisyllabic nonsense words. Since increases in MLU in the stage of development that Gerken studied predominantly reflect the increasing use of closed class morphemes, this correlation corroborates the idea that the omission of both closed class morphemes and unstressed syllables reflects the same underlying mechanism.

In this study, we analyse the patterns of realization and omission of unstressed syllables in content words, and of a particular class of function words, viz. determiners, in two Dutch children's spontaneous utterances, recorded between 1;6 and 2;11. Our aim was to establish whether these children adapt the rhythmic form of words and phrases in such a way that trochaic patterns result. We also wanted to find out whether the subjects only use omission (truncation) as a means to this end, or whether they are also able to deploy other strategies, such as stress shifting or the addition of ‘filler’ syllables.

In addition to our attempt to establish whether or not children adapt the rhythmic form of words and phrases, we also address the question of whether the composition of early vocabularies reflects a bias towards words that match the rhythmic constraint. Child phonologists have found that the intake of words from the ambient language may be affected by certain segmental biases. Children may avoid words that contain sounds they are unable to pronounce, or they may show a preference for structures that fall within their capabilities (Waterson, 1971; Schwartz, 1988). It is conceivable that a similar ‘lexical filtering’ occurs with regard to rhythmic forms.

[pagina 65]

Finally, should the early utterances indeed be bounded by the trochaic constraint, we will examine whether later there are any signs of the abandonment of this early pattern in favour of the stress rules of Dutch.

Method

Subjects

This study is based on a longitudinal corpus of utterances produced by two children acquiring Dutch: Daan (male) and Marloes (female). Both are first-born children, growing up in middle-class families. At age 2;0, Marloes was enrolled in a daycare centre.

At the start of the observation period, Daan was 1;7, Marloes 1;6. At the end, their ages were 2;11 and 2;10, respectively. Daan's MLUs were 1·0 and 3·6 in the first and last samples, respectively. The values for Marloes were 1·0 and 3·2. The average length of words (in syllables) in the first sample were 1·69 for Daan and 1·49 for Marloes. These figures indicate that both children were capable of producing multisyllabic words.

Data collection

The children were recorded in their homes, at monthly intervals (one session with Marloes, at age 2;8, was missed due to a holiday). Each recording session lasted for approximately 60 minutes. Usually, the children were interacting with their mother. Mother and child were often engaged in play. Also, picture books were used to stimulate naming. In some of the later recordings of Marloes, her father was present. Furthermore, two researchers (the second and third authors) were present, one handling the recording equipment, the other observing and taking notes. The subjects were recorded on both video- and audiotape. The audio equipment was of high quality (Marantz CP-43O cassette recorder and two Sennheiser MD21 microphones), so as to allow for acoustic analysis of the children's speech (see Den Os, 1990a, b).

Transcription and coding

The first 100 child utterances in each session were fully transcribed, both orthographically and phonetically. The rhythmic form of all content words (nouns, verbs and adjectives) was coded, in terms of strong (S) and weak (W) syllables. The syllable bearing primary stress was coded as S, all other syllables were coded as W. It appeared that the data set thus obtained contained too few tokens of W-initial words and words longer than two syllables to allow for quantitative analyses. Therefore, we added all tokens of W-initial and multisyllabic S-initial words that occurred in the remaining utterances to the data base. We did not distinguish between morphologically simple and complex words, since this appeared unnecessary for our purposes.

[pagina 66]

Generally, all words that were obviously imitated were discarded from the analyses.

The assignment of stress in the children's speech was evaluated perceptually by a phonetically trained person (the second author). To check for reliability, a sample of 314 word tokens taken from the two corpora (95 from Marloes, 219 from Daan) was evaluated by a second judge, a speech therapist. The rhythmic classifications assigned by the two judges agreed in 84% of the cases.

For all noun phrases in the first 100 utterances of each sample, we decided whether, according to adult standards, a determiner would have been obligatory, and we registered the appearance of a determiner or filler syllable (schwa).

Results

The composition of the early lexicon: filtering?

If a rhythmic constraint affects the intake of lexical items, it may be expected that words with initial S-syllables are preferred to words with initial W-syllables, and that words in which strong and weak syllables alternate are preferred to words with a non-alternating pattern. To test these predictions, we classified all content words in our subjects' speech according to their rhythmic pattern in adult usage (irrespective of the actually realized forms). Thus, in this section and throughout the paper, classifications of words (and phrases) in terms of rhythmic structure refer to their appearance in adult speech. Fig. 1 is a cumulative graph of the absolute numbers of S, SW, SWW, WS, WSW and WWS words (types) in the children's corpora as a function of age. Note that SW and WS, and SWW, WSW and WWS, constitute a minimally different pair and triplet, respectively, with respect to the position of the stressed syllable and the presence or absence of alternation, while the number of syllables, which might have an additional influence on the intake of words from the ambient language, is constant. If lexical filtering in accordance with the trochaic constraint occurs, we would expect a predominance of SW over all other word types. Moreover, we would not expect a clear difference between WS and WSW, except for a possible preference for the shorter words. Finally, we suspect that initial stress is more important than stress alternation, so that SWW should be preferred to WSW, which, in turn would be preferred to WWS.

The patterns in Fig. 1 show a close correspondence to these predictions SW-words are predominant in both children, throughout the period of observation, except for the very first sample of Marloes, they even outnumber monosyllabic words. The proportions of WS- and WSW-words are quite low, but, apparently, approximately equal. SWW-words are more frequent than WSW-words, and, finally, WWS are only marginally present at best.

[pagina 67]

In order to conclude, however, that these trends reflect rhythm-based filtering in lexical acquisition, we are required to show that the children's proportions of rhythmic types differ from the pattern that is prevalent in the speech addressed to them. Hence, we classified and tallied all content words in the speech addressed to the subjects by their mothers in the first four samples, and compared the resulting numbers to the distributions of rhythmic types in the children's corpora up to 1;11. We chose these samples because we suspected that a putative lexical learning bias would present itself

[pagina 68]

table 1. The numbers and percentages of content words in six rhythmic form classes as derived from the speech of the subjects' mothers in the first four samples

	Marloes' mother	Marloes' mother	Daan's mother	Daan's mother
Word type	Types	Tokens	Types	Tokens
S	102	245	54	121
	(42·5)Ga naar voetnoota	(45·5)	(25·6)	(20·6)
SW	97	209	100	351
	(40·4>	(38·8)	(47·4)	(59·8)
SWW	21	37	18	39
	(8·8)	(6·9)	(8·5)	(6·6)
WS	10	26	9	13
(4·2)	(4·8)	(4·3)	(2·2)
WSW	4	11	23	41
	(1·7)	(2·0)	(10·9)	(7)
WWS	6	10	7	22
	(2·5)	(1·9)	(3·3)	(3·7)
Total	240	538	211	587
	(100)	(100)	(100)	(100)

most clearly during the early stages. Additionally, until this age the children's mothers were the primary source of input language. Marloes, for instance, started to go to a day care centre at age two. The categorization of the input vocabularies was based on the words' expected rhythmic form, rather than transcriptions of the actual pronunciations, which were not available. This appears to be acceptable, however, in view of the fact that the researchers who recorded the mother-child interactions did not notice any systematic deviations from standard Dutch in the rhythmic structure of either mother's speech.

Table 1 shows that both mothers' choice of words in their child-directed speech quite closely matches the composition - in terms of rhythmic patterns - of their children's vocabularies. Chi-squared tests do not yield significant differences between the distributions of rhythmic types in the children's corpora and those in their respective mother's speech. In contrast, the children's distributions differ significantly from those in the speech of their counterpart's mother. Thus, the proportions of rhythmic types in Daan's corpus differ significantly from the proportions in Marloes' mother's speech (χ²(5) = 27·08, p < 0·001), and Marloes' proportions differ from those of Daan's mother (χ²(5) = 21·96, p < 0·005).

[pagina 69]

Realizations of content words

Chi-squared tests indicate that the overall proportion of correctly realized monosyllabic S tokens (94%) is significantly higher than the proportion of correctly realized polysyllabic, S-initial words (Marloes: 88%; χ²(1) = 13·15, p < 0·001; Daan: 86%; χ²(1) = 21·20, p < 0·001). The graphs in Fig. 2 suggest that the first two (Marloes) or three (Daan) sections of the observation period are largely responsible for this difference.

Polysyllabic S-initial tokens appear to be significantly more often correct than W-initial tokens (Marloes: 46%; χ²(1) = 203·41, p < 0·001; Daan:

[pagina 70]

68%; χ²(1) = 53·63, p < 0·001). Fig. 2 gives the impression that the effect in Daan is brought about mainly by the data from the first four months (1;7-1;10). In Marloes, on the other hand, the difference is present throughout the full period of observation. In other words, the children appear to show different developmental trends. In Marloes, we see a rather gradual increase of correct production of W-initial words, but she does not reach the level of 70% correct in the last four months of the observation period (2;7-2;10). Daan, on the other hand, shows a rather abrupt course of development, with below 20% scores at ages 1;7 to 1;10, and a jump to the near 80% level around age two.

In the following sections we will focus on the realizations of words that - in adult usage - can be described as S, SW, SWW, WS, WSW and WWS. Although words with more than three syllables (in the target form) occurred as well, their numbers were too small to allow for quantitative analyses. We will concentrate on the realizations that deviate from the target form, since these will allow us to gain insight into the child's rhythm rules; particularly, whether a trochaic constraint is operative.

 

Words with an initial S-syllable. All correct and incorrect realizations of S-and SW-content words in the first hundred utterances of all samples were tallied. Throughout the observation period, the majority of the S-words (94%) were correctly produced. Marloes made 33 errors, Daan 27. Most of these (24 in Marloes; 19 in Daan) involved schwa epenthesis in the context of a liquid and a non-coronal obstruent, which is practically standard in child Dutch:

(2a)	jurk	/jœrk/	‘dress’:	['jœrək]	(D 2;3)Ga naar voetnoot2
(b)	melk	/mεlk/	‘milk’:	['mεlək]	(M 1;10)

In a few cases (4 in each child), a schwa was appended at the end of a word:

(3a)	oom	/om/	‘uncle’:	['omə]	(M 2;3)
(b)	bal	/bαl/	‘ball’:	['bαlə]	(D 1;9)

And, finally, in the remaining cases (9 in all), we see the insertion or addition of a schwa at the beginning of a word, analogous to the cases of word-final epenthesis, resulting in a WS-structure.

(4 a)	klok	/klɔk/	‘clock’:	[kə'lɔk]	(M 1;10)
(b)	trein	/trn/	‘train’:	[tə'rειn]	(D 2;7)

Of the 1531 SW words (tokens), 94% were realized in a rhythmically correct way. The majority of errors (67%) were deletions of the final (unstressed) syllable:

[pagina 71]

(5a)	centjes	/'sεncjəs/	‘coins’:	[sε]	(M 1;10)
(b)	meisje	/'mειſə/	‘girl’:	[mειs]	(M 1;10)
(c)	molen	/'mo:lə/	‘windmill’:	[mo:l]	(M 1;10)
(d)	koffie	/'kɔfi/	‘coffee’:	[kɔf]	(D 1;10)
(e)	puzzel	/'pœsəl/	‘puzzle’:	[pœs]	(D 1;10)
(f)	onder	/'ɔndər/	‘under’:	[ɔn]	(D 1;10)

Other errors, in order of frequency, involved epenthetic schwa-insertion, resulting in SWW or WSW structures (26%, ex. 6), and stress shift (7%, ex. 7).

(6a)	varken	/'fαrkə/	‘pig’:	['xαrəkə]	(M 2;0)
(b)	vlinder	/'flιndər/	‘butterfly’:	[xə'lιnə]	(M 1;10)
(c)	zebra	/'sebra/	‘zebra’:	['sebəra]	(D 1;11)
(7a)	baby	/'bebi/	‘baby’:	[be'be:]	(M 1;10)
(b)	ijsbeer	/'ειsbe:r/	‘polar bear’:	[ειs'be:r]	(M 1;10)

All SWW word tokens were included in the data base, 126 for Marloes, and 169 for Daan. Correct realization was considerably less frequent in these words than in the two categories described above; the overall scores were 64·3% for Marloes and 62·1% for Daan. However, performance improved with increasing age. Approximately one quarter (25·4% in Daan, 23% in Marloes) of all renditions involved deletions of one of the two weak syllables, which results in an SW structure. The words from which one weak syllable is deleted can be divided into two types: those in which both weak syllables are equivalent in terms of stress (both contain schwas), and those in which one of the two weak syllables is stronger than the other. In the second type of word (ex. 8), it is mostly the weakest syllable (usually containing a schwa) that is omitted (88% of all cases). The other weak syllable, which in fact bears secondary stress, is retained. In the first type of word (ex. 9), it is not possible to tell which of the two weak syllables is lost, because both contain schwas. A reasonable interpretation is that either syllable 2 or 3 can be dropped, or that the two are conflated.

(8a)	ziekenhuis	/'sikə,h∧ys/	‘hospital’:	['sikh∧ys]	(M 1;10)
(b)	olifant	/'oli,fαnt/	‘elephant’:	['o:xαnt]	(M 1;11)
(c)	kangoeroe	/'kαηxu,ru/	‘kangeroo’:	['kαxru]	(D 1;11)
(d)	stofzuiger	/'stɔf,s∧yxər/	‘vacuum cleaner’:	[‘kɔkh∧yx]	(M 1;11)
(e)	vrachtauto	/'frαxt,αuto/	‘lorry’:	['frαxαu]	(M 1;11)
(f)	neushoorn	/'nøs,hɔ:rən/	‘rhinoceros’:	[‘nø:shɔ:r]	(D 1;11)

(9a)	poppetje	/'pɔpəcjə/	‘doll’:	[pɔpə]	(D 1;8)
(b)	andere	/'αndərə/	‘other’:	['αnrə]	(D 2;4)

[pagina 72]

(c)	andere	/'αndərə/	‘other’:	['αndə]	(M 2;4)
(d)	Janneke	/'jαnəkə/	:	['zαηkə]	(M 2;3)

Words with initial W-syllables. From all samples in both children, all words that, according to their target form, have initial W-syllables were extracted. Table 2 summarizes the various realizations of WS-words as a function of

 

table 2. Realizations of WS words

	Marloes	Marloes	Marloes	Marloes	Daan	Daan	Daan	Daan
AgeGa naar voetnoot*	CorGa naar voetnoota	Trn	S-s	Tot	Cor	Trn	S-s	Tot
A	9	10	0	19	4	11	1	16
	(47·4)Ga naar voetnootb	(52·6)	(0)	(100)	(25)	(68·8)	(6·3)	(100)
B	3	5	11	19	21	1	1	23
	(15·8	(26·3)	(57·9)	(100)	(91·3)	(4·3)	(4·3)	(100)
C	10	4	22	36	32	6	0	38
	(27·8)	(11·1)	(61·1)	(100)	(84·2)	(15·8)	(0)	(100)
D	11	1	9	21	53	3	0	56
	(52·4)	(4·8)	(42·9)	(100)	(94·6)	(5·4)	(0)	(100)
Total	33	20	42	95	110	21	2	133
	(34·7)	(21·1)	(44·2)	(100)	(82·7)	(15·8	(1·5	(100)

age. Aggregated over all samples, Marloes produces considerably more incorrect realizations than Daan (65% vs. 17%). This difference is primarily related to a difference in the pattern of development parallel to the one noted before. The percentage of Marloes’ correct realizations barely increases over age. Daan, by contrast, shows an abrupt improvement: from 25% correct realizations in the first period to around 90% after age 1;11. Most of Marloes' incorrect realizations are stress-shifts of the type WS → SW. However, all of these cases concern the same word, viz. the child's own name. The reason for this idiosyncracy is not clear. Daan most frequently truncates (ex. 10); his stress-shifts affect only two words, lawaai (‘noise’) and Ireen. Overall, 19·2% of the WS tokens in both corpora are realized with an SW rhythm, and 17·9% are realized as monosyllables. Monosyllabic realization is predominant in the first four months of the observation period (Daan: 68·8%, Marloes: 52·6%).

(10a)	ballon	/bα'lɔn/	‘balloon’:	[lɔn]	(D 1;7)
(b)	papier	/pα'pi:r/	‘paper’:	[pi:r]	(D 1;10)
(c)	konijn	/ko'nειn	/‘rabbit’:	[nειn]	(D 2;5)

[pagina 73]

A sudden increase in the amount of correct realizations in Daan's corpus, similar to the one observed for the WS words, is present in the WSW words, as Table 3 shows. In Marloes' data, again, the improvement appears to be

 

table 3. Realizations of WSW words

AgeGa naar voetnoot*	CorGa naar voetnoota	Ts	Tsw	Rst	Tot
			Marloes
A	0	1	2	0	3
	(0)Ga naar voetnootb	(33·3)	(66·7)	(0)	(100)
B	5	0	13	0	18
	(27·8)	(0)	(72·2)	(0)	(100)
C	10	0	14	1	25
	(40)	(0)	(56)	(4)	(100)
D	17	0	1	0	18
	(94·4)	(0)	(5·6)	(0)	(100)
Total	32	1	30	1	64
	(50)	(1·6)	(46·9)	(1·6)	(100)
			Daan
A	1	0	19	0	20
	(5)	(0)	(95)	(0)	(100)
B	17	0	6	0	23
	(73·9)	(0)	(26·1)	(0)	(100)
C	11	0	3	1	15
	(73·3)	(0)	(20)	(6·7)	(100)
D	17	0	3	2	22
	(77·3)	(0)	(13·6)	(9·1)	(100)
Total	46	0	31	3	80
	(57·5)	(0)	(38·8)	(3·8)	(100)

more gradual. Overall, 54% of all word tokens are incorrectly realized. Most of these cases involve deletion of the initial weak syllable (ex. 11). It should be noted that truncations to WS do not occur. Furthermore, the frequency of monosyllabic realization is nearly zero. Overall, 42·4% of all WSW words in both children are realized as SW. During the first four months, this percentage is higher, 66·7% in Marloes and 95% in Daan.

(11a)	kastanje	/,kαs'tαnjə/	‘chestnut’:	['tαnə]	(M 1;10)
(b)	pantoffel	/,pan'tɔfəl/	‘slipper’:	[‘tɔfət]	(M 2;1)
(c)	giraffen	/,ſi'rαfə/	‘giraffes’:	['sαfə]	(D 1;11)
(d)	verwarming	/fər'υαr,mιη/	‘heater’:	['fυαmιη]	(D 2;6)

As can be seen in Table 4, the percentages of incorrectly realized WWS

[pagina 74]

Table 4. Realizations of WWS words

AgeGa naar voetnoot*	CorGa naar voetnoota	Ts	Tws	S-s	Rst	Tot
			Marloes	Marloes
A	0	0	3	1	0	4
	(0)Ga naar voetnootb	(0)	(75)	(25)	(0)	(100)
B	17	0	4	8	0	29
	(58·6)	(0)	(13·8)	(27·6)	(0)	(100)
C	11	0	7	3	0	21
	(52·4)	(0)	(33·3)	(14·3)	(0)	(100)
D	10	0	2	1	0	13
	(76·9)	(0)	(15·4)	(7·7)	(0)	(100)
Total	38	0	16	13	0	67
	(56·7)	(0)	(23·9)	(19·4)	(0)	(100)
			Daan	Daan
A	3	13	3	1	0	20
	(15)	(65)	(15)	(5)	(0)	(100)
B	12	1	2	2	1	18
	(66·7)	(5·6)	(11·1)	(11·1)	(9·6)	(100)
C	10	0	0	3	1	14
	(71·4)	(0)	(0)	(21·4)	(7·1)	(100)
D	7	1	1	2	0	11
	(63·6)	(9·1)	(9·1)	(18·2)	(0)	(100)
Total	32	15	6	8	2	63
	(50·8)	(23·8)	(9·5)	(12·7)	(3·2)	(100)

words in the two children do not differ very much: 43% in Marloes and 49% in Daan. One-syllable truncations (ex. 12) make up 55% of all incorrect renderings in Marloes' corpus and 19% in Daan's corpus. In most of these cases, the syllable bearing secondary stress is retained (12 a, b, c). In some other cases, it was impossible to determine which syllable was removed (cf. 12 d, e). Daan produces relatively many two-syllable truncations (ex. 13). Stress shifts, with or without syllable deletions, occur more often in Marloes than in Daan (ex. 14). Overall, 11·5% of all WWS words are realized as monosyllables, and 3·8% as SW patterns.

(12a)	paraplu	/,para'ply/	‘umbrella’:	[pa'ply]	(M 2;1)
(b)	boerderij	/,burdə'rει/	‘farm’:	[bu'rει]	(D 2;0)
(c)	Marjolijn	/,mαrjo'lειn/:		[mα'lειn]	(D 2;11)
(d)	microfoon	/,mikro'fon/	‘microphone’:	[mi'kron]	(M 1;9)
(e)	Mickey Mouse	/,mιki'mαus/:		[mi'j∧ys]	(M 2;0)

(13a)	Evelien	/,efə'lin/:		[li:n]	(D 1;7)
(b)	chocola	/,ſoko'la/	‘chocolate’:	[la]	(D 1;7)

[pagina 75]

(c)	papegaai	/,pαpə'xaj/	‘parrot’:	[pα:j]	(D 1;7)
(14a)	telefoon	/,telə'fon/	‘telephone’:	['te:te:]	(M 1;7)
(b)	Sinterklaas	/,sιntər'klas/:		['sιntəklas]	(M 1;7)
(c)	Zwarte Piet	/,sυαrtə'pit/:		['sυαrtəpit]	(D 2;2)
(d)	krokodil	/,kroko'dιl/	‘crocodile’:	['krokədιl]	(D 2;6)

The proportions of correct and incorrect realizations presented thus far all refer to tokens. Thus, it is not yet revealed whether words (types) are consistently realized in a particular way, or whether there is variation. Particularly, in order to decide at which level of representation a putative rhythmic constraint takes effect (lexical representations or phonological representations constructed during utterance planning), it is relevant to decide whether there is free variation. Free variation is defined as the appearance of different rhythmic forms - either correct or incorrect - throughout the period of observation, without a discernable approximation towards the adult form. If the latter should occur, it would be labelled systematic variation.

To explore this issue, we looked at instances of polysyllabic words (other than SW) which were used throughout the whole observation period. To minimize the effect of putative speech errors or other transitory problems, as well as misperceptions on the part of the transcribers of our data set, we added the criteria that at least two different types of realization should be present, and that each of these should be represented by at least two tokens. We found 13 words in Daan's corpus, and 9 in Marloes' corpus that met these criteria. Fourteen of these words showed free variation, which is illustrated by the examples in Table 5. These findings appear to contradict the conjecture that it is the lexical representations that are rhythmically constrained.

Determiners

In order to decide whether the rhythmic form of a putative phrase affects the (non)realization of unstressed, closed-class words, we took stock of all locations in the children's first 100 utterances in each sample where a determiner (usually an article) would have been obligatory in adult speech. We determined the proportion of cases in which a determiner was present under two conditions: (i) the post-determiner content word had an initial S-syllable, or (ii) the content word had an initial W-syllable.

This analysis was complicated by two problems. First, as we have seen above, the realized forms of nouns and other content words may (often quite regularly) deviate from the target form. Hence, the question arises whether we should consider the realized form of the post-determiner content word, or the target form. In the ensuing analysis we have opted for the former, to avoid possibly ill-grounded assumptions about underlying forms in the

[pagina 76]

table 5. Examples of free variation of the rhythmic structure in successive renderings of words

Target form	Word	Age	Utt.no.	Realized form
		Marloes
WS	banaan	1;10	36	ma:n
	‘banana’	1;10	37	ma:n
		1;10	38	ma:n
		1;10	39	na:n
		1;10	40	mə'na:n
		1;10	42	ma:n
		1;10	43	mə'na:n
		1;10	44	ma:n
		1;10	47	na:n
WWS	microfoon	1;9	8	mi'krɔ:n
	‘microphone‘	1;9	12	mə'kro:n
		1;10	159	mə'xo:n
		2;2	1	miko'xo:n
		2;2	3	mιko'xo:n
		2;2	86	kιη'xo:n
		2;2	87	miko'xo:n
		2;2	88	mikə'xo:n
		2;2	89	miko'xo:n
		2;9	87	miko'fo:n
		2;9	88	miko'fo:n
		Daan
WS	kapot	1;7	130	pɔt
	‘broken‘	1;7	134	dα'bɔt
		1;9	38	pot
		2;0	151	kə'pɔt
		2;0	152	ka'pɔt
		2;1	127	ka'pɔt
		2;1	128	ka'pɔt
SWW	olifant	1;8	35	'ojəf…fαnt
	‘elephant‘	1;9	46	'o:fαnt
		1;9	55	'o:lifαnt
		1;11	129	o:lə'fαnt
		1;11	135	o:lə'fαnt
		2;0	40	'o:ləfαnt
		2;0	41	o:lə'fαnt
		2;0	52	'o:ləfαnt
		2;1	155	o:lə'fαnt
		2;1	161	o:lə'fαnt

child's mental lexicon. A parallel analysis was also performed on the - much smaller - subset of noun phrases in which the realized rhythmic form of the content word (noun or adjective) corresponded to the adult model. However, the results of this analysis do not differ from those presented below.

A second problem is that our subjects (particularly Marloes) often produce filler sounds (schwas) at putative determiner positions, which cannot be

[pagina 77]

readily identified as one of the Dutch article forms. However, on the basis of our impression that the distribution of these fillers equals the distribution of determiners in corresponding adult sentences, we decided to count them as determiners. A preliminary analysis of the much smaller number of cases in which the determiner was phonetically distinct yielded the same pattern of results as the one reported below.

Fig. 3 depicts the percentage of cases in which a determiner was produced where it would have been obligatory according to adult standards. It is obvious that the proportion of realized determiners increase over age. If the rhythmic constraint determines the form not only of words, but also of noun phrases, it can be expected that fewer articles will be produced before words

[pagina 78]

table 6. Articles in sentence-initial and non-sentence-initial positions

	Initial positions	Initial positions	Initial positions	Other positions	Other positions	Other positions
AgeGa naar voetnoot*	Absent	Present	Tot	Absent	Present	Tot
			Marloes
A	225	50	275	60	13	73
(81·8)Ga naar voetnoota	(18·2)	(100)	(82·2)	(17·8)	(100)
B	119	46	165	105	79	184
	(72·1)	(27·9)	(100)	(57·1)	(42·9)	(100)
C	58	39	97	66	69	135
(60)	(40)	(100)	(48·9)	(51·1)	(100)
D	30	26	56	51	67	118
	(53·6)	(46·4)	(100)	(43·2)	(56·8)	(100)
Total	432	161	593	282	228	510
	(72·8)	(27·2)	(100)	(55·3)	(44·7)	(100)
			Daan
A	279	4	283	25	9	34
	(98·6)	(1·4)	(100)	(73·5)	(26·5)	(100)
B	137	16	153	57	22	79
	(89·5)	(10·5)	(100)	(72·2)	(27·8)	(100)
C	80	33	113	48	93	141
	(70·8)	(29·2)	(100)	(34)	(66)	(100)
D	61	31	92	42	129	171
	(66·3)	(33·7)	(100)	(24·6)	(75·4)	(100)
Total	557	84	641	172	253	425
	(86·9)	(12·1)	(100)	(40·5)	(59·5)	(100)

with an initial W-syllable than before words with an initial S-syllable. In the former case the article-content word string would involve two adjacent W-syllables, whereas in the latter case, the article (W) and the first syllable of the content word would constitute an alternating pattern, which follows the requirement of rhythmic alternation. Marloes' results agree with this prediction: there are significantly more determiner omissions before W-initial words than before S-initial words (χ²(1) = 4·3, p < 0·05). This result is not replicated for Daan.

The figures in Table 6 show that for both children, articles are significantly less often realized in sentence-initial positions than in sentence-medial positions (Marloes: χ²(1) = 188·33, p < 0·001; Daan: χ²(1) = 254·77, p < 0·0001). This finding suggests that it may be difficult for these children to start an utterance with an unstressed word, which may be a sentence-level parallel of the tendency to delete word-initial weak syllables shown before.

[pagina 79]

Discussion

We have studied two children's language production in terms of rhythmic patterns, i.e. the alternation of weak (unstressed) and strong (stressed) syllables. Our aim was to determine whether the realized forms of words and noun phrases meet the requirements that constitute a trochaic constraint, as proposed by Allen & Hawkins (1978, 1980). Furthermore, we examined whether the intake of words showed signs of a rhythmically based ‘lexical filtering’. With regard to the latter point, we found that our subjects' early lexical inventories appear to be dominated by S-initial words. Moreover, WS and WSW words are quite infrequent, and words with an alternation of weak and strong syllables (i.e. SW, WS, and WSW) are preferred to words with a non-alternating pattern, so that, as a result, WWS words are marginal. However, we could not decide whether a rhythmical bias was effective in the intake of words from the input, since we were unable to show that the proportions of rhythmic types differ from those of the mothers' input vocabularies. Additionally, we observed that the children's vocabularies - in terms of rhythmic patterns - diverge from those of their counterparts' mothers. Taken together, the results may mean that the composition of the children's vocabularies simply reflects the statistical properties of the input. That is, since words that are frequently heard presumably have a higher chance of being acquired than words that are infrequently heard, the distribution of rhythmic types in the child's lexicon will reflect that of the input vocabulary. On the other hand, the structure of the mother's speech might encourage the child to focus on particular types of words. To decide empirically between these two interpretations appears to be rather difficult, however, At any rate, both assumptions may explain why monosyllabic words are considerably more frequent in Marloes than in Daan, since Marloes' mother uses more monosyllabic words than Daan's mother.

As regards the realizations of words, some rather systematic tendencies were observed. SW-words are nearly always produced correctly. SWW-words are considerably more often deformed, and incorrect realizations of W-initial words are very frequent, particularly in the early samples. Generally, the observed deformations agree with the patterns reported by Allen & Hawkins (1978, 1980): SWW words lose one of their weak syllables (though not necessarily the first), or the weak syllables may be conflated. WS and WSW words most often lose their initial syllable. The trochaic constraint does not appear to be all-embracing, however, which also corresponds to Allen & Hawkins' (1980) observations. Marloes, for instance, shows a preference to delete only one weak syllable from WWS words (which produces WS). Moreover, neither of the children shows a tendency to append filler syllables to monosyllabic words (S → SW), and the pattern of truncations also indicates that monosyllables are acceptable renderings.

[pagina 80]

Nevertheless, the tendency toward the trochaic pattern is quite clear. During the first four months of observation, 30% of the words that (in their adult form) have a structure other than S or SW are realized as monosyllables, and 31% are realized with an SW-pattern. In sum, the data suggest that the children do adapt the word forms to a SW-template, in which, however, the weak part is optional. Truncation appears to be the preferred strategy by which this adaptation is effected. Stress shifting is observed, but appears to play a minor and presumably idiosyncratic role (but see Fikkert, 1993, who claims that stress-shifting becomes more important in a later stage of development). It appears that the adaptation of words to the trochaic pattern is a stochastic process. In particular, the data suggest that one target word may be associated with various alternating realizations. Similar observations have been made in the segmental domain (Ferguson & Farwell, 1975).

Unfortunately, Dutch is not the ideal testing ground for the hypothesis that a trochaic rhythm in speech reflects a prosodic developmental universal. In view of the preponderance of trochaic patterns in Dutch, our subjects' tendency to produce trochaic patterns may very well be the result of their initial hypothesis concerning the metrical structure of Dutch (see Demuth 1992, for an elaboration of this idea). To gain more insight into the issue of universality, one should look at the behaviour of young children who acquire languages in which S-initial patterns are infrequent, such as French.

The results with regard to the realization of determiners suggest that rhythmic requirements may affect the form of phrases, as well as words, in production. We found that Marloes more often produced determiners before S-initial words than before W-initial words, which may be ascribed to the fact that the WS…-string resulting in the former case meets the requirement of alternation, whereas a WW…-string does not. Unfortunately, this result was not replicated in Daan, which, however, may be ascribed to Daan's faster abandonment of the rhythmic constraint, as indicated by the abrupt increase of correct realizations of W-initial words around age two. Nevertheless, in both children a tendency to omit sentence-initial determiners was observed, which accords with Gerken's (1990, 1991) experimental findings regarding the omission of sentence-initial closed class words. This pattern appears to be a sentence-level parallel of the omission of word-initial weak syllables. This finding corroborates the suggestion that particular phenomena in child language that are traditionally associated with syntactic or semantic immaturity, such as the omission of sentence-initial subject pronouns (Hyams, 1986), or, for that matter, the optionality of grammatical morphemes, may in fact be the result of a phonological mechanism.

The determiner results, as well as the observed free variation in the realization of word forms are compatible with the suggestion that the rhythmic constraint is implemented in the ‘transduction’ mechanism, rather than in the lexicon. We surmise that the observed behaviour can be

[pagina 81]

understood as the result of a process by which ‘melodic’ information, stored in the lexicon, is associated with representations that encode metrical rhythm. This process can be seen as a part of the utterance planning mechanism (see also Gerken, 1990, 1991). The results of this and related studies appear to give some clues as to the details of this process.

First, the ‘merging’ of segments or features from two neighbouring syllables (see, for instance, examples 1, 11c, 11d, 12d, 12e, 13c), seems to rule out a mechanism proposed by Gerken (1990, 1991), in which the mapping of melodic information onto syllabic templates, and the mapping of syllables onto prosodic templates are distinct. Such a mechanism would predict the omission/preservation of complete syllables, rather than syllable constituents. Rather, it appears that syllabic and prosodic mapping are conflated, as is suggested by some versions of nonlinear phonology (Fee, 1991). We could speculate that the melodic (segmental or featural) information from the lexicon is associated with the independently addressable onset, peak and coda slots of disyllabic templates, of which the leftmost peak slot is marked ‘strong’, and the rightmost ‘weak’. Second, the order in which melodic information is mapped onto this template seems to be correlated with relative prominence. Of course, the clearest instantiation of this principle is that primary-stress-bearing vowels are systematically preserved in children's productions. Furthermore, we have observed, as others have before us (Lohuis-Weber, 1991), that if a multisyllabic word containing a non-reducible vowel as well as a reducible vowel or schwa is truncated, the non-reducible vowel is retained. These observations suggest that relative prominence is an attribute of vowels as they are stored in the lexicon. This hypothesis may explain why stress-shifting is only a marginal phenomenon in our subjects’ output. It would entail the alteration of inherent attributes of the stored information, whereas truncation appears to be the ‘coincidental’ outcome of a mechanism whereby as much melodic information as possible is linked to the most appropriate slots in a restricted template (but see Fikkert 1993, for a competing view).

Given these - speculative - hypotheses, the attainment of the specific rhythmic structures of the target language would imply the gradual dissolution of the rhythmic restrictions on output representations, or rather, the acquisition of rhythm and stress rules that allow for the construction of other than SW-templates. Thus, we would expect to find a connection between the acquisition of language-specific metrical rules and the disappearance of the SW-constraint. One of the characteristics of the Dutch stress system is quantity sensitivity: a heavy syllable, i.e. a syllable with a branching rime, cannot occupy a weak position in a metrical foot. Some of our observations are suggestive of the acquisition of this knowledge, viz. Daan's successive renderings of the word olifant /'oli,fαnt/ (‘elephant’, see Table 5). The nice thing about olifant is, that it is an exception to the rule of quantity sensitivity.

[pagina 82]

The final syllable /fαnt/ is heavy (superheavy, in fact), and since Dutch stress is assigned from right to left (Kager, 1989), it should have had primary stress (compare ledikant /ledi'kαnt/ ‘bed’). Daan's mispronunciations of olifant as [oli'fαnt], which begin to occur at age 1;11, seem to imply that he has acquired the quantity sensitivity rule and is actually overregularizing. At approximately the same time, the proportion of correct productions of W-initial words, WWS among them, shows a sudden increase. This observation provides some support for our assumption that the trochaic patterns in child speech are the outcome of a grammar-based restriction on phonological representations. Furthermore, it agrees with Dresher & Kaye's (1990) suggestion that quantity sensitivity is one of the first subsystems to be grasped by the child in the course of acquiring language-specific stress rules.

References

Allen, G.D. & Hawkins, S. (1978). The development of phonological rhythm. In A. Bell & J.B. Hooper (eds). Syllables and segments. Amsterdam: North Holland.

Allen, G.D. & Hawkins, S. (1980). Phonological rhythm: definition and development. In G.H. Yeni-Komshian, J.F. Kavanagh & C.A. Ferguson (eds). Child Phonology: Vol. I. Production, New York: Academie Press.

Brown, R. (1973). A first language: the early stages. Cambridge, MA: Harvard University Press.

Clark, H.H. & Clark, E.V. (1977). Psychology and language: an introduction to psycholinguistics. New York: Harcourt Brace Jovanovich.

Cutler, A. & Carter, D.M. (1987). The predominance of strong initial syllables in the English vocabulary. Computer Speech and Language 2, 133-42.

Cutler, A. & Norris, D. (1988). The role of strong syllables in segmentation for lexical access. Journal of Experimental Psychology: Human Perception and Performance 14, 113-21.

Dell, G.S. (1986). A spreading-activation model of retrieval in sentence production. Psychological Review 93, 283-321.

Demuth, K. (1992). On the ‘underspecification’ of functional categories in early grammars. In B. Lust (ed.), Syntactic theory and first language acquisition. Hillsdale: Erlbaum.

Den Os, E. (1990a). The development of prosody. Poster presented at the Fifth International Congress for the Study of Child Language, Budapest, July.

Den Os, E. (1990b). Development of temporal properties in the speech of one child between one and three years of age. Proceedings of the Institute of Phonetic Sciences of the University of Amsterdam 14, 39-52.

Dresher, B.E. & Kaye, J.D. (1990). A computational learning model for metrical phonology. Cognition 34, 137-95.

Fee, E.J. (1991). Segments and syllables in early language acquisition. Paper presented at the Boston University Conference on Language Acquisition, October.

Ferguson, C.A. & Farwell, C.B. (1975). Words and sounds in early language acquisition. Language 51, 419-39.

Fikkert, P. (1991). Child phonology from a theoretical perspective. Paper presented to the OTS Linguistics Colloquium, November.

Fikkert, P. (1993). The acquisition of Dutch stress. In M. Verrips & F. Wijnen (eds), Amsterdam Series in Child Language Development, Vol. 1.

Gerken, L.A. (1990). Performance constraints in early language: the case of subjectless sentences. Papers and Reports in Child Language Development 29, 54-61.

Gerken, L.A. (1991). The metrical basis for children's subjectless sentences. Journal of Memory and Language 30, 431-51.

[pagina 83]

Gerken, L.A. (1992). Young children's representations of prosodic phonology: evidence from English-speakers' weak syllable omissions. Unpublished paper, State University of New York at Buffalo.

Gerken, L.A., Landau, B. & Remez, R.E. (1990). Function morphemes in young children's speech perception and production. Developmental Psychology 26, 204-16.

Gleitman, L.R., Gleitman, H., Landau, B. & Wanner, E. (1988). Where learning begins: initial representations for language learning. In F.J. Newmeyer (ed.), Linguistics: the Cambridge survey. Cambridge: C.U.P.

Hochberg, J.G. (1988). First steps in the acquisition of Spanish stress. Journal of Child Language 15, 273-92.

Hyams, N. (1986). Language acquisition and the theory of parameters. Dordrecht: Reidel.

Kager, R.W.J. (1989). A metrical theory of stress and destressing in Dutch. Unpublished doctoral dissertation, University of Utrecht.

Kelly, M.H. & Bock, J.K. (1988). Stress in time. Journal of Experimental Psychology: Human Perception and Performance 14, 389-403.

Levelt, W.J.M. (1989). Speaking: from intention to articulation. Cambridge, MA: MIT Press.

Lohuis-Weber, H. (1991). Kinder-Stress: een onderzoek naar de rol van klemtoon bij moedertaalverwerving. Unpublished M.A. Thesis, University of Utrecht, Dept. of Dutch Language and Literature.

Menyuk, P., Menn, L. & Silber, R. (1986). Early strategies for the perception and production of words and sounds. In P. Fletcher & M. Garman (eds), Language acquisition (2nd ed.). Cambridge: C.U.P.

Petretic, P.A. & Tweney, R.D. (1977). Does comprehension precede production? The development of children's responses to telegraphic sentences of varying grammatical adequacy. Journal of Child Language 4, 201-209.

Pye, C. (1983). Mayan telegraphese: intonational determinants of inflectional development in Quiché Mayan. Language 59, 583-604.

Quené, H. (1992). Integration of acoustic-phonetic cues in word segmentation. In: M.E.H. Schouten (ed), The auditory processing of speech: from sounds to words. Berlin: Mouton-De Gruyter.

Shipley, E.F., Smith, C.S. & Gleitman, L.R. (1969). A study in the acquisition of language: free responses to commands. Language 45, 322-42.

Radford, A. (1990). Syntactic theory and the acquisition of English syntax. Oxford: Blackwell.

Schwartz, R.G. (1988). Phonological factors in early lexical acquisition. In M.D. Smith and J.L. Locke (eds), The emergent lexicon: the child's development of a linguistic vocabulary. New York: Academic Press.

Waterson, N. (1971). Child phonology: a prosodic view. Journal of Linguistics 7, 179-211.

Wijnen, F. (1990). The development of sentence planning. Journal of Child Language 17, 651-75.

Wijnen, F. (1992). Incidental word and sound errors in young speakers. Journal of Memory and Language 31, 734-55.