When is a word a word?

J. Child Lang. 21 (1994), 517-542. Copyright © 1994 Cambridge University Press.

When is a word a word?* MARILYN MAY VIHMAN Southeastern Louisiana University AND

LORRAINE McCUNE Rutgers University (Received 25 June 1991. Revised 16 April 1993)

ABSTRACT

Although adult-based words co-occur in the period of transition to speech with a variety of non-word vocalizations, little attention has been given to the formidable problem of identifying these earliest words. This paper specifies explicit, maximally 'inclusive' identification procedures, with criteria based on both phonetic and contextual parameters. A formal system for evaluating phonetic match is suggested, as well as a set of child-derived functional categories reflecting use in context. Analysis of word use across two samples of 10 children each, followed from o;o to i;4, provides evidence to suggest that context-bound words can be 'trained' by focusing on eliciting language, but that the timing of context-flexible word use remains independent of such training.

INTRODUCTION In studies of early child language the word plays a pivotal role. Recognition of the 'first word' preoccupied clinical practitioners long before the advent of contemporary psycholinguistic research (McCarthy, 1954), while interest continues to focus on the size and typology of the earliest vocabularies of more or less rapidly advancing infants (Goldfield & Reznick, 1990; [•] This research was supported by funding from the National Science Foundation (BNS8209695 and 85 20048). We would like to thank Fumiko Arao (Stanford University), Catherine Durand (C.N.R.S., Paris), and Liselotte Roug-Hellichius and Ingrid Landberg (Institute of Linguistics, Stockholm University) for their help in applying our criteria to crosslinguistic data; we also thank Charles A. Ferguson for his thoughtful and constructive comments. Address for correspondence: Marilyn May Vihman, Department of Special Education, Southeastern Louisiana University, P.O. Box 879, Hammond, LA 70402, USA. 517

VIHMAN & MCCUNE

Huttenlocher, Haight, Bryk, Seltzer & Lyons, 1991). Such current debates as the relative precocity of early language in the oral vs. the gestural mode (Orlansky & Bonvillian, 1988; Petitto, 1988; Meier & Newport, 1990) and resolution of the issue of the ' nominal insight' and its relation to referential word use (McShane, 1979; Harris, Barrett, Jones & Brookes, 1988) depend crucially on recognition of the child's very first words and on establishing the developmental significance of different categories of early words. Yet the problem of word definition and identification has received relatively little attention among influential works addressing the onset of word use. For example, neither Halliday's (1975) diary study nor Bates, Benigni, Bretherton, Camaioni & Volterra's (1979) broad-based study of 25 Italianand English-learning infants provided any discussion of the problem. When explicit mention is made of this central methodological issue, two complementary criteria are characteristically invoked: (a) resemblance of phonetic form to an adult word and (b) situational consistency in use (e.g. Lewis, 1936; Leopold, 1939; Nelson, 1973; Ferguson & Farwell, 1975; Greenfield & Smith, 1976). The recognition of words via the broad criterion of conventional form with a sound-meaning link is far from simple, however, particularly in the earliest period of potential word production, when children make a gradual transition from babbling to adult-like word use, producing a variety of vocalizations of differing degrees of 'wordiness' (Menn, 1978; Bates et al. 1979; Vihman & Miller, 1988). Meanings may be unconventionally narrow or limited, on the one hand, or global and diffuse, on the other (Rescorla, 1980; Griffiths, 1986; Nelson, 1988); forms may be indistinguishable from concurrent babble (Labov & Labov, 1978; Vihman, Macken, Miller, Simmons & Miller, 1985), and both the sound and meaning consistencies may be idiosyncratic to a given child (Halliday, 1975; Dore, Franklin, Miller & Ramer, 1976; Ferguson, 1978). Balanced attention to form and function combined with detailed specification of the problems involved or the criteria used is rare. To allow comparability across studies investigators would need, at a minimum, to address such specific questions as the degree and type of phonetic resemblance to an adult word and the nature of the factors which they take to constitute evidence for the existence of a usage pattern which is situationally consistent. One problem in comparing infant words to adult language is the essentially syntactic character of the latter. Some authors have considered the single word to be a sentence or HOLOPHRASE (e.g. de Laguna, 1927; see also Dore, 1985), while others have been unwilling even to accord the term 'word' to presentational verbalizations (e.g. Werner & Kaplan, 1963, who coined the term VOCABLE). Different categories have been proposed for children's productions in the single word period. Nelson (1973) categorized children's words as reported by mothers partly on the basis of their apparent function for the child (e.g.

WHEN IS A WORD A WORD?

'personal-social words', 'action words'), but partly on the basis of the syntactic function of the word in adult sentences ('modifiers', 'function words'). Bates et al. (1979) described a continuum from context-bound performative uses of words to context-flexible referential use. Spontaneous use of a word in a variety of circumstances rather than in restricted contexts only is a qualitatively different behaviour which has been shown to follow from representational abilities also evident in symbolic play. McCuneNicolich (1981) demonstrated that relational words (such as allgone, here, more, up) depend upon representational development. They demonstrate the child's capacity to consider and comment on reversible perceptual events, including attention to the immediate past and future (cf. also Sinclair, 1970; Bloom, 1973). Attention to developmentally distinct categories of early words is an essential aspect of word identification which seriously affects the interpretation of results. This paper provides a methodology for word identification which takes account of developmental differences in word use and presents results demonstrating the importance of such distinctions. Some prior studies have employed the 'exclusive approach' of eliminating any word-candidates which fail to occur more than once, in different contexts, as recognized by independent transcribers (cf. Bloom, 1973; Harris et al. 1988). This allows relative confidence that errors of over-interpretation will be avoided ('false positives'). On the other hand, it is less well adapted to the apprehension of the 'ragged beginnings' of word use, particularly the production of the first 10 or 15 words. That is, it may result in more 'false negatives' than an inclusive set of criteria intended to capture to the greatest possible extent those vocalizations which might be broadly conceived as phonetic patterns 'either borrowed from the adult language or influenced by its forms' (Lewis, 1936: 124). In fact our data demonstrate that these differences in methodology are confounded with developmental differences marking categories of early words. Our approach to decisions of word identification has a dialogic and hermeneutical character (Packer, 1985). We began by posing a somewhat artificial question: Is a given vocal behaviour a 'word' or not? In attempting to answer this question we first developed an exhaustive inventory of vocal shapes recorded in the course of the session. Next, we ascertained the context of use of all formally relatable vocal shapes. It was then possible to marshall the evidence bearing on a particular instance to arrive at a preliminary quantification, weighing the acceptability of a potential word candidate. A point-for-point match with the presumed adult model sharpened our attention to the potential for evidence based on form alone; categorization in terms of function of use clarified our understanding of the possible meaning base of the word candidate. Where the quantitative evidence was weak but the subjective impression remained strong, we

VIHMAN & MCCUNE

consulted external evidence, both from the mother's report and from the longitudinal record, before making a judgement. We returned periodically to the data in context to ask how strong an impression of ' wordiness' we could gain from the textually embedded vocalization itself. A final decision was eventually made in each case, although we remain keenly aware of the inherent ambiguity of child behaviour which, like adult behaviour, is subject to the progressive evolution and expansion or re-definition of meaning in the course of use. In short, although the procedures described here make explicit the rationale for including each accepted word and thus provide a useful meta-analysis for comparison with previous work and for the design of future methodologies, one theme of this paper is that judgement is necessarily at the core of studies of the earliest words. METHOD

Data collection, transcription, and reliability Data were collected from five boys and five girls each at Rutgers and at Stanford, from o;o. to i ;6 (referred to here as the 'Rutgers sample' and the ' Stanford sample'). Six Rutgers subjects and all Stanford subjects were firstborn. The Stanford children were largely recruited through infant-care classes, the Rutgers children through newspaper advertisements. All of the parents had at least a high-school education; several of the parents in both groups held postgraduate degrees. The Rutgers children were video-recorded monthly in the home during half-hour free-play mother-child interactions with a standard set of toys. The Stanford children were audio- and videorecorded weekly in the home during half-hour unstructured interactions involving mother, child, and, to a lesser extent, the observer, using the toys at hand. A Sony microphone was placed near the Rutgers mother and child dyads. Two Sony Electret microphones were hidden in a cloth vest [i.e. sleeveless jacket] worn by the Stanford subjects. In both samples recordings were supplemented by maternal reports regarding the child's lexicon. Differences in project goals were reflected in the instructions and explanations given to mothers in the course of the recordings. At Rutgers mothers were told that the researchers were primarily interested in the development of play. Mothers were encouraged to allow their children to take the lead in playing with the toys, and were given no specific instructions about language use since language was expected to occur naturally in the play sessions. The two observers present, silently occupied with camera or note-taking, rarely addressed the mother and attempted to deflect any communicative advances made by the child. Mothers' lists of words were collected monthly without comment from the experimenters. At Stanford, mothers were interviewed weekly regarding the children's new words and other communicative behaviour. Mothers were asked to 520


prepare for the sessions in such a way as to encourage the child to produce any new words or other behaviours of interest (e.g. by bringing in a bowl of flowers when the child had been using the word flower over the previous week). Stanford mothers frequently engaged the children in book-reading and some of them elicited imitations as a way of demonstrating new words. The one observer typically present interacted verbally with the mother from time to time and responded to the child's advances but did not normally initiate activities with the child. For the analyses reported here, one session per child, at i;s or i;6, was transcribed phonetically from the videotapes for each of the 20 children by a single transcriber, with contextual information regarding child activity, gestures, and direction of gaze in connection with vocalizations, and also relevant maternal or observer actions and talk directed to the child. In addition, five to seven more monthly sessions per child were transcribed from the Rutgers videotapes and the Stanford audiotapes, supplemented for the Stanford sample by contextual video notes or, for a few sessions, full videoscripts. Reliability of phonetic transcription from the video-recordings was checked for three of the Stanford samples against full earlier transcriptions of the audio-recordings of the same sessions by the same transcriber (cf. Vihman et al. 1985, for discussion of intertranscriber reliability based on audio only). Agreement as to length in syllables and complete consonant identity for all supraglottal consonants, disregarding differences in voicing and in place of articulation for sibilants, averaged 83-6% (based on 805 vocalizations and 498 consonants). A partial retranscription of a single Rutgers videotape by the same transcriber, after a two-year interval, yielded 81-7% agreement as to length in syllables and identity of supraglottal consonants.

Procedure for identifying words Following transcription of the tapes, an initial screening involved 'nomination' of a maximally inclusive group of word candidates, that is, picking out any vocalizations which resembled adult words ('plausible phonetic shape') and which also seemed potentially relevant to the ongoing situation ('plausible context of use'). We included such non-dictionary words as animal sounds (meow, grrr) or pretend sounds relating to vehicles (vroom) or eating (yum). In short, we expect early words to be 'like recipes for skilfully "knowing how" to perform some roughly appropriate sound in some apparently appropriate context' (Dore, 1985; 35). Apparent imitation by the child or reformulation of the child's vocalization as a word by the mother (see Veneziano, 1981) were also sufficient to treat a form as a wordcandidate.

VIHMAN & MCCUNE

Word candidates were then evaluated for specific evidence of word status using the following criteria: I. Criteria based on context (1) Determinative context. Does at least one use occur in a context which strongly suggests that word and no other? Applies only to words with specific meanings easily identifiable in context, including most concrete nouns and many relational words. Does not apply to an imitative response to a purely verbal stimulus. (2) Maternal identification. Does the mother identify at least one instance of the form as a token of the hypothesized word ? Identification need not be explicitly intended as such; it could involve the mother acknowledging a particular word by continuing the conversation or by rejecting the child's word choice as an error. (3) Multiple use. Does the child use the word more than once ? (4) Multiple episodes. Is there more than one episode of use ? Multiple uses are identified only in determinative contexts, and with similar phonological shapes across different uses. II. Criteria based on vocalization shape (5) Complex match: Does the child form match more than two segments of the adult form ? Credited if three segments match, or two non-nasal segments plus nasality match a model which includes a nasal, or vowel length or an offglide match the complex nucleus of the model, in addition to the basic two-segment match. Also applies if a second consonant matches in manner of articulation but not in place, or vice versa. (6) Exact match. Is there at least one instance that even an untrained ear would recognize as an instance of the word ? Credited if the child form neither clearly omits, adds nor substitutes segments in relation to the model (again disregarding voicing). Reflects the probable judgement of a non-specialist that a particular word is intended. (7) Prosodic match. (i) To model: is there a tuneful match with the adult target ? (it) Across tokens: is there a characteristic tune which fits the wordmeaning and which occurs across all suspected tokens ? Credited when the child uses a special vocal effect (growl, squeak) repeatedly, in pragmatically plausible contexts, for the same probable word (lion, mouse). III. Relation to other vocalizations (8) Imitated tokens. Is at least one instance imitated ? 522


Credited if imitation is produced with apparent understanding. (9) Invariant. Do all instances of the word exhibit the same phonological shape ? Phonetic identity ('invariance') evaluated by the principles applied for phonetic match in general and for 'exact' in particular. (10) No inappropriate uses. Do all uses occur in contexts which plausibly suggest the same word ? Scored if the candidate form is not used in conflicting contexts (no homonymy) or outside of any plausible context (no 'favourite sound pattern' uses). Each word candidate was rated for the presence or absence of each type of evidence, yielding a word status profile which provided the framework for the dialogic process of evaluation, discussion and re-evaluation. Since a variety of irrelevant factors may influence (and even bias) the initial screening - such as the observer's familiarity with the child, the child's volubility and apparent intelligibility, or maternal evaluation and level of interaction with the observer - the word status profile and decision-making based on it maximize comparability in word identification across subjects. The word identification criteria are intended to provide an explicit record of the factors entering into a decision regarding the word status of each wordlike vocalization, or word candidate, as opposed to such relatively loose procedures as ' [determining] recognition [of a word] by global characteristics of the sound sequence and, inevitably, by appropriateness of the conditions of usage' (Plunkett, 1993). A team of coders could independently apply our criteria to a subset of data as a test of inter-observer reliability in judging word status. Alternatively, decisions for one or more subjects could be made by a single observer, with use of the word identification criteria as documentation. In either case, identification would be based on an explicit procedure, but a final pass would also be needed in which the subjective impression of the observer/transcriber was compared against the mechanical results of applying criteria. Each team of investigators would ultimately have to make decisions based on a complex understanding of the material, and then try to articulate and illustrate those decisions as we do here, especially when dealing with the earliest words.

RESULTS AND DISCUSSION

Words identified: sessions at 1; 3/1; 4 Whereas the Rutgers subjects produced a mean of 137 vocalizations each in these sessions, the Stanford subjects produced a mean of 243. Out of these vocalizations, 11 % were identified as word candidates in the Rutgers sample, 9% in the Stanford sample. Of the candidates so identified, 77% were 523

VIHMAN & McCUNE

accepted as words for Rutgers, 82 % for Stanford, yielding a mean of 11 words per Rutgers subject (8 % mean word production), and 17 per Stanford subject (7% mean word production). Thus, the 'yield' of potential and ultimately validated words for the two samples was proportional to relative volubility. Table 1 lists the subjects in the order of the size of their lexicon in TABLE

1. Vocalizations, word candidates and words accepted at 1; 3 / 1 ; 4 (Subjects listed in order of number of words accepted) Total Word Words vocalizations candidates accepted I. RUTGERS

Alice Aurie Rala Rick Vido Jase Ronny [1; 3] Kari Danny Nenny Mean

178 154 181

35

34

24

21 16 11

181

17 20

283

16

116

74

9 13

9 8 8 4

51

6

74 77

2

2

5

1369

147

0 113

II. STANFORD

Molly Sean Deborah Timmy Emily [1; 3] Thomas Andrew Jonah Camille Susie [1; 3] Mean

5°

49

38 38

27 26 23

162

32 15 18

165

12

126

7 5 5

7 5 4 3 168

433 368 264 239

.85

iS° 334 2426

205

12 12

the session sampled; this order roughly corresponds to the order of 'volubility', or total vocalizations produced. The list of words accepted after exhaustive application of the full word-identification procedure, together with the phonetic shape used for checking phonetic match, is given in the Appendix. The word identification criteria specified the evidence that a vocalization or set of vocalizations represented an attempt at an adult word. Review of the evidence actually used for each word candidate accepted in the initial analysis indicated that, in general, candidates supported by no more than two criteria failed to be accepted as words, while candidates supported by as many as four criteria were accepted (see Table 2). Notice that no mechanical counting of 524


TABLE 2. Application of internal (within-session) and external criteria for accepting words Subjects: Targets: Shapes: N uses Decision Criteria Det. ctxt Mat. id. Mult, use Mult.eps. Complex Exact Prosodic Imitated Invariant No inapprop. Total

Deborah

Alice

Sean

Rick

flower tea yum no down outside sky berries yellow [p'adi] [ti:] [?m] [nae] [ta] [Jja] [ks:] [bebij] [hawa] 3 Y

i

i

i

Y

N

Y"

4 Y

i

i

N

N

3 Y

spoon

i

N

Y = yes, accepted; N = no, rejected. Context suggests that the child is rejecting a suggestion by the mother, but is too vague to be credited as Determinative context. Both maternal report and earlier recorded use confirm word status. b Child vocalizes as he reaches for the spoon, after previously picking up the coffee pot. Maternal report confirms word status.

a

points will yield the final 'yes' or ' n o ' decisions actually arrived at, however, any more than any one set of 'necessary and sufficient' criteria would adequately specify the construct 'word' in adult language (cf., for example, Matthews, 1974). Instead, we tested the 'count' of applicable criteria against our subjective impression of wordiness; when the two failed to agree, we reconsidered, using external criteria (as specified in the notes to Table 2) as well as a return to the transcripts or even the videotapes (as discussed below). Word-like forms which lack word status Word candidates which were apparently well-supported but remained doubtful required a review of the evidence of use within the session and of our sum of knowledge of the child's lexicon. Problems of this type included cases of high-frequency 'global' or ill-defined (deictic) use of a range of similar forms, which we take to be protowords, and cases of word-based forms with no identifiable meaning. We illustrate two of the latter: Thomas responded to his mother's attempts to elicit the name of his soon-to-be adopted sister, (Baby) Molly, with [babmo], [maebma]. We 525

VIHMAN & MCCUNE

rejected the form, judging it to be a trained sound association without meaning for the child, despite credit for Maternal identification, Complex match, Imitated tokens, No inappropriate uses. Timmy produced the forms [gagga] twice, while pointing to a picture. The second use followed his mother's response, 'where's the dog?', and so could have been imitated. Timmy made frequent use of forms like gaga in other contexts as well, however; his phonological repertoire was still small while his vocal output was unusually large (Vihman, 1992). Also, his mother interpreted his vocalizations freely. We rejected the form, despite credit for Determinative context, Maternal identification, Multiple use, Imitated tokens, Invariant. Criteria used to identify words in the two samples Consideration of the manner in which the criteria characterized word candidates in the two samples demonstrates their general applicability. Determinative context was the criterion used the most often for both samples (Rutgers, 76 % use, Stanford 93 %). Multiple use and Maternal identification also applied for well over half the words accepted. About one-third of the words accepted in both samples were used in more than one episode. No inappropriate use was the second most used criterion for the Rutgers sample (65%), but applied to only 49% of the Stanford words. This difference reflects the higher volubility of the Stanford subjects, resulting in more frequent use of the same vocalization both in and out of a plausible word-use context. The other criteria based on vocalization shape played a role in identifying less than half the accepted words for either sample. Of these criteria, Complex match was used the most (Rutgers 4 7 % , Stanford 43 %)• Finally, the incidence of Imitation was the most striking dissimilarity between the two samples, with only 29 % imitated use at Rutgers as against 47 % at Stanford. The proportion of accepted words which were used only imitatively is much smaller for both samples, but the asymmetry remains (Rutgers 16%, Stanford 27%). Evaluation of phonetic match In order to go beyond global judgements in specifying what constitutes 'plausible agreement' in phonetic shape, we made a point-for-point comparison between child form and adult model, basing our evaluation of 'agreement' (or sufficient similarity to suggest intentional modelling) on certain assumptions regarding the 'expected' child realization of adult targets. The procedure is designed to take into account typical discrepancies between child and adult forms, reflecting, for example, the likelihood of low accuracy in child production of unstressed syllables and the possibility of temporal sequencing errors. We also take into consideration the limits of 526

WHEN IS A WORD A WORD:

TABLE 3. Evaluating phonetic match Example a: baby SUBJECTS

ADULT FORM: # 'b CHILD FORMS

Alice Deborah Vido

[beibi] [p'e:bi] [bobap]

e

+ + + + + [o]

i

b

i

+

+

+ +

(+) +

+

[a]

[p] u

Example b: cock-a-doodle-doo ADULT FORM : # k CHILD FORMS

a

k

3

d

SUBJECTS

Aurie Jonah Sean

[kakijali::] [skakekia] [a] [dalodalu::]

+ +

+ +

[i]

[CV] [CV]

[d]

+

+ +

M

d

a

1

'd

u: #

[CV]

[1]

Matching segments indicated as +, feature-match (but not full segment) as (+). Mismatching segments set off by square brackets; matching syllable-count but mismatching segments indicated as [CV]. (All of these illustrative words were accepted and are included in the children's 1; 4 word lists given in the Appendix.)

phonetic transcription when applied to infant vocalizations. Vowels occupying neighbouring spaces in the IPA transcription chart (e.g. [a] and [as]) were not treated as distinct for these purposes; similarly, consonants differing only in voicing or aspiration and sibilants differing in place of articulation ([s] vs. [J] vs. [5]) were treated as 'matching' for these purposes. Table 3 illustrates the point-for-point comparison of adult model and child form.1 Of the word candidates across both samples, 33 showed less than a twosegment match to the adult model (Rutgers, 13; Stanford, 20). Of these poorly matching forms, 18 were accepted (8 for Rutgers, 10 for Stanford). That is, 6 % of the children's words were accepted despite a phonetic match insufficient to pass more conservative criteria (cf. Huttenlocher et al. 1991). In each sample, 85 % of the child forms taken as word-candidates were vowel-final monosyllables, the most common word-like form occurring in this period for children acquiring English (cf. Vihman, Ferguson & Elbert, 1986, and contrast French, Japanese and Swedish: de Boysson-Bardies, Vihman, Roug-Hellichius, Durand, Landberg & Arao, 1992). In all but one case of such minimally matching forms ultimately accepted as words, there was a good match for either the initial consonant (e.g. Alice: nose [n:ae]) or the stressed vowel (e.g. Alice: key [ci]). In the one exception, only prosody matched (Deborah: [bon], with low pitch and extra length, for moo). Five words corresponded to the adult model or 'matched' only globally, with reordering of matching segments or only an approximation of both consonant [1] For a more complete discussion of the criteria used for word identification or of the procedure for determining degree of phonetic match, together with examples drawn from the data reported here, please write to the authors. 527

VIHMAN & MCCUNE and vowel (e.g. Sean: [?ij] horse; Timmy: [s: ae] fish).2 In such cases no phonetic criteria were taken to aid in word identification; contextual criteria were strong enough to lead to acceptance of the word. Evaluation of use in context In order to determine the appropriate categories for describing children's word use we explored the range of contexts accompanying a word form for a given child. In thus attempting to infer the meaning of a word for the child, we kept in mind distinctions previously reported in the literature. The taxonomy presented in Table 4 reflects the minimum number of categories T A B L E 4. Word use categories 1. Nominals: words referring to animals, objects and other adult-noun referents. Context-flexible: nominal forms used with reference to a range of entities, suggesting child awareness of type/token relationships. Context-limited: nominal forms used in a limited way, to refer to a single referent or as part of a routinized context, such as labelling animals with their characteristic sounds while 'reading' with Mother. Specific: nominals used to refer to particular persons or entities. The category corresponds to the adult sub-class of 'proper nouns', but may refer to terms with broader adult use which are treated as proper nouns by the child (e.g. mommy, numnum for a favourite blanket). 2. Relationals: words referring to reversible temporal or spatial transformations in the environment: all-gone, back, more, up. More than one use is required, to provide evidence of context-flexible application. Single uses of potential relational words are generally interpreted as ' event' words (which may be context-bound). 3. Event: used in relation to events which do not exhibit a reversible character. Includes words marking pretend events (feeding do\\-yum; rolling vehicle - vroom; serving teatea [to refer to a range of tea-related objects and actions]; quaffing a drink - ah; sleeping — sh; cleaning — clean) as well as real-life events (hurt finger - ow; sliding — whee). 4. Social expression: words used to mark (real or pretend) social interactions (please, hi, yay). 5. Routine/game: words used as part of verbal rituals or routines not supported by a larger situational context, including animal sounds in response to questioning out of context (baa), or games such as peek-a-boo, ' how big is baby ?' etc. 6. Deictic: words used to point out people, entities or events of interest (this, that), or to mark interest in general (aha, look, oh).

which we found necessary to account for the children's word uses. When a word was produced spontaneously in varying contexts its use was considered referential, whereas production in limited contexts was taken to reflect use as [2] Acceptance of the plausibility of such 'global' matches depends on familiarity with the phonetic patterns used by a particular child. For further discussion of Timmy's incipient phonological system, which supports recognition of ' fricative + low vowel' as an attempt at fish, for example, see Vihman, Velleman & McCune (1994). Compare also Waterson (1971), who cites such globally matching forms as [wae] forfly,[oj] for vest, for her son at i ; 5 .

528


a routine vocal response or accompaniment to a specific type of event. Nominal forms, which may or may not meet criteria for referential use, were variously categorized as Context-flexible, Context-limited or Specific (i.e. proper nouns). Relational words can only be identified by multiple occurrences which allow the inference that the child is referencing a reversible temporal or spatial phenomenon across contexts; thus, words categorized as 'relational' met our criteria for referential use by definition. The categories Event, Social expression and Routine/game rely on specific situational frames for their application. These categories differ in the circumstances of their use and eliciting conditions, but share with contextlimited nominals a restriction in application which does not characterize relational and flexible nominals. Deictic single words have functional significance, expressing the child's attention to external reality and interest in having others attend as well. Such words are limited to indicating aspects of context, although they will later play syntactic roles in sentences. We did not further classify the function or use of words occurring only as imitations. In a few cases otherwise plausible word-candidates resisted classification in terms of a 'use' or 'meaning' for the child in the given situation. Such words were eliminated from further consideration. Comparison of the Stanford and Rutgers samples We have already noted one difference in word use across the two samples: imitations account for a larger proportion of the Stanford than the Rutgers sample. In order to achieve a clearer perspective on the overall distribution of word use in the two samples we collapsed the spontaneous word use categories of Table 4 into three groups: flexible nominals, which reflect referential use, relationals, which similarly constitute referential word use, and context-bound words. The results of this re-grouping, summarized in Table 5, are striking. Whereas the two samples differ sharply in numbers of TABLE 5. Word use at 1 53/1 ;4 Rutgers

Stanford

Total Percent Total Percent Flexible nominals Relationals Context-bound Imitations Total

45 10

41 18 114

529

39 9 36 16

43 9 7' 45 168

26 5 42 27

VIHMAN & MCCUNE TABLE Months

6. Word use over time

9

IO

— —

2

2

3

3 S 3

6 3

i-7

2O

I I

12

13

14

15

16

3

16

20

44

2

5

14

24

28

28

I. Rutgers sample

Flexible nominals Relational Context-bound Imitations Total No. of subjects Mean no. of words per subject II. Stanford sample Flexible nominals Relational Context-bound Imitations Total No. of subjects Mean no. of words per subject

3 I

i

IO

10

12

6

2

20

12

21

4

7

30

30

5 47 7 67

1

4

13

17

3

3 33 25

6 46 37

6 33

S

—

I

—

—

—

i

—

5 4

14 11

17

39

2

7

21

25 6

67

2

IO

26

2

4

7

ro

25

3'7

—

41

7

95

74 8 93

9

19

62

i°5

9 69

10

105

106

47 8 84 46 185

10

10

106

i8'5

Raw totals are given for each month. The 'number of subjects' refers to the subjects with identifiable words in each month sampled.

context-bound spontaneous words as well as in imitations, they show virtually identical numbers of referential productions (flexible nominals + relational). Since nearly half as many words again were identified for the Stanford sample, the proportion of all words represented by the 52 contextflexible words (31%) is naturally smaller than the proportion represented by the 55 context-flexible words in the Rutgers sample (48%). It is likely that the differential experiences the research afforded the parents at each site are relevant to the differences in numbers of early words produced. The Stanford mothers became close collaborators in monitoring their child's language acquisition. They focused on vocal production, encouraged by weekly inquiries from the observer. They also engaged in relatively frequent verbal interchange with the observer in the course of the session, resulting in a 'chatty' atmosphere which probably contributed to the children's higher vocal output. The Rutgers mothers were not encouraged to attend to any particular aspects of the children's development. The research team provided general information about child development upon request and did collect written lists of new words, but there was an explicit intention to minimize parental awareness of detailed research goals, lest this influence their interactions with the children. These mothers attended to their children's play and communication without attempting to elicit any specific behaviour relevant to the research. They attempted to maintain the children's interest in the toy set and rarely spoke with the observer. It is certainly 53°

WHEN IS A WORD A WORD? Rutgers sample

Stanford sample

40

20

10 11 12 13 14 15 16

10 11 12 13 14 15 16

Age in months Fig. i. Word use over time in two samples. D, Flexible; +, other spontaneous; *, imitated. N = io each.

plausible that the weekly visits and emphasis on language contributed to a generally greater focus on language in the daily lives of the Stanford sample. Such an emphasis may have yielded an unintended 'training effect' on the children's language acquisition. If indeed the difference in the total number of words produced in the two samples reflects differences in parental experience, which thus had differential influence on the language environments of the home, then the most notable finding here is the ABSENCE of a difference in flexible word use in the two samples. This supports the possibility that the greater overall word production in the Stanford sample (Table i) reflected an unintended 'training effect' due to the focus on language, and at the same time suggests that such training is constrained by developmental limitations. Contextbound word use is susceptible to variation in parental efforts to encourage language, but adult-like referential usage will emerge in relation to developmental variables less susceptible to variation based on parental attention, participation in verbal games, book-reading and so on (see McCune, 1992). Longitudinal findings In order to explore further the interpretation of a training effect, we applied the same four-way grouping to the words identified across all eight months sampled. Table 6 shows the results of this analysis; Fig. 1 offers a graphic representation of the findings.

VIHMAN. & MCCUNE

The longitudinal perspective confirms our interpretation of the lexicons presented in the Appendix. The two samples begin to diverge, in both numbers of words produced per subject and in numbers of children contributing words to the pool, by o; 11 or i ;o. The Stanford sample shows a sharply ascending curve for both context-bound words and imitations, while the rising curve for these categories is greatly attenuated in the Rutgers samples. The trajectory for the acquisition of referential words, on the other hand, is virtually identical in the two samples. Such words are meagrely represented in either group until i;2. At that point, both samples show a sudden increase in the occurrence of flexible nominals - the kind of increase that might be expected to accompany an abrupt cognitive revelation, such as the 'nominal insight'. Relational words double in both samples in the following month, and double again in the Rutgers sample at i;4. In the Rutgers sample, the shift to flexible words at i; 2 is also accompanied by a doubling of word production in the 'other spontaneous' category, which suggests a new level of attention to the linguistic forms available in the environment. CONCLUSION Our data sets were collected independently, using different methods in the service of different immediate goals, but in the course of years of collaboration we have developed a common, maximally explicit and maximally inclusive procedure for identifying early word use. Two consequences of our approach can be brought out here. First, if the very earliest words-the 'ragged beginnings' of speech - are to be recognized, close attention to phonetic patterns is needed, as well as awareness of a variety of types of ways of linking sound and meaning. Secondly, if we are interested in identifying relationships between language and early cognitive development, it is essential to separate out the 'dribble' of earliest, context-bound words used by some children from the flexible usage which emerges somewhat later. These two kinds of word, with their differing degrees of symbolic and referential value, continue to co-exist, as they characterize not only the language of children but the usage of adults as well, for example such discourse markers as well, with no 'meaning' or reference outside of the immediate pragmatic context (Schiffrin, 1985), or formulas and conversational routines, which fit globally into a discourse event, but which, again, cannot be said to bear referential 'meaning' (Ferguson, 1976; Coulmas, 1979). Some children begin contextbound word production before they are ready for context-flexible language use; others begin both at the same time. The notion of a 'sound-meaning link' is far from simple. The value of the 'word' as a unit of form and meaning varies across the world's languages and even within a single language, given the problematic status of formulaic expressions and marginal words, for example. Special care and attention to 532


definition are all the more necessary in recognizing the emergence of word use in the vocal production of young children. As linguist and psychologist in collaboration we found it essential to specify phonetic and contextual criteria in order to clearly identify and communicate our own underlying assumptions. In recounting our attempts to define and evaluate both formal and functional criteria ever more closely we have attempted to provide for other investigators an instrument for assessing the reliability or degree of confidence that can be assumed in connection with different approaches to word identification, and thus for selecting methods appropriate to individual goals and areas of expertise.

REFERENCES Bates, E., Benigni, L., Bretherton, I., Camaioni, L. & Volterra, V. (1979). The emergence of symbols: cognition and communication in infancy. New York: Academic Press. Bloom, L. (1973). One word at a time. The Hague: Mouton. de Boysson-Bardies, B., Vihman, M. M., Roug-Hellichius, L., Durand, C , Landberg, I. & Arao, F. (1991). Material evidence of infant selection from the target language: a crosslinguistic phonetic study. In C. A. Ferguson, L. Menn & C. Stoel-Gammon (eds), Phonological development: models, research, implications. Parkton, MD: York Press. Coulmas, F. (1979). On the sociolinguistic relevance of routine formulae. Journal of Pragmatics 3, 239-66. Dore, J. (1985). Holophrases revisited: their 'logical' development from dialog. In M. D. Barrett (ed.), Children's single-word speech. New York: Wiley. Dore, J., Franklin, M. B., Miller, R. R. & Ramer, A. L. H. (1976). Transitional phenomena in early language acquisition. Journal of Child Language 3, 13—28. Ferguson, C. A. (1976). The structure and use of politeness formulas. Language in Society 5, 137-5'(1978). Learning to pronounce: the earliest stages of phonological development in the child. In F. D. Minifie & L. L. Lloyd (eds), Communicative and cognitive abilities — early behavioral assessment. Baltimore, M D : University Park Press. Ferguson, C. A. & Farwell, C. B. (1975). Words and sounds in early language acquisition. Language 51, 419-39. Goldfield, B. A. & Reznick, J. S. (1990). Early lexical acquisition: rate, content, and the vocabulary spurt. Journal of Child Language 17, 171-83. Greenfield, P. M. & Smith, J. H. (1976). The structure of communication in early language development. New York: Academic Press. Griffiths, P. (1986). Early vocabulary. In P. Fletcher & M. Garman (eds), Language acquisition : studies in first language development. (2nd edn.) Cambridge: C.U.P. Halliday, M. A. K. (1975). Learning how to mean - explorations in the development of language. London: Edward Arnold. Harris, M., Barrett, M., Jones, D. & Brookes, S. (1988). Linguistic input and early word meaning. Journal of Child Language 15, 77-94. Huttenlocher, J., Haight, W., Bryk, A., Seltzer, M. & Lyons, T. (1991). Early vocabulary growth: relation to language input and gender. Developmental Psychology 27, 236-48. Labov, W. & Labov, T. (1978). The phonetics of cat and mama. Language 54, 816-52. de Laguna, G. A. (1927). Speech: its function and development. New Haven, CO: Yale University Press. Leopold, W. F. (1939). Speech development of a bilingual child: a linguist's record, I: vocabulary growth in the first two years. New York: Northwestern University Press. Lewis, M. M. (1936). Infant speech : a study of the beginnings of language. New York: Harcourt Brace.

533

VIHMAN & MCCUNE McCarthy, D. (1954). Language development in children. In L. Carmichael (ed.), Manual of child psychology. (2nd edn.) New York: Wiley. McCune, L. (1992). First words: a dynamic systems view. In C. A. Ferguson, L. Menn & C. Stoel-Gammon (eds), Phonological development: models, research, implications. Parkton, MD: York Press. McCune-Nicolich, L. (1981). The cognitive bases of relational words in the single word period. Journal of Child Language 8, 15-34. McNeill, D. (1970). The development of language. New York: Harper & Row. McShane, J. (1979). The development of naming. Linguistics 17, 879-905. Matthews, P. H. (1974). Morphology: an introduction to the theory of word structure. Cambridge: C.U.P. Meier, R. P. & Newport, E. L. (1990). Out of the hands of babes: on a possible sign advantage. Language 66, 1-23. Menn, L. (1978). Pattern, control, and contrast in beginning speech: a case study in the development of word form and word function. Bloomington, Indiana: Indiana University Linguistics Club. Nelson, K. (1973). Structure and strategy in learning to talk. Monographs of the Society for Research in Child Development 38. (1988). Constraints on word learning? Cognitive Development 3, 221—46. Orlansky, M. D. & Bonvillian, J. D. (1988). Early sign language acquisition. In M. D. Smith & J. L. Locke (eds), The emergent lexicon : the child's development of a linguistic vocabulary. New York : Academic Press. Packer, M. J. (1985). Hermeneutic inquiry in the study of human conduct. American Psychologist 40, 1081-93. Petitto, L. A. (1988). 'Language' in the pre-linguistic child. In F. S. Kessel (ed.), The development of language and language researchers. Hillsdale, NJ: Erlbaum. Plunkett, K. (1993). Lexical segmentation and vocabulary growth in early language acquisition. Journal of Child Language 20, 43-60. Rescorla, L. A. (1980). Overextension in early language development. Journal of Child Language 7, 321-35. Schiffrin, D. (1985). Conversational coherence: the role of well. Language 61, 640—67. Sinclair, H. (1970). The transition from sensory-motor behavior to symbolic activity. Interchange 1, 119—25.

Veneziano, E. (1981). Early language and nonverbal representation: a reassessment. Journal of Child Language 8, 541-63. Vihman, M. M. (1992). Early syllables and the construction of phonology. In C. A. Ferguson, L. Menn & C. Stoel-Gammon (eds), Phonological development: models, research, implications. Parkton, MD: York Press. Vihman, M. M., Ferguson, C. A. & Elbert, M. (1986). Phonological development from babbling to speech: common tendencies and individual differences. Applied Psycholinguistics 7, 3—40.

Vihman, M. M., Macken, M. A, Miller, R., Simmons, H. & Miller, J. (1985). From babbling to speech: a re-assessment of the continuity issue. Language 61, 397—445. Vihman, M. M. & Miller, R. (1988). Words and babble at the threshold of lexical acquisition. In M. D. Smith & J. L. Locke (eds), The emergent lexicon: the child's development of a linguistic vocabulary. New York: Academic Press. Vihman, M. M., Velleman, S. L. & McCune, L. (1994). How abstract is child phonology? Toward an integration of linguistic and psychological approaches. In M. Yavas (ed.), First and second language phonology. San Diego: Singular Press. Waterson, N. (1971). Child phonology: a prosodic view. Journal of Linguistics 7, 179-211. Werner, H. & Kaplan, B. (1963). Symbol formation : an organismic-developmental approach to language and the expression of thought. New York: Wiley. (Reprinted 1984. Hillsdale, NJ: Erlbaum.)

534


APPENDIX Word forms at i;4, organized by lexical category. The most common spontaneous phonetic variant is given or, if variants are equally distributed, the variant best matching the adult form. Frequency of use is given in parentheses. Three subjects are sampled at i; 3 as indicated below. One Rutgers subject produced no words.

I. RUTGERS SUBJECTS

Alice

Nominate:

flexible

limited

specific

apple (4) [?ae] bang (1) [pai] elephant (4) [5:IJA] baby (7) [beibi] nose (1) [n:ae] belly (1) [vei] blanket (1) [k'et] bottle (14) [badj] bunny (3) [bAnn:i] daddy (2) [da:di] duck (1) [tae?] egg (8) [?ei] eye (4) [?ai] flower (3) [p'adi] hat (2) [?a] iron (2) [?aiji] key (1) [ci] lady (4) [je:ji] man (1) [ma;:] meat (1) [mi?] milk (1) [mms] Oscar (= puppet) (6) [?a?] plate (4) [p'ei] shoe (1) [ci] tea (1) [p:]

Relational down (2) [dau] no (1) [nae] Event shiny (3) [ta:ji]

up (2) [?A:p] clean (2) [ti:ni]

Social bye (1) [bai]

Imitation hello (1) [lou] 535

Grandpa (3) [p'a] mommy (10) [ma:ni]

VIHMAN & McCUNE

Aurie Nominate: flexible limited apple (2) [Paebi] cock-a-doodle-doo sound (6) [kakijali::] book (1) [buk:ts] dolly (2) [?adi] shoe (2) [tsi] Relational bye (1) [baebaei] up (5) [?ap] peek-a-boo (2) [p'iljhu:] Event yum sound (2) [vAm:i] ow (1) [?aeo] Social hi (4) [haei] hello (12) [hA:o] Routine/game tickle-tickle (1) [tikatik] open (2) [thA:ppu] Imitation stuck (1) [tae] orange (1) [Pawis] uh-oh (1) [hA?u] out (1) [Paut] whoop (1) [wu::] pretty (1) [pli] yoohoo (5) [huho] Danny Deictic aha (8) [?&hsb]

that (6) [da::]

Jase limited Nominals: flexible woof sound (5) [wAp] ball (1) [bo::a] dog (4) [du:] Grover (puppet) (3) [go:] juice (4) [cfeis] Relational more (5) [mo] Social no (2) [nou] Kari Event car sound (5) [?m] Social uh-oh (1) [?A?a] Imitation no (1) [nA]

up (1) [Pap]

536

specific mommy (7) [mam]


Rala Nominate: flexible baby (5) [p'epi] ball (10) [bo] beads (2) [?ebi] bottle (6) [hAbau] car (7) [[hek'a] Relational uh-oh (7) [?A?] Event tea (11) [ti:] do (6) [nhto] Deictic see (20) [si:] Imitation bye (3) [bai] toe (2) [hethoo] Rick Nominals: flexible ball (3) [p'o:] doggie (2) [go:di] eyes (3) [?ai] spoon (1) [p'o:p'i] Relational no (2) [nou] Event yum sound (1) [?Am] Social hi (6) [hai]

comb (8) [hekom] mouth (4) [mao] shoes (2) [aja:/] spoon (5) [hebu:]

X go[es] (6) [Pik'ou]

limited beads (1)

specific daddy (2) [daegi] mommy (6) [mam:]

open (17) [oup'i]

Ronny (i;3) Nominals: flexible baby (7) [p'ebi] woof sound (1) [wa[3a] Event vroom sound (1) [vf:um] Social hello (1) [[?al:o] Routine/game sit (1) [sit] Deictic this (1) [di]

specific mama (1) [mama]

537

VIHMAN & MCCUNE

Imitation yellow (i) [ijaeb] Vido Nominals: Social thank you (8) [de:ide] Imitation apple (2) [bapba] daddy (1) [darn] flower (3) [fa:|3wa]

limited baby (2) [bobap] unh-hunh (5) [[?5h good (1) [go:n] keys (2) [se:s] yellow (1) [deblou]

I I . STANFORD SUBJECTS

Andrew Nominals:

Limited spoon (2)

Relational up (3) [?aep] Social no (1) [ne::] Imitation bird (7) [be] clock (3) [kaek] tail (4) [t'e]

[p'o]

mountain (2) [ITIA]

Camille Relational up (7) [Pap] Event car sound (3) [djhis] Deictic (what's) this (7) [?9zis] Imitation hello (5) [haha] Deborah Nominals: flexible baby (7) [p'e:bi] ball (7) [boa] bird (2) [bwa] bottle (2) [ba:] cheese (2) [§i:] corn (5) [kho:]

limited moo sound (8) [bo::]

538

specific monkey (1) [hmmae:]


duck (i) [tae] eye (6) [?ai] kitty (4) [k'ekji] Relational down (4) [ta] Event water (5) [wawoa] hug sound (4) [?m:m] yum sound (12) [?m] Social bye (1) [pan] Routine/game A (1) [?e:] three (8) [§i::] Imitation carrot (1) [kheiwi] cow (1) [kh?:] ear (1) [?i:?]

up (3) [?ah]

hi (i S ) [hai] two (2) [thi]

hair (1) [?ea?] crashing rock sound (2) [kx::]

Emily (153) Nominals: flexible beads (1) [phi:ge] box (2) [phat:] Relational more (17) [mo:] Event yum sound (17) [?Am] daddy (7) [daedae] Routine/game tickle (5) [thithi] Deictic oh (2) [?ou] Imitation night-night (1) [na|naet] patty-cake (3) [p'ak'ae] open (5) [?appi] up (2) [?Ap] Jonah Event rock-rock (3) [wa:hwa] Routine/game roar sound (9) [kfi] Imitation cock-a-doodle-doo sound (2) [akakek:a] cookie (2) [k'ak'a] no (2) [na::u] 539

specific Marilyn (2) [meme]

VIHMAN & McCUNE

Molly

Nominate: flexible

limited

specific

baby (2) [pebi] block (8) [pak] book (13) [puk] camera (4) [kama] ear (4) [he:] glasses (29) [kaki] meow sound (8) [miA] nose (1) [no:]

bead (8) [pi:] piano (2) [pau] picture (4) [popo]

Brett (9) [pat] Graham (9) [kon:i] Grandma (1) [meuwA] Grandpa (1) [kaepo] mama (7) [mama] Nicky (2) [m:i] Nonny (10) [nanni] Ruth (6) [wut]

Relational stuck (1) [kak] Event vroom sound (12) [b?Am] bang (4) [pan:a] walk-walk (8) [wokae] choochoo sound (10) [tutu] down (10) [tanna]

Social hi (4) [?ai:]

Routine/game baa sound (2) [pae:] cluck sound (11) [bA?bA?] neigh sound (8) [p'A] moo sound (13) [mu:] peek-a-boo (7) [pik]

snort sound (14) [?e?] three (2) [wi:] two (2) [tu] woof sound (17) [wu]

Imitation click (11) [kik] good girl (2) [guga] green (3) [kyn:i] house (2) [haut] in (1) [ihni] name (1) [nem:i] oink sound (1) [ho:k]

open (1) [hopo] pig (1) [pik] red (2) [wat] rug (3) [wa:k] tail (26) [teu] that (2) [tat] work (6) [hAk]

Sean

Nominals: flexible

limited

specific

bird (12) [bwi:ts] block (16) [pak] book (4) [bik] bug (2) [mbAkl] butterfly (3) [pAJA:] cracker (2) [djak] dog (3) [tao]

cock-adoodle-doo sound (2) quack sound (5) [?a?] vroom sound (3) [(3::] woof sound (2) [wau]

mama (9) [ma:] [daludalu::]

54O


duck (4) [tAk1] fish (4) [f0tj] horse (12) [?ij] moo sound (6) [?6:h] mouse (9) [ma1] rabbit (3) [phaets] Relational more (29) [mo] Routine/game baa sound (3) [bah] Deictic look (4) [jekh] oh (9) [?oh] Imitation berries (3) [bsbij] Ian (2) [nih] stuck (1) [na:k'] Susie (i;3) Event kiss sound (7) [?m] Imitation ice (cream) (1) [?a?i] Thomas Nominals: flexible ball (23) [pakhi] glasses (6) [gae:ti] Relational no (5) [no] Event down (1) [ta:] Social hi (4) [hai] Imitation baby (2) [bi:] string (1) [tih] Timmy Nominals: flexible baby (12) [paepae] block (17) [gmbae] boat (2) [pae]

blue (3) [boh] this (6) [dis]

out there (2) [atats] limited truck (4) [tAkis] woof sound (2) [?A?A?A] up (2) [Pap]

oh yeah (1) [ha:jah] toe (1) [tou] zoom (2) [zu::i] limited balloon (5) [bei] toottoot sound (2) [dAits:]

specific daddy (2) [tadae] mama (2) [mAmae] Nana (1) [Paemenae]

VIHMAN & MCCUNE

bracelet (6) [paepae] car (2) [kai] eye (5) [?aei] fish (2) [s:ae] light (15) [aija] round & round (15) [AligAligae] ( < helicopter) Social hiya (3) [aijae] Routine/game D (1) [di::] Imitation boy (1) [bAi] goodbye (2) [kaebae] please (3) [pai] cookie (1) [kaki] flower (1) [Aoae]

542

Ruth (4) [hA|3ae] Simon (6) [nAmae]