'referential' children? Relationships between

2 downloads 0 Views 1MB Size Report
Feb 17, 2009 - 100-word measures on the seven children who went on to acquire 100 words .... usage also seem to be implicated, at least at 100 words. 82 ...
Journal of Child Language http://journals.cambridge.org/JCL Additional services for Journal of Child 

Language: Email alerts: Click here Subscriptions: Click here Commercial reprints: Click here Terms of use : Click here

How referential are ‘referential’ children?  Relationships between maternal­report and  observational measures of vocabulary  composition and usage Julian M. Pine Journal of Child Language / Volume 19 / Issue 01 / February 1992, pp 75 ­ 86 DOI: 10.1017/S0305000900013635, Published online: 17 February 2009

Link to this article: http://journals.cambridge.org/abstract_S0305000900013635 How to cite this article: Julian M. Pine (1992). How referential are ‘referential’ children? Relationships  between maternal­report and observational measures of vocabulary composition  and usage. Journal of Child Language, 19, pp 75­86 doi:10.1017/ S0305000900013635 Request Permissions : Click here

Downloaded from http://journals.cambridge.org/JCL, IP address: 138.253.100.121 on 17 Jul 2013

J. Child Lang. 19 (1992), 75-86. Printed in Great Britain

How referential are 'referential' children? Relationships between maternal-report and observational measures of vocabulary composition and usage* JULIAN M. PINE University of Dundee (Received 5 November 1990. Revised 23 April 1991)

ABSTRACT

This study examines the relationship between maternal-report measures of referential vocabulary and observational measures of referential vocabulary and usage in eight first-born middle-class children at 50 and 100 words. The results indicate that although maternal-report measures of vocabulary composition can be reasonably reliable, provided some attempt is made to restrict variation in the number of vocabulary items upon which they are based, such measures tend to exaggerate the relative importance of common nouns for the child in two ways; firstly, in the sense that they reflect differential maternal sensitivity to such wordtypes in comparison with other less-'referential' items; and, secondly, in the sense that they overestimate the extent to which such word-types actually occur in the child's spontaneous speech.

INTRODUCTION Despite the increasing use of diary and checklist methodologies for the investigation of individual differences in early vocabulary development (e.g. Nelson, 1973; Snyder, Bates & Bretherton, 1981; Olson, Bayles & Bates, 1986; Bates, Bretherton & Snyder, 1988; Dale, Bates, Reznick & Morisset, 1989), the question of the relationship between differences in vocabulary composition and differences in children's actual vocabulary use is one which has received very little attention in the literature, or has, at most, been dealt with exclusively in terms of issues of reliability between maternal-report and observational data (e.g. Bates et al. 1988; Hampson, 1989). This state of [•] Address for correspondence: Julian M. Pine, Department of Psychology, University of Dundee, DDi 4HN, Scotland. 75

CHILD LANGUAGE

affairs would seem to reflect the implicit assumption that vocabulary measures of the type used in Nelson's original study bear an essentially simple and direct relationship to children's language use, and that, provided maternal reports can be trusted, such measures represent a trouble-free means of assessing early linguistic behaviour which is much less labourintensive than collecting and transcribing naturalistic speech samples. However, while the use of diary and checklist methodologies may well be the only practical way of collecting early vocabulary data for large samples of children, the assumption that maternal-report vocabulary measures and measures of actual usage are somehow equivalent is difficult to justify, confusing, as it does, issues of reliability - are maternal-report measures accurate ? - and issues about the relationship between vocabulary and usage - to what extent do children make use of the different words which make up their early vocabularies ? The result is that neither of these questions has been satisfactorily addressed in the literature and hence the precise nature of the relationship between differences in vocabulary composition and differences in vocabulary use is far from clear. The reliability of maternal-report data Of the two issues outlined above, the question of the reliability of maternalreport data has received by far the most attention in the literature, with several studies documenting good agreement between maternal-report or interview measures and observational measures of early language (e.g. Bretherton, McNew, Snyder & Bates, 1983; Tomasello & Todd, 1983; Bates et al. 1988). At first sight, such findings would appear to provide strong support for the claim that maternal-report measures represent a valid and reliable tool for the study of early language acquisition. However, it is important to consider the range and limits of the analyses performed before accepting such a conclusion. The most obvious point to make, in this respect, is that, where vocabulary acquisition has been the focus of study, reliability analyses have tended to be restricted to measures of vocabulary size alone. For example, Tomasello & Todd (1983) report a correlation of 0-92 for vocabulary size at 18 months and Bates et al. (1988) report a correlation of 0-72 for vocabulary size at 13 months, but neither study reports reliability measures for the proportions of different word-classes used by the child. Thus, although there is clearly evidence that maternal-report measures are useful for assessing the relative size of children's vocabularies at particular age-points, there is much less evidence to support the assumption that they are accurate in picking up variation in the actual make-up of these vocabularies. Indeed, the results of the only analysis which bears directly on this issue - Hampson's (1989) comparison of maternal-report and observational measures of referential 76

MEASURES OF VOCABULARY COMPOSITION

vocabulary at 13 months - reveal somewhat lower correlations of 0-47 (p < 0-03) for the sample as a whole (N = 45) and 0-65 (p < 0-007) f° r a subgroup of the 18 most advanced children, suggesting that maternal-report measures of vocabulary make-up may be less reliable than previous studies have tended to assume. This conclusion is not altogether surprising when one considers the obvious difference between the two types of reliability measures. Proportional measures of vocabulary composition are, after all, inherently more finegrained than absolute measures of vocabulary size, particularly when the latter are taken at age-points when variation in vocabulary size is effectively enormous, and there is no guarantee that mothers who are able to give accurate reports of one are also going to be accurate in reporting the other. On the other hand, given that it is upon measures of vocabulary composition rather than vocabulary size that claims about stylistic differences actually depend, and that the vast majority of such claims are based on data derived from just such maternal report measures, the implications of these findings for studies of individual differences are rather disconcerting. The picture is complicated, however, by the fact that, as Pine & Lieven (1990) have shown, 13-month measures of variation in vocabulary composition are themselves problematic. That is to say, wide variation in the size of the vocabularies upon which they are based tends to reduce their reliability relative to measures based on a fixed number of vocabulary items (e.g. 50 or 100 words). It is therefore unclear to what extent the lower reliability of Hampson's maternal-report measures is due to maternal inaccuracy per se as opposed to the poor reliability of the proportional measures themselves. That is to say, since Hampson's data show a clear increase in reliability (from 0-47 to 0-65) when some attempt is made to control variation in vocabulary size by restricting the analysis to a more homogeneous subgroup of children, it might well be that the reliability of maternal-report measures would increase still further if measures were based on variation at 50 and 100 words rather than 13 months. The present study aims to clarify this issue by examining the relationship between maternal-report and observational measures of vocabulary composition at 50 and 100 words. It is predicted that the reliability of the former will be somewhat higher than that revealed by both of Hampson's (1989) analyses, including not only that based on her entire sample of 45 children but also that based on the 18 children who had acquired 15 or more words by 13 months. Vocabulary and usage Of course another potential reason for the relatively low reliability of maternal report measures of vocabulary composition is the possibility that the actual composition of children's vocabulary may be only weakly related 77

CHILD LANGUAGE

to children's actual language use. That is to say, some children may have large numbers of common nouns in their early vocabularies, but use them seldom, whereas others may have relatively few, but use them much more extensively. Although generally ignored in discussions of individual differences in vocabulary composition, this possibility is not only a methodological but also an empirical issue. That is to say, despite the tendency to interpret variation in vocabulary composition in terms of differences in language use, it has never been shown in any reliable way that such a direct relationship exists between the child's underlying vocabulary and the uses to which it is put. This point is illustrated by Lieven's (1980) longitudinal study of three children, Beth, Kate and Jane. Unfortunately, Lieven's study did not involve the use of maternal-report data. However, she does present data for the distribution of both word-types and word-tokens in the children's spontaneous speech. Interestingly, these two sets of data are related to each other in rather complex ways. In terms of observed vocabulary distribution, general nominals were found to be the most frequent word-type for all three children, with Kate scoring highest and Beth scoring lowest, though for none of the children did the proportion of general nominals reach Nelson's criterial value of 50 per cent. In terms of word-frequency, however, although Jane still used general nominals more frequently than any other word-class, Kate used almost as many personal-social words as general nominals, and Beth used many more specific nominals and non-classifiable utterances. The most obvious point to make about these results is that they reveal differences in the rank ordering of children across the different measures, such that Kate would be classified as the most referential child according to the distribution of words in her observed vocabulary, whereas Jane would be classified as the most referential child in terms of her actual frequency of word-use, casting doubt on the assumption that vocabulary measures, even when based entirely on observation, are directly related to the frequency of actual word-use. Lieven discusses this point at some length, showing that the large discrepancy between Kate's vocabulary and usage is due to the frequent use of a small number of personal-social words, particularly yes and no, whereas the large discrepancy between Beth's vocabulary and usage is due to frequent use of a small number of formally heterogeneous words (Mummy, Julian, look, there, more) used primarily to gain attention from others. It is worth noting, however, that this analysis not only underlines the problems with assuming a direct relationship between vocabulary and usage, but also suggests that vocabulary measures of referentiality may greatly overestimate the relative importance of referential items in the child's spontaneous speech (by an average of over 16% in Lieven's study), a conclusion which, though not particularly surprising, raises the question of whether it is sensible to discuss individual differences purely in terms of referentiality and non78

MEASURES OF VOCABULARY COMPOSITION

referentiality, given that the referential category may account for a much smaller proportion of children's actual speech than it does of their underlying vocabularies. The present study aims to investigate the relationship between vocabulary, word-type and word-token usage by looking at the relationship between these three sets of measures at 50 and 100 words. It is hypothesized that measures of referentiality based on cumulative vocabulary records will be significantly higher than measures based on observational data alone, and that the proportion of referential word-types observed will be significantly higher than the proportion of referential word-tokens which children actually use.

METHOD

The present study formed part of a more extensive longitudinal study of the early language development of 13 first-born middle-class children (seven girls and six boys) which involved the collection of monthly 60-minute audiorecordings and cumulative maternal-report diary records of children's speech between o; 11 and 1 ;8. Subjects Subjects for the present study included the eight children (six girls and two boys) who reached 50 words before recording ceased at i ; 8 . Procedure At the beginning of the study mothers were informed that the experimenter was interested in their child's language development and instructed in how to keep a continuous diary record of his or her speech. They were told to record any spontaneous occurrence of a new expression, together with its meaning and situational context. Maternal-report referential vocabulary measures were thus derived by calculating the proportion of common nouns present in the children's first 50 and 100 words from cumulative diary records collected month by month throughout the study. The corresponding observational measures, together with measures of the proportion of common noun tokens actually used by children, were derived from 60-minute audiorecordings made in the children's homes at the age-points at which 50 and 100 words were reached. Fifty-word measures were based on the eight children who reached 50 words before recording ceased at 20 months, and 100-word measures on the seven children who went on to acquire 100 words before this cut-off point. No attempt was made to restrict recording to the mother-infant dyad. That is to say, the researcher behaved as a participant 79

CHILD LANGUAGE

observer throughout, and in some cases fathers were also present. However, despite these attempts to reduce the artificiality of the situation, recording sessions typically involved prolonged periods of mother-infant interaction in joint play situations.

RESULTS

Maternal-report and observational vocabulary measures Descriptive statistics for the maternal-report and observational measures of referential vocabulary at 50 and 100 words are presented in Table 1 and show TABLE 1. Maternal report and observational measures of vocabulary composition at 50 and 100 words Mean

S.D.

Range

16-6 42-8

2O

140-200

8-6

280-540

331

123

200-51-2

i8- 3 48-0 34'3

17 107 I 12

350-670 222-537

50-word measures

Age Reported noun types (%) Observed noun types (%) i oo-word measures Age Reported noun types (%) Observed noun types (%)

150-20-0

substantial variation in the two types of measures at both measurement points. Moreover, although the maternal-report measures at 50 words are, on average, somewhat lower than those reported by Nelson (42-8 vs. Si'o), it is important to note that this is, at least in part, a result of the present study's adoption of Bates et al.'s (1988) more narrowly defined common-noun category which explicitly excludes certain vocabulary items treated as 'general nominals' in Nelson's original study, including pronouns such as he and that, abstractions such as God and birthday, letters and numbers such as E and two and also possibly animal noises such as Woof and Moo (Nelson, !973: i7)' Thus, although only two of the eight children in the present study would be classified as referential on the basis of having more than 25 common nouns in their first 50 words, this is at least partly because such a criterion is considerably more conservative than the 50% general nominal criterion originally used by Nelson. Certainly, there is at least one additional child in the sample who is close enough to the 50 % boundary with 24 common nouns in her first 50 words to have been classed as referential if the present study were to have adopted Nelson's (1973) practice of treating pronouns as referential items. 80

MEASURES OF VOCABULARY COMPOSITION

Relationships between the pairs of maternal-report and observational measures presented in Table i were investigated using Pearson productmoment correlation coefficients and Wilcoxon signed rank tests. The results of this analysis are presented in Table 2. Both the 50- and 100-word TABLE 2. Relationships between maternal-report and observational measures of vocabulary composition at 50 and 100 words

50 words Common nouns (%)

o-66*

2-38**

100 words Common nouns (%)

0-89***

2-37**

* p < 0-05 ; ** p < 0 0 1 ; *** p < 0-005.

correlations are, as predicted, substantially higher than Hampson's reliability coefficient at 0-47 (r = o-66, p < 0-05; and r = 0-89, p < 0-005, respectively), confirming the hypothesis that maternal-report measures of vocabulary composition tend to be more reliable when some attempt is made to restrict variation in the size of the vocabularies upon which these scores are based. However, there is clearly very little difference between the correlation obtained at 50 words (r = o-66) and the correlation obtained by Hampson for her subgroup of 18 more advanced children (r = 0-65). This might be taken to suggest that cross-sectional measures can be as reliable as measures based on a fixed number of vocabulary items provided they exclude, as Hampson did, children with very small vocabularies (of 14 words or less). However, it is clear that, as a reliability coefficient, neither of these correlations is particularly impressive and both contrast sharply with that obtained at 100 words (r = 0-89, p < 0-005), raising the possibility that maternal-report referential vocabulary scores at 100 words may be superior to both 13-month and 50-word measures as indicators of stylistic differences in children's actual language use. Certainly, this correlation is the highest reported in the literature to date between maternal-report and observational referential vocabulary scores, though it is obviously based on a much smaller sample than either of those reported by Hampson at 13 months. A further feature of the results presented in Table 2, however, is the significant differences in referential vocabulary scores across the two types of measures. That is to say, it is clear from Table 1 that there is a systematic drop in the relative numerical importance of common nouns between maternalreport and observational measures of vocabulary composition (9-7 % and I 3 ' 7 % o n average at 50 and 100 words respectively), suggesting either that mothers systematically overestimate the relative proportion of common 81

CHILD LANGUAGE

nouns in their children's vocabularies or else that their children significantly under-use the common nouns present in their underlying vocabularies in their actual speech. Of course it is impossible to distinguish between these two alternative explanations on the basis of the results presented above. However, one way of resolving this issue is to tackle it more directly by looking for specific disagreements between the observational and maternalreport measures (i.e. cases where word-types occur on tape but go unreported on the diary form) and adjusting the original vocabulary measures accordingly (i.e. by adding in the unreported items). The new derived measures (Diary + Tape) could then be compared with the original vocabulary measures (Diary only) and the original observational measures (Tape only) in order to determine where the original differences actually lay. For example, if there were no differences between the two sets of vocabulary measures, but significant differences between vocabulary and observational measures, this would tend to suggest that the original differences could be accounted for in terms of differences in children's actual usage, whereas significant differences between the two sets of vocabulary measures would tend to suggest significant effects due to maternal biases in reporting. Such an analysis was performed by adding unreported spontaneous items occurring on relevant recordings to the end of diary records collected on corresponding visits to derive new 50- and 100-word measures. The new vocabulary measures were then compared with the original vocabulary and observational measures using Wilcoxon signed rank tests. The results of this analysis are presented in Table 3. They suggest that maternal reporting TABLE 3. Differences between corrected maternal-report and observational measures of vocabulary composition at 30 and 100 words Maternal-report (2)

Observational (*)

50 words Common nouns (%)

2^37**

o-oo

100 words Common nouns (%)

2'O2*

2 '37**

* p < 0-05; ** p < o-oi.

biases are indeed an important factor in explaining the differences between maternal-report and observational scores at both measurement points, though differences between children's underlying vocabularies and actual usage also seem to be implicated, at least at 100 words.

82

MEASURES OF VOCABULARY COMPOSITION

Vocabulary and usage Descriptive statistics for the observational measures of referential token use at 50 and 100 words are presented in Table 4. The relationship between these TABLE 4. Referential token use at j o and 100 words

50-word measures Observed noun tokens (%) 100-word measures Observed noun tokens (%)

Mean

s.D.

Range

261

152

9'6~49'i

270

129

142-491

measures and maternal-report measures of vocabulary composition was investigated using Pearson product moment correlation coefficients. The results of this analysis are presented in Table 5. Again these correlations TABLE 5. Relationships between maternal-report measures of vocabulary composition and observational measures of token usage at 50 and 100 words R coefficients 50 words Common nouns (%)

052

100 words Common nouns (%)

099*

*p < 00005.

suggest a difference in the reliability of the 50- and 100-word measures, though in this case the correlation at 50 words (r = 0-52) is not only substantially lower than that at 100 words (r = 0-90, p < C0005), but also fails to reach significance, suggesting that the widespread assumption of a direct relationship at 50 words between individual differences in vocabulary composition and individual differences in word use may not, in fact, be justified. On the other hand these results, which remain substantially the same when Spearman rank correlations are used (rho = 0-52 and 0-96 respectively), provide further support for the view that maternal-report referential vocabulary scores at 100 words may be superior to both the 13month and 50-word measures as indicators of stylistic differences in children's actual language use. Indeed the reliability of the maternal-report vocabulary scores at 100 words is extremely impressive in this respect, not only supporting the claims of previous research about the potential reliability of maternal-report measures (e.g. Snyder et al. 1981; Bretherton et al. 1983), 83

CHILD LANGUAGE

but also suggesting that there may be very strong relationships between vocabulary and usage at this point in development. It is also clear, however, from a comparison of children's referential usage and observational measures of their underlying referential vocabularies that even the latter significantly overestimate the extent to which common nouns are actually used by the child (by 7% and 6 7 % on average at 50 and 100 words respectively). These differences both proved significant at p < 0-05 when examined using Wilcoxon signed rank tests, suggesting that, although maternal-report referential vocabulary measures may be good predictors of children's actual word-use in relative terms, common nouns may be systematically under-used in children's speech relative to other word-types in the child's vocabulary. DISCUSSION

The results of the present study have several important implications for the study of individual differences in early vocabulary development. Firstly, although they indicate that maternal-report measures of vocabulary composition can be reasonably reliable provided some attempt is made to restrict variation in the number of vocabulary items upon which they are based, they also suggest that 100-word measures may be superior to both 13month and 50-word measures as indicators of stylistic differences in children's actual language use. Secondly, they show that, despite their reliability in relative terms, maternal-report measures at both 50 and 100 words are susceptible to reporting biases which may result in a relative overemphasis on the commonnoun category. It is perhaps worth noting at this point what the reportingbias effects documented in the present study mean in absolute terms. That is to say, since the analysis from which they emerged was based entirely on information about maternal 'errors of omission', it is clearly not the case that the differences observed reflect a tendency in the mothers to report common nouns which the child has yet to acquire, but rather that they reflect a tendency to under-report instances of other word-classes and thereby underestimate their relative contribution to the child's vocabulary. While this may be less of a problem for methodologies based on checklists as opposed to cumulative diary records, it could well be that this tendency reflects commonsense maternal beliefs about the nature of language as a referential system, beliefs which have tended to be reinforced and reproduced in the literature by an almost exclusive emphasis on common nouns as the defining feature of stylistic variation. This is clearly a tendency which should be borne in mind when interpreting maternal-report measures of early vocabulary development since, while it is unlikely to account for the fact of variation itself, it may interact with early rate-effects; for example, if mothers of slower children were to report non-referential words for lack of anything 84

MEASURES OF VOCABULARY COMPOSITION

else to say, whereas mothers of more precocious children were to ignore them because they have plenty of their preferred category of common nouns to report. If this is indeed the case it implies the need to distinguish between relatively quick and relatively slow subgroups of children when considering variation in early vocabulary composition (e.g. Hampson, 1989). That is to say, given that the only real criterion for labelling a child as 'expressive' is a low proportion of common nouns in the child's early vocabulary, it is possible that there may be qualitative differences between relatively slow and relatively advanced ' expressive' children such that only the latter (who were strongly represented in Hampson's (1989) subgroup of early talkers) may exhibit positive characteristics of an alternative language style. Thirdly, these results indicate that measures of vocabulary composition significantly overestimate the relative numerical importance of common nouns in children's actual speech both in terms of the proportion of referential word-types and the proportion of referential word-tokens which are actually used. Thus, although common nouns are still, on average, the most frequently used word-class at 50 and 100 words, they account for less than 30 % of the children's actual utterances, a substantially smaller proportion than they do of children's underlying vocabularies, at least according to maternal report (42-8% and 4 8 % respectively at 50 and 100 words). Of course it is not at all surprising that children should use some of the words in their underlying vocabularies more frequently than others, nor indeed that the more frequently used words should tend not to be common nouns, since many common nouns (e.g. body-parts and the more exotic animal names) are likely to be of limited use in day-to-day interaction. However, this finding does raise the question of whether it is sensible to continue to discuss individual differences purely in terms of referentiality and non-referentiality, given that the referential category may account for a much smaller proportion of children's actual speech than has tended to be assumed as a result of studies based entirely on measures of vocabulary composition. Finally, the failure to find a significant correlation between 50-word measures of underlying vocabulary composition and measures of actual word use tends to cast doubt on the assumption that the former can be viewed as more or less directly related to measures of the child's spontaneous speech. This suggests that claims about children's language use based solely on information about the type of words present in their underlying vocabularies should be treated with a certain amount of caution at least until the child is beyond the 50-word stage. That is to say, although vocabulary data obviously provides valuable information about the kind of words which children are extracting from their input, it is important to remember that the relationship between differences in vocabulary composition and differences in children's actual speech is not necessarily a simple one and requires further research if it is to be properly understood. 85

CHILD LANGUAGE

To summarize, although maternal-report measures of vocabulary composition would seem to be reasonably reliable provided some attempt is made to restrict variation in the number of vocabulary items upon which they are based, it appears that such measures tend to exaggerate the relative importance of common nouns for the child in two ways; firstly in the sense that they reflect differential maternal sensitivity to such word-types in comparison with other less 'referential' items; and, secondly, in the sense that they overestimate the extent to which such word-types actually occur in the child's spontaneous speech. While it is clearly not possible to 'explain away' individual differences in early vocabulary development by reference to such effects - the maternal-report measures are, after all, reasonably reliable in relative terms - these tendencies would seem to underline the need not only for a more detailed investigation of the relationship between differences in vocabulary composition and differences in actual language use, but also for a characterization of stylistic variation which goes beyond simple measures of so-called 'referential vocabulary' to provide a positive account not only of 'non-referential' children, but also of children's 'non-referential' vocabulary and the functions which it serves. REFERENCES Bates, E., Bretherton, I. & Snyder, L. (1988). From first words to grammar: individual differences and dissociable mechanisms. Cambridge: C.U.P.

Bretherton, I., McNew, S., Snyder, L. & Bates, E. (1983). Individual differences at twenty months: analytic and holistic strategies in language acquisition. Journal of Child Language 10, 293-320. Dale, P., Bates, E., Reznick, J. & Morisset, C. (1989). The validity of a parent report instrument of child language at twenty months. Journal of Child Language 16, 239-49. Hampson, J. (1989). Elements of style: maternal and child contributions to referential and expressive styles of language acquisition. Unpublished doctoral dissertation. City University of New York. Lieven, E. V. M. (1980). Language development in young children: children's speech and speech to children. Unpublished doctoral dissertation, University of Cambridge. Nelson, K. (1973). Structure and strategy in learning to talk. Monographs of the Society for Research in Child Development 38 (1-2).

Olson, S. L., Bayles, K. & Bates, J. E. (1986). Mother and child interaction and children's speech progress: a longitudinal study of the first two years. Merrill-Palmer Quarterly 32, 1-20.

Pine, J. M. & Lieven, E. V. M. (1990). Referential style at thirteen months: why age-defined cross-sectional measures are inappropriate for the study of strategy differences in early language development. Journal of Child Language 17, 625—31. Snyder, L., Bates, E. & Bretherton, I. (1981). Content and context in early lexical development. Journal of Child Language 8, 565-82. Tomasello, M. & Todd, J. (1983). Joint attention and lexical acquisition style. First Language 4, 197-212.

86