Assessing Age of Onset Effects in (Early) Child L2 Acquisition

Abstract

This study compares the development of three different types of bilingual/second language children in their acquisition of gender-marking on adjectives in Dutch to investigate whether there is evidence for age-of-onset effects in early childhood as proposed by Meisel (2009). The three groups of children are: simultaneous bilingual children, exposed to Dutch and English from birth; early successive bilingual children, first exposed to Dutch between the ages of 1 and 3 years; and second language children, whose age at first exposure ranged from 4 to 10 years. In an initial analysis that included all children, early successive bilingual and in particular second language children produced qualitatively different errors from the other bilingual and monolingual groups. It is argued, however, that these errors resulted from transfer from the children's other language, English. Once children's knowledge of gender attribution is taken into account, similar error profiles were observed across all groups (contra Meisel 2009).

ACKNOWLEDGMENTS

This research was supported by the Netherlands Organisation for Scientific Research with a VENI Innovational Research Incentives Scheme award to the author and an international programme award to Leonie Cornips. I wish to thank the participants and research assistants, as well as Leonie Cornips, Aafke Hulk, Antonella Sorace, and Ianthi Tsimpli for discussion of the issues in this paper, and audiences at GALA 2011, ISB8, and Indiana University for feedback on a presentation version. I also thank two anonymous reviewers and Elma Blom for their comments.

Notes

¹The null morpheme also has another entry for [−attributive] (Blom, Polišenskà & Weerman 2006).

²In fact, on the analysis given in (2), what children need to do is to specify the gender of neuter nouns, rather than all nouns (L. Cornips, p.c.).

³However, note that Unsworth et al. (in press) also examined the acquisition of Greek gender, which is more salient and systematic than gender in Dutch and is acquired much earlier. There, too, for comparable groups of children, age of onset was not found to be a significant predictor of accuracy scores for gender-marking on definite determiners. This suggests that also for the acquisition of more transparent gender systems, age of onset may not be the crucial predictive factor for children's accuracy scores.

⁴These are the same children as in Unsworth et al. (in press).

⁵Note that using a stricter definition of child L2, for example, age of onset between 4 and 7 (following Schwartz 2004), does not significantly alter the results presented in the analysis presented here.

⁶Cumulative length of exposure differs from the traditional way of calculating length of exposure, that is, age at testing minus age of onset, in that it takes into account the intensity of the child's exposure over time (for complete overview see Unsworth 2013).

⁷For the reader unfamiliar with boxplots: the thick black line indicates the median value, the length of the box gives the interquartile range, that is, all scores between the 25th and 75th percentile, circles indicate outliers and stars represent extreme outliers. It has been argued that boxplots are more informative than the commonly used barplots (Larson-Hall & Herrington 2010).

⁸We cannot of course rule out the possibility that some children, especially the older children, most of whom have received L2 Dutch instruction, have rote-learned the gender of a given noun. For present purposes, however, the most important aspect of this analysis is its consequences for the reanalysis of the adjective data.

⁹It is of course possible that the lack of a significant effect of group in the second analysis is due to reduced statistical power resulting from fewer children being included in this analysis; data from more proficient children would be needed to completely rule out this explanation.

¹⁰ Meisel (2009) also speculates that there may be a second critical period ending at age 6; using this age rather than age 4 as the basis for categorizing the children in the present study does not significantly alter the findings.

¹¹Interestingly, as Blom et al. (2008:316, fn. 12) note, phonological properties of the L1 for the adults in their study, that is, Moroccan Arabic, where unstressed word-final vowels are sometimes elided, may also contribute to the production of bare adjectives by these learners.

¹²I thank an anonymous reviewer for pointing this out to me.

¹³It would also be necessary to control for other factors which may cause children to produce bare forms (cf. fn. 11). It is furthermore assumed that the gender system in the L1 is already established when the L2 is acquired (because otherwise, there will be nothing to transfer, as is the case for L1 English children). Likewise, it would be wise to select a language as L1 where gender-marking on adjectives is not spelled out as a null morpheme.

¹⁴An anonymous reviewer points out that there is considerable variation in terms of age at testing and that this may affect children's performance between and within groups. Although this variation is indeed present in the sample—and to a certain extent it is inevitable—age of testing (in months) does not correlate significantly with any of the children's scores.