Caudate nucleus
3. Age effects and theoretical approaches to morphosyntactic learning in SLA
3.3 L2 proficiency and rate of learning in instructed contexts 1 Classroom-based longitudinal studies
3.3.2 Age-effects in L2 morphosyntactic learning
Given the linguistic focus of the present investigation (word order, case marking, and learning of form-meaning relationships) it is useful to review some of the literature that found age effects in the L2 learning of specific morphosyntactic targets. In one of these studies García Mayo (2003) looked at how age of onset interacted with the accurate grammaticality judgment of three syntactic constructions.
The study compared early and late L1 Basque learners of English (AO = 8/9 and AO = 11/12) after 396 hours (Time 1) and 594 hours of exposure (Time 2) on their attainment in a GJT. The GJT probed three constructions whose ungrammaticality in English has been related to the lack of pro-drop4 in this language (sentences with null subjects, V-S inversion and that-trace effects; Rizzi, 1982).
Overall the author reports an advantage for late starters. In particular late starters were better already at Time 1 on the correct identification of the
ungrammaticality of null-subject sentences and verb-subject inversion, a difference that became statistically significant at Time 2. Longitudinally, there were significant gains for both groups in all three aspects of the pro-drop parameter under
investigation.
Comparing the two focal groups of the BAF project and an adult group, Muñoz (2006b) set out to investigate the relationship between the order and rate of acquisition of a set of morphological functors (morphemes, articles, and irregular forms) studied in previous research on developmental sequences (Krashen, Sferlazza, Feldman, & Fathman, 1976). Overall the study found that the order of acquisition strongly correlated with those proposed by Krashen et al. and by Pica (1983) for instructed contexts and that there was no interaction between age and order of acquisition. Also in this case, however, age of onset did predict accuracy with an advantage for later starters, although for the two younger groups the proficiency gap found at Time 1 (200 hours) had closed by Time 3 (726 hours). By the time of the last test battery, child late starters (adults were not tested at Time 3) were more proficient
4
In generative linguistics a language is described as pro-drop if it allows null pronouns in subject position.
in all functors considered compared to child early starters, except in the correct use of regular past participle endings, where the latter were more accurate.
Looking at a similar selection of morphemes Pfenninger (2011), cited in Pfenninger and Singleton (2017), compared early starters and late starters recruited for her Switzerland-based seven-year longitudinal study. At Time 1 the two groups were tested on morphology production in two writing tasks and on a written GJT. The results showed that the performance of early and late starters was comparable, except for the production and judgment of irregular past forms where late starters had a marked advantage. Pfenninger also observed that, somewhat similarly to the results of Muñoz (2006b), early starters tended to over-regularize past tense forms and
suggested distinct patterns of learning for the two age groups, with the early starters relying more on rule-based patterns and the late starters relying more on memory- based associative learning (Pfenninger and Singleton, 2017, p. 73).
Overall these studies indicate an advantage for late starters compared to early starters in the learning of morphosyntax. However, for early starters, two of the studies reported a better attainment on regular morphemes and a tendency to regularize morphological patterns (see 4.3.2 for further evidence of morpheme regularization in 6-year-old children).
3.4 Pili-Moss (2017)
Pili-Moss (2017) investigated age-related differences in attainment and rate of L2 learning in laboratory conditions and since it served as a preliminary study for the present thesis it is briefly discussed here. Six 8-9 year olds and eight young adults (all L1 English monolinguals) were trained in the miniature language BrocantoJ for six blocks on three consecutive days (one training block in session 1, two training blocks in session 2, and three training blocks in session 3, cf. 6.3.4). Although the vocabulary
was entirely comprised of nonsense words, the structure of the language followed the word order and morphological pattern of Japanese main clauses (cf. Chapter 5). A computer game (Morgan-Short, 2007, Morgan-Short et al., 2010; 2012) provided a meaningful environment for language comprehension and, although participants were aware they were exposed to a new language, instruction on language rules (word order and relationships between NP positions/case markers and thematic roles) was
incidental. In order to make points in the game, in each trial participants had to perform correctly the move that was described by an aural sentence stimulus, so that accuracy in the game was a measure of accuracy in language comprehension. Unlike the instructed longitudinal L2 studies reviewed previously, this study focused on the investigation of rate of learning in the very first hours of exposure.
Beside the overall measure of comprehension, two subsets of the game trials provided respectively a measure of the understanding of the linking between the position of syntactic arguments and their thematic interpretation and a measure of the linking between accusative markers and the interpretation of the related NP as patient. Both children and adults scored significantly above chance on all measures, although the proportion of adult correct responses was significantly higher overall, as well as with respect to the linking trials. Specifically, the significant adult advantage was found in session 1 and 2. By the third day, however, the children appeared to have bridged the accuracy gap both in terms of overall performance and in terms of
accuracy in the processing of form-meaning relationships. In terms of rate of learning the study found that, overall, adult rate significantly increased earlier during practice compared to children (session 2 vs. session 3).
The overall results of the game task, essentially a listening-comprehension task, are broadly compatible with those of longitudinal studies that found that late
learners attained significantly better than children in this linguistic skill (e.g., Cenoz, 2002). However, since children were able to bridge an initial gap with adults in a matter of days, the results also show that in the long run age differences in this linguistic area may level off. Also, significant increases in rate of learning happened in both adults and children, but at different points during exposure. For the set of stimuli measuring form-meaning relationships, it is interesting to note that whilst accuracy in adults increased incrementally, with no significant between-block
differences, accuracy in children significantly increased in spurts between sessions. A possibility is that a between-session variable (e.g., sleep) had a significant role in the consolidation of learning in children compared to adults.
This said, it is important to note that a direct comparison between the present study and previous longitudinal research on age differences requires caution. First of all, it is entirely possible that the patterns of proficiency development recorded over an extended time span may differ from or obscure the fine-grained trajectory of attainment variation recorded over a few days of exposure. Also, in longitudinal rate- of-learning studies the tasks used as tests are designed to assess linguistic knowledge that has been trained and consolidated during some time prior to the test. In Pili-Moss (2017) participants had only had a very short exposure to the novel language prior to the game task, so that the game in itself offered a further opportunity for learning. Consequently, what was measured by the game task was to a large extent attainment during learning (i.e., language learning ability).
Further, whilst previous studies used natural languages, Pili-Moss (2017) trained participants in an artificial language, which, even if meaningful, is a miniature system with limited syntactic and semantic complexity. Finally, in a training study like the one described here, i.e. one with a very short instruction period, age of onset
and age at testing practically coincide. This is not the case in longitudinal studies spanning many years where potential confounds between the two variables inevitably arise.