A critical period for second language acquisition: Evidence from 2/3 million English speakers

Abstract

Children learn language more easily than adults, though when and why this ability declines have been obscure for both empirical reasons (underpowered studies) and conceptual reasons (measuring the ultimate attainment of learners who started at different ages cannot by itself reveal changes in underlying learning ability). We address both limitations with a dataset of unprecedented size (669,498 native and non-native English speakers) and a computational model that estimates the trajectory of underlying learning ability by disentangling current age, age at first exposure, and years of experience. This allows us to provide the first direct estimate of how grammar-learning ability changes with age, finding that it is preserved almost to the crux of adulthood (17.4 years old) and then declines steadily. This finding held not only for "difficult" syntactic phenomena but also for "easy" syntactic phenomena that are normally mastered early in acquisition. The results support the existence of a sharply-defined critical period for language acquisition, but the age of offset is much later than previously speculated. The size of the dataset also provides novel insight into several other outstanding questions in language acquisition.

Keywords: Critical period; L2 acquisition; Language acquisition.

Fig. 1

(A–D) Schematic depictions of four theories of how language learning ability might change with age. (E–H) Schematic depictions of four theories of how ultimate attainment might vary with age of first exposure to the language. Note: While the curves hypothesized for learning ability and ultimate attainment resemble one another, there is little systematic relationship between the two; see the main text.

Fig. 2

Simulation results showing how the mapping between hypothetical changes in underlying learning rate (the left graph in each pair) and empirically measured changes in ultimate attainment is many-to-many. These quantitative predictions were derived from the ELSD model, described below, but the basic point is model-independent.

Fig. 3

(A) Current country of residence of participants (excluding participants with multiple residences). (B) Histogram of participants by age of first exposure to English. (C) Native languages of the bilinguals (excluding English). (D) Histogram of participants by current age.

Fig. 4

(A and B) Performance curves for monolinguals and immersion learners (A) and non-immersion learners (B) under 70 years old, smoothed with five-year floating windows. (C and D) Corresponding curves for the best-fitting model. (E) Learning rate for the best-fitting model (black), with examples of the many hypotheses for how learning rate changes with age that were considered in model fitting (grey). For additional detail, see Fig. 7, S3, and S6.

Fig. 5

(A) Histogram of cutoffs used for minimum years of experience to asymptotic learning in previous studies of syntax (Abrahamsson, 2012; Birdsong & Molis, 2001; DeKeyser, 2000; DeKeyser et al., 2010; Flege et al., 1999; Granena & Long, 2013; Jia et al., 2002; Johnson & Newport, 1989, 1991; Mayberry & Lock, 2003; Mayberry, Lock, & Kazmi, 2002; McDonald, 2000; Weber-Fox & Neville, 1996). Papers with multiple studies are included only once, except for McDonald (2000), which used different cutoffs in two different studies. (B) Accuracy for monolinguals (N = 246,497) and simultaneous bilinguals (N = 30,397). Shadowed area represents ± 1 SE. This highlights information also available in Fig. 4A.

Fig. 6

Ultimate attainment for monolinguals, immersion learners, and non-immersion learners, smoothed with a three-year floating window. Shadowed areas represent ± 1 SE. Attainment for monolinguals was significantly higher than that of simultaneous bilinguals (immersion learners with exposure age = 0) (p < .01).

Fig. 7

Accuracy as a function of years of experience, by age of first exposure for immersion learners (A) and non-immersion learners (B). Color scheme is same as in Fig. 4. Red: monolinguals. Orange: AoFE < 11. Green: 10 < AoFE < 21. Blue: AoFE > 20. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Fig. 8

We conducted 2500 simulated experiments of monolingual and immersion learners with each of three sample sizes: N = 69 (equivalent to Johnson & Newport, 1989), N = 275 (larger than the largest prior lab-based study), and N = 11,371 (equivalent to the present study). Three analyses were considered. Left: Correlation between age of first exposure and ultimate attainment prior to 16 years old minus after 16 years old. Middle: First subgroup of subjects to be significantly worse than monolinguals in a t-test (note: the top graph uses the same age bins as Johnson & Newport, 1989). Right: age of first exposure at which performance begins to decline more rapidly, if any. Blue: estimates from Bialystok and Miller (1999), Birdsong and Molis (2001), DeKeyser (2000), DeKeyser et al. (2010), Flege et al. (1999), Johnson and Newport (1989), and Weber-Fox and Neville (1996). While many other papers addressed similar issues, these papers provide the closest analog to Johnson & Newport in that they used a broad-spectrum test of syntax, defined the onset of learning as the age at immigration, and (crucially) report comparable statistics. Red: estimates from current study. Full details available in Supplementary Materials. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)