What predicts successful literacy acquisition in a second language?

Most modern societies today include as an integral part of their educational curriculum the learning of a second or foreign language, whether because they are multilingual societies, or simply because globalization emphasizes the need for a Lingua Franca. Beginning with Carroll’s seminal work (e.g., Carroll & Sapon, 1959), studies tracking the process of L2 acquisition have suggested that individuals substantially differ in their capacity to learn another language (e.g., Sparks et al., 2009; Lundberg, 2002). Whereas some seem to assimilate a new linguistic system with relative ease, for others this task involves significant difficulty. What determines this variability? What predicts successful and fast acquisition of an additional language?

The question of the source of individual differences is seldom straightforward (see Dewaele, 2009). Many factors can be offered to explain success or failure in any given task, and pinpointing a single mental capacity to account for the population variance in performance is difficult in most cases. For example, research on L2 spoken language and literacy acquisition has suggested that success is moderated by a variety of individual differences in factors such as context of language exposure, phonological abilities, working memory, executive functions, motivation, or personality traits (e.g., Kroll & Linck, 2007; Bialystok et al., 2008; DeKeyser, 2000; Dewaele, 2009). In the present context, however, we were concerned with adjudicating between two qualitatively different theoretical approaches to L2 acquisition in young adults: the first considers it to be a domain-specific capacity belonging to the modular faculty of language, while the second considers it to reflect a general and non-linguistic cognitive capacity for statistical learning.

The view that second language acquisition is a subset of the language faculty of cognitive architecture argues that L2 learning is mediated by the same linguistic capacities as L1 learning, and by similar neurocircuitry. Indeed, it has been shown that L1 linguistic capacities such as phonological awareness, syntactic abilities, orthographic knowledge or vocabulary, generally predict success in L2 acquisition (e.g., Ganschow et al., 1998). Similarly, the level of literacy achieved in one’s native language has been shown to determine, to a large extent, the acquisition of literacy in a foreign language (e.g., Koda, 2007). Moreover, L2 neurocircuitry has been shown to converge with L1 neurocircuitry as L2 proficiency increases (e.g., Abutalebi et al., 2001). According to this approach, mastering literacy in a second language draws upon specific capacities that are specific to language, and therefore linguistic measures are the best predictors of success in this task.

The theoretical approach underlying the claim that second language learning mainly reflects a general capacity for statistical learning considers language learning to be primarily a process of picking up and implicitly assimilating the statistical properties of a linguistic environment. By this view, language is regarded as yet another example of a well-structured environment, and learning it draws upon the general cognitive ability to perceive and identify systematic structures and correlations. Thus, words in each language, whether morphologically simple or complex, are characterized by statistical correlations and transitional probabilities that constrain and determine their internal structure, and by correlations between their sublinguistic components and semantic features (see Frost, 2012a, for a discussion). Mastering a lexicon, therefore, involves the implicit learning of these correlations. More specifically for literacy acquisition, each language presents to the reader a writing system that is characterized by a set of correlations that determine the possible co-occurrences of letter sequences, which eventually result in establishing orthographic representations. Moreover, each writing system is characterized by a different level of consistency (i.e., high or low correlations) in the mapping of graphemes to phonemes, and these consistent mappings eventually result in mapping orthographic representations to phonological ones. Morphological structure is also reflected by systematic correlations, where repeated letter clusters consistently convey features of semantic meaning (e.g., Plaut & Gonnerman, 2000).

This approach suggests that each language implicates a differential tuning to statistical structure, given its idiosyncratic linguistic characteristics. By this view, native speakers who are proficient readers implicitly develop differential sensitivities to the statistical properties of their own language, and while learning to read in L2, speakers (and readers) acquire a new lexical system, mainly by picking up and assimilating a new set of statistical regularities. Hence, the fundamental cognitive faculty of implicit correlation-learning which underlies any form of learning plays a primary role in second language literacy acquisition. Note that two main factors may impede this process. First, the degree of similarity or dissimilarity between the statistical properties of the L1 and the L2 (e.g., Bialystok et al., 2005). Second, like any cognitive ability, there are individual differences in the sensitivity to the correlations in the environment. The question at hand is, therefore, whether individual differences in statistical learning predict individual differences in second language literacy acquisition. This is the focus of the present paper.

To address this question, we examined the process of assimilating the writing system of Hebrew by native English speakers. Hebrew and English were chosen for this study as an L1-L2 pair because they are contrasting languages characterized by qualitatively different statistical properties. These concern the structure of base words, given how morphological complexity is determined, and consequently, the characteristics of the writing systems. English -- an Indo-European language with a concatenated morphology -- imposes few constraints on the internal structure of base words, and in general, the predictive value of a given phoneme regarding the identity of the subsequent one is relatively low. For example, word onsets can consist of any consonant or any vowel, and since the permissible syllables are numerous, in principle, phonemes could be located in any position within the spoken word at more or less equal probability. Prefixes and suffixes are then linearly appended to the base word and are characterized by high distributional properties (Frost, 2012a). In Hebrew, a Semitic language, words -- whether spoken or printed -- are structured very differently. Semitic words are normally composed by intertwining a tri-consonantal root morpheme, which carries the core-meaning of the word, with word-pattern morphemes -- abstract phonological structures, consisting of vowels, or of vowels and consonants, in which there are “open slots” for the root’s consonants to fit into. Semitic words, therefore, have a recognizable, predictable, and well-defined internal structure. Word patterns can begin with a very restricted number of consonants, and these determine a set of transitional probabilities regarding the order and identity of subsequent consonants and vowels. Because the language is root-based, and the roots convey the core meaning of words, these need to be easily extracted and recognized by the speaker. The only clue regarding their location within the world is a well-defined phonological word structure, which on the one hand constrains dramatically the permissible phonological word-forms in the language, but on the other hand, allows the root consonants to stand out. Given the bi-morphemic structure of Hebrew words, the writing system of Hebrew consists of letters that mostly represent consonants (thereby root information), while most of the vowels can optionally be superimposed on the consonants as diacritical marks (“points”). These points, however, are omitted from most adult reading material (Frost et al., 1987; Bentin & Frost, 1987). Since the structure of spoken words is highly constrained by the permissible Semitic word patterns, readers can converge on the correct vowel configuration quite effortlessly (Frost, 1995; 2006).

In the last decade, the reading of Hebrew has been extensively investigated by Frost and his colleagues in a variety of experimental paradigms (Feldman, Frost & Pnini, 1995; Frost et al., 1997, 2000, 2005; Deutsch et al., 1998, 2000, 2003, 2005; Velan & Frost 2007; 2011). One consistent finding in all of these studies is that root primes facilitate both lexical decision and naming of target words that are derived from these roots. Moreover, proficient Hebrew readers have been shown to be tuned to the statistical properties that characterize the Semitic structure of words, so that the extraction of the embedded root consonants is fast and automatic (e.g., Frost et al., 1997; Deutsch et al, 2003), and the retrieval of vowel information is governed by the knowledge of word patterns (Frost, 2006). In other words, the sensitivity of Hebrew readers to the skewed frequency distribution of phonemes and letters in their language determines their processing of printed information; consequently, the typical reading marker of reading Hebrew -- sensitivity to the location of root consonants -- emerges (Frost, 2012; Velan, Deutsch, & Frost, in press).

In the present study we tracked the process of assimilating the linguistic properties of Hebrew by native speakers of English during their first year of learning Hebrew as a second language. We focused on reading as an output measure, because most research on processing morphological information in Hebrew has been carried out in the visual modality, and the markers of learning Hebrew have been mostly documented in the domain of visual word recognition. More importantly, as Hebrew print contains mainly consonantal information whereas vowels, for the most part, are not represented in print, reading Hebrew fluently cannot be carried out by simply applying grapheme-phoneme conversion rules, but requires a deep understanding of the language (Frost, 2006). Reading proficiency in Hebrew was assessed, therefore, in three tasks. The first monitored speed of decoding of pointed nonwords and reflected the assimilation of the characteristics of the Hebrew writing system. The second monitored accuracy in naming unpointed words, reflecting the implicit learning of Hebrew phonological word patterns; that is, the transitional probabilities of permissible phoneme sequences in Hebrew. The third -- cross-modal morphological priming -- directly tapped the main marker of reading in Hebrew: the assimilation of the morphological root-based composition of words. Since our sample of English speakers, not surprisingly, presented a varied distribution of relative success in learning Hebrew, our aim was to examine whether this success (or non-success) in assimilating the writing system and the Semitic structure of words could be predicted by general statistical learning abilities that are non-linguistic in nature.

We therefore employed a visual statistical learning (VSL) task in which 24 relatively complex visual shapes (taken from Turke-Brown et al., 2005; See also Kim et al., 2009; Glickson & Cohen, 2011) were presented in a consecutive stream that lasted about ten minutes. The 24 shapes were organized in eight triplets, which were presented in the 10-minute stream in random order. The transitional probabilities between shapes within triplets was p=1.0. Participants were not told that the stream was constructed of triplets. However, following the familiarization phase, participants were presented with 32 two-alternative-forced-choice (2AFC) trials, each trial consisting of two triplets: 1) a sequence of three shapes that appeared in the stream in the original order (transitional probabilities of p=1.0), and 2) foils made of a sequence of three shapes that were taken from the stream, but did not appear in that order (transitional probabilities of p=0.0). Their success rate in distinguishing the base triplets from the foils reflected their implicit learning of the transitional probabilities of the visual shapes within the stream. The visual shapes used for the experiment and examples of the transitional probabilities of triplets are presented in Figure 1.

We then correlated relative success in the VSL task with relative success in the Hebrew reading tasks. We hypothesized that if a general statistical learning ability underlies learning to read in a new language, which is characterized by a novel set of statistical regularities, then relative success in learning the transitional probabilities of random visual shapes would predict the speed and success of learning to read a new language. Note that the theoretical implications of this possibility cannot be overestimated. It would suggest that a general non-linguistic faculty of statistical learning accounts, at least to some extent, for success in second language acquisition when L1 and L2 differ in their basic statistical properties. Such an outcome would also imply that a simple and short test involving visual shapes could predict the speed of assimilating a new linguistic environment, even before the first foreign word has been learnt.