JCPSLP Vol 22 No 1 2020

Witteloostuijn, Boersma, Wijnen, and Rispens (2017) reached a similar conclusion about age-related effects in their meta-analysis of artificial grammar learning studies but also raised questions about publication bias favouring the reporting of statistically significant group differences. In line with a view of SL as a property emerging from multiple sub abilities, individuals within the dyslexic population may differ in the efficiency of particular components that underpin SL. This complicates matters substantially. For instance, while Henderson and Warmington (2017) found that English-speaking adults with dyslexia performed similarly to a control group on a serial reaction time task, this was not the case when it came to performance on the Hebb repetition task. The way that artificial grammar learning tasks are administered can affect results in these kinds of comparisons across dyslexic and non-dyslexic individuals (e.g., the study of children by Schiff, Katan, Sasson, and Kahta [2017]; the study of adults by Schiff, Sasson, Star, and Kahta [2017]; and the study of adults by Kahta and Schiff [2019]). Likewise, W. He and Tong (2017) showed that the number of times Chinese children with dyslexia were exposed to regularities affected SL performance as measured by a serial reaction time task. A move away from only examining group comparisons towards inclusion of an individual differences approach may be fruitful in discovering the extent to which SL contributes to variability in reading in dyslexic and non-dyslexic individuals (e.g., Gabay, Thiessen, & Holt, 2015). In a departure from previous studies, Kleij, Groen, Segers, and Verhoeven (2019) examined SL in dyslexic and non-dyslexic individuals longitudinally from grade 5 to grade 6, and in the context of response to intervention for dyslexic children. Their study of 74 Dutch children with dyslexia and 39 typically developing peers showed that although there were no statistically significant group differences in the capacity for sequential SL (as measured by a serial reaction time task) or spatial SL (as measured by a spatial contextual cueing task), sequential SL predicted reading ability in both groups. Moreover, during a literacy intervention, dyslexic children who exhibited faster learning on the sequential SL task tended to show greater improvement in pseudoword reading. A recent study of Chinese by Tong and colleagues (2019) included 35 children with dyslexia and 37 children without dyslexia. A triplet task was used to measure visual SL. Their results, like Kleij et al.’s (2019), revealed a relationship between the capacity for SL and reading ability in Chinese in an analysis that included both groups of children. Unlike Kleij et al. they also found statistically significant group differences with dyslexic children showing poorer SL. In considering these different findings across SL studies – some indicating a link between dyslexia and poorer SL ability while others show no such link – it is worth considering methodological differences across studies with regard to participant-related variables, task-related variables, and language-related variables. The next section discusses studies of individual differences in SL in the general population. Individual differences in SL One of the first studies of whether individual differences in the capacity for SL relate to variability in reading proficiency in the general population was conducted by Arciuli and Simpson (2012) who utilised a visual SL task created earlier by Arciuli and Simpson (2011). The SL task is a version of the classic triplet paradigm originally devised to examine infants’ ability to learn aurally presented sequences of syllables (Saffran et al., 1996). Following adaptations of the triplet task using visually presented stimuli in adult studies (e.g.,

Brady & Oliva, 2008; Fiser & Aslin, 2002; Turk-Browne, Jungé, & Scholl, 2005), Arciuli and Simpson (2011) created a child-friendly task of SL using unfamiliar cartoon-like characters described as aliens. It is important that the characters are unfamiliar and cannot be described by a pre-existing label (i.e., in the way a picture of a chair is inescapably linked with the label chair ) or easily distinguished by a single feature (e.g., a green alien versus a blue alien). This allows us to ascertain learning of novel stimuli. This type of SL task usually has a familiarisation phase, which may include a cover task, followed immediately by a surprise test phase. In the task developed by Arciuli and Simpson (2011) participants are provided with a back- story that explains the purpose of the (cover) task at the beginning of the familiarisation phase. They are told that they will be playing a game where aliens from different planets are queuing to enter a spaceship – their task is to detect when two aliens from the same planet appear together (i.e., one after the other) by pressing a button on the keyboard as quickly as possible. This task is an example of a very common children’s card game played in many parts of the world (https://en.wikipedia.org/wiki/ Snap_(card_game)). Not all SL tasks include a cover task but a compelling back-story conceals the aim of learning while also encouraging children to attend to the visual stimuli during familiarisation. As described in several published studies that have used this task, 12 individual stimuli (i.e., the aliens) are grouped into four base triplets (ABC, DEF, GHI, JKL). Each base triplet is presented 24 times during familiarisation. For 6 of the 24 instances, one alien is presented twice in a row (constituting the cover task which simulates the card game commonly known as “snap”). The repetitions are counterbalanced among the aliens within the triplet so that there are no cues to triplet boundaries (e.g., for triplet ABC there are instances of AABC, ABBC, ABCC). Following Turk-Browne et al. (2005) triplet order is randomised after the following constraints – no repeated triplets (i.e., no instances of ABCABC) and no repeated pairs of triplets (e.g., ABCDEFABCDEF). In the surprise test phase, the learning of the base triplets is assessed via forced choice trials where base triplets are pitted against foil triplets. Aliens in the foil triplets never appeared together in sequence in the familiarisation phase (AEI, DHL, GKC, JBF). It is important to use a substantial number of trials as this impacts task reliability. In the task created by Arciuli and Simpson (2011; 2012) the number of test-trials is 64. For every test trial, participants are asked to identify which of the two triplets had appeared previously during familiarisation by verbal indication (the verbal indication reduces cognitive load on children to remember which button on the keyboard to press). There are no time limits on responding. The dependent variable reflecting learning of triplets is an accuracy measure: percent correct score for the 64 trials. Arciuli and Simpson (2012) used this visual SL task alongside a standardised test of word reading accuracy (WRAT-4; Wilkinson & Robertson, 2006) and reported a modest but significant relationship between individuals’ capacity for SL and reading proficiency in 38 children and in 37 adults. A regression analysis of combined child and adult data showed that SL was a predictor of reading ability after accounting for age and attention (i.e., correct hits during the cover task). Torkildsen et al. (2019) used exactly the same SL task in their study of Norwegian children, with instructions translated from English to Norwegian. The Norwegian version of the TOWRE was used as the

5

JCPSLP Volume 22, Number 1 2020

www.speechpathologyaustralia.org.au