The serial reaction time task (SRT)

Support for the procedural deficit view comes, in the main, from impaired performance of language-disordered participants on implicit serial learning tasks, such as the serial reaction time task (Nissen & Bullemer, 1987). In the serial reaction time task participants respond to a stimulus appearing in 1 of 4 locations as fast as possible. Faster response times to trials that follow a covert sequence compared to random trials are taken as evidence of implicit learning (Seger, 1994). Successful implicit learning performance on the serial reaction time task has been shown to recruit the prefrontal

64 cortex, striatum and cerebellum, all structures implicated in the implicit learning circuitry of the brain (Pascual-Leone, 1993; Keele, Ivry, Mayr, Hazeltine, & Heuer, 2003; Laforce & Doyon, 2001).

The original deterministically structured serial reaction time task has been criticized for not fully dissociating implicit and explicit learning (Shanks & Johnstone, 1998; Shanks & St John, 1994). Shanks, Green & Kolodny (1994) demonstrated that unaware subjects performed above chance in post-task “generate” tests that tested awareness of the sequence by asking participants to predict future locations. Therefore, they could not be said to be properly unaware. More complex, probabilistically structured (Schvaneveldt & Gomez, 1998) or alternating versions (Howard & Howard, 1997) that render the sequence indiscernible have since been developed to minimize the risk of explicit learning. The proportion of unaware participants on these tasks is far higher, yet they still show significant implicit learning. For example, participants on a complex probabilistic serial reaction time task (Cleeremans & McClelland, 1991) spanning 60,000 trials over 20 sessions were shown to develop progressive sensitivity to the sequential dependencies within the task up to and including three elements of context, but developed limited reportable knowledge of the sequences. Although they were significantly better able to predict strings that followed the sequential structure of the task than those that did not during a post-task generation test, this effect was small.

Most research using the serial reaction time task to investigate the procedural deficit hypothesis uses extreme group designs of language-disordered and control participants. In these studies, language-disordered children have been shown to perform poorly both on deterministic serial reaction time tasks (developmental language disorder: Conti-Ramsden, Ullman, & Lum, 2015; Gabriel, Maillart, Stefaniak, Lejeune, Demottes, & Meulemans, 2013; Hsu & Bishop, 2014; Lum, Conti- Ramsden, Page, & Ullman, 2012; Lum, Gelgic, & Conti-Ramsden, 2010; Lukacs & Kemeny, 2014; Sengottuvel & Rao, 2013; 2014; Sengottuvel, Rao, & Bishop, 2016; Jiménez-Fernández, Vaquero, Jiménez, & Defior, 2011; dyslexia: Stoodley, Ray, Jack, & Stein, 2008; Vicari et al., 2005; Vicari, Marotta, Menghini, Molinari, & Petrosini,

65 2003) and on more complex alternating task versions (Desmottes, Meulemans, & Maillart, 2016a; 2016b; Hedenius, 2013; Howard, Howard, Japikse, & Eden, 2006). However, other papers do not support these results with null findings on deterministic tasks in children with developmental language disorder (Gabriel, Maillart, Guillaume, Stefaniak, & Meulemans, 2011; Gabriel, Meulemans, Parisse, & Maillart, 2015; Gabriel, Stefaniak, Maillart, Schmitz, & Meulemans, 2012; Lum & Bleses, 2012; Mayor-Dubois, Zesiger, Van der Linden, & Roulet-Perez, 2014); and in children with dyslexia (Menghini et al., 2010; Vakil, Lowe, & Goldfus, 2015; Yang & Hong-Yan, 2011).

In adults with dyslexia the story is similarly inconsistent, with null findings on several deterministic tasks (Rüsseler, Gerth, & Münte, 2006; Kelly, Griffiths, & Frith, 2002; Laasonen et al., 2014), as well as a recent study using an alternating version of the serial reaction time task (Henderson & Warmington, 2017). However, several studies do find poorer implicit learning in dyslexic adults on deterministic tasks (Gabay et al., 2012a; Menghini et al., 2006; 2008; Stoodley, Harrison, & Stein, 2006) and on complex alternating versions (Howard, Howard, Japikse, & Eden, 2006). Only two studies examine procedural learning deficits in adults with developmental language disorder and both report null results (Lee & Tomblin, 2015; Lee, Mueller, & Tomblin, 2016).

Few correlational studies examining serial reaction time learning and language ability exist. Those that do are in children and all report a predominantly null result for the relationship, either between implicit learning and knowledge of past tense morphology or with receptive vocabulary on either the BPVS-II or the PPVT (Lum & Kidd, 2012; Kidd & Kirjavainen, 2011), as well as with reading ability (Waber et al., 2003). The lack of significant findings in large-scale correlational studies raises a red-flag about the legitimacy of significant findings in the extreme group studies, which will be explored further in Chapter 4.

66

3.3.1.1 Noise within the data

There are a number of possible explanations for the inconsistent results in the literature. Mayor-Dubois et al. (2014) found implicit learning impaired in children with developmental language disorder who were comorbid for developmental coordination disorder, but not for children with developmental language disorder alone. This suggests that impairments on the task may not be related to language difficulties, but to motor deficits instead. In line with this suggestion, Gabriel et al. (2012) reported that children with developmental language disorder made more learning-related errors than typically developing children when using a keyboard, but not a touch-screen, suggesting that implicit learning measures may be affected by task- specific response mechanisms.

3.3.1.2 Sequence complexity

It is also possible that differences between groups may be related to sequence complexity. Learning of first order conditional (FOC) sequences can be based on knowledge of a single preceding location, while second order conditional (SOC) sequences are more complex, with learning requiring knowledge of the two preceding sequence locations. Robertson (2007) suggested that any differences in implicit learning would relate purely to the greater computational complexity of SOC sequences, not the type of learning they engender. However, it is possible that FOC sequences are easier to learn explicitly (Curran, 1997) and that SOC sequences, therefore, may provide a purer measure of implicit learning.

In support of this, Kelly, Jahanshahi, & Dirnberger (2004) found Parkinson disease patients predominantly impaired for SOC not FOC sequence learning, concluding that FOC sequence learning is aided by declarative memory mechanisms. At a first glance the use of tasks with SOC sequences supports the procedural deficit hypothesis, with significant group differences found in dyslexic adults (Howard et al., 2006) and in children with developmental language disorder (Gabriel et al., 2013). However, studies that analyse FOC and SOC learning side by side are once again inconsistent. While Du & Kelly (2013) also found dyslexic adults impaired for higher order, not lower order, transitions within the same task, Clark & Lum (2017) found the opposite,

67 with children with developmental language disorder impaired on the FOC version only. Finally, Deroost et al. (2010) found dyslexic children unimpaired on both FOC and SOC versions of the same deterministic serial reaction time task.

3.3.1.3 Consolidation

The extent to which consolidation of implicit learning may explain the difference between language-disordered and control groups has also been examined. Two studies have suggested that the implicit learning impairment in developmental language disorder is confined to consolidation rather than acquisition of implicit learning (Desmottes, Meulemans, & Maillart, 2016a; Desmottes, Maillart, & Meulemans, 2017). Initial procedural learning was intact for both children with developmental language disorder and controls, but the developmental language disorder group were impaired during a second attempt at the same task. Similar results were also reported on an alternating serial reaction time task in dyslexic children (Hedenius et al., 2013). However, Gabay et al. (2012a) found the opposite, with dyslexic adults performing comparably with controls during later learning stages, but showing impaired initial learning, while Henderson & Warmington (2017) found no implicit learning in either group across initial and consolidation sessions of their task. Once again, therefore, the answer is not clear-cut.