Research Article: Dissociable neural systems of sequence learning

Date Published: May 21, 2012

Publisher: University of Finance and Management in Warsaw

Author(s): Freja Gheysen, Wim Fias.


Although current theories all point to distinct neural systems for sequence
learning, no consensus has been reached on which factors crucially define this
distinction. Dissociable judgment-linked versus motor-linked and implicit versus
explicit neural systems have been proposed. This paper reviews these two
distinctions, yet concludes that these traditional dichotomies prove
insufficient to account for all data on sequence learning and its neural
organization. Instead, a broader theoretical framework is necessary providing a
more continuous means of dissociating sequence learning systems. We argue that a
more recent theory, dissociating multidimensional versus unidimensional neural
systems, might provide such framework, and we discuss this theory in relation to
more general principles of associative learning and recent imaging findings.

Partial Text

A fundamental characteristic of human cognition is the ability to learn sequence
information and to adapt to the environment based on this newly acquired knowledge,
reflecting the remarkable plasticity of the human brain. Sequence knowledge is
crucial for efficient daily functioning and therefore, omnipresent during life
(Clegg, DiGirolamo, & Keele, 1998).
For instance, in the morning we get dressed, go down the stairs, quickly reach for a
cup, pour some coffee, drive the car and then start working on the computer. Imagine
how inadequate our life would be when these actions, or parts of these actions, were
performed in a different serial order. Frequently, the sequential regularities
composing these actions are acquired through repeated practice and are often
difficult if not impossible to describe, suggesting that sequence knowledge can be
acquired in a procedural and unconscious way (Destrebecqz & Cleeremans, 2001; Stadler & Frensch, 1998). Sequence learning thus provides an
intriguing example of implicit skill learning. Moreover, the apparent ease with
which these skills can be performed is extremely remarkable because most of these
actions actually entail complex sequence structures. Computer skills such as typing
require a complex coordination of visuo-motor components which need to be sequenced
properly in time.

Sequence learning paradigms differ considerably with respect to the dependent
measures that are used to assess sequence knowledge: Judgment-linked and
motor-linked measures have been dissociated (Seger,
1997, 1998).
Judgment-linked tasks measure sequence knowledge based on
participants’ ability to make correct judgments about the stimulus sequences;
a model task is the artificial grammar (AG) learning task (Reber, 1967). In a typical instantiation of this task, series
of letter strings (e.g., VXVS) are presented that are constructed
according to a finite-state grammar. This grammar represents a rule system defining
the serial order in which letters can follow each other. In the learning phase,
participants have to observe and memorize these meaningless letter strings. In a
subsequent test phase, they are asked to judge whether new strings conform to the
grammar or not. Successful, better-than-chance judgment suggests that participants
have acquired sequence knowledge (Knowlton &
Squire, 1996; Reber, 1967; Seger, 1998). On the other hand,
motor-linked tasks assess sequence knowledge via the reaction
time (RT) of motor responses, that is, the extent to which motor responses become
more facilitated to sequenced stimuli compared to random stimuli. The most studied
task that measures sequence learning via RT performance is the serial reaction time
(SRT) task (Nissen & Bullemer, 1987). In
a basic SRT task, a visual stimulus appears at one of the four horizontally aligned
positions on a computer screen. Participants have to react as fast and as accurately
as possible to the location of the stimulus by pressing the spatially corresponding
key. The succession of the stimuli (and hence responses) follows a repeating
sequential pattern. With continued practice, RTs become much faster on trials
following the sequence than on trials violating the sequence. The RT differences
between sequenced and non-sequenced (random) trials suggest that participants have
learned the sequence. The SRT paradigm is an optimal task to study sequence learning
given the relatively simple experimental implementation, the typically fast
acquisition of sequence knowledge and the objective (RT) measurements to assess
sequence learning (Clegg et al., 1998).

For a long time, consciousness has been regarded as a crucial factor dissociating the
neural networks of learning and memory (Shanks &
St. John, 1994; Squire, 1992,
2009; Tulving, 1987). The idea that implicit (unconscious) sequence learning
is neurally independent of explicit (conscious) sequence learning partly comes from
neuropsychological studies on amnesic patients (suffering from a dysfunction of the
medial temporal lobe/hippocampus) and PD patients (suffering from a basal ganglia
dysfunction due to damage of dopaminergic cells in the substantia nigra). A
traditional theory in the field of learning and memory is that the hippocampus and
basal ganglia systems are divided by consciousness, with the hippocampus subserving
explicit learning and the basal ganglia subserving implicit forms of learning (Squire, 2009). Using the SRT task, some studies
indeed demonstrated that amnesic patients with hippocampal lesions retain the
ability to learn sequences despite having no explicit knowledge (e.g., Gagnon, Foster, Turcotte, & Jongenelis,
2004; Nissen & Bullemer, 1987;
Reber & Squire, 1994) whereas
implicit SRT learning has been found to be impaired in PD patients (Jackson et al., 1995; Siegert, Taylor, Weatherall, & Abernethy, 2006; Wilkinson et al., 2009).

Although these previous frameworks have provided useful insights in how neural
networks of sequence learning might be dissociated, the literature overview
indicates that the judgment versus motor linked and the implicit versus explicit
dichotomy prove insufficient. It appears that a broader theoretical framework is
necessary to explain the various findings on sequence learning and its neural
organization. One such framework has been proposed by Keele, Ivry, Mayr, Hazeltine,
and Heuer (2003). Their theory posits that
the human brain supports two broad systems of sequence learning: a multidimensional
and a unidimensional system.

Research on the neural basis of sequence learning has led to the consensus that
sequence learning is not unitary in nature but is mediated by distinct neural
systems. However, how exactly these learning systems are divided remains unclear. A
distinction has been proposed between sequence learning assessed through
judgment-linked (e.g., artificial grammar) and motor-linked (e.g., SRT) tasks (Seger, 1997, 1998). However, neuropsychological and neuroimaging findings have not
been straightforward, suggesting that other factors must be taken into account to
fully understand the operating principles characterizing the distinct sequence
learning systems. Another classical distinction, which has been proposed for more
than two decades, is between implicit and explicit learning processes (Shanks & St.John, 1994; Tulving, 1987). Neuroscientists have linked
this distinction frequently to basal ganglia versus hippocampal learning systems
(Squire, 2009). In the field of sequence
learning, the same distinction has been made; yet, results from neuroscience
research have not been univocal and accumulating data show that this
implicit-explicit distinction proves insufficient.




Leave a Reply

Your email address will not be published.