Elsevier

Acta Psychologica

Volume 157, May 2015, Pages 195-199
Acta Psychologica

Is an attention-based associative account of adjacent and nonadjacent dependency learning valid?

https://doi.org/10.1016/j.actpsy.2015.03.002Get rights and content

Highlights

  • Adjacent but not nonadjacent dependencies were learned under incidental instructions.

  • Participants were as sensitive to spacing violations as to order violations.

  • The same pattern was observed for sequences of digits and syllables.

  • The role of attention in learning the two types of dependencies is discussed.

  • The presence/absence of an intermediate event is a part of the learned association.

Abstract

Pacton and Perruchet (2008) reported that participants who were asked to process adjacent elements located within a sequence of digits learned adjacent dependencies but did not learn nonadjacent dependencies and conversely, participants who were asked to process nonadjacent digits learned nonadjacent dependencies but did not learn adjacent dependencies. In the present study, we showed that when participants were simply asked to read aloud the same sequences of digits, a task demand that did not require the intentional processing of specific elements as in standard statistical learning tasks, only adjacent dependencies were learned. The very same pattern was observed when digits were replaced by syllables. These results show that the perfect symmetry found in Pacton and Perruchet was not due to the fact that the processing of digits is less sensitive to their distance than the processing of syllables, tones, or visual shapes used in most statistical learning tasks. Moreover, the present results, completed with a reanalysis of the data collected in Pacton and Perruchet (2008), demonstrate that participants are highly sensitive to violations involving the spacing between paired elements. Overall, these results are consistent with the Pacton and Perruchet's single-process account of adjacent and nonadjacent dependencies, in which the joint attentional processing of the two events is a necessary and sufficient condition for learning the relation between them, irrespective of their distance. However, this account should be completed to encompass the notion that the presence or absence of an intermediate event is an intrinsic component of the representation of an association.

Introduction

Nonadjacent dependencies refer to the cases where a statistical association exists between two events that are not immediately contiguous in space or time, due to the occurrence of one or several intervening events. This pattern is quite frequent in natural languages (e.g., between auxiliaries and inflectional morphemes, as in “is writing”, irrespective of the verb stem). Nonadjacent dependencies are also present in other domains of high-level knowledge such as music. In Western music, for instance, two structurally important tones are often separated by other, less important tones (the ornaments). If the nonadjacent dependency between the two structurally important tones was not captured by the listener, the musical structure would not be perceived. Likewise, capturing the relationships between distant objects seems essential. As claimed by Turk-Browne, Jungé, and Scholl (2005), “People are constantly bombarded with noise in space and time that needs to be segregated in order to extract a coherent representation of the world, and people rarely encounter a sequence of relevant stimuli without any interruptions” (p. 562).

There is increasing evidence that the learning of nonadjacent dependencies is possible, but only under specific conditions (for a review: Perruchet, Poulin-Charronnat, & Pacton, 2012). Let us refer to a nonadjacent structure as AXC, where A and C stand for the associated events and X stands for a variable event, statistically independent from both A and C. A non-exhaustive list of conditions includes: (1) the high level of variability of the X event (Gómez, 2002, Gómez, 2006, Onnis et al., 2003). (2) The high level of similarity between A and C. Similarity can be assessed on an acoustic dimension. Using musical tone sequences, Creel, Newport, and Aslin (2004) showed that nonadjacent dependencies were not acquired when all elements differed equally one another, whereas learning was successful when A and C were similar in pitch or timbre, and different from X. Likewise, Onnis, Monaghan, Richmond, and Chater (2005) showed that no learning was obtained without some degree of phonological similarity between A and C syllables. (3) The membership of A and C to the same category, itself differing from the category of X. For instance, Newport and Aslin (2004) failed to observe learning with nonadjacent syllables (i.e., A, X, and C were syllables), whereas learning occurred when A and C were consonants and X was a vowel and, conversely, when A and C were vowels and X was a consonant. (4) The introduction of short pauses between the AXC sequences during the familiarization phase (Peña, Bonatti, Nespor, & Mehler, 2002). (5) The occurrence of an earlier training phase during which the to-be-associated pairs have been studied in adjacent conditions. Introducing structural complexity progressively during learning would meet the general learning principle known as the “starting small” effect (Lai & Poletiek, 2011).

Because adjacent dependencies are remarkably easy to learn in a large array of experimental settings, as shown throughout the associative learning literature, the restrictive conditions listed above have led some researchers to claim that learning adjacent and nonadjacent dependencies rely on different processes (e.g., Peña et al., 2002). In contrast with this dual-process view, Pacton and Perruchet (2008) proposed to account for both adjacent and nonadjacent dependencies within an integrated framework, grounded on the role of attention in associative learning (e.g., Hoffmann and Sebald, 2005, Hsiao and Reber, 1998, Jiménez and Méndez, 1999). Several authors have suggested that associative learning is an automatic process that links together all of the components that are present in the attentional focus at a given point (e.g., Frensch and Miner, 1994, Logan and Etherton, 1994, Perruchet and Vinter, 2002, Stadler, 1995, Treisman and Gelade, 1980) and that the joint attention given to a pair of events would be a necessary, but also a sufficient condition for the emergence of associative learning and memory. It is reasonable to postulate that, by default, the mental content composing the attentional focus at a given moment has a high chance of representing events that are close on spatial and/or temporal dimensions in the environment. This would account for the overt precedence to the condition of contiguity in the conventional associative learning literature. However, crucially, the attentional content may also encompass events that are not adjacent in the environment, although only if some specific conditions lead to pay joint attention to those events. In this regard, it is noteworthy that all the five experimental conditions listed above as beneficial for learning nonadjacent dependencies can be viewed as facilitating the attentional processing of the relevant events (i.e., A and C). Thus an attention-based account is seemingly able to retrospectively account for earlier data. More compellingly, this account is also able to generate testable predictions.

A straightforward prediction of the Pacton and Perruchet (2008) integrated framework is that conditions ensuring the very same amount of attentional processing for both adjacent and nonadjacent dependencies would result in a perfect symmetry between the two forms of learning. The authors reported a set of five experiments in which attentional processing was manipulated through the instructions given to the learners. Participants were faced with a set of problems, each comprising a sequence of digits embedding both adjacent and nonadjacent regularities. They were asked to perform an arithmetic task that involved the joint processing of two selected digits. These two digits were adjacent for a first group of participants, and nonadjacent for a second group. A subsequent recognition test explored how well participants from the two groups learned both adjacent and nonadjacent dependencies. The results were clear-cut. Participants who were asked to process adjacent elements learned adjacent dependencies but did not learn nonadjacent dependencies. Much more interestingly, participants who were asked to process nonadjacent elements learned nonadjacent dependencies but did not learn adjacent dependencies. It is noteworthy that the recognition score for nonadjacent dependencies was not significantly lower than the recognition score for adjacent dependencies reached by the participants who focused on adjacent dependencies. Thus, the objective adjacency of the events in the display played no role of its own when the attentional processing of A and C, as prompted by the task demand, was the same for each type of dependency (see also Jahn, 2012, for a replication).

Although Pacton and Perruchet (2008) data provided a strong support for an attention-based, unitary account of adjacent and nonadjacent dependencies, a potential limitation could be that their conclusion was based on experiments relying exclusively on digits as stimuli. Using digits instead of syllables or visual shapes, as commonly exploited in the statistical learning literature, was dictated by the need for creating a task that allows to focus on either adjacent or nonadjacent elements in a meaningful way. An arithmetic task is especially well-suited for this objective because processing nonadjacent digits is what anyone does while performing the most basic arithmetic calculations in real-world settings. However, the downside is that the processing of digits could be quite specific. The perfect symmetry found between the processing of adjacent and nonadjacent dependencies in Pacton and Perruchet could be restricted to this material. In particular, a possibility is that when there is no specific reason to pay joint attention to either adjacent or nonadjacent events, as in most standard tasks of incidental learning, the processing of digits would be insensitive to the contiguity condition which has been shown to be so important throughout the associative learning literature, or at least, less sensitive to the contiguity of the items than the processing of other stimuli such as syllables, tones, or visual shapes. A control condition using standard incidental instructions, such as listening to oral language or tones, or watching visual shapes, was not implemented in Pacton and Perruchet, and as a consequence, there is currently no evidence that the natural processing of digits would exhibit a strong preference for adjacent relationships, as regularly found for other stimuli.

The first objective of the present study was to explore the pattern of performance in a condition using neutral instructions, which was missing in the study of Pacton and Perruchet (2008). Neutral instructions refer here to a task demand that does not require the selective processing of either adjacent or nonadjacent dependencies, as in most incidental learning tasks. Participants were exposed to sequences of digits embedding both adjacent and nonadjacent dependencies, as in Pacton and Perruchet, but they were simply asked to read aloud the items. To examine further whether the processing of digits is endowed with particular properties, half of the participants performed the very same reading task with syllables. Our hypothesis was that under neutral instructions, the usual asymmetry between adjacent and nonadjacent dependency should be found for digits as for syllables, with easier, if not exclusive learning of adjacent dependencies. If this hypothesis turned out to be wrong, then the unified attentional model proposed in Pacton and Perruchet, grounded on the exclusive use of digits as stimuli, should be reconsidered.

Whereas most studies on nonadjacent dependencies deal with the conditions making learning easier, the “what is learned?” issue has not yet been extensively explored. The second objective of the present study was to shed preliminary light on a particular aspect of this issue, namely the status of the intermediate event (X) within the AXC sequence. In a nutshell, the question is: While learning AXC, does the learner simply code that A is followed by C, or is X a mandatory component of the learner's representation? The framework of Pacton and Perruchet (2008) does not deal explicitly with this question, but its emphasis on the role of the attention paid to the target stimuli, A and C, suggests that the intermediate event, X, must receive no, or only a minimal amount of attentional processing during training.

A recognition test including the correct sequence AXC and a distractor like AXD has often been used (e.g., Gómez, 2002). However, such a test is inappropriate to investigate whether X is a mandatory component of the learner's representation. Indeed, participants could express a preference for AXC over AXD simply because A and C, contrary to A and D, have formed the mental content of their attentional focus at a given time during the study phase, without having learned whether an intermediate element is located between A and C. In order to address this issue, distractors must include a spacing violation (e.g., AXC vs. ACX). Pacton and Perruchet (2008) used distractors of this kind along with distractors that include an order violation (e.g., AXC vs. CXA). However, they pooled them together for analyses. Thus, the score used by Pacton and Perruchet could be above chance because participants have coded the order of elements A and C (i.e., scores above chance for items with an order violation) even though they have not learned whether an intermediate element is located between A and C (i.e., scores at chance level for items with a spacing violation).

A huge amount of data collected in various subareas of the associative learning literature, such as classical conditioning, paired-associate paradigms, and serial learning, concur to show that the order of the elements in a pair or a sequence is actively coded and remembered. Although these studies investigated the order between adjacent elements, it seems reasonable to expect that the order is also coded in the case of nonadjacent elements and therefore that forced-choice recognition was above chance for items comprising an order violation in the study of Pacton and Perruchet (2008). Analyzing performance on recognition tests with distractors comprising a spacing violation alone is essential to investigate whether X is a mandatory component of the learner's representation. However, the comparison of the effect of a spacing violation with the better known effect of an order violation may be especially instructive because the stimuli including order violations can serve as a benchmark in evaluating the effect of spacing violations.

In the present study, as in that of Pacton and Perruchet (2008), the distractors used in the recognition tests are composed of two kinds of violations, namely spacing (e.g., AXC vs. ACX) and order (e.g., AXC vs. CXA). We analyzed forced-choice recognition for each kind of violation for both the new collected data and Pacton and Perruchet's data that we reanalyzed. If learning proceeds from the joint attentional focus on A and C, as proposed in the Pacton and Perruchet's framework, the presence or absence of the unattended event X in the test items should be relatively inconsequential. In any case, order violations, which involve the relevant events A and C, should be easier to detect than spacing violations, which involve the unattended event X.

Section snippets

Participants

Forty-eight undergraduate students, 39 females, aged 17–24 years (M = 20 years and 2 months) from the Université Paris Descartes, Paris, France, participated in the experiment in partial fulfillment of a course requirement. All participants were native French speakers. Participants were assigned to one of the four experimental conditions defined below, with n = 12 for each group.

Material

The stimuli were digits for half of the participants, and syllables for the other half, everything else being equal. For

Discussion

Pacton and Perruchet (2008) proposed an attention-based single-process account of adjacent and nonadjacent dependency learning. They showed that adjacent and nonadjacent dependencies were learned to the same extent when the current focus of attention is oriented towards the relevant events by the task demand. However, this account is based on experiments relying on digits as stimuli, and it is possible that the processing of digits differs from the processing of syllables or pictures with

References (22)

  • A.T. Hsiao et al.

    The role of attention in implicit sequence learning: Exploring the limits of the cognitive unconscious

  • Cited by (13)

    • Gradual development of non-adjacent dependency learning during early childhood

      2021, Developmental Cognitive Neuroscience
      Citation Excerpt :

      In line with the proposal of an age-related decrease, German adults have been shown to struggle to learn NADs from Italian sentences through passive listening (Friederici et al., 2013), but to successfully learn the NADs under active conditions, that is, in the presence of a task (Mueller et al., 2009; see also Pacton et al., 2015; Pacton and Perruchet, 2008). Together, these findings imply that during development, there is an initial stage of NAD learning, during which young children are able to learn NADs associatively, and a later stage, during which older children and adults need additional cues (e.g. Gómez, 2002; Grama and Wijnen, 2018; Newport and Aslin, 2004; for a review, see Wilson et al., 2018) or a task (Mueller et al., 2012; Pacton et al., 2015; Pacton and Perruchet, 2008) to guide their attention to successfully learn NADs. Our current findings contribute to this notion and, moreover, provide evidence that the transition between these developmental stages of NAD learning occurs in the form of a gradual decrease of associative learning during early childhood.

    • Linguistic and non-linguistic non-adjacent dependency learning in early development

      2020, Developmental Cognitive Neuroscience
      Citation Excerpt :

      For example, perceptual cues were shown to enhance behavioral NAD learning in the non-linguistic domain in adults for sequences of flute and violin timbres (Creel et al., 2004), sequences of musical tones (Endress, 2010), and sequences of Macintosh alert sounds (Gebhart et al., 2009), which is in line with the findings on linguistic NAD learning discussed above. Furthermore, guiding attention to dependent elements (e.g., by re-typing sequences) improves both linguistic and non-linguistic NAD learning (Pacton et al., 2015 - sequences of digits and syllable sequences; Pacton and Perruchet, 2008 - sequences of digits; Uddén et al., 2017 - sequences of letters). At the neural level, processing of learned NADs in both syllable sequences (Friederici et al., 2006) and sequences of letters (Uddén et al., 2017), involves the left Broca’s area in adults, while processing learned NADs in music primarily recruits the right inferior frontal gyrus, specifically the right homologue of Broca’s area (Cheung et al., 2018).

    View all citing articles on Scopus

    This work has been supported by the Institut National de la Santé et de la Recherche Médicale (INSERM, UMR S 894), the Centre National de la Recherche Scientifique (CNRS, UMR 5022), the Université Paris Descartes, Sorbonne Paris Cité, the Université de Bourgogne, and the Institut Universtaire de France.

    View full text