The word segmentation process as revealed by click detection

Assessing the impact of attention fluctuations on statistical learning

Attention fluctuates between optimal and suboptimal states. However, whether these fluctuations affect how we learn visual regularities remains untested. Using web-based real-time triggering, we investigated the impact of sustained attentional state on statistical learning using online and offline measures of learning. In three experiments (N = 450), participants performed a continuous performance task (CPT) with shape stimuli. Unbeknownst to participants, we measured response times (RTs) preceding each trial in real time and inserted distinct shape triplets in the trial stream when RTs indicated that a participant was attentive or inattentive. We measured online statistical learning using changes in RTs to regular triplets relative to random triplets encountered in the same attentional states. We measured offline statistical learning with a target detection task in which participants responded to target shapes selected from the regular triplets and with tasks in which participants explicitly re-created the regular triplets or selected regular shapes from foils. Online learning evidence was greater in high vs. low attentional states when combining data from all three experiments, although this was not evident in any experiment alone. On the other hand, we saw no evidence of impacts of attention fluctuations on measures of statistical learning collected offline, after initial exposure in the CPT. These results suggest that attention fluctuations may impact statistical learning while regularities are being extracted online, but that these effects do not persist to subsequent tests of learning about regularities.

Online measurement of learning temporal statistical structure in categorization tasks

The ability to grasp relevant patterns from a continuous stream of environmental information is called statistical learning. Although the representations that emerge during visual statistical learning (VSL) are well characterized, little is known about how they are formed. We developed a sensitive behavioral design to characterize the VSL trajectory during ongoing task performance. In sequential categorization tasks, we assessed two previously identified VSL markers: priming of the second predictable image in a pair manifested by a reduced reaction time (RT) and greater accuracy, and the anticipatory effect on the first image revealed by a longer RT. First, in Experiment 1A, we used an adapted paradigm and replicated these VSL markers; however, they appeared to be confounded by motor learning. Next, in Experiment 1B, we confirmed the confounding influence of motor learning. To assess VSL without motor learning, in Experiment 2 we (1) simplified the categorization task, (2) raised the number of subjects and image repetitions, and (3) increased the number of single unpaired images. Using linear mixed-effect modeling and estimated marginal means of linear trends, we found that the RT curves differed significantly between predictable paired and control single images. Further, the VSL curve fitted a logarithmic model, suggesting a rapid learning process. These results suggest that our paradigm in Experiment 2 seems to be a viable online tool to monitor the behavioral correlates of unsupervised implicit VSL.

Statistical Learning of Language: A Meta-Analysis Into 25 Years of Research

Statistical learning is a key concept in our understanding of language acquisition. Ample work has highlighted its role in numerous linguistic functions-yet statistical learning is not a unitary construct, and its consistency across different language properties remains unclear. In a meta-analysis of auditory-linguistic statistical learning research spanning the last 25 years, we evaluated how learning varies across different language properties in infants, children, and adults and surveyed the methodological trends in the literature. We found robust learning across stimuli (syllables, words, etc.) in infants, and across stimuli and structures (adjacent dependencies, non-adjacent dependencies, etc.) in adults, with larger effect sizes when multiple cues were present. However, the analysis also showed significant publication bias and revealed a tendency toward using a narrow range of simplified language properties, including in the strength of the transitional probabilities used during training. Bayes factor analyses revealed prevalent data insensitivity of moderators commonly hypothesized to impact learning, such as the amount of exposure and transitional probability strength, which contradict core theoretical assumptions in the field. Methodological factors, such as the tasks used at test, also significantly impacted effect sizes in adults and children, suggesting that choice of task may critically constrain current theories of how statistical learning operates. Collectively, our results suggest that auditory-linguistic statistical learning has the kind of robustness needed to play a foundational role in language acquisition, but that more research is warranted to reveal its full potential.

Assessing the Sensitivity of EEG-Based Frequency-Tagging as a Metric for Statistical Learning

Statistical learning (SL) is hypothesized to play an important role in language development. However, the measures typically used to assess SL, particularly at the level of individual participants, are largely indirect and have low sensitivity. Recently, a neural metric based on frequency-tagging has been proposed as an alternative measure for studying SL. We tested the sensitivity of frequency-tagging measures for studying SL in individual participants in an artificial language paradigm, using non-invasive electroencephalograph (EEG) recordings of neural activity in humans. Importantly, we used carefully constructed controls to address potential acoustic confounds of the frequency-tagging approach, and compared the sensitivity of EEG-based metrics to both explicit and implicit behavioral tests of SL. Group-level results confirm that frequency-tagging can provide a robust indication of SL for an artificial language, above and beyond potential acoustic confounds. However, this metric had very low sensitivity at the level of individual participants, with significant effects found only in 30% of participants. Comparison of the neural metric to previously established behavioral measures for assessing SL showed a significant yet weak correspondence with performance on an implicit task, which was above-chance in 70% of participants, but no correspondence with the more common explicit 2-alternative forced-choice task, where performance did not exceed chance-level. Given the proposed ubiquitous nature of SL, our results highlight some of the operational and methodological challenges of obtaining robust metrics for assessing SL, as well as the potential confounds that should be taken into account when using the frequency-tagging approach in EEG studies.

Statistically Induced Chunking Recall: A Memory-Based Approach to Statistical Learning

The computations involved in statistical learning have long been debated. Here, we build on work suggesting that a basic memory process, chunking, may account for the processing of statistical regularities into larger units. Drawing on methods from the memory literature, we developed a novel paradigm to test statistical learning by leveraging a robust phenomenon observed in serial recall tasks: that short-term memory is fundamentally shaped by long-term distributional learning. In the statistically induced chunking recall (SICR) task, participants are exposed to an artificial language, using a standard statistical learning exposure phase. Afterward, they recall strings of syllables that either follow the statistics of the artificial language or comprise the same syllables presented in a random order. We hypothesized that if individuals had chunked the artificial language into word-like units, then the statistically structured items would be more accurately recalled relative to the random controls. Our results demonstrate that SICR effectively captures learning in both the auditory and visual modalities, with participants displaying significantly improved recall of the statistically structured items, and even recall specific trigram chunks from the input. SICR also exhibits greater test-retest reliability in the auditory modality and sensitivity to individual differences in both modalities than the standard two-alternative forced-choice task. These results thereby provide key empirical support to the chunking account of statistical learning and contribute a valuable new tool to the literature.

Jacques Mehler's early psycholinguistic days in Paris

This article first describes Jacques Mehler's initial efforts to make psycholinguistics and, more generally, the cognitive sciences better known during his first years in Paris. Two lines of research on sentence perception, that we conducted in collaboration with Jacques, are then presented to illustrate his focus. In the Seventies, sentence perception was a central topic in psycholinguistics, with contrasting proposals of syntactic autonomy and interactivity being confronted. A first series of experiments aimed at defining the role of syntax in lexical selection process as revealed by the rapid serial visual presentation (RSVP) of the words in a sentence. The second series, using the phoneme monitoring technique, examined the clause as a processing unit during the auditory perception of sentences. These results confirm the fundamental role played by syntax in language processing.

The role of metacognition in recognition of the content of statistical learning

Despite theoretical debate on the extent to which statistical learning is incidental or modulated by explicit instructions and conscious awareness of the content of statistical learning, no study has ever investigated the metacognition of statistical learning. We used an artificial language-learning paradigm and a segmentation task that required splitting a continuous stream of syllables into discrete recurrent constituents. During this task, statistical learning potentially produces knowledge of discrete constituents as well as about statistical regularities that are embodied in familiarization input. We measured metacognitive sensitivity and efficiency (using hierarchical Bayesian modelling to estimate metacognitive sensitivity and efficiency) to probe the role of conscious awareness in recognition of constituents extracted from the familiarization input and recognition of novel constituents embodying the same statistical regularities as these extracted constituents. Novel constituents are conceptualized to represent recognition of statistical structure rather than recognition of items retrieved from memory as whole constituents. We found that participants are equally sensitive to both types of learning products, yet subject them to varying degrees of conscious processing during the postfamiliarization recognition test. The data point to the contribution of conscious awareness to at least some types of statistical learning content.

Assessing Visual Statistical Learning in Early-School-Aged Children: The Usefulness of an Online Reaction Time Measure

Visual statistical learning (VSL) was traditionally tested through offline two-alternative forced choice (2-AFC) questions. More recently, online reaction time (RT) measures and alternative offline question types have been developed to further investigate learning during exposure and more adequately assess individual differences in adults (Siegelman et al., 2017b, 2018). We assessed the usefulness of these measures for investigating VSL in early-school-aged children. Secondarily, we examined the effect of introducing a cover task, potentially affecting attention, on children’s VSL performance. Fifty-three children (aged 5–8 years) performed a self-paced VSL task containing triplets, in which participants determine the presentation speed and RTs to each stimulus are recorded. Half of the participants performed a cover task, while the other half did not. Online sensitivity to the statistical structure was measured by contrasting RTs to unpredictable versus predictable elements. Subsequently, participants completed 2-AFC (choose correct triplet) and 3-AFC (fill blank to complete triplet) offline questions. RTs were significantly longer for unpredictable than predictable elements, so we conclude that early-school-aged children are sensitive to the statistical structure during exposure, and that the RT task can measure that. We found no evidence as to whether children can perform above chance on offline 2-AFC or 3-AFC questions, or whether the cover task affects children’s VSL performance. These results show the feasibility of using an online RT task when assessing VSL in early-school-aged children. This task therefore seems suitable for future studies that aim to investigate VSL across development or in clinical populations, perhaps together with behavioral tasks.

Opposing Timing Constraints Severely Limit the Use of Pupillometry to Investigate Visual Statistical Learning

Majority of visual statistical learning (VSL) research uses only offline measures, collected after the familiarization phase (i.e., learning) has occurred. Offline measures have revealed a lot about the extent of statistical learning (SL) but less is known about the learning mechanisms that support VSL. Studies have shown that prediction can be a potential learning mechanism for VSL, but it is difficult to examine the role of prediction in VSL using offline measures alone. Pupil diameter is a promising online measure to index prediction in VSL because it can be collected during learning, requires no overt action or task and can be used in a wide-range of populations (e.g., infants and adults). Furthermore, pupil diameter has already been used to investigate processes that are part of prediction such as prediction error and updating. While the properties of pupil diameter have the potentially to powerfully expand studies in VSL, through a series of three experiments, we find that the two are not compatible with each other. Our results revealed that pupil diameter, used to index prediction, is not related to offline measures of learning. We also found that pupil differences that appear to be a result of prediction, are actually a result of where we chose to baseline instead. Ultimately, we conclude that the fast-paced nature of VSL paradigms make it incompatible with the slow nature of pupil change. Therefore, our findings suggest pupillometry should not be used to investigate learning mechanisms in fast-paced VSL tasks.

Towards a theory of individual differences in statistical learning

In recent years, statistical learning (SL) research has seen a growing interest in tracking individual performance in SL tasks, mainly as a predictor of linguistic abilities. We review studies from this line of research and outline three presuppositions underlying the experimental approach they employ: (i) that SL is a unified theoretical construct; (ii) that current SL tasks are interchangeable, and equally valid for assessing SL ability; and (iii) that performance in the standard forced-choice test in the task is a good proxy of SL ability. We argue that these three critical presuppositions are subject to a number of theoretical and empirical issues. First, SL shows patterns of modality- and informational-specificity, suggesting that SL cannot be treated as a unified construct. Second, different SL tasks may tap into separate sub-components of SL that are not necessarily interchangeable. Third, the commonly used forced-choice tests in most SL tasks are subject to inherent limitations and confounds. As a first step, we offer a methodological approach that explicitly spells out a potential set of different SL dimensions, allowing for better transparency in choosing a specific SL task as a predictor of a given linguistic outcome. We then offer possible methodological solutions for better tracking and measuring SL ability. Taken together, these discussions provide a novel theoretical and methodological approach for assessing individual differences in SL, with clear testable predictions.This article is part of the themed issue 'New frontiers for statistical learning in the cognitive sciences'.

Redefining "Learning" in Statistical Learning: What Does an Online Measure Reveal About the Assimilation of Visual Regularities?

From a theoretical perspective, most discussions of statistical learning (SL) have focused on the possible "statistical" properties that are the object of learning. Much less attention has been given to defining what "learning" is in the context of "statistical learning." One major difficulty is that SL research has been monitoring participants' performance in laboratory settings with a strikingly narrow set of tasks, where learning is typically assessed offline, through a set of two-alternative-forced-choice questions, which follow a brief visual or auditory familiarization stream. Is that all there is to characterizing SL abilities? Here we adopt a novel perspective for investigating the processing of regularities in the visual modality. By tracking online performance in a self-paced SL paradigm, we focus on the trajectory of learning. In a set of three experiments we show that this paradigm provides a reliable and valid signature of SL performance, and it offers important insights for understanding how statistical regularities are perceived and assimilated in the visual modality. This demonstrates the promise of integrating different operational measures to our theory of SL.

Parsing Complex Noun Phrases: Effects of Hierarchical Structure and Sentence Position on Memory Load

In this paper, we report two experiments in Spanish designed to find out what kind of processes underlie the online parsing of complex noun phrases (NPs). To that end, we used a 'click detection' paradigm coupled with an oral comprehension task with sentences made up of complex NPs comprising embedded prepositional phrases PPs or coordinate NPs. The critical NPs consisted of words or pseudowords, and were inserted either at subject position (Experiment 1) or at object position (Experiment 2) in the sentence. Results show an opposite pattern of RTs to clicks when the complex NP is located at subject (vs. object) position, with the former case showing heavier processing demands as the parser delves deeper into the complex NP, regardless of the internal constituency of the target NP and its lexical content, and the latter yielding the opposite pattern. These results suggest that structural complexity by itself does not determine an increase in processing costs during sentence parsing, which is only apparent in cases involving deferred operations like subject-verb agreement.

Disentangling perceptual and psycholinguistic factors in syntactic processing: Tone monitoring via ERPs

Franco, Gaillard, Cleeremans, and Destrebecqz (Behavior Research Methods, 47, 1393-1403, 2015), in a study on statistical learning employing the click-detection paradigm, conclude that more needs to be known about how this paradigm interacts with statistical learning and speech perception. Past results with this monitoring technique have pointed to an end-of-clause effect in parsing-a structural effect-but we here show that the issues are a bit more nuanced. Firstly, we report two Experiments (1a and 1b), which show that reaction times (RTs) are affected by two factors: (a) processing load, resulting in a tendency for RTs to decrease across a sentence, and (b) a perceptual effect which adds to this tendency and moreover helps neutralize differences between sentences with slightly different structures. These two factors are then successfully discriminated by registering event-related brain potentials (ERPs) during a monitoring task, with Experiment 2 establishing that the amplitudes of the N1 and P3 components-the first associated with temporal uncertainty, the second with processing load in dual tasks-correlate with RTs. Finally, Experiment 3 behaviorally segregates the two factors by placing the last tone at the end of sentences, activating a wrap-up operation and thereby both disrupting the decreasing tendency and highlighting structural effects. Our overall results suggest that much care needs to be employed in designing click-detection tasks if structural effects are sought, and some of the now-classic data need to be reconsidered.

Assessing segmentation processes by click detection: online measure of statistical learning, or simple interference?

Statistical learning can be used to extract the words from continuous speech. Gómez, Bion, and Mehler (Language and Cognitive Processes, 26, 212-223, 2011) proposed an online measure of statistical learning: They superimposed auditory clicks on a continuous artificial speech stream made up of a random succession of trisyllabic nonwords. Participants were instructed to detect these clicks, which could be located either within or between words. The results showed that, over the length of exposure, reaction times (RTs) increased more for within-word than for between-word clicks. This result has been accounted for by means of statistical learning of the between-word boundaries. However, even though statistical learning occurs without an intention to learn, it nevertheless requires attentional resources. Therefore, this process could be affected by a concurrent task such as click detection. In the present study, we evaluated the extent to which the click detection task indeed reflects successful statistical learning. Our results suggest that the emergence of RT differences between within- and between-word click detection is neither systematic nor related to the successful segmentation of the artificial language. Therefore, instead of being an online measure of learning, the click detection task seems to interfere with the extraction of statistical regularities.

Rapid Serial Auditory Presentation

Abstract. The Rapid Serial Visual Presentation procedure is a method widely used in visual perception research. In this paper we propose an adaptation of this method which can be used with auditory material and enables assessment of statistical learning in speech segmentation. Adult participants were exposed to an artificial speech stream composed of statistically defined trisyllabic nonsense words. They were subsequently instructed to perform a detection task in a Rapid Serial Auditory Presentation (RSAP) stream in which they had to detect a syllable in a short speech stream. Results showed that reaction times varied as a function of the statistical predictability of the syllable: second and third syllables of each word were responded to faster than first syllables. This result suggests that the RSAP procedure provides a reliable and sensitive indirect measure of auditory statistical learning.