Several recent empirical findings have reinforced the notion that a basic learning and memory skill—chunking—plays a fundamental role in language processing. Here, we provide evidence that chunking shapes sentence processing at multiple levels of linguistic abstraction, consistent with a recent theoretical proposal by Christiansen and Chater (2016). Individual differences in chunking ability at two different levels is shown to predict on-line sentence processing in separate ways: i) phonological chunking ability, as assessed by a variation on the non-word repetition task, predicts processing of complex sentences featuring phonological overlap; ii) multiword chunking ability, as assessed by a variation on the serial recall task, is shown to predict reading times for sentences featuring long-distance number agreement with locally distracting number-marked nouns. Together, our findings suggest that individual differences in chunking ability shape language processing at multiple levels of abstraction, consistent with the notion of language acquisition as learning to process.

A common assumption in the cognitive sciences is thatartificial and natural language learning rely on sharedmechanisms. However, attempts to bridge the two haveyielded ambiguous results. We suggest that an empiricaldisconnect between the computations employed duringlearning and the methods employed at test may explain thesemixed results. Further, we propose statistically-basedchunking as a potential computational link between artificialand natural language learning. We compare the acquisition ofnon-adjacent dependencies to that of natural languagestructure using two types of tasks: reflection-based 2AFCmeasures, and processing-based recall measures, the latterbeing more computationally analogous to the processes usedduring language acquisition. Our results demonstrate thattask-type significantly influences the correlations observedbetween artificial and natural language acquisition, withreflection-based and processing-based measures correlatingwithin – but not across – task-type. These findings havefundamental implications for artificial-to-natural languagecomparisons, both methodologically and theoretically.

Achieving linguistic proficiency requires identifying words from speech, and discovering the constraints that govern the way those words are used. In a recent study of non-adjacent dependency learning, Frost and Monaghan (2016) demonstrated that learners may perform these tasks together, using similar statistical processes — contrary to prior suggestions. However, in their study, non-adjacent dependencies were marked by phonological cues (plosive- continuant-plosive structure), which may have influenced learning. Here, we test the necessity of these cues by comparing learning across three conditions; fixed phonology, which contains these cues, varied phonology, which omits them, and shapes, which uses visual shape sequences to assess the generality of statistical processing for these tasks. Participants segmented the sequences and generalized the structure in both auditory conditions, but learning was best when phonological cues were present. Learning was around chance on both tasks for the visual shapes group, indicating statistical processing may critically differ across domains

Attempts to connect individual differences in statisticallearning with broader aspects of cognition have receivedconsiderable attention, but have yielded mixed results. Apossible explanation is that statistical learning is typicallytested using the two-alternative forced choice (2AFC) task.As a meta-cognitive task relying on explicit familiarityjudgments, 2AFC may not accurately capture implicitlyformed statistical computations. In this paper, we adapt theclassic serial-recall memory paradigm to implicitly teststatistical learning in a statistically-induced chunking recall(SICR) task. We hypothesized that artificial languageexposure would lead subjects to chunk recurring statisticalpatterns, facilitating recall of words from the input.Experiment 1 demonstrates that SICR offers more fine-grained insights into individual differences in statisticallearning than 2AFC. Experiment 2 shows that SICR hashigher test-retest reliability than that reported for 2AFC. Thus,SICR offers a more sensitive measure of individualdifferences, suggesting that basic chunking abilities mayexplain statistical learning.