Adults often encounter difficulty perceiving and processing sounds of a second language (L2). In order to acquire word-meaning mappings, learners need to determine what the language-relevant phonological contrasts are in the language. In this study, we examined the influence of phonology on non-native word learning, determining whether the language-relevant phonological contrasts could be acquired by abstracting over multiple experiences, and whether awareness of these contrasts could be related to learning. We trained English- and Mandarin-native speakers with pseudowords via a cross-situational statistical learning task (CSL). Learners were able to acquire the phonological contrasts across multiple situations, but similar-sounding words (i.e., minimal pairs) were harder to acquire, and words that contrast in a non-native suprasegmental feature (i.e., Mandarin lexical tone) were even harder for English-speakers, even with extended exposure. Furthermore, awareness of the non-native phonology was not found to relate to learning.

Children typically produce high-frequency phonotactic sequences, such as the /st/ in “toaster,” more accurately than the lower frequency /mk/ in “tomcat.” This high-frequency advantage can be simulated experimentally with a statistical learning paradigm, and when 4-year-old children are familiarized with many examples of a sequence like /mk/, they generally produce it more accurately than if they are exposed to just a few examples. Here, we sought to expand our understanding of the high-frequency advantage, but surprisingly, we instead uncovered an exception. Twenty-nine children between 4 and 5 years of age completed a phonotactic statistical learning experiment, but they also completed a separate experiment focused on statistical learning of prosodic contours. The order of the experiments was randomized, with the phonotactic statistical learning experiment occurring first for half of the children. For the children who completed the phonotactic learning experiment first, the results were consistent with previous research and a high-frequency advantage. However, children who completed the phonotactic learning experiment second produced low-frequency sequences more accurately than high-frequency sequences. There is little precedent for the latter effect, but studies of multistream statistical learning may provide some context for unpacking and extending the result.

We compare two frameworks for the segmentation of words in child-directed speech, PHOCUS and MULTICUE. PHOCUS is driven by lexical recognition, whereas MULTICUE combines sub-lexical properties to make boundary decisions, representing differing views of speech processing. We replicate these frameworks, perform novel benchmarking and confirm that both achieve competitive results. We develop a new framework for segmentation, the DYnamic Programming MULTIple-cue framework (DYMULTI), which combines the strengths of PHOCUS and MULTICUE by considering both sub-lexical and lexical cues when making boundary decisions. DYMULTI achieves state-of-the-art results and outperforms PHOCUS and MULTICUE on 15 of 26 languages in a cross-lingual experiment. As a model built on psycholinguistic principles, this validates DYMULTI as a robust model for speech segmentation and a contribution to the understanding of language acquisition.

How much information do language users need to differentiate potentially absolute synonyms into near-synonyms? How consistent must the information be? We present two simple experiments designed to investigate this. After exposure to two novel verbs, participants generalized them to positive or negative contexts. In Experiment 1, there was a tendency across conditions for the verbs to become differentiated by context, even following inconsistent, random, or neutral information about context during exposure. While a subset of participants matched input probabilities, a high proportion did not. As a consequence, the overall pattern was of growth in differentiation that did not closely track input distributions. Rather, there were two main patterns: When each verb had been presented consistently in a positive or negative context, participants overwhelmingly specialized both verbs in their output. When this was not the case, the verbs tended to become partially differentiated, with one becoming specialized and the other remaining less specialized. Experiment 2 replicated and expanded on Experiment 1 with the addition of a pragmatic judgment task and neutral contexts at test. Its results were consistent with Experiment 1 in supporting the conclusion that quality of input may be more important than quantity in the differentiation of synonyms.

We examined how noun frequency and the typicality of surrounding linguistic context contribute to children’s real-time comprehension. Monolingual English-learning toddlers viewed pairs of pictures while hearing sentences with typical or atypical sentence frames (Look at the… vs. Examine the…), followed by nouns that were higher- or lower-frequency labels for a referent (horse vs. pony). Toddlers showed no significant differences in comprehension of nouns in typical and atypical sentence frames. However, they were less accurate in recognizing lower-frequency nouns, particularly among toddlers with smaller vocabularies. We conclude that toddlers can recognize nouns in diverse sentence contexts, but their representations develop gradually.

Statistical learning (SL) is assumed to lead to long-term memory representations. However, the way that those representations influence future learning remains largely unknown. We studied how children’s existing distributional linguistic knowledge influences their subsequent SL on a serial recall task, in which 49 German-speaking seven- to nine-year-old children repeated a series of six-syllable sequences. These contained either (i) bisyllabic words based on frequently occurring German syllable transitions (naturalistic sequences), (ii) bisyllabic words created from unattested syllable transitions (non-naturalistic sequences), or (iii) random syllable combinations (unstructured foils). Children demonstrated learning from naturalistic sequences from the beginning of the experiment, indicating that their implicit memory traces derived from their input language informed learning from the very early stages onward. Exploratory analyses indicated that children with a higher language proficiency were more accurate in repeating the sequences and improved most throughout the study compared to children with lower proficiency.

Statistical learning—the skill to pick up probability-based regularities of the environment—plays a crucial role in adapting to the environment and learning perceptual, motor, and language skills in healthy and clinical populations. Here, we developed a new method to measure statistical learning without any manual responses. We used the Alternating Serial Reaction Time (ASRT) task, adapted to eye-tracker, which, besides measuring reaction times (RTs), enabled us to track learning-dependent anticipatory eye movements. We found robust, interference-resistant learning on RT; moreover, learning-dependent anticipatory eye movements were even more sensitive measures of statistical learning on this task. Our method provides a way to apply the widely used ASRT task to operationalize statistical learning in clinical populations where the use of manual tasks is hindered, such as in Parkinson’s disease. Furthermore, it also enables future basic research to use a more sensitive version of this task to measure predictive processing.

We introduce hierarchically regularized entropy balancing as an extension to entropy balancing, a reweighting method that adjusts weights for control group units to achieve covariate balance in observational studies with binary treatments. Our proposed extension expands the feature space by including higher-order terms (such as squared and cubic terms and interactions) of covariates and then achieves approximate balance on the expanded features using ridge penalties with a hierarchical structure. Compared with entropy balancing, this extension relaxes model dependency and improves the robustness of causal estimates while avoiding optimization failure or highly concentrated weights. It prevents specification searches by minimizing user discretion in selecting features to balance on and is also computationally more efficient than kernel balancing, a kernel-based covariate balancing method. We demonstrate its performance through simulations and an empirical example. We develop an open-source R package, hbal, to facilitate implementation.

Critical cascades are found in many self-organizing systems. Here, we examine critical cascades as a design paradigm for logic and learning under the linear threshold model (LTM), and simple biologically inspired variants of it as sources of computational power, learning efficiency, and robustness. First, we show that the LTM can compute logic, and with a small modification, universal Boolean logic, examining its stability and cascade frequency. We then frame it formally as a binary classifier and remark on implications for accuracy. Second, we examine the LTM as a statistical learning model, studying benefits of spatial constraints and criticality to efficiency. We also discuss implications for robustness in information encoding. Our experiments show that spatial constraints can greatly increase efficiency. Theoretical investigation and initial experimental results also indicate that criticality can result in a sudden increase in accuracy.

Several studies have signaled grammatical difficulties in individuals with developmental dyslexia. These difficulties may stem from a phonological deficit, but may alternatively be explained through a domain-general deficit in statistical learning. This study investigates grammar in children with and without dyslexia, and whether phonological memory and/or statistical learning ability contribute to individual differences in grammatical performance. We administered the CELF “word structure” and “recalling sentences” subtests and measures of phonological memory (digit span, nonword repetition) and statistical learning (serial reaction time, nonadjacent dependency learning) among 8- to 11-year-old children with and without dyslexia (N = 50 per group). Consistent with previous findings, our results show subtle difficulties in grammar, as children with dyslexia achieved lower scores on the CELF (word structure: p = .0027, recalling sentences: p = .053). While the two phonological memory measures were found to contribute to individual differences in grammatical performance, no evidence for a relationship with statistical learning was found. An error analysis revealed errors in irregular morphology (e.g., plural and past tense), suggesting problems with lexical retrieval. These findings are discussed in light of theoretical accounts of the underlying deficit in dyslexia.

Personalised nutrition (PN) is an emerging field that bears great promise. Several definitions of PN have been proposed and different modelling approaches have been used to claim PN effects. We tentatively propose to group these approaches into two categories, which we term outcome-based and population reference approaches, respectively. Understanding the fundamental differences between these two types of modelling approaches may allow a more realistic appreciation of what to expect from PN interventions presently and may be helpful for designing and planning future studies investigating PN interventions.

Studies of statistical learning have shaped our understanding of the processes involved in the early stages of language acquisition. Many of these advances were made using experimental paradigms with artificial languages that allow for careful manipulation of the statistical regularities in the input. This article summarizes how these paradigms have begun to inform bilingualism research. We focus on two complementary goals that have emerged from studies of statistical learning in bilinguals. The first is to identify whether bilinguals differ from monolinguals in how they track distributional regularities. The second is determining how learners are capable of tracking multiple inputs, which arguably is an important facet of becoming proficient in more than one language.

We select three word segmentation models with psycholinguistic foundations – transitional probabilities, the diphone-based segmenter, and PUDDLE – which track phoneme co-occurrence and positional frequencies in input strings, and in the case of PUDDLE build lexical and diphone inventories. The models are evaluated on caregiver utterances in 132 CHILDES corpora representing 28 languages and 11.9 m words. PUDDLE shows the best performance overall, albeit with wide cross-linguistic variation. We explore the reasons for this variation, fitting regression models to performance scores with linguistic properties which capture lexico-phonological characteristics of the input: word length, utterance length, diversity in the lexicon, the frequency of one-word utterances, the regularity of phoneme patterns at word boundaries, and the distribution of diphones in each language. These properties together explain four-tenths of the observed variation in segmentation performance, a strong outcome and a solid foundation for studying further variables which make the segmentation task difficult.