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

Prosody enhances learning of statistical dependencies from continuous speech streams in adults

Foreign languages sound like seamless streams of speech sounds without pauses between words and phrases. This makes it challenging for the listener to discover the underlying structure of a new language. However, all spoken languages have a melody, and changes in pitch, syllable duration and stress can provide prosodic cues about word and phrase boundaries. It is currently underspecified how adults use prosodic cues to crack the structure of a new language. Here, we investigated how pitch patterns affect the ability to learn adjacent and nonadjacent statistical dependencies from novel, artificial speech streams. In a series of eight online experiments along two studies, we presented native Finnish speakers with short, two-minute speech streams with a hidden probabilistic structure that did or did not include prosodic pitch patterns. We measured learning outcomes using a forced choice recognition task along with confidence ratings. In Study 1, we found that learning adjacent dependencies was boosted with familiar-to-listener (i.e., typical for Finnish language) prosodic pitch patterns but not with unfamiliar-to-listener or random prosodic pitch patterns. In Study 2, we found that more complex nonadjacent dependencies were only learned with the presence of familiar-to-listener prosodic patterns. Intriguingly, prosodic patterns also enabled concurrent learning of multiple adjacent and nonadjacent dependencies in speech. Moreover, they enhanced participants' confidence in remembering adjacent, but not nonadjacent, dependencies. Together, the results suggest that adults use language-background-dependent prosodic patterns to acquire novel linguistic knowledge from speech streams in a fast and efficient manner. The findings support the idea that prosody has an important role in language learning, making the underlying statistical structure of spoken languages more accessible and learnable for listeners.

Parental Input and Its Relationship With Language Outcomes in Children With (Suspected) Developmental Language Disorder: A Systematic Review

PURPOSE

It is widely acknowledged that parental input plays an important role in typical language development. Less is known about the input provided to children with (suspected) developmental language disorder (DLD) or those at risk for DLD. These children may not benefit from parental input in the same way as their typically developing peers, and different aspects of parental input may be more important for them. The present systematic literature review aimed to uncover (a) how parental input has been assessed in the context of DLD, (b) which aspects of parental input have been studied, and (c) what the reported associations are with language outcomes in children with DLD (age 0-6 years).

METHOD

A systematic literature review of 67 articles was conducted following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines. Parental input was coded based on Rowe and Snow's (2020) categorization of input, divided into interactive, linguistic, and conceptual aspects of parental input.

RESULTS

The majority of studies used audio or video recordings of naturalistic parent-child interaction during play. Aspects within the interactive dimension (parental responsiveness and turn-taking) were most often assessed, followed by the linguistic and conceptual dimensions. Thirty-eight studies investigated the relationship between parental input and child language outcomes. Findings show that aspects within the interactive dimension are positively associated with language outcomes in children with DLD. Additionally, the sheer quantity of language input was not associated with language outcomes, although replication of this observation is necessary.

CONCLUSIONS

The results of this review highlight the importance of parents' responsive behaviors as well as the child's involvement in parent-child interaction. The findings also signal gaps in the existing literature and provide associated directions for future research on effects of parental input in children with DLD.

Memory representations are flexibly adapted to orthographic systems: A comparison of English and Hebrew

Across languages, speech unfolds in the same temporal order, constrained by the forward flow of time. But the way phonology is spatially mapped onto orthography is language-specific, ranging from left-to-right, right-to-left, and top-to-bottom, among others. While the direction of writing systems influences how known words are visually processed, it is unclear whether it influences learning and memory for novel orthographic regularities. The present study tested English and Hebrew speakers on an orthographic word-referent mapping task in their native orthographies (written left-to-right and right-to-left, respectively), where the onsets and offsets of words were equally informative cues to word identity. While all individuals learned orthographic word-referent mappings significantly above chance, the parts of the word that were most strongly represented varied. English monolinguals false alarmed most to competing foils that began with the same bigram as the target, representing word onsets most strongly. However, Hebrew bilinguals trained on their native orthography showed no difference between false alarm rates to onset and offset competitors, representing the beginning and ends of words equally strongly. Importantly, Hebrew bilinguals tested on English words displayed a more English-like false alarm pattern (although not a full switch), suggesting that memory biases adapt to the opposite directionality of encountered text while retaining traces of native language biases. These findings demonstrate that experience with different writing systems influences how individuals represent novel orthographic words, starting in the earliest stages of learning.

Sensitive periods in language development: Do children outperform adults on auditory word-form segmentation?

Children are more successful language learners than adults, yet the nature and cause of this phenomenon are still not well understood. Auditory statistical learning from speech has been a prominent focus of research in the field of language development because it is regarded as a fundamental learning mechanism underlying word segmentation in early language acquisition. However, a handful of studies that investigated developmental trajectories for auditory statistical learning found no clear child advantages. The degree to which the learning task measures explicit rather than implicit mechanisms might obscure a potential advantage for younger learners, as suggested by recent findings. In the present study, we compared children aged 7-12 years and young adults on an adapted version of the task that disentangles explicit and implicit contributions to learning. They were exposed to a continuous stream of speech sounds comprising four repeating trisyllabic pseudowords. Learning of the hidden words was tested (a) online through a target-detection task and (b) offline via a forced-choice word recognition test that included a memory judgement procedure. Both measures revealed comparable learning abilities. However, children's performance on the recognition task showed evidence for both explicit and implicit word knowledge while adults appeared primarily sensitive to explicit memory. Since implicit memory is more stable in time than explicit memory, we suggest that future work should focus more on developmental differences in the nature of the memory that is formed, rather than the strength of learning, when trying to understand child advantages in language acquisition.

Statistical learning ability at 17 months relates to early reading skills via oral language

Statistical learning ability has been found to relate to children's reading skills. Yet, statistical learning is also known to be vital for developing oral language skills, and oral language and reading skills relate strongly. These connections raise the question of whether statistical learning ability affects reading via oral language or directly. Statistical learning is multifaceted, and so different aspects of statistical learning might influence oral language and reading skills distinctly. In a longitudinal study, we determined how two aspects of statistical learning from an artificial language tested on 70 17-month-old infants-segmenting sequences from speech and generalizing the sequence structure-related to oral language skills measured at 54 months and reading skills measured at approximately 75 months. Statistical learning segmentation did not relate significantly to oral language or reading, whereas statistical learning generalization related to oral language, but only indirectly related to reading. Our results showed that children's early statistical learning ability was associated with learning to read via the children's oral language skills.

Assessing processing-based measures of implicit statistical learning: Three serial reaction time experiments do not reveal artificial grammar learning

Implicit statistical learning, whereby predictable relationships between stimuli are detected without conscious awareness, is important for language acquisition. However, while this process is putatively implicit, it is often assessed using measures that require explicit reflection and conscious decision making. Here, we conducted three experiments combining an artificial grammar learning paradigm with a serial reaction time (SRT-AGL) task, to measure statistical learning of adjacent and nonadjacent dependencies implicitly, without conscious decision making. Participants viewed an array of six visual stimuli and were presented with a sequence of three auditory (nonsense words, Expt. 1; names of familiar objects, Expt. 2) or visual (abstract shapes, Expt. 3) cues and were asked to click on the corresponding visual stimulus as quickly as possible. In each experiment, the final stimulus in the sequence was predictable based on items earlier in the sequence. Faster responses to this predictable final stimulus compared to unpredictable stimuli would provide evidence of implicit statistical learning, without requiring explicit decision making or conscious reflection. Despite previous positive results (Christiansen et al. 2009 and Misyak et al. 2010) we saw little evidence of implicit statistical learning in any of the experiments, suggesting that in this case, these SRT-AGL tasks were not an effective measure implicit statistical learning.

The relationship between neural phase entrainment and statistical word-learning: A scoping review

Statistical language-learning, the capacity to extract regularities from a continuous speech stream, arguably involves the ability to segment the stream before the discrete constituents can be stored in memory. According to recent accounts, the segmentation process is reflected in the alignment of neural activity to the statistical structure embedded in the input. However, the degree to which it can predict the subsequent leaning outcome is currently unclear. As this is a relatively new avenue of research on statistical learning, a scoping review approach was adopted to identify and explore the current body of evidence on the use of neural phase entrainment as a measure of online neural statistical language-learning and its relation to the learning outcome, as well as the design characteristics of these studies. All included studies (11) observed entrainment to the underlying statistical pattern with exposure to the structured speech stream. A significant association between entrainment and learning outcome was observed in six of the studies. We discuss these findings in light of what neural entrainment in statistical word-learning experiments might represent, and speculate that it might reflect a general auditory processing mechanism, rather than segmentation of the speech stream per se. Lastly, as we find the current selection of studies to provide inconclusive evidence for neural entrainment's role in statistical learning, future research avenues are proposed.

Perceptual Awareness in Human Infants: What is the Evidence?

Perceptual awareness in infants during the first year of life is understudied, despite the philosophical, scientific, and clinical importance of understanding how and when consciousness emerges during human brain development. Although parents are undoubtedly convinced that their infant is conscious, the lack of adequate experimental paradigms to address this question in preverbal infants has been a hindrance to research on this topic. However, recent behavioral and brain imaging studies have shown that infants are engaged in complex learning from an early age and that their brains are more structured than traditionally thought. I will present a rapid overview of these results, which might provide indirect evidence of early perceptual awareness and then describe how a more systematic approach to this question could stand within the framework of global workspace theory, which identifies specific signatures of conscious perception in adults. Relying on these brain signatures as a benchmark for conscious perception, we can deduce that it exists in the second half of the first year, whereas the evidence before the age of 5 months is less solid, mainly because of the paucity of studies. The question of conscious perception before term remains open, with the possibility of short periods of conscious perception, which would facilitate early learning. Advances in brain imaging and growing interest in this subject should enable us to gain a better understanding of this important issue in the years to come.

Constraints on novel word learning in heritage speakers

Introduction

Recent research on word learning has found that adults can rapidly learn novel words by tracking cross-situational statistics, but learning is greatly influenced by the phonological properties of the words and by the native language of the speakers. Mandarin-native speakers could easily pick up novel words with Mandarin tones after a short exposure, but English-native speakers had specific difficulty with the tonal components. It is, however, unclear how much experience with Mandarin is needed to successfully use the tonal cue in word learning. In this study, we explored this question by focusing on the heritage language population, who typically are exposed to the target language at an early age but then develop and switch to another majority language. Specifically, we investigated whether heritage Mandarin speakers residing in an English-speaking region and speaking English as a dominant language would be able to learn novel Mandarin tonal words from statistical tracking. It helps us understand whether early exposure to the target feature is sufficient to promote the use of that feature in word learning later in life.

Methods

We trained 30 heritage Mandarin speakers with Mandarin pseudowords via a cross-situational statistical word learning task (CSWL).

Results and discussion

Heritage Mandarin speakers were able to learn the pseudowords across multiple situations, but similar-sounding words (i.e., minimal pairs) were more difficult to identify, and words that contrast only in lexical tones (i.e., Mandarin lexical tone) were distinguished at chance level throughout learning. We also collected information about the participants' heritage language (HL) experience and usage. We did not observe a relationship between HL experience/usage and performance in tonal word learning, suggesting that HL exposure does not necessarily lead to an advantage in learning the target language.

Evaluating the Relative Importance of Wordhood Cues Using Statistical Learning

Identifying wordlike units in language is typically done by applying a battery of criteria, though how to weight these criteria with respect to one another is currently unknown. We address this question by investigating whether certain criteria are also used as cues for learning an artificial language-if they are, then perhaps they can be relied on more as trustworthy top-down diagnostics. The two criteria for grammatical wordhood that we consider are a unit's free mobility and its internal immutability. These criteria also map to two cognitive mechanisms that could underlie successful statistical learning: learners might orient themselves around the low transitional probabilities at unit boundaries, or they might seek chunks with high internal transitional probabilities. We find that each criterion has its own facilitatory effect, and learning is best where they both align. This supports the battery-of-criteria approach to diagnosing wordhood, and also suggests that the mechanism behind statistical learning may not be a question of either/or; perhaps the two mechanisms do not compete, but mutually reinforce one another.

Statistical learning of syllable sequences as trajectories through a perceptual similarity space

Learning from sequential statistics is a general capacity common across many cognitive domains and species. One form of statistical learning (SL) - learning to segment "words" from continuous streams of speech syllables in which the only segmentation cue is ostensibly the transitional (or conditional) probability from one syllable to the next - has been studied in great detail. Typically, this phenomenon is modeled as the calculation of probabilities over discrete, featureless units. Here we present an alternative model, in which sequences are learned as trajectories through a similarity space. A simple recurrent network coding syllables with representations that capture the similarity relations among them correctly simulated the result of a classic SL study, as did a similar model that encoded syllables as three dimensional points in a continuous similarity space. We then used the simulations to identify a sequence of "words" that produces the reverse of the typical SL effect, i.e., part-words are predicted to be more familiar than Words. Results from two experiments with human participants are consistent with simulation results. Additional analyses identified features that drive differences in what is learned from a set of artificial languages that have the same transitional probabilities among syllables.

Rhythmically Modulating Neural Entrainment during Exposure to Regularities Influences Statistical Learning

The ability to discover regularities in the environment, such as syllable patterns in speech, is known as statistical learning. Previous studies have shown that statistical learning is accompanied by neural entrainment, in which neural activity temporally aligns with repeating patterns over time. However, it is unclear whether these rhythmic neural dynamics play a functional role in statistical learning or whether they largely reflect the downstream consequences of learning, such as the enhanced perception of learned words in speech. To better understand this issue, we manipulated participants' neural entrainment during statistical learning using continuous rhythmic visual stimulation. Participants were exposed to a speech stream of repeating nonsense words while viewing either (1) a visual stimulus with a "congruent" rhythm that aligned with the word structure, (2) a visual stimulus with an incongruent rhythm, or (3) a static visual stimulus. Statistical learning was subsequently measured using both an explicit and implicit test. Participants in the congruent condition showed a significant increase in neural entrainment over auditory regions at the relevant word frequency, over and above effects of passive volume conduction, indicating that visual stimulation successfully altered neural entrainment within relevant neural substrates. Critically, during the subsequent implicit test, participants in the congruent condition showed an enhanced ability to predict upcoming syllables and stronger neural phase synchronization to component words, suggesting that they had gained greater sensitivity to the statistical structure of the speech stream relative to the incongruent and static groups. This learning benefit could not be attributed to strategic processes, as participants were largely unaware of the contingencies between the visual stimulation and embedded words. These results indicate that manipulating neural entrainment during exposure to regularities influences statistical learning outcomes, suggesting that neural entrainment may functionally contribute to statistical learning. Our findings encourage future studies using non-invasive brain stimulation methods to further understand the role of entrainment in statistical learning.

The association between statistical learning and language development during childhood: A scoping review

The statistical account of language acquisition asserts that language is learned through computations on the statistical regularities present in natural languages. This type of account can predict variability in language development measures as arising from individual differences in extracting this statistical information. Given that statistical learning has been attested across different domains and modalities, a central question is which modality is more tightly yoked with language skills. The results of a scoping review, which aimed for the first time at identifying the evidence of the association between statistical learning skills and language outcomes in typically developing infants and children, provide preliminary support for the statistical learning account of language acquisition, mostly in the domain of lexical outcomes, indicating that typically developing infants and children with stronger auditory and audio-visual statistical learning skills perform better on lexical competence tasks. The results also suggest that the relevance of statistical learning skills for language development is dependent on sensory modality.