Information from multiple modalities helps 5‐month‐olds learn abstract rules

By 7 months of age, infants are able to learn rules based on the abstract relationships between stimuli ( Marcus et al ., 1999 ), but they are better able to do so when exposed to speech than to some other classes of stimuli. In the current experiments we ask whether multimodal stimulus information will aid younger infants in identifying abstract rules. We habituated 5‐month‐olds to simple abstract patterns (ABA or ABB) instantiated in coordinated looming visual shapes and speech sounds (Experiment 1), shapes alone (Experiment 2), and speech sounds accompanied by uninformative but coordinated shapes (Experiment 3). Infants showed evidence of rule learning only in the presence of the informative multimodal cues. We hypothesize that the additional evidence present in these multimodal displays was responsible for the success of younger infants in learning rules, congruent with both a Bayesian account and with the Intersensory Redundancy Hypothesis.     

The profile of abstract rule learning in infancy: Meta‐analytic and experimental evidence

Everyone agrees that infants possess general mechanisms for learning about the world, but the existence and operation of more specialized mechanisms is controversial. One mechanism—rule learning—has been proposed as potentially specific to speech, based on findings that 7‐month‐olds can learn abstract repetition rules from spoken syllables (e.g. ABB patterns: wo‐fe‐fe, ga‐tu‐tu…) but not from closely matched stimuli, such as tones. Subsequent work has shown that learning of abstract patterns is not simply specific to speech. However, we still lack a parsimonious explanation to tie together the diverse, messy, and occasionally contradictory findings in that literature. We took two routes to creating a new profile of rule learning: meta‐analysis of 20 prior reports on infants’ learning of abstract repetition rules (including 1,318 infants in 63 experiments total), and an experiment on learning of such rules from a natural, non‐speech communicative signal. These complementary approaches revealed that infants were most likely to learn abstract patterns from meaningful stimuli. We argue that the ability to detect and generalize simple patterns supports learning across domains in infancy but chiefly when the signal is meaningfully relevant to infants’ experience with sounds, objects, language, and people.     

Infants’ representations of same and different in match- and non-match-to-sample

Three experiments investigated the representations that underlie 14-month-old infants’ and adults’ success at match-to-sample (MTS) and non-match-to-sample (NMTS) tasks. In Experiment 1, 14-month-old infants were able to learn rules based on abstract representations of sameness and/or difference. When presented with one of eighteen sample stimuli (A) and a choice between a stimulus that was the same as the sample (A) and a different stimulus (B), infants learned to choose A in MTS and B in NMTS. In Experiments 2 and 3, we began to explore the nature of the representations at play in these paradigms. Experiment 2 confirmed that abstract representations were at play, as infants generalized the MTS and NMTS rules to stimuli unseen during familiarization. Experiment 2 also showed that infants tested in MTS learned to seek the stimulus that was the same as the sample, whereas infants tested in NMTS did not learn to seek the different stimulus, but instead learned to avoid the stimulus that was the same as the sample. Infants appeared to only use an abstract representation of the relation same in these experiments. Experiment 3 showed that adult participants, despite knowing the words “same” and “different”, also relied on representations of sameness in both MTS and NMTS in a paradigm modeled on that of Experiment 2. We conclude with a discussion of how young infants may possibly represent the abstract relation same.     

Dog is a dog is a dog: infant rule learning is not specific to language

Human infants possess powerful learning mechanisms used for the acquisition of language. To what extent are these mechanisms domain specific? One well-known infant language learning mechanism is the ability to detect and generalize rule-like similarity patterns, such as ABA or ABB [Marcus, G. F., Vijayan, S., Rao, S. B., & Vishton, P. M. (1999). Rule learning by seven-month-old infants. Science, 283, 77-80.]. The results of three experiments demonstrate that 7-month-old infants can detect and generalize these same patterns when the elements consist of pictures of animals (dogs and cats). These findings indicate that rule learning of this type is not specific to language acquisition.     

Infants' ability to categorise on the basis of rhyme

Two experiments are described, which investigated 7 - to 13-month-old infants' abilities to categorise syllables according to their ending sounds, or rhymes. Using the Conditioned Headturn (CHT) Procedure, infants were conditioned to turn their head when one set of rhyming CVCs changed to another set of rhyming CVCs. Even the 7 -month-old infants demonstrated an ability to categorise according to rimes. The infants could be separated into three different groups: those who conditioned but did not succeed in a rhyming task; those who succeeded at one rhyming task but not a second; and those infants who succeeded on two rhyming tasks. The infants in the second group were showing an early sensitivity to rhyme. However, they appeared to be learning a very specific rule, listening only for one category of rhyme. The third group of infants was able to extend the initial rule they learned to include new rhymes. This group learned not only to highlight one particular rhyme, but were also able to abstract the previously learned rule to new rhymes, thereby demonstrating a higher level of sophistication in their categorisation of rhymes. These results are discussed in relation to the ontogeny of language learning, and work demonstrating a relationship between rhyming ability and competence in reading and writing.     

A direct replication: Unconscious arithmetic processing

Across two experiments involving four presentation times in total, Sklar et al. (Proceedings of the National Academy of Sciences of the United States of America, 109, 2012, 19614) found that complex subtraction equations can be solved without awareness of the equations. These findings challenge the current position that consciousness is necessary for performing abstract, rule-following tasks. Given the important implications of their work, we aimed to directly replicate Sklar's findings using a larger sample (n = 94) from a different population. Using continuous flash suppression, we investigated if people were able to solve an equation after subliminal (1,300 ms) exposure to it. We found evidence of unconscious subtraction consistent with Sklar et al., albeit the effect is weak. Critical review of our results and implications for further research are discussed.     

Three ideal observer models for rule learning in simple languages

Children learning the inflections of their native language show the ability to generalize beyond the perceptual particulars of the examples they are exposed to. The phenomenon of “rule learning”—quick learning of abstract regularities from exposure to a limited set of stimuli—has become an important model system for understanding generalization in infancy. Experiments with adults and children have revealed differences in performance across domains and types of rules. To understand the representational and inferential assumptions necessary to capture this broad set of results, we introduce three ideal observer models for rule learning. Each model builds on the next, allowing us to test the consequences of individual assumptions. Model 1 learns a single rule, Model 2 learns a single rule from noisy input, and Model 3 learns multiple rules from noisy input. These models capture a wide range of experimental results—including several that have been used to argue for domain-specificity or limits on the kinds of generalizations learners can make—suggesting that these ideal observers may be a useful baseline for future work on rule learning.     

An investigation of mental rotation in infancy using change detection

Two experiments were conducted to examine mental rotation in 6- to 12-month-old infants (N = 166) using a change detection task. These experiments were replications of Lauer and Lourenco (Lauer et al., 2015; Lauer & Lourenco, 2016), using identical stimuli and variations of their procedure, including an exact replication conducted in a laboratory setting (Experiment 1), and an online assessment using Lookit (Scott et al.,2017; Scott & Schulz, 2017) (Experiment 2). Both experiments failed to replicate the results of the original study; in neither experiment did infants’ behavior provide evidence that they mentally rotated the object. Results are discussed in terms of the robustness of mental rotation in infancy and about limits in our experimental procedures for uncovering perceptual and cognitive abilities in infants.     

Emotional facial expressions affect visual rule learning in 7- to 8-month-old infants

Rule learning (RL) is an implicit learning mechanism that allows infants to detect and generalize rule-like repetition-based patterns (such as ABB and ABA) from a sequence of elements. Increasing evidence shows that RL operates both in the auditory and the visual domain and is modulated by the perceptual expertise with the to-be-learned stimuli. Yet, whether infants’ ability to detect a high-order rule from a sequence of stimuli is affected by affective information remains a largely unexplored issue. Using a visual habituation paradigm, we investigated whether the presence of emotional expressions with a positive and a negative value (i.e., happiness and anger) modulates 7- to 8-month-old infants’ ability to learn a rule-like pattern from a sequence of faces of different identities. Results demonstrate that emotional facial expressions (either positive and negative) modulate infants’ visual RL mechanism, even though positive and negative facial expressions affect infants’ RL in a different manner: while anger disrupts infants’ ability to learn the rule-like pattern from a face sequence, in the presence of a happy face infants show a familiarity preference, thus maintaining their learning ability. These findings show that emotional expressions exert an influence on infants’ RL abilities, contributing to the investigation on how emotion and cognition interact in face processing during infancy.     

“Mummy,keep it steady”: phonetic variation shapes word learning at 15 and 17 months

Fifteen‐month‐olds have difficulty detecting differences between novel words differing in a single vowel. Previous work showed that Australian English (AusE) infants habituated to the word‐object pair DEET detected an auditory switch to DIT and DOOT in Canadian English (CanE) but not in their native AusE (Escudero et al., 2014 ). The authors speculated that this may be because the vowel inherent spectral change variation (VISC) in AusE DEET is larger than in CanE DEET. We investigated whether VISC leads to difficulty in encoding phonetic detail during early word learning, and whether this difficulty dissipates with age. In Experiment 1, we familiarized AusE‐learning 15‐month‐olds to AusE DIT, which contains smaller VISC than AusE DEET. Unlike infants familiarized with AusE DEET (Escudero et al., 2014 ), infants detected a switch to DEET and DOOT. In Experiment 2, we familiarized AusE‐learning 17‐month‐olds to AusE DEET. This time, infants detected a switch to DOOT, and marginally detected a switch to DIT. Our acoustic analysis showed that AusE DEET and DOOT are differentiated by the second vowel formant, while DEET and DIT can only be distinguished by their changing dynamic properties throughout the vowel trajectory. Thus, by 17 months, AusE infants can encode highly dynamic acoustic properties, enabling them to learn the novel vowel minimal pairs that are difficult at 15 months. These findings suggest that the development of word learning is shaped by the phonetic properties of the specific word minimal pair.     

Analyzing Machine-Learned Representations: A Natural Language Case Study

As modern deep networks become more complex, and get closer to human-like capabilities in certain domains, the question arises as to how the representations and decision rules they learn compare to the ones in humans. In this work, we study representations of sentences in one such artificial system for natural language processing. We first present a diagnostic test dataset to examine the degree of abstract composable structure represented. Analyzing performance on these diagnostic tests indicates a lack of systematicity in representations and decision rules, and reveals a set of heuristic strategies. We then investigate the effect of training distribution on learning these heuristic strategies, and we study changes in these representations with various augmentations to the training set. Our results reveal parallels to the analogous representations in people. We find that these systems can learn abstract rules and generalize them to new contexts under certain circumstances—similar to human zero-shot reasoning. However, we also note some shortcomings in this generalization behavior—similar to human judgment errors like belief bias. Studying these parallels suggests new ways to understand psychological phenomena in humans as well as informs best strategies for building artificial intelligence with human-like language understanding.     

Abstract Rule Learning in 11- and 14-Month-Old Infants

This study tests the hypothesis that distributional information can guide infants in the generalization of word order movement rules at the initial stage of language acquisition. Participants were 11- and 14-month-old infants. Stimuli were sentences in Russian, a language that was unknown to our infants. During training the word order of each sentence was transformed following a consistent pattern (e.g., ABC–BAC). During the test phase infants heard novel sentences that respected the trained rule and ones that violated the trained rule (i.e., a different transformation such as ABC–ACB). Stimuli words had highly variable phonological and morphological shapes. The cue available was the positional information of words and their non-adjacent relations across sentences. We found that 14-month-olds, but not 11-month-olds, showed evidence of abstract rule generalization to novel instances. The implications of this finding to early syntactic acquisition are discussed.     

Bimodal emotion congruency is critical to preverbal infants’ abstract rule learning

Extracting general rules from specific examples is important, as we must face the same challenge displayed in various formats. Previous studies have found that bimodal presentation of grammar‐like rules (e.g. ABA) enhanced 5‐month‐olds’ capacity to acquire a rule that infants failed to learn when the rule was presented with visual presentation of the shapes alone (circle‐triangle‐circle) or auditory presentation of the syllables (la‐ba‐la) alone. However, the mechanisms and constraints for this bimodal learning facilitation are still unknown. In this study, we used audio‐visual relation congruency between bimodal stimulation to disentangle possible facilitation sources. We exposed 8‐ to 10‐month‐old infants to an AAB sequence consisting of visual faces with affective expressions and/or auditory voices conveying emotions. Our results showed that infants were able to distinguish the learned AAB rule from other novel rules under bimodal stimulation when the affects in audio and visual stimuli were congruently paired (Experiments 1A and 2A). Infants failed to acquire the same rule when audio‐visual stimuli were incongruently matched (Experiment 2B) and when only the visual (Experiment 1B) or the audio (Experiment 1C) stimuli were presented. Our results highlight that bimodal facilitation in infant rule learning is not only dependent on better statistical probability and redundant sensory information, but also the relational congruency of audio‐visual information. A video abstract of this article can be viewed at https://m.youtube.com/watch?v=KYTyjH1k9RQ     

Rule learning by cotton-top tamarins

Previous work suggests that human infants are capable of rapidly generalizing patterns that have been characterized as abstract algebraic rules (Science 283 (1999) 77), a process that may play a pivotal role in language acquisition. Here we explore whether this capacity is uniquely human and evolved specifically for the computational problems associated with language, or whether this mechanism is shared with other species, and therefore evolved for problems other than language. We used the same materials and methods that were originally employed in tests of human infants to assess whether cotton-top tamarin monkeys can extract abstract algebraic rules. Specifically, we habituated subjects to sequences of consonant-vowel syllables that followed one of two patterns, AAB (e.g. wi wi di) or ABB (le we we). Following habituation, we presented subjects with two novel test items, one with the same pattern as that presented during habituation and one with a different pattern. Like human infants, tamarins were more likely to dishabituate to the test item with a different pattern. We conclude that the capacity to generalize rule-like patterns, at least at the level demonstrated, did not evolve specifically for language acquisition, though it remains possible that infants might use such rules during language acquisition.     

Infant learning is influenced by local spurious generalizations

In previous work, 11‐month‐old infants were able to learn rules about the relation of the consonants in CVCV words from just four examples. The rules involved phonetic feature relations (same voicing or same place of articulation), and infants' learning was impeded when pairs of words allowed alternative possible generalizations (e.g. two words both contained the specific consonants p and t). Experiment 1 asked whether a small number of such spurious generalizations found in a randomly ordered list of 24 different words would also impede learning. It did – infants showed no sign of learning the rule. To ask whether it was the overall set of words or their order that prevented learning, Experiment 2 reordered the words to avoid local spurious generalizations. Infants showed robust learning. Infants thus appear to entertain spurious generalizations based on small, local subsets of stimuli. The results support a characterization of infants as incremental rather than batch learners.     

Can infants map meaning to newly segmented words? Statistical segmentation and word learning

The present experiments investigated how the process of statistically segmenting words from fluent speech is linked to the process of mapping meanings to words. Seventeen-month-old infants first participated in a statistical word segmentation task, which was immediately followed by an object-label-learning task. Infants presented with labels that were words in the fluent speech used in the segmentation task were able to learn the object labels. However, infants presented with labels consisting of novel syllable sequences (nonwords; Experiment 1) or familiar sequences with low internal probabilities (part-words; Experiment 2) did not learn the labels. Thus, prior segmentation opportunities, but not mere frequency of exposure, facilitated infants' learning of object labels. This work provides the first demonstration that exposure to word forms in a statistical word segmentation task facilitates subsequent word learning.     

Learning General Phonological Rules From Distributional Information: A Computational Model

Phonological rules create alternations in the phonetic realizations of related words. These rules must be learned by infants in order to identify the phonological inventory, the morphological structure, and the lexicon of a language. Recent work proposes a computational model for the learning of one kind of phonological alternation, allophony (Peperkamp, Le Calvez, Nadal, & Dupoux, 2006). This paper extends the model to account for learning of a broader set of phonological alternations and the formalization of these alternations as general rules. In Experiment 1, we apply the original model to new data in Dutch and demonstrate its limitations in learning nonallophonic rules. In Experiment 2, we extend the model to allow it to learn general rules for alternations that apply to a class of segments. In Experiment 3, the model is further extended to allow for generalization by context; we argue that this generalization must be constrained by linguistic principles.     

8─11个月婴儿迂回行为发展特点与学习能力的研究

采用程序标准化的婴儿迂回行为测验,探讨了２４４名８─１１个月婴儿的迂回行为发展特点及学习能力。结果表明：（１）８─１１个月婴儿的迂回行为水平表现出随婴儿月龄的增长而提高的趋势。（２）８─１１个月的婴儿具有从反复尝试中学会迂回行为的可能性,其中９、１０个月婴儿的学习可能性较大,而８个月婴儿的学习可能性较小。（３）８──１１个月婴儿的学习速度随月龄增大而提高     

Processing abstract sequence structure: learning without knowing, or knowing without learning?

Constant interaction with a dynamic environment—from riding a bicycle to segmenting speech—makes sensitivity to the sequential structure of the world a fundamental dimension of information processing. Accounts of sequence learning vary widely, with some authors arguing that parsing and segmentation processes are central, and others proposing that sequence learning involves mere memorization. In this paper, we argue that sequence knowledge is essentially statistical in nature, and that sequence learning involves simple associative prediction mechanisms. We focus on a choice reaction situation introduced by Lee (1997), in which participants were exposed to material that follows a single abstract rule, namely that stimuli are selected randomly, but never appear more than once in a legal sequence. Perhaps surprisingly, people can learn this rule very well. Or can they? We offer a conceptual replication of the original finding, but a very different interpretation of the results, as well as simulation work that makes it clear how highly abstract dimensions of the stimulus material can in fact be learned based on elementary associative mechanisms. We conclude that, when relevant, memory is optimized to facilitate responding to events that have not occurred recently, and that sequence learning in general always involves sensitivity to repetition distance.     

Structural generalizations over consonants and vowels in 11-month-old infants

Recent research has suggested consonants and vowels serve different roles during language processing. While statistical computations are preferentially made over consonants but not over vowels, simple structural generalizations are easily made over vowels but not over consonants. Nevertheless, the origins of this asymmetry are unknown. Here we tested if a lifelong experience with language is necessary for vowels to become the preferred target for structural generalizations. We presented 11-month-old infants with a series of CVCVCV nonsense words in which all vowels were arranged according to an AAB rule (first and second vowels were the same, while the third vowel was different). During the test, we presented infants with new words whose vowels either followed or not, the aforementioned rule. We found that infants readily generalized this rule when implemented over the vowels. However, when the same rule was implemented over the consonants, infants could not generalize it to new instances. These results parallel those found with adult participants and demonstrate that several years of experience learning a language are not necessary for functional asymmetries between consonants and vowels to appear.