统计学习的认知神经机制及其与语言的关系

统计学习是指个体在连续刺激流中发现转移概率等统计规律的过程, 在Saffran等(1996)的经典婴儿语音切分研究中首次被提出。大量研究证实了统计学习的普遍存在, 近期学界开始关注统计学习的特异性及其对认知的影响, 尤其是从学习过程及其特异性两个方面阐述统计学习的认知神经机制并揭示其和语言的交互作用。未来应从脑和行为的多模态数据视角, 丰富统计学习结果的行为和神经指标, 考察不同类型统计学习过程的动态神经活动模式, 建立统计学习行为和脑的关联, 深化对统计学习认知神经机制的认识, 在统计学习与语言交互作用的基础上, 从成人二语学习切入结合音乐统计学习训练探讨促进语言学习的统计学习干预手段。     

Visual statistical learning in infancy: evidence for a domain general learning mechanism

The rapidity with which infants come to understand language and events in their surroundings has prompted speculation concerning innate knowledge structures that guide language acquisition and object knowledge. Recently, however, evidence has emerged that by 8 months, infants can extract statistical patterns in auditory input that are based on transitional probabilities defining the sequencing of the input's components (Science 274 (1996) 1926). This finding suggests powerful learning mechanisms that are functional in infancy, and raises questions about the domain generality of such mechanisms. We habituated 2-, 5-, and 8-month-old infants to sequences of discrete visual stimuli whose ordering followed a statistically predictable pattern. The infants subsequently viewed the familiar pattern alternating with a novel sequence of identical stimulus components, and exhibited significantly greater interest in the novel sequence at all ages. These results provide support for the likelihood of domain general statistical learning in infancy, and imply that mechanisms designed to detect structure inherent in the environment may play an important role in cognitive development.     

Effects of retrieval practice and presentation modality on verbal learning: testing the limits of the testing effect

ABSTRACT

The testing effect refers to improved memory after retrieval practice and has been researched primarily with visual stimuli. In two experiments, we investigated whether the testing effect can be replicated when the to-be-learned information is presented auditorily, or visually?+?auditorily. Participants learned Swahili-English word pairs in one of three presentation modalities – visual, auditory, or visual?+?auditory. This was manipulated between-participants in Experiment 1 and within-participants in Experiment2. All participants studied the word pairs during three study trials. Half of participants practiced recalling the English translations in response to the Swahili cue word twice before the final test whereas the other half simply studied the word pairs twice more. Results indicated an improvement in final test performance in the repeated test condition, but only in the visual presentation modality (Experiments 1 and 2) and in the visual?+?auditory presentation modality (Experiment 2). This suggests that the benefits of practiced retrieval may be limited to information presented in a visual modality.     

Domain specificity versus expertise: factors influencing distinct processing of faces

To explore face specificity in visual processing, we compared the role of task-associated strategies and expertise on the N170 event-related potential (ERP) component elicited by human faces with the ERPs elicited by cars, birds, items of furniture, and ape faces. In Experiment 1, participants performed a car monitoring task and an animacy decision task. In Experiment 2, participants monitored human faces while faces of apes were the distracters. Faces elicited an equally conspicuous N170, significantly larger than the ERPs elicited by non-face categories regardless of whether they were ignored or had an equal status with other categories (Experiment 1), or were the targets (in Experiment 2). In contrast, the negative component elicited by cars during the same time range was larger if they were targets than if they were not. Furthermore, unlike the posterior-temporal distribution of the N170, the negative component elicited by cars and its modulation by task were more conspicuous at occipital sites. Faces of apes elicited an N170 that was similar in amplitude to that elicited by the human face targets, albeit peaking 10 ms later. As our participants were not ape experts, this pattern indicates that the N170 is face-specific, but not specie-specific, i.e. it is elicited by particular face features regardless of expertise. Overall, these results demonstrate the domain specificity of the visual mechanism implicated in processing faces, a mechanism which is not influenced by either task or expertise. The processing of other objects is probably accomplished by a more general visual processor, which is sensitive to strategic manipulations and attention.     

Constraint on the semantic flexibility in visual statistical learning

We examined whether the expression of visual statistical learning (VSL) is flexible at the superordinate-categorical level. In the familiarization phase, participants viewed a sequence of line drawings. In the test phase, participants observed two test sequences (statistically related triplets versus unrelated foils) that consisted of the same objects as those presented during the familiarization phase (the same condition) or different objects that shared the same categorical information with drawings during the familiarization (the different condition). They then decided whether the first or the second sequence was more familiar. The results of Experiment 1 showed greater familiarity above chance levels for statistically related triplets only in the same condition. In Experiment 2, even where the word stimuli representing each superordinate-level category were included in the test phase (the categorical condition), the results showed VSL only in the same condition. Our findings suggest that the semantic flexibility of VSL is limited to the basic-level category.     

通道呈现方式与感知学习风格的匹配性对粤-普双言切换代价的影响

选取不同感知学习风格的熟练粤-普双言者被试, 比较在不同通道呈现方式下他们刺激命名任务的表现, 由此考察感知学习风格与通道呈现方式的匹配性对熟练双言者双言切换代价的影响。结果发现, 被试在视觉线索呈现条件下要比在听觉线索条件下的切换代价小; 当感知学习风格与通道呈现方式匹配时, 双言切换代价更低。表明感知学习风格与通道呈现方式的匹配性对于双言切换代价有调节作用。     

Comparing statistical learning across perceptual modalities in infancy: An investigation of underlying learning mechanism(s)

Statistical learning (SL), sensitivity to probabilistic regularities in sensory input, has been widely implicated in cognitive and perceptual development. Little is known, however, about the underlying mechanisms of SL and whether they undergo developmental change. One way to approach these questions is to compare SL across perceptual modalities. While a decade of research has compared auditory and visual SL in adults, we present the first direct comparison of visual and auditory SL in infants (8–10 months). Learning was evidenced in both perceptual modalities but with opposite directions of preference: Infants in the auditory condition displayed a novelty preference, while infants in the visual condition showed a familiarity preference. Interpreting these results within the Hunter and Ames model (1988), where familiarity preferences reflect a weaker stage of encoding than novelty preferences, we conclude that there is weaker learning in the visual modality than the auditory modality for this age. In addition, we found evidence of different developmental trajectories across modalities: Auditory SL increased while visual SL did not change for this age range. The results suggest that SL is not an abstract, amodal ability; for the types of stimuli and statistics tested, we find that auditory SL precedes the development of visual SL and is consistent with recent work comparing SL across modalities in older children.     

Prediction during statistical learning, and implications for the implicit/explicit divide

Accounts of statistical learning, both implicit and explicit, often invoke predictive processes as central to learning, yet practically all experiments employ non-predictive measures during training. We argue that the common theoretical assumption of anticipation and prediction needs clearer, more direct evidence for it during learning. We offer a novel experimental context to explore prediction, and report results from a simple sequential learning task designed to promote predictive behaviors in participants as they responded to a short sequence of simple stimulus events. Predictive tendencies in participants were measured using their computer mouse, the trajectories of which served as a means of tapping into predictive behavior while participants were exposed to very short and simple sequences of events. A total of 143 participants were randomly assigned to stimulus sequences along a continuum of regularity. Analysis of computer-mouse trajectories revealed that (a) participants almost always anticipate events in some manner, (b) participants exhibit two stable patterns of behavior, either reacting to vs. predicting future events, (c) the extent to which participants predict relates to performance on a recall test, and (d) explicit reports of perceiving patterns in the brief sequence correlates with extent of prediction. We end with a discussion of implicit and explicit statistical learning and of the role prediction may play in both kinds of learning.     

Auditory statistical learning: predictive frequency information affects the deployment of contextually mediated attentional resources on perceptual tasks

Statistical learning is a general phenomenon in which environmental regularities are implicitly acquired through repeated exposure to those environments. Sometimes, that information can be utilised to affect various aspects of cognitive performance (e.g. reaction time) on tasks that utilise selective attention (e.g. visual search). In the current study, we examined the effect of passively listening to predictive auditory contexts in facilitating attention to a certain frequency or frequency range. In doing so, we found that there is a general tendency for attentional resources to be negatively affected when the context sequences are made novel after context–target associations have been formed (Experiment 1), and when the context no longer reliably cues the previous target (Experiment 2). The experiments are framed to contrast Associative Learning and Memory Hypothesis perspectives.     

Sensory modality and stimulus control in the pigeon: Cross-species generality of single-incentive selective-association effects

In earlier studies with rats, the effectiveness of the auditory element of a tone–light discriminative stimulus was enhanced when the conditioned incentive value of the compound was negative rather than positive. The present experiment systematically replicated these results in pigeons trained to press a treadle in the presence of a tone–light compound under food-reinforcement or shock-avoidance schedules. Positive incentive value was conditioned to the compound by associating it with either food or relative safety from shock. The compound was made negative in other groups by associating it with shock or the absence of food. When tone and light were presented separately following this training, control by the auditory element was significantly enhanced in the conditions designed to make the compound negative rather than positive. The similarity of this constraint on learning in rats and pigeons suggests that it involves a fundamental attentional and incentive-motivational process with widespread species generality.     

Modelling unsupervised online-learning of artificial grammars: Linking implicit and statistical learning

Humans rapidly learn complex structures in various domains. Findings of above-chance performance of some untrained control groups in artificial grammar learning studies raise questions about the extent to which learning can occur in an untrained, unsupervised testing situation with both correct and incorrect structures. The plausibility of unsupervised online-learning effects was modelled with n-gram, chunking and simple recurrent network models. A novel evaluation framework was applied, which alternates forced binary grammaticality judgments and subsequent learning of the same stimulus. Our results indicate a strong online learning effect for n-gram and chunking models and a weaker effect for simple recurrent network models. Such findings suggest that online learning is a plausible effect of statistical chunk learning that is possible when ungrammatical sequences contain a large proportion of grammatical chunks. Such common effects of continuous statistical learning may underlie statistical and implicit learning paradigms and raise implications for study design and testing methodologies.     

Visual learning of statistical relations among nonadjacent features: Evidence for structural encoding

Recent results suggest that observers can learn, unsupervised, the co-occurrence of independent shape features in viewed patterns (e.g., Fiser & Aslin, 2001). A critical question with regard to these findings is whether learning is driven by a structural, rule-based encoding of spatial relations between distinct features or by a pictorial, template-like encoding, in which spatial configurations of features are embedded in a “holistic” fashion. In two experiments, we test whether observers can learn combinations of features when the paired features are separated by an intervening spatial “gap”, in which other, unrelated features can appear. This manipulation both increases task difficulty and makes it less likely that the feature combinations are encoded simply as larger unitary features. Observers exhibited learning consistent with earlier studies, suggesting that unsupervised learning of compositional structure is based on the explicit encoding of spatial relations between separable visual features. More generally, these results provide support for compositional structure in visual representation.     

The importance of understanding: Model space moderates goal specificity effects

The three-space theory of problem solving predicts that the quality of a learner's model and the goal specificity of a task interact on knowledge acquisition. In Experiment 1 participants used a computer simulation of a lever system to learn about torques. They either had to test hypotheses (nonspecific goal), or to produce given values for variables (specific goal). In the good- but not in the poor-model condition they saw torque depicted as an area. Results revealed the predicted interaction. A nonspecific goal only resulted in better learning when a good model of torques was provided. In Experiment 2 participants learned to manipulate the inputs of a system to control its outputs. A nonspecific goal to explore the system helped performance when compared to a specific goal to reach certain values when participants were given a good model, but not when given a poor model that suggested the wrong hypothesis space. Our findings support the three-space theory. They emphasize the importance of understanding for problem solving and stress the need to study underlying processes.     

The metamorphosis of the statistical segmentation output: Lexicalization during artificial language learning

This study combined artificial language learning (ALL) with conventional experimental techniques to test whether statistical speech segmentation outputs are integrated into adult listeners’ mental lexicon. Lexicalization was assessed through inhibitory effects of novel neighbors (created by the parsing process) on auditory lexical decisions to real words. Both immediately after familiarization and post-one week, ALL outputs were lexicalized only when the cues available during familiarization (transitional probabilities and wordlikeness) suggested the same parsing (Experiments 1 and 3). No lexicalization effect occurred with incongruent cues (Experiments 2 and 4). Yet, ALL differed from chance, suggesting a dissociation between item knowledge and lexicalization. Similarly contrasted results were found when frequency of occurrence of the stimuli was equated during familiarization (Experiments 3 and 4). Our findings thus indicate that ALL outputs may be lexicalized as far as the segmentation cues are congruent, and that this process cannot be accounted for by raw frequency.     

Testing the limits of long-distance learning: learning beyond a three-segment window

Traditional flat-structured bigram and trigram models of phonotactics are useful because they capture a large number of facts about phonological processes. Additionally, these models predict that local interactions should be easier to learn than long-distance ones because long-distance dependencies are difficult to capture with these models. Long-distance phonotactic patterns have been observed by linguists in many languages, who have proposed different kinds of models, including feature-based bigram and trigram models, as well as precedence models. Contrary to flat-structured bigram and trigram models, these alternatives capture unbounded dependencies because at an abstract level of representation, the relevant elements are locally dependent, even if they are not adjacent at the observable level. Using an artificial grammar learning paradigm, we provide additional support for these alternative models of phonotactics. Participants in two experiments were exposed to a long-distance consonant-harmony pattern in which the first consonant of a five-syllable word was [s] or [∫] ("sh") and triggered a suffix that was either [-su] or [-∫u] depending on the sibilant quality of this first consonant. Participants learned this pattern, despite the large distance between the trigger and the target, suggesting that when participants learn long-distance phonological patterns, that pattern is learned without specific reference to distance.     

The acquisition of allophonic rules: statistical learning with linguistic constraints

Phonological rules relate surface phonetic word forms to abstract underlying forms that are stored in the lexicon. Infants must thus acquire these rules in order to infer the abstract representation of words. We implement a statistical learning algorithm for the acquisition of one type of rule, namely allophony, which introduces context-sensitive phonetic variants of phonemes. This algorithm is based on the observation that different realizations of a single phoneme typically do not appear in the same contexts (ideally, they have complementary distributions). In particular, it measures the discrepancies in context probabilities for each pair of phonetic segments. In Experiment 1, we test the algorithm's performances on a pseudo-language and show that it is robust to statistical noise due to sampling and coding errors, and to non-systematic rule application. In Experiment 2, we show that a natural corpus of semiphonetically transcribed child-directed speech in French presents a very large number of near-complementary distributions that do not correspond to existing allophonic rules. These spurious allophonic rules can be eliminated by a linguistically motivated filtering mechanism based on a phonetic representation of segments. We discuss the role of a priori linguistic knowledge in the statistical learning of phonology.     

Relational learning in a context of transposition: a review

In a typical transposition task, an animal is presented with a single pair of stimuli (for example, S3+ S4−, where plus and minus denote reward and nonreward and digits denote stimulus location on a sensory dimension such as size). Subsequently, an animal is presented with a testing pair that contains a previously reinforced or nonreinforced stimulus and a novel stimulus (for example, S2–S3 and S4–S5). Does the choice of a novel S2 instead of previously reinforced S3 in a testing pair S2–S3 indicate that the animal has learned a relation (i.e., “select smaller”)? This review of empirical evidence and theoretical accounts shows that an organism''s behavior in a transposition task is undoubtedly influenced by prior reinforcement history of the training stimuli (Spence, 1937). However, it is also affected by two other factors that are relational in nature—a similarity of two testing stimuli to each other and an overall similarity of the testing pair as a whole to the training pair as a whole. The influence of the two latter factors is especially evident in studies that use multiple pairs of training stimuli and a wide range of testing pairs comprising nonadjacent stimuli (Lazareva, Miner, Young, & Wasserman, 2008; Lazareva, Wasserman, & Young, 2005). In sum, the evidence suggests that both prior reinforcement history and relational information affect an animal''s behavior in a typical transposition task.     

Use of phonetic specificity during the acquisition of new words: differences between consonants and vowels

The present study explores the issue of the use of phonetic specificity in the process of learning new words at 20 months of age. The procedure used follows Nazzi and Gopnik [Nazzi, T., & Gopnik, A. (2001). Linguistic and cognitive abilities in infancy: When does language become a tool for categorization? Cognition, 80, B11-B20]. Infants were first presented with triads of perceptually dissimilar objects, which were given made-up names, two of the objects receiving the same name. Then, word learning was evaluated through object selection/categorization. Tests involved phonetically different words (e.g. [pize] vs. [mora], Experiment 1), words differing minimally on their onset consonant (e.g. [pize] vs. [tize], Experiment 2a), and conditions which had never been tested before: non-initial consonantal contrasts (e.g. [pide] vs. [pige], Experiment 2b), and vocalic contrasts (e.g. [pize] vs. [pyze]; [pize] vs. [paze]; [pize] vs. [pizu], Experiments 3a-c). Results differed across conditions: words could be easily learnt in the phonetically different condition, and were learnt, though to a lesser degree, in both the initial and non-initial minimal consonant contrast; however, infants' global performance on all three vocalic contrasts was at chance level. The present results shed new light regarding the specificity of early words, and raise the possibility of different contributions for vowels and consonants in early word learning.