Object‐Label‐Order Effect When Learning From an Inconsistent Source

Learning in natural environments is often characterized by a degree of inconsistency from an input. These inconsistencies occur, for example, when learning from more than one source, or when the presence of environmental noise distorts incoming information; as a result, the task faced by the learner becomes ambiguous. In this study, we investigate how learners handle such situations. We focus on the setting where a learner receives and processes a sequence of utterances to master associations between objects and their labels, where the source is inconsistent by design: It uses both “correct” and “incorrect” object‐label pairings. We hypothesize that depending on the order of presentation, the result of the learning may be different. To this end, we consider two types of symbolic learning procedures: the Object‐Label (OL) and the Label‐Object (LO) process. In the OL process, the learner is first exposed to the object, and then the label. In the LO process, this order is reversed. We perform experiments with human subjects, and also construct a computational model that is based on a nonlinear stochastic reinforcement learning algorithm. It is observed experimentally that OL learners are generally better at processing inconsistent input compared to LO learners. We show that the patterns observed in the learning experiments can be reproduced in the simulations if the model includes (a) an ability to regularize the input (and also to do the opposite, i.e., undermatch) and (b) an ability to take account of implicit negative evidence (i.e., interactions among different objects/labels). The model suggests that while both types of learners utilize implicit negative evidence in a similar way, there is a difference in regularization patterns: OL learners regularize the input, whereas LO learners undermatch. As a result, OL learners are able to form a more consistent system of image‐utterance associations, despite the ambiguous learning task.     

Adaptive categorization in unsupervised learning

In 3 experiments, the authors provide evidence for a distinct category-invention process in unsupervised (discovery) learning and set forth a method for observing and investigating that process. In the 1st 2 experiments, the sequencing of unlabeled training instances strongly affected participants' ability to discover patterns (categories) across those instances. In the 3rd experiment, providing diagnostic labels helped participants discover categories and improved learning even for instance sequences that were unlearnable in the earlier experiments. These results are incompatible with models that assume that people learn by incrementally tracking correlations between individual features; instead, they suggest that learners in this study used expectation failure as a trigger to invent distinct categories to represent patterns in the stimuli. The results are explained in terms of J. R. Anderson's (1990, 1991) rational model of categorization, and extensions of this analysis for real-world learning are discussed.     

Curiosity‐based learning in infants: a neurocomputational approach

Infants are curious learners who drive their own cognitive development by imposing structure on their learning environment as they explore. Understanding the mechanisms by which infants structure their own learning is therefore critical to our understanding of development. Here we propose an explicit mechanism for intrinsically motivated information selection that maximizes learning. We first present a neurocomputational model of infant visual category learning, capturing existing empirical data on the role of environmental complexity on learning. Next we “set the model free”, allowing it to select its own stimuli based on a formalization of curiosity and three alternative selection mechanisms. We demonstrate that maximal learning emerges when the model is able to maximize stimulus novelty relative to its internal states, depending on the interaction across learning between the structure of the environment and the plasticity in the learner itself. We discuss the implications of this new curiosity mechanism for both existing computational models of reinforcement learning and for our understanding of this fundamental mechanism in early development.     

A Bayesian Model of Biases in Artificial Language Learning: The Case of a Word‐Order Universal

In this article, we develop a hierarchical Bayesian model of learning in a general type of artificial language‐learning experiment in which learners are exposed to a mixture of grammars representing the variation present in real learners’ input, particularly at times of language change. The modeling goal is to formalize and quantify hypothesized learning biases. The test case is an experiment ( Culbertson, Smolensky, & Legendre, 2012 ) targeting the learning of word‐order patterns in the nominal domain. The model identifies internal biases of the experimental participants, providing evidence that learners impose (possibly arbitrary) properties on the grammars they learn, potentially resulting in the cross‐linguistic regularities known as typological universals. Learners exposed to mixtures of artificial grammars tended to shift those mixtures in certain ways rather than others; the model reveals how learners’ inferences are systematically affected by specific prior biases. These biases are in line with a typological generalization—Greenberg's Universal 18—which bans a particular word‐order pattern relating nouns, adjectives, and numerals.     

Regularization of languages by adults and children: A mathematical framework

The fascinating ability of humans to modify the linguistic input and “create” a language has been widely discussed. In the work of Newport and colleagues, it has been demonstrated that both children and adults have some ability to process inconsistent linguistic input and “improve” it by making it more consistent. In Hudson Kam and Newport (2009), artificial miniature language acquisition from an inconsistent source was studied. It was shown that (i) children are better at language regularization than adults and that (ii) adults can also regularize, depending on the structure of the input. In this paper we create a learning algorithm of the reinforcement-learning type, which exhibits patterns reported in Hudson Kam and Newport (2009) and suggests a way to explain them. It turns out that in order to capture the differences between children’s and adults’ learning patterns, we need to introduce a certain asymmetry in the learning algorithm. Namely, we have to assume that the reaction of the learners differs depending on whether or not the source’s input coincides with the learner’s internal hypothesis. We interpret this result in the context of a different reaction of children and adults to implicit, expectation-based evidence, positive or negative. We propose that a possible mechanism that contributes to the children’s ability to regularize an inconsistent input is related to their heightened sensitivity to positive evidence rather than the (implicit) negative evidence. In our model, regularization comes naturally as a consequence of a stronger reaction of the children to evidence supporting their preferred hypothesis. In adults, their ability to adequately process implicit negative evidence prevents them from regularizing the inconsistent input, resulting in a weaker degree of regularization.     

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.     

When global structure "Explains Away" local grammar: a Bayesian account of rule-induction in tone sequences

While many constraints on learning must be relatively experience-independent, past experience provides a rich source of guidance for subsequent learning. Discovering structure in some domain can inform a learner’s future hypotheses about that domain. If a general property accounts for particular sub-patterns, a rational learner should not stipulate separate explanations for each detail without additional evidence, as the general structure has “explained away” the original evidence. In a grammar-learning experiment using tone sequences, manipulating learners’ prior exposure to a tone environment affects their sensitivity to the grammar-defining feature, in this case consecutive repeated tones. Grammar-learning performance is worse if context melodies are “smooth” — when small intervals occur more than large ones — as Smoothness is a general property accounting for a high rate of repetition. We present an idealized Bayesian model as a “best case” benchmark for learning repetition grammars. When context melodies are Smooth, the model places greater weight on the small-interval constraint, and does not learn the repetition rule as well as when context melodies are not Smooth, paralleling the human learners. These findings support an account of abstract grammar-induction in which learners rationally assess the statistical evidence for underlying structure based on a generative model of the environment.     

Learning diphone-based segmentation

This paper reconsiders the diphone-based word segmentation model of Cairns, Shillcock, Chater, and Levy (1997) and Hockema (2006), previously thought to be unlearnable. A statistically principled learning model is developed using Bayes' theorem and reasonable assumptions about infants' implicit knowledge. The ability to recover phrase-medial word boundaries is tested using phonetic corpora derived from spontaneous interactions with children and adults. The (unsupervised and semi-supervised) learning models are shown to exhibit several crucial properties. First, only a small amount of language exposure is required to achieve the model's ceiling performance, equivalent to between 1 day and 1 month of caregiver input. Second, the models are robust to variation, both in the free parameter and the input representation. Finally, both the learning and baseline models exhibit undersegmentation, argued to have significant ramifications for speech processing as a whole.     

Effects of visual information on adults' and infants' auditory statistical learning

Infant and adult learners are able to identify word boundaries in fluent speech using statistical information. Similarly, learners are able to use statistical information to identify word-object associations. Successful language learning requires both feats. In this series of experiments, we presented adults and infants with audio-visual input from which it was possible to identify both word boundaries and word-object relations. Adult learners were able to identify both kinds of statistical relations from the same input. Moreover, their learning was actually facilitated by the presence of two simultaneously present relations. Eight-month-old infants, however, do not appear to benefit from the presence of regular relations between words and objects. Adults, like 8-month-olds, did not benefit from regular audio-visual correspondences when they were tested with tones, rather than linguistic input. These differences in learning outcomes across age and input suggest that both developmental and stimulus-based constraints affect statistical learning.     

What infants know about syntax but couldn't have learned: experimental evidence for syntactic structure at 18 months

Generative linguistic theory stands on the hypothesis that grammar cannot be acquired solely on the basis of an analysis of the input, but depends, in addition, on innate structure within the learner to guide the process of acquisition. This hypothesis derives from a logical argument, however, and its consequences have never been examined experimentally with infant learners. Challenges to this hypothesis, claiming that an analysis of the input is indeed sufficient to explain grammatical acquisition, have recently gained attention. We demonstrate with novel experimentation the insufficiency of this countervailing view. Focusing on the syntactic structures required to determine the antecedent for the pronoun one, we demonstrate that the input to children does not contain sufficient information to support unaided learning. Nonetheless, we show that 18-month-old infants do have command of the syntax of one. Because this syntactic knowledge could not have been gleaned exclusively from the input, infants' mastery of this aspect of syntax constitutes evidence for the contribution of innate structure within the learner in acquiring a grammar.     

The implicit benefit of learning without errors

Two studies examined whether the number of errors made in learning a motor skill, golf putting, differentially influences the adoption of a selective (explicit) or unselective (implicit) learning mode. Errorful learners were expected to adopt an explicit, hypothesis-testing strategy to correct errors during learning, thereby accruing a pool of verbalizable rules and exhibiting performance breakdown under dual-task conditions, characteristic of a selective mode of learning. Reducing errors during learning was predicted to minimize the involvement of explicit hypothesis testing leading to the adoption of an unselective mode of learning, distinguished by few verbalizable rules and robust performance under secondary task loading. Both studies supported these predictions. The golf putting performance of errorless learners in both studies was unaffected by the imposition of a secondary task load, whereas the performance of errorful learners deteriorated. Reducing errors during learning limited the number of error-correcting hypotheses tested by the learner, thereby reducing the contribution of explicit processing to skill acquisition. It was concluded that the reduction of errors during learning encourages the use of implicit, unselective learning processes, which confer insusceptibility to performance breakdown under distraction.     

TRACX: a recognition-based connectionist framework for sequence segmentation and chunk extraction

Individuals of all ages extract structure from the sequences of patterns they encounter in their environment, an ability that is at the very heart of cognition. Exactly what underlies this ability has been the subject of much debate over the years. A novel mechanism, implicit chunk recognition (ICR), is proposed for sequence segmentation and chunk extraction. The mechanism relies on the recognition of previously encountered subsequences (chunks) in the input rather than on the prediction of upcoming items in the input sequence. A connectionist autoassociator model of ICR, truncated recursive autoassociative chunk extractor (TRACX), is presented in which chunks are extracted by means of truncated recursion. The performance and robustness of the model is demonstrated in a series of 9 simulations of empirical data, covering a wide range of phenomena from the infant statistical learning and adult implicit learning literatures, as well as 2 simulations demonstrating the model's ability to generalize to new input and to develop internal representations whose structure reflects that of the items in the input sequence. TRACX outperforms PARSER (Perruchet & Vintner, 1998) and the simple recurrent network (SRN, Cleeremans & McClelland, 1991) in matching human sequence segmentation on existing data. A new study is presented exploring 8-month-olds' use of backward transitional probabilities to segment auditory sequences.     

Structural packaging in the input to language learning: Contributions of prosodic and morphological marking of phrases to the acquisition of language

The linguistic input to language learning is usually thought to consist of simple strings of words. We argue that input must also include information about how words group into syntactic phrases. Natural languages regularly incorporate correlated cues to phrase structure, such as prosody, function words, and concord morphology. The claim that such cues are necessary for successful acquisition of syntax was tested in a series of miniature language learning experiments with adult subjects. In each experiment, when input included some cue marking the phrase structure of sentences, subjects were entirely successful in learning syntax; in contrast, when input lacked such a cue (but was otherwise identical), subjects failed to learn significant portions of syntax. Cues to phrase structure appear to facilitate learning by indicating to the learner those domains within which distributional analyses may be most efficiently pursued, thereby reducing the amount and complexity of required input data. More complex target systems place greater premiums on efficient analysis; hence, such cues may be even more crucial for acquisition of natural language syntax. We suggest that the finding that phrase structure cues are a necessary aspect of language input reflects the limited capacities of human language learners; languages may incorporate structural cues in part to circumvent such limitations and ensure successful acquisition.     

Statistical learning in a serial reaction time task: access to separable statistical cues by individual learners.

The ability of adult learners to exploit the joint and conditional probabilities in a serial reaction time task containing both deterministic and probabilistic information was investigated. Learners used the statistical information embedded in a continuous input stream to improve their performance for certain transitions by simultaneously exploiting differences in the predictability of 2 or more underlying statistics. Analysis of individual learners revealed that although most acquired the underlying statistical structure veridically, others used an alternate strategy that was partially predictive of the sequences. The findings show that learners possess a robust learning device well suited to exploiting the relative predictability of more than I source of statistical information at the same time. This work expands on previous studies of statistical learning, as well as studies of artificial grammar learning and implicit sequence learning.     

Visual artificial grammar learning by rhesus macaques (<Emphasis Type="Italic">Macaca mulatta</Emphasis>): exploring the role of grammar complexity and sequence length

Humans and nonhuman primates can learn about the organization of stimuli in the environment using implicit sequential pattern learning capabilities. However, most previous artificial grammar learning studies with nonhuman primates have involved relatively simple grammars and short input sequences. The goal in the current experiments was to assess the learning capabilities of monkeys on an artificial grammar-learning task that was more complex than most others previously used with nonhumans. Three experiments were conducted using a joystick-based, symmetrical-response serial reaction time task in which two monkeys were exposed to grammar-generated sequences at sequence lengths of four in Experiment 1, six in Experiment 2, and eight in Experiment 3. Over time, the monkeys came to respond faster to the sequences generated from the artificial grammar compared to random versions. In a subsequent generalization phase, subjects generalized their knowledge to novel sequences, responding significantly faster to novel instances of sequences produced using the familiar grammar compared to those constructed using an unfamiliar grammar. These results reveal that rhesus monkeys can learn and generalize the statistical structure inherent in an artificial grammar that is as complex as some used with humans, for sequences up to eight items long. These findings are discussed in relation to whether or not rhesus macaques and other primate species possess implicit sequence learning abilities that are similar to those that humans draw upon to learn natural language grammar.     

The proportion valid effect in covert orienting: Strategic control or implicit learning?

It is well known that the difference in performance between valid and invalid trials in the covert orienting paradigm (i.e., the cueing effect) increases as the proportion of valid trials increases. This proportion valid effect is widely assumed to reflect “strategic” control over the distribution of attention. In the present experiments we determine if this effect results from an explicit strategy or implicit learning by probing participant’s awareness of the proportion of valid trials. Results support the idea that the proportion valid effect in the covert orienting paradigm reflects implicit learning not an explicit strategy.     

A mathematical theory of instruction: Instructor/learner interaction and instruction pacing

In this paper, a mathematical theory of instruction applicable in the educational environment is developed from concepts of psychological learning theory. Within the framework of optimization and control theory, the dynamics of the interaction between instructor and learner are modelled, and the trade-off between instruction cost and learner achievement is formulated so that optimal instruction inputs can be determined. One important aspect of the classroom environment that is characterized by the theory is the interaction between an instructor and a group of learners with various learning abilities.A basic dynamic model that relates learner achievement and instruction cost is developed from learning theory concepts. This model, which applies to the individual learner situation, is analyzed in detail to determine instruction intensity inputs that match the learner's characteristics in order to maximize an objective that measures both achievement and cost.This basic model is used as a building block to describe how individual learner achievement depends on instruction pacing. To determine optimal instruction pacing the concept of gain, which is essentially learner achievement per unit time, is introduced. In this extended model, instruction pacing is intimately related with the concept of learner aptitude. This relationship leads immediately to the consideration of instruction pacing for a group of learners with various aptitudes and thus optimal instruction pacing is determined for nonhomogenous groups.Throughout the development of the theory, hypothetical examples are presented to demonstrate many of the implications of the theory. One of the contributions of the theory is the definition of the concepts of learner aptitude and instruction pacing within a framework that structures the empirical investigation of these concepts by means of experimental research.     

Incidental Learning of Melodic Structure of North Indian Music

Musical knowledge is largely implicit. It is acquired without awareness of its complex rules, through interaction with a large number of samples during musical enculturation. Whereas several studies explored implicit learning of mostly abstract and less ecologically valid features of Western music, very little work has been done with respect to ecologically valid stimuli as well as non‐Western music. The present study investigated implicit learning of modal melodic features in North Indian classical music in a realistic and ecologically valid way. It employed a cross‐grammar design, using melodic materials from two modes (rāgas) that use the same scale. Findings indicated that Western participants unfamiliar with Indian music incidentally learned to identify distinctive features of each mode. Confidence ratings suggest that participants' performance was consistently correlated with confidence, indicating that they became aware of whether they were right in their responses; that is, they possessed explicit judgment knowledge. Altogether our findings show incidental learning in a realistic ecologically valid context during only a very short exposure, they provide evidence that incidental learning constitutes a powerful mechanism that plays a fundamental role in musical acquisition.     

两种学习模式下类别学习的结果：原型和样例

利用“学习－迁移”的任务范式和单一特征类别判断技术,探讨了分类和推理两种类别学习模式的结果,比较了两种学习模式的效果和策略。研究表明：两种学习模式产生了不同的结果,分类学习的结果是样例,推理学习的结果是原型;在学习效果方面,分类学习比推理学习在达标比例上更高,但在进度上差异不显著;在策略运用方面,分类学习比推理学习更快地使用单维度策略,而在高水平策略的运用上,两者差异不显著     

Effects of grammar complexity on artificial grammar learning

The present study identified two aspects of complexity that have been manipulated in the implicit learning literature and investigated how they affect implicit and explicit learning of artificial grammars. Ten finite state grammars were used to vary complexity. The results indicated that dependency length is more relevant to the complexity of a structure than is the number of associations that have to be learned. Although implicit learning led to better performance on a grammaticality judgment test than did explicit learning, it was negatively affected by increasing complexity: Performance decreased as there was an increase in the number of previous letters that had to be taken into account to determine whether or not the next letter was a grammatical continuation. In particular, the results suggested that implicit learning of higher order dependencies is hampered by the presence of longer dependencies. Knowledge of first-order dependencies was acquired regardless of complexity and learning mode.