Contextual cueing of visual attention

Visual context information constrains what to expect and where to look, facilitating search for and recognition of objects embedded in complex displays. This article reviews a new paradigm called contextual cueing, which presents well-defined, novel visual contexts and aims to understand how contextual information is learned and how it guides the deployment of visual attention. In addition, the contextual cueing task is well suited to the study of the neural substrate of contextual learning. For example, amnesic patients with hippocampal damage are impaired in their learning of novel contextual information, even though learning in the contextual cueing task does not appear to rely on conscious retrieval of contextual memory traces. We argue that contextual information is important because it embodies invariant properties of the visual environment such as stable spatial layout information as well as object covariation information. Sensitivity to these statistical regularities allows us to interact more effectively with the visual world.     

Age and experience shape developmental changes in the neural basis of language-related learning

Very little is known about the neural underpinnings of language learning across the lifespan and how these might be modified by maturational and experiential factors. Building on behavioral research highlighting the importance of early word segmentation (i.e. the detection of word boundaries in continuous speech) for subsequent language learning, here we characterize developmental changes in brain activity as this process occurs online, using data collected in a mixed cross-sectional and longitudinal design. One hundred and fifty-six participants, ranging from age 5 to adulthood, underwent functional magnetic resonance imaging (fMRI) while listening to three novel streams of continuous speech, which contained either strong statistical regularities, strong statistical regularities and speech cues, or weak statistical regularities providing minimal cues to word boundaries. All age groups displayed significant signal increases over time in temporal cortices for the streams with high statistical regularities; however, we observed a significant right-to-left shift in the laterality of these learning-related increases with age. Interestingly, only the 5- to 10-year-old children displayed significant signal increases for the stream with low statistical regularities, suggesting an age-related decrease in sensitivity to more subtle statistical cues. Further, in a sample of 78 10-year-olds, we examined the impact of proficiency in a second language and level of pubertal development on learning-related signal increases, showing that the brain regions involved in language learning are influenced by both experiential and maturational factors.     

Enhanced Verbal Statistical Learning in Glossolalia

Glossolalia (“speaking in tongues”) is a rhythmic utterance of word-like strings of sounds, regularly occurring in religious mass gatherings or various forms of private religious practices (e.g., prayer and meditation). Although specific verbal learning capacities may characterize glossolalists, empirical evidence is lacking. We administered three statistical learning tasks (artificial grammar, phoneme sequence, and visual-response sequence) to 30 glossolalists and 30 matched control volunteers. In artificial grammar, participants decide whether pseudowords and sentences follow previously acquired implicit rules or not. In sequence learning, they gradually draw out rules from repeating regularities in sequences of speech sounds or motor responses. Results revealed enhanced artificial grammar and phoneme sequence learning performances in glossolalists compared to control volunteers. There were significant positive correlations between daily glossolalia activity and artificial grammar learning. These results indicate that glossolalists exhibit enhanced abilities to extract the statistical regularities of verbal information, which may be related to their unusual language abilities.     

Variations in intensity statistics for representational and abstract art, and for art from the Eastern and Western hemispheres

Two recent studies suggest that natural scenes and paintings show similar statistical properties. But does the content or region of origin of an artwork affect its statistical properties? We addressed this question by having judges place paintings from a large, diverse collection of paintings into one of three subject-matter categories using a forced-choice paradigm. Basic statistics for images whose caterogization was agreed by all judges showed no significant differences between those judged to be 'landscape' and 'portrait/still-life', but these two classes differed from paintings judged to be 'abstract'. All categories showed basic spatial statistical regularities similar to those typical of natural scenes. A test of the full painting collection (140 images) with respect to the works' place of origin (provenance) showed significant differences between Eastern works and Western ones, differences which we find are likely related to the materials and the choice of background color. Although artists deviate slightly from reproducing natural statistics in abstract art (compared to representational art), the great majority of human art likely shares basic statistical limitations. We argue that statistical regularities in art are rooted in the need to make art visible to the eye, not in the inherent aesthetic value of natural-scene statistics, and we suggest that variability in spatial statistics may be generally imposed by manufacture.     

How implicit is visual statistical learning?

In visual statistical learning, participants learn the statistical regularities present in a sequence of visual shapes. A recent study (Kim, Seitz, Feenstra, & Shams, 2009) suggests that visual statistical learning occurs implicitly, as it is not accompanied by conscious awareness of these regularities. However, that interpretation of the data depends on 2 unwarranted assumptions concerning the nature and sensitivity of the tasks used to measure learning. In a replication of this study, we used a 4-choice completion task as a direct measure of learning, in addition to an indirect measure consisting of a rapid serial visual presentation task. Moreover, binary confidence judgments were recorded after each completion trial. This way, we measured systematically the extent to which sequence knowledge was available to consciousness. Supporting the notion that the role of unconscious knowledge was overestimated in Kim et al.'s study, our results reveal that participants' performance cannot be exclusively accounted for by implicit knowledge. (PsycINFO Database Record (c) 2012 APA, all rights reserved).     

Cognitive phenomenology and conscious thought

How does mental content feature in conscious thought? I first argue that for a thought to be conscious the content of that thought must conscious, and that one has to appeal to cognitive phenomenology to give an adequate account of what it is for the content of a thought to be conscious. Sensory phenomenology cannot do the job. If one claims that the content of a conscious thought is unconscious, one is really claiming that there is no such thing as conscious thought. So one must either accept that there is such a thing as cognitive phenomenology, or deny the existence of conscious thought. Once it is clear that conscious thought requires cognitive phenomenology, there is a pressing question about the exact relationship between a thought’s cognitive phenomenological properties and its content. I conclude with a discussion of the nature of this relationship.     

Co‐occurrence statistics as a language‐dependent cue for speech segmentation

To what extent can language acquisition be explained in terms of different associative learning mechanisms? It has been hypothesized that distributional regularities in spoken languages are strong enough to elicit statistical learning about dependencies among speech units. Distributional regularities could be a useful cue for word learning even without rich language‐specific knowledge. However, it is not clear how strong and reliable the distributional cues are that humans might use to segment speech. We investigate cross‐linguistic viability of different statistical learning strategies by analyzing child‐directed speech corpora from nine languages and by modeling possible statistics‐based speech segmentations. We show that languages vary as to which statistical segmentation strategies are most successful. The variability of the results can be partially explained by systematic differences between languages, such as rhythmical differences. The results confirm previous findings that different statistical learning strategies are successful in different languages and suggest that infants may have to primarily rely on non‐statistical cues when they begin their process of speech segmentation.     

A Symbolic Model of the Nonconscious Acquisition of Information

This article presents counter evidence against Smolensky's theory that human intuitive/nonconscious congnitive processes can only be accurately explained in terms of subsymbolic computations carried out in artificial neural networks. We present symbolic learning models of two well-studied, complicated cognitive tasks involving nonconscious acquisition of information: learning production rules and artificial finite state grammars. Our results demonstrate that intuitive learning does not imply subsymbolic computation, and that the already well-established, perceived correlation between “conscious” and “symbolic” on the one hand, and between “nonconscious” and “subsymbolic” on the other, does not exist.     

How Many Mechanisms Are Needed to Analyze Speech? A Connectionist Simulation of Structural Rule Learning in Artificial Language Acquisition

Some empirical evidence in the artificial language acquisition literature has been taken to suggest that statistical learning mechanisms are insufficient for extracting structural information from an artificial language. According to the more than one mechanism (MOM) hypothesis, at least two mechanisms are required in order to acquire language from speech: (a) a statistical mechanism for speech segmentation; and (b) an additional rule-following mechanism in order to induce grammatical regularities. In this article, we present a set of neural network studies demonstrating that a single statistical mechanism can mimic the apparent discovery of structural regularities, beyond the segmentation of speech. We argue that our results undermine one argument for the MOM hypothesis.     

Isolated words enhance statistical language learning in infancy

Infants are adept at tracking statistical regularities to identify word boundaries in pause-free speech. However, researchers have questioned the relevance of statistical learning mechanisms to language acquisition, since previous studies have used simplified artificial languages that ignore the variability of real language input. The experiments reported here embraced a key dimension of variability in infant-directed speech. English-learning infants (8-10 months) listened briefly to natural Italian speech that contained either fluent speech only or a combination of fluent speech and single-word utterances. Listening times revealed successful learning of the statistical properties of target words only when words appeared both in fluent speech and in isolation; brief exposure to fluent speech alone was not sufficient to facilitate detection of the words' statistical properties. This investigation suggests that statistical learning mechanisms actually benefit from variability in utterance length, and provides the first evidence that isolated words and longer utterances act in concert to support infant word segmentation.     

Excluded middle and bivalence

I consider two related objections to the claim that the law of excluded middle does not imply bivalence. One objection claims that the truth predicate captured by supervaluation semantics is not properly motivated. The second objection says that even if it is, LEM still implies bivalence. I show that LEM does not imply bivalence in a supervaluational language. I also argue that considering supertruth as truth can be reasonably motivated.     

iMinerva: A Mathematical Model of Distributional Statistical Learning

Statistical learning refers to the ability to identify structure in the input based on its statistical properties. For many linguistic structures, the relevant statistical features are distributional: They are related to the frequency and variability of exemplars in the input. These distributional regularities have been suggested to play a role in many different aspects of language learning, including phonetic categories, using phonemic distinctions in word learning, and discovering non‐adjacent relations. On the surface, these different aspects share few commonalities. Despite this, we demonstrate that the same computational framework can account for learning in all of these tasks. These results support two conclusions. The first is that much, and perhaps all, of distributional statistical learning can be explained by the same underlying set of processes. The second is that some aspects of language can be learned due to domain‐general characteristics of memory.     

Word Meanings Evolve to Selectively Preserve Distinctions on Salient Dimensions

Words refer to objects in the world, but this correspondence is not one‐to‐one: Each word has a range of referents that share features on some dimensions but differ on others. This property of language is called underspecification. Parts of the lexicon have characteristic patterns of underspecification; for example, artifact nouns tend to specify shape, but not color, whereas substance nouns specify material but not shape. These regularities in the lexicon enable learners to generalize new words appropriately. How does the lexicon come to have these helpful regularities? We test the hypothesis that systematic backgrounding of some dimensions during learning and use causes language to gradually change, over repeated episodes of transmission, to produce a lexicon with strong patterns of underspecification across these less salient dimensions. This offers a cultural evolutionary mechanism linking individual word learning and generalization to the origin of regularities in the lexicon that help learners generalize words appropriately.     

Learning at a distance II. Statistical learning of non-adjacent dependencies in a non-human primate

In earlier work we have shown that adults, infants, and cotton-top tamarin monkeys are capable of computing the probability with which syllables occur in particular orders in rapidly presented streams of human speech, and of using these probabilities to group adjacent syllables into word-like units. We have also investigated adults' learning of regularities among elements that are not adjacent, and have found strong selectivities in their ability to learn various kinds of non-adjacent regularities. In the present paper we investigate the learning of these same non-adjacent regularities in tamarin monkeys, using the same materials and familiarization methods. Three types of languages were constructed. In one, words were formed by statistical regularities between non-adjacent syllables. Words contained predictable relations between syllables 1 and 3; syllable 2 varied. In a second type of language, words were formed by statistical regularities between non-adjacent segments. Words contained predictable relations between consonants; the vowels varied. In a third type of language, also formed by regularities between non-adjacent segments, words contained predictable relations between vowels; the consonants varied. Tamarin monkeys were exposed to these languages in the same fashion as adults (21 min of exposure to a continuous speech stream) and were then tested in a playback paradigm measuring spontaneous looking (no reinforcement). Adult subjects learned the second and third types of language easily, but failed to learn the first. However, tamarin monkeys showed a different pattern, learning the first and third type of languages but not the second. These differences held up over multiple replications, using different sounds instantiating each of the patterns. These results suggest differences among learners in the elementary units perceived in speech (syllables, consonants, and vowels) and/or the distance over which such units can be related, and therefore differences among learners in the types of patterned regularities they can acquire. Such studies with tamarins open interesting questions about the perceptual and computational capacities of human learners that may be essential for language acquisition, and how they may differ from those of non-human primates.     

Statistical computations over a speech stream in a rodent

Statistical learning is one of the key mechanisms available to human infants and adults when they face the problems of segmenting a speech stream (Saffran, Aslin, & Newport, 1996) and extracting long-distance regularities (G6mez, 2002; Pe?a, Bonatti, Nespor, & Mehler, 2002). In the present study, we explore statistical learning abilities in rats in the context of speech segmentation experiments. In a series of five experiments, we address whether rats can compute the necessary statistics to be able to segment synthesized speech streams and detect regularities associated with grammatical structures. Our results demonstrate that rats can segment the streams using the frequency of co-occurrence (not transitional probabilities, as human infants do) among items, showing that some basic statistical learning mechanism generalizes over nonprimate species. Nevertheless, rats did not differentiate among test items when the stream was organized over more complex regularities that involved nonadjacent elements and abstract grammar-like rules.     

On medicine as a human science

All the powerful influences exertedby the subjective-interpersonal dimension onthe organic or technical-functional dimensionof sickness and health do not make anintersubjective test concerning medicaltherapeutic results impossible. Theseinfluences are not arbitrary; on the contrary,they obey laws that are de facto sufficientlystable to allow predictions and explanationssimilar to those of experimental sciences.While, in this respect, the rules concerninghuman action are analogous to the scientificlaws of nature, they can at any time be revokedby becoming aware of them. Law-like andreproducible regularities in the sciences ofman are by no means separated from a patient'spersonal-hermeneutic mediation. This makes itpossible for human beings to modify, improve orsometimes even entirely (or better almostentirely) suspend these psychological,sociological, ethnological, medical,regularities. For this reason the sciences ofman including medicine are under the obligationof constantly inspecting the continuingvalidity of the rules on which theirpredictions and explanations are based, namelyby indirect, statistical methods. Thisrequires a synergistic collaboration ofextra-clinical and clinical tests through whichmedicine can obtain a good level ofintersubjective testability.     

Generation of hypotheses in Wason's 2-4-6 task: an information theory approach.

We explored the "context of discovery" in Wason's 2-4-6 task, focusing on how the first hypothesis is generated. According to Oaksford and Chater (1994a) people generate hypotheses extracting "common features", or regularities, from the available triples, but their model does not explain why some regularities contribute to the hypothesis more than do other regularities. Our conjecture is that some regularities contribute to the hypothesis more than do other regularities because people estimate the amount of information in the perceived regularities and try to preserve as much information as possible in their initial hypotheses. Experiment 1, which used two initial triples, showed that the presence of high-information relational regularities in the initial triples affected the information in the initial hypotheses more than did the presence of low-information object regularities. Experiment 2 extended the results to the classic situation in which only one initial triple is given. It also suggested that amount of information is the only aspect of the structure of the triple that affects hypotheses generation. Experiment 3 confirmed the latter finding: Although relations are commonly distinguished between first-order and higher order relations, the latter being most important for generating hypotheses (Gentner, 1983), higher order relations do have an effect on Wason's 2-4-6 task only if their presence incre ases information. In the conclusion we discuss the statistical soundness of human hypotheses generation processes, and we ask an unanswered question: Amount of information explains why some regularities are preferred to others, but only within a set of "nonarbitrary" regularities; there are object regularities that are rich in information content, but are considered "arbitrary", and are not used in generating hypotheses. Which formal property can distinguish between these two sets of regularities?     

Phonological encoding in the lexical decision task

In lexical decision experiments, subjects have difficulty in responding NO to non-words which are pronounced exactly like English words (e.g. BRANE). This does not necessarily imply that access to a lexical entry ever occurs via a phonological recoding of a visually-presented word. The phonological recoding procedure might be so slow that when the letter string presented is a word, access to its lexical entry via a visual representation is always achieved before phonological recoding is completed. If prelexical phonological recodings are produced by using grapheme-phoneme correspondence rules, such recodings can only occur for words which conform to these rules (regular words), since applications of the rules to words which do not conform to the rules (exception words) produce incorrect phonological representations. In two experiments, it was found that time to achieve lexical access (as measured by YES latency in a lexical decision task) was equivalent for regular words and exception words. It was concluded that access to lexical entries in lexical decision experiments of this sort does not proceed by sometimes or always phonologically recoding visually-presented words.     

Language-specific listening

Languages differ in their phonological structure and physcholinguists have begun to explore the conseqence, of this fact for speech perception. We review research documenting that listeners attune their perceptual processes finaly to exploit the phonological regularities of their nativ language. As a consequence, these perceptual process are fill-adapted to listening to languages that do not display such, regularities. Thus, not only do late language-learners have trouble speaking a second language, also they do not hear it as native speakers do; worse, they apply their native language listening prosedures which may actually interfere with successful processing of the non-native input. We also present data from studies on infants showing that the initial attuning occurs early in life; very yong infants are sensitive to the relevant phonological regularities which distinguish different languages, and quickly distinguish the native language of their environment from languages with different regularities.