No description available.

No description available.

No description available.

No description available.

No description available.

This thesis explores implicit learning in children with developmental dyslexia. While specific cognitive abilities such as phonology and memory have been extensively explored in developmental dyslexia more global, fundamental abilities are rarely studied. A literature review is reported, which indicates that there is a gap in the study of more generic abilities highlighting at the same time, the need of investigating developmental dyslexia in the kind of contemporary context that learning literature provides. Implicit learning seems a suitable paradigm case to explore global abilities in developmental dyslexia since there have been suggestions that learning becomes more implicit in nature after explicit instruction. Based on the proposed relationship between implicit learning and reading, it is argued that impairments in the mechanisms of implicit learning could mediate selective weaknesses in reading performance in developmental dyslexia. The present thesis tests this argument in a series of three studies that are composed of five linked experiments. Together the three studies reported in the present thesis provide evidence for the implicit learning abilities in children with and without developmental dyslexia. The results suggest that while implicit learning abilities are found intact in typically developing children, children with developmental dyslexia on the other hand, might be facing an implicit learning deficit that could affect their reading performance and inhibit them from reaching their full learning potential.

Cette thèse pour but de développer des méthodes de segmentation pour des scènes fortement structurées (ex. bâtiments et environnements urbains) ou faiblement structurées (ex. paysages ou objets naturels). En particulier, les images de bâtiments peuvent être décrites en termes d'une grammaire de formes, et une dérivation de cette grammaire peut être inférée pour obtenir une segmentation d'une image. Cependant, il est difficile et long d'écrire de telles grammaires. Pour répondre à ce problème, nous avons développé une nouvelle méthode qui permet d'apprendre automatiquement une grammaire à partir d'un ensemble d'images et de leur segmentation associée. Des expériences montrent que des grammaires ainsi apprises permettent une inférence plus rapide et produisent de meilleures segmentations. Nous avons également étudié une méthode basée sur les auto-contextes pour segmenter des scènes fortement structurées et notamment des images de bâtiments. De manière surprenante, même sans connaissance spécifique sur le type de scène particulier observé, nous obtenons des gains significatifs en qualité de segmentation sur plusieurs jeux de données. Enfin, nous avons développé une technique basée sur les réseaux de neurones convolutifs (CNN) pour segmenter des images de scènes faiblement structurées. Un filtrage adaptatif est effectué à l'intérieur même du réseau pour permettre des dépendances entre zones d'images distantes. Des expériences sur plusieurs jeux de données à grande échelle montrent là aussi un gain important sur la qualité de segmentation / The aim of this thesis is to develop techniques for segmenting strongly-structuredscenes (e.g. building images) and weakly-structured scenes (e.g. natural images). Buildingimages can naturally be expressed in terms of grammars and inference is performed usinggrammars to obtain the optimal segmentation. However, it is difficult and time consum-ing to write such grammars. To alleviate this problem, a novel method to automaticallylearn grammars from a given training set of image and ground-truth segmentation pairs isdeveloped. Experiments suggested that such learned grammars help in better and fasterinference. Next, the effect of using grammars for strongly structured scenes is explored.To this end, a very simple technique based on Auto-Context is used to segment buildingimages. Surprisingly, even with out using any domain specific knowledge, we observedsignificant improvements in terms of performance on several benchmark datasets. Lastly,a novel technique based on convolutional neural networks is developed to segment imageswithout any high-level structure. Image-adaptive filtering is performed within a CNN ar-chitecture to facilitate long-range connections. Experiments on different large scale bench-marks show significant improvements in terms of performance

Previous research on speakers' abilities to track non-adjacent dependencies (e.g., vowels or consonants that co-occur across syllables) in artificial grammar learning (AGL) tasks has shown that the acquisition of these patterns is extremely difficult (e.g. Newport&Aslin 2004; Gómez 2002; Bonatti, PenÞa, Nespor&Mehler 2005). One assumption made in this literature is that all speakers of all languages should be capable of tracking these patterns even when the native language of those speakers contains no such non-adjacent dependencies. This dissertation questions this assumption by testing whether native Khalkha Mongolian speakers attend to and track the frequency of vowel patterns and harmonic class size in their language. It also tests their ability to acquire non-adjacent vocalic dependencies in AGL tasks.Because Khalkha displays [ATR] vowel harmony (Svantesson, Tsendina, Karlsson&Franzén 2005) which restricts vowel co-occurrences, it was hypothesized that Khalkha speakers are biased towards attending to the frequency and form with which these vowel patterns occur. The results of three experiments indicated that Khalkha speakers both attend to and track the frequency with which vowel patterns occur. These results also indicate that Khalkha speakers build abstract categories based on the relative token numbers of [+ATR] and [-ATR]harmonic spans.Khalkha speakers were further tested in three experiments which focused on speakers' ability to acquire novel non-adjacent vocalic dependencies in AGL tasks. The results indicated that participants successfully acquired vocalic dependencies (both harmonic and disharmonic) in all three experiments. These results indicate that Khalkha speakers' attention is biased towards vowels, regardless of harmonic status of the item.Collectively, these results highlight the role of language-specific phonology in the ways that speakers abstract and utilize phonological information. The special status of harmonic vowel patterns and harmonic class size are new variables with which to conduct future research on vowel harmonic languages and with vowel harmonic language speakers. The effects of language-specific phonology on speech perception and lexical access must be considered a crucial aspect in future psycholinguistic research, particularly in regards to the aspects of language toward which speakers attend.

In this thesis, we compare the description length of different grammars, and extend the research of automatic grammar learning to the grammar production of Stanford parser. In our research before, we have introduced that how to minimize the description length of the grammar which is generated from the Academia Sinica Balanced Corpus. Based on the concept of data compression, the encoding method in our research is effective in reducing the description length of a text corpus. Moreover, we further discussed about the description length of two special cases of context-free grammars: exhaustive and recursive. The exhaustive grammar is that for every distinct sentence in the corpus is derived, and the recursive one covers all strings. In our research of this thesis, we use a parsing tool called "Stanford parser" to parse sentences and generate grammar rules. We also compare the description length of the grammar parsed by machine with the grammar fixed by artificial. In one of the experiments, we use Stanford parser to parse ASBC corpus, and the description length is 53.0Mb. The description length of rule is only 52,683. In the other experiment, we use Stanford parser to parse Sinica Treebank and compare the description length of the generated grammar with the origin. The result shows that the description length of grammar of the Sinica Treebank is 2.76Mb, and the grammar generated by Stanford parser is 4.02Mb.

abstract: Incidental learning of sequential information occurs in visual, auditory and tactile domains. It occurs throughout our lifetime and even in nonhuman species. It is likely to be one of the most important foundations for the development of normal learning. To date, there is no agreement as to how incidental learning occurs. The goal of the present set of experiments is to determine if visual sequential information is learned in terms of abstract rules or stimulus-specific details. Two experiments test the extent to which interaction with the stimuli can influence the information that is encoded by the learner. The results of both experiments support the claim that stimulus and domain specific details directly shape what is learned, through a process of tuning the neuromuscular systems involved in the interaction between the learner and the materials. / Dissertation/Thesis / M.A. Psychology 2013