Downscaling of technology causes signal integrity problems due to crosstalk between closely-spaced interconnect lines. Existing crosstalk estimation and optimization techniques operate at the layout-level of circuits and fail to utilize the efficient design-space exploration at the high-level. To address this, we propose word-level statistical techniques which estimate crosstalk between bus lines: (1) Given a data stream, the first technique simply counts the number of crosstalk events on each bus line. The drawback of this technique is that the execution time is proportional to the stream length. This is overcome by the second enumerative technique which is purely statistical in nature. (2) Given word-level statistics, we estimate the bit-level crosstalk probability of bus lines. (3) We further speedup the statistical method using a non-enumerative technique by linearizing its complexity with respect to the bus width. Average errors of less than 15% are obtained for bus-widths ranging from 8b to 32b while execution times are reduced by two orders of magnitude, compared to HSPICE. We then measure the crosstalk susceptibility of nets in the post global routing phase (performed using CADENCE Silicon Ensemble), prior to detailed routing using (1) Pt , the probability of crosstalk on victims in different regions along their route; and (2) Vpeak, the maximum crosstalk noise amplitude experienced by victims along their route. Pt is estimated using the fast and accurate statistical estimator we previously proposed. Vpeak is estimated by predicting the cross-coupling capacitances between neighboring wires, using their global routing information. Average errors are less than 8%, compared to HSPICE. We combine the crosstalk susceptibility values from individual regions along a victim wire’s route, to obtain a single susceptibility value for the entire wire. Further, we propose a register binding technique during high-level synthesis to minimize crosstalk at the register outputs in the RT-level design. It involves modification of the cliquepartitioning algorithm to make crosstalk-aware choices of edges to be mapped to the same register. RT-level comparisons between the regular and crosstalk-aware designs show upto 16% reduction in crosstalk activity at the register outputs.

The aim of this study is to investigate the occurrence of sexist language-use in the British online newspaper The Daily Mail. The material consists of 162 articles that were analysed by using feminist stylistics. The scope of the study was limited to selected features from feminist stylistics at word- and discourse-level. The features of linguistic sexism analysed were the use of gendered generic words, naming of females and males and how female and male characters are described. The gender of the journalists was also analysed to examine if it affected the language-use in terms of sexism. The results show that linguistic sexism is expressed to some extent at both word-level and discourse-level. At word-level linguistic sexism is expressed inthe generic use of some masculine words, the difference of how first name and surname are used to refer to women and men and in the use of titles. At the level of discourse linguistic sexism is expressed in the difference of how women and men are referred to in terms of their relationship to others and in terms of appearance. The gender of the journalist did not show any significance for the language-use in terms of sexism. Considering the limited material of the study, the results might not be suitable for generalisations. The results are nonetheless interesting and it can be concluded that the toolkit of feminist stylistic is relevant to this day and that linguistic sexism exists to some extent in the online version of The Daily Mail.

Cette thèse est centrée sur l'évaluation des capacités de lecture et reliées à la lecture des collégiens normo-lecteurs français. Le premier objectif est d'identifier les capacités engagées dans la compréhension écrite. Dans les premiers grades (6 et 7), les capacités de compréhension écrite sont surtout expliquées par celles de compréhension orale, de conscience morphologique et de lecture de mots irréguliers, mais pas par la lecture de pseudomots, quelle que soit la mesure (précision ou temps). Ce dernier résultat se retrouve dans les grades supérieurs (8 et 9) dans lesquels la lecture de mots irréguliers n'a plus d'incidence sur la compréhension écrite. Le second objectif est d'élaborer un outil diagnostique des capacités de lecture, et des capacités reliées, chez les collégiens (6 à 9). L'examen des capacités d'identifications des mots écrits montre que les effets de régularité et de lexicalité sont significatifs, quelle que soit la mesure, et le niveau scolaire. Les effets de longueur varient selon la lexicalité: quel que soit le niveau scolaire, les pseudomots longs sont lus moins précisément et moins rapidement que les courts alors que les mots irréguliers longs ne sont jamais pénalisés. Le niveau de lecture (mesuré à l'aide d'un test Français de référence) est déterminé surtout par les capacités de lecture de pseudomots, et également par celles de conscience phonémique mais uniquement lorsque les temps sont pris en compte. La construction cohérente de l'outil est vérifiée à l'aide d'une classification hiérarchique de variables. L'ensemble des résultats souligne l'importance de considérer pour toutes évaluations les temps de traitement en plus de la précision. / The aim of this thesis was to evaluate the reading and reading-related skills of French middle school students (Grades 6 to 9). The first objective was to identify the capacity involved in reading comprehension at the middle school level. In the first grades (Grades 6 and 7), the results show that reading comprehension skills are largely predicted by listening comprehension, morphological awareness, and irregular word reading, but not pseudoword reading skills, whatever the measure (accuracy or time). This latter result is found in the higher grades (Grades 8 and 9) in which irregular word reading has no impact on reading comprehension, again regardless of the measure. The second objective was to develop a diagnostic evaluation tool for word-level reading and reading-related skills in middle school students. The examination of word-level reading skills at middle school showed significant effects of both regularity and lexicality, whatever the measure and independently of grade. The effect of length depends on lexicality: long pseudowords are read less accurately and more slowly than short ones, whereas long irregular words are read as well as short ones. Reading level (assessed by a French "gold standard" test) is mainly predicted by pseudoword reading skills, and also by phonemic awareness, for response times. The consistent structure (validity) of the tool is verified using a new statistical method: hierarchical classification of variables. The overall results underline the importance of considering response times in addition to accuracy in all assessments.

The purpose of the study was to investigate the influence of THRASS (Teaching Handwriting, Reading and Spelling Skills) on the word level accuracy skills of a group of grade 2 learners. Word level accuracy is one sub skill in learning to read and is an indicator of the word recognition abilities of the child. THRASS is a program that has been designed to systematically teach phonics and, thus, teaches the basic building blocks of word sounds and structure so as to improve the child’s decoding ability and word recognition ability. The research took place within the positivist paradigm and the methodology is quantitative in nature. The data collection method took the form of a one group pretest-posttest design, where a standardised reading test was administered prior to exposing the participants to the THRASS Program and then readministered one year later on the same group of learners. Data analysis took the form of statistical analysis to investigate any statistical significant difference in the word level accuracy skills of those Grade 2 learners. The result showed that over the period of a year the average reading accuracy age for the target population increased by four months. However, after statistical analysis the difference was not statistically significant. The Null Hypothesis that; exposing a group of Grade 2 learners to the THRASS Program for a period of one year will have no statistically significant influence on their word level accuracy skills cannot be rejected . However, the changes both in average reading accuracy as well as error patterns have inspired recommendations for further research. Copyright / Dissertation (MEd)--University of Pretoria, 2010. / Educational Psychology / unrestricted

Second language (L2) learners are a growing population in Canadian school systems, and acquisition of literacy skills is critical to their success in Canadian society. While much research has been devoted to writing development in first language (L1) learners, text-level writing remains relatively underexplored in L2 populations. The present study sought to address this gap by considering the relative contributions of component oral language and word-level literacy skills to writing in elementary students speaking English as a first (EL1) or second (EL2) language. A sample of 124 kindergarten students (56 EL1, 68 EL2) and 112 grade three students (51 EL1, 61 EL2) completed a battery of standardized measures assessing oral language, word-level literacy, and writing skills. An ordinary least squares (OLS) regression-based mediation path analysis was used to test associations among oral language, word-level literacy, and writing skills in each group. Results indicated that word-level literacy skills had a significant direct effect on writing in all groups, but that oral language had no significant direct effect on writing in any groups. Instead, the effect of oral language on writing was significantly mediated by word-level skills in the kindergarten EL1 and EL2 groups, and the grade three EL1 group. The indirect effect of oral language on writing through word-level skills was not significant in the grade three EL2 group. Despite this, no significant differences in variable associations were found between EL1 and EL2 groups in either grade. Oral language skills were additionally found to have a significant effect on word-level literacy skills in the kindergarten EL1 and EL2 groups and the grade three EL1 group; the significance of this effect in the grade three EL2 group was unclear. Results of this study are discussed in relation to existing literature, and existing theories of L1 and L2 writing. / Graduate

Text in Afrikaans / Die doel van hierdie studie was om te bepaal tot watter mate ’n suiwer patroongebaseerde benadering tot woordafbreking bevredigende resultate lewer. Die masjienleertegnieke kunsmatige neurale netwerke, beslissingsbome en die TEX-algoritme is ondersoek aangesien dit met letterpatrone uit woordelyste afgerig kan word om lettergreep- en saamgesteldewoordverdeling te doen. ’n Leksikon van Afrikaanse woorde is uit ’n korpus van elektroniese teks genereer. Om lyste vir lettergreep- en saamgesteldewoordverdeling te kry, is woorde in die leksikon in lettergrepe verdeel en saamgestelde woorde is in hul samestellende dele verdeel. Uit elkeen van hierdie lyste van ±183 000 woorde is ±10 000 woorde as toetsdata gereserveer terwyl die res as afrigtingsdata gebruik is. ’n Rekursiewe algoritme is vir saamgesteldewoordverdeling ontwikkel. In hierdie algoritme word alle ooreenstemmende woorde uit ’n verwysingslys (die leksikon) onttrek deur stringpassing van die begin en einde van woorde af. Verdelingspunte word dan op grond van woordlengte uit die samestelling van begin- en eindwoorde bepaal. Die algoritme is uitgebrei deur die tekortkominge van hierdie basiese prosedure aan te spreek. Neurale netwerke en beslissingsbome is afgerig en variasies van beide tegnieke is ondersoek om die optimale modelle te kry. Patrone vir die TEX-algoritme is met die OPatGen-program gegenereer. Tydens toetsing het die TEX-algoritme die beste op beide lettergreep- en saamgesteldewoordverdeling presteer met 99,56% en 99,12% akkuraatheid, respektiewelik. Dit kan dus vir woordafbreking gebruik word met min risiko vir afbrekingsfoute in gedrukte teks. Die neurale netwerk met 98,82% en 98,42% akkuraatheid op lettergreep- en saamgesteldewoordverdeling, respektiewelik, is ook bruikbaar vir lettergreepverdeling, maar dis meer riskant. Ons het bevind dat beslissingsbome te riskant is om vir lettergreepverdeling en veral vir woordverdeling te gebruik, met 97,91% en 90,71% akkuraatheid, respektiewelik. ’n Gekombineerde algoritme is ontwerp waarin saamgesteldewoordverdeling eers met die TEXalgoritme gedoen word, waarna die resultate van lettergreepverdeling deur beide die TEXalgoritme en die neurale netwerk gekombineer word. Die algoritme het 1,3% minder foute as die TEX-algoritme gemaak. ’n Toets op gepubliseerde Afrikaanse teks het getoon dat die risiko vir woordafbrekingsfoute in teks met gemiddeld tien woorde per re¨el ±0,02% is. / The aim of this study was to determine the level of success achievable with a purely pattern based approach to hyphenation in Afrikaans. The machine learning techniques artificial neural networks, decision trees and the TEX algorithm were investigated since they can be trained with patterns of letters from word lists for syllabification and decompounding. A lexicon of Afrikaans words was extracted from a corpus of electronic text. To obtain lists for syllabification and decompounding, words in the lexicon were respectively syllabified and compound words were decomposed. From each list of ±183 000 words, ±10 000 words were reserved as testing data and the rest was used as training data. A recursive algorithm for decompounding was developed. In this algorithm all words corresponding with a reference list (the lexicon) are extracted by string fitting from beginning and end of words. Splitting points are then determined based on the length of reassembled words. The algorithm was expanded by addressing shortcomings of this basic procedure. Artificial neural networks and decision trees were trained and variations of both were examined to find optimal syllabification and decompounding models. Patterns for the TEX algorithm were generated by using the program OPatGen. Testing showed that the TEX algorithm performed best on both syllabification and decompounding tasks with 99,56% and 99,12% accuracy, respectively. It can therefore be used for hyphenation in Afrikaans with little risk of hyphenation errors in printed text. The performance of the artificial neural network was lower, but still acceptable, with 98,82% and 98,42% accuracy for syllabification and decompounding, respectively. The decision tree with accuracy of 97,91% on syllabification and 90,71% on decompounding was found to be too risky to use for either of the tasks A combined algorithm was developed where words are first decompounded by using the TEX algorithm before syllabifying them with both the TEX algoritm and the neural network and combining the results. This algoritm reduced the number of errors made by the TEX algorithm by 1,3% but missed more hyphens. Testing the algorithm on Afrikaans publications showed the risk for hyphenation errors to be ±0,02% for text assumed to have an average of ten words per line. / Decision Sciences / D. Phil. (Operational Research)

Title: Hybrid Deep Question Answering Author: Ahmad Aghaebrahimian Institute: Institute of Formal and Applied Linguistics Supervisor: RNDr. Martin Holub, Ph.D., Institute of Formal and Applied Lin- guistics Abstract: As one of the oldest tasks of Natural Language Processing, Question Answering is one of the most exciting and challenging research areas with lots of scientific and commercial applications. Question Answering as a discipline in the conjunction of computer science, statistics, linguistics, and cognitive science is concerned with building systems that automatically retrieve answers to ques- tions posed by humans in a natural language. This doctoral dissertation presents the author's research carried out in this discipline. It highlights his studies and research toward a hybrid Question Answering system consisting of two engines for Question Answering over structured and unstructured data. The structured engine comprises a state-of-the-art Question Answering system based on knowl- edge graphs. The unstructured engine consists of a state-of-the-art sentence-level Question Answering system and a word-level Question Answering system with results near to human performance. This work introduces a new Question An- swering dataset for answering word- and sentence-level questions as well. Start- ing from a...

This study was designed to provide information on the reading instructional practices of 36 first grade teachers in nine schools that serve predominantly Spanish-speaking and ELL students in a southeastern U.S. school district. The purpose of this investigation was to describe teaching practices employed during English language arts instruction and to examine their use in relation to instructional grouping strategies, teacher language use, and student engagement. Participating classrooms were observed three times throughout the 2006-07 school year. Data were collected via the Timed Observations of Student Engagement/Language (TO/SEL) classroom observation instrument (Foorman & Schatchneider, 2003). Paired sample t-tests, multivariate analyses of variance (MANOVA), and multiple regression analyses were employed to investigate the relationship among the following observed variables: allocation of reading instructional time, grouping strategies, teacher language use and student engagement. Several key findings emerged. Participating teachers spent a greater amount of time on meaning-focused reading instruction (i.e., writing, reading texts, reading comprehension) than on code-focused reading instruction (i.e., word work, spelling, reading fluency, phonemic awareness), both during all four observed grouping strategies and after controlling for individual student seat work. In addition, of five key collapsed instructional variables (word work/spelling, oral language, writing, reading texts, and reading comprehension), teachers spent most time on word work/spelling (19%) and writing (18%). Reading texts and reading comprehension instruction together comprised 26% of total instructional time. Whole class instruction was the grouping strategy of choice among teachers (65% of total observed time); in sharp contrast, teachers spent 11% of observed time engaged in small group instruction, despite research findings supporting the effectiveness of sound grouping instruction. In addition, as little as 1% of teachers' total instructional time was spent in oral language/discussion, and 6% of total instructional time was spent in vocabulary instruction. The results also demonstrated little variation in teacher language use. Thus, evidence of "codeswitching" was not significant. Student engagement was high- 91% of total time students were observed; and was highest during writing and word work/spelling instruction. More longitudinal research is warranted that further explores precisely documented teacher reading instructional practices in relation to student outcomes with culturally and linguistically diverse student populations. Implications for practice include teacher training and professional development on managing small group instruction, and incorporating additional oral language/discussion, vocabulary and meaningful tasks into daily classroom activities.

Text in Afrikaans / Die doel van hierdie studie was om te bepaal tot watter mate ’n suiwer patroongebaseerde benadering tot woordafbreking bevredigende resultate lewer. Die masjienleertegnieke kunsmatige neurale netwerke, beslissingsbome en die TEX-algoritme is ondersoek aangesien dit met letterpatrone uit woordelyste afgerig kan word om lettergreep- en saamgesteldewoordverdeling te doen. ’n Leksikon van Afrikaanse woorde is uit ’n korpus van elektroniese teks genereer. Om lyste vir lettergreep- en saamgesteldewoordverdeling te kry, is woorde in die leksikon in lettergrepe verdeel en saamgestelde woorde is in hul samestellende dele verdeel. Uit elkeen van hierdie lyste van ±183 000 woorde is ±10 000 woorde as toetsdata gereserveer terwyl die res as afrigtingsdata gebruik is. ’n Rekursiewe algoritme is vir saamgesteldewoordverdeling ontwikkel. In hierdie algoritme word alle ooreenstemmende woorde uit ’n verwysingslys (die leksikon) onttrek deur stringpassing van die begin en einde van woorde af. Verdelingspunte word dan op grond van woordlengte uit die samestelling van begin- en eindwoorde bepaal. Die algoritme is uitgebrei deur die tekortkominge van hierdie basiese prosedure aan te spreek. Neurale netwerke en beslissingsbome is afgerig en variasies van beide tegnieke is ondersoek om die optimale modelle te kry. Patrone vir die TEX-algoritme is met die OPatGen-program gegenereer. Tydens toetsing het die TEX-algoritme die beste op beide lettergreep- en saamgesteldewoordverdeling presteer met 99,56% en 99,12% akkuraatheid, respektiewelik. Dit kan dus vir woordafbreking gebruik word met min risiko vir afbrekingsfoute in gedrukte teks. Die neurale netwerk met 98,82% en 98,42% akkuraatheid op lettergreep- en saamgesteldewoordverdeling, respektiewelik, is ook bruikbaar vir lettergreepverdeling, maar dis meer riskant. Ons het bevind dat beslissingsbome te riskant is om vir lettergreepverdeling en veral vir woordverdeling te gebruik, met 97,91% en 90,71% akkuraatheid, respektiewelik. ’n Gekombineerde algoritme is ontwerp waarin saamgesteldewoordverdeling eers met die TEXalgoritme gedoen word, waarna die resultate van lettergreepverdeling deur beide die TEXalgoritme en die neurale netwerk gekombineer word. Die algoritme het 1,3% minder foute as die TEX-algoritme gemaak. ’n Toets op gepubliseerde Afrikaanse teks het getoon dat die risiko vir woordafbrekingsfoute in teks met gemiddeld tien woorde per re¨el ±0,02% is. / The aim of this study was to determine the level of success achievable with a purely pattern based approach to hyphenation in Afrikaans. The machine learning techniques artificial neural networks, decision trees and the TEX algorithm were investigated since they can be trained with patterns of letters from word lists for syllabification and decompounding. A lexicon of Afrikaans words was extracted from a corpus of electronic text. To obtain lists for syllabification and decompounding, words in the lexicon were respectively syllabified and compound words were decomposed. From each list of ±183 000 words, ±10 000 words were reserved as testing data and the rest was used as training data. A recursive algorithm for decompounding was developed. In this algorithm all words corresponding with a reference list (the lexicon) are extracted by string fitting from beginning and end of words. Splitting points are then determined based on the length of reassembled words. The algorithm was expanded by addressing shortcomings of this basic procedure. Artificial neural networks and decision trees were trained and variations of both were examined to find optimal syllabification and decompounding models. Patterns for the TEX algorithm were generated by using the program OPatGen. Testing showed that the TEX algorithm performed best on both syllabification and decompounding tasks with 99,56% and 99,12% accuracy, respectively. It can therefore be used for hyphenation in Afrikaans with little risk of hyphenation errors in printed text. The performance of the artificial neural network was lower, but still acceptable, with 98,82% and 98,42% accuracy for syllabification and decompounding, respectively. The decision tree with accuracy of 97,91% on syllabification and 90,71% on decompounding was found to be too risky to use for either of the tasks A combined algorithm was developed where words are first decompounded by using the TEX algorithm before syllabifying them with both the TEX algoritm and the neural network and combining the results. This algoritm reduced the number of errors made by the TEX algorithm by 1,3% but missed more hyphens. Testing the algorithm on Afrikaans publications showed the risk for hyphenation errors to be ±0,02% for text assumed to have an average of ten words per line. / Decision Sciences / D. Phil. (Operational Research)