No description available.

No description available.

Grapheme-based speech recognition systems are faster to develop, but typically do not reach the same level of performance as phoneme-based systems. Using Afrikaans speech recognition as a case study, we first analyse the reasons for the discrepancy in performance, before introducing a technique for improving the performance of standard grapheme-based systems. It is found that by handling a relatively small number of irregular words through phoneme-to-grapheme (P2G) transliteration – transforming the original orthography of irregular words to an ‘idealised’ orthography – grapheme-based accuracy can be improved. An analysis of speech recognition accuracy based on word categories shows that P2G transliteration succeeds in improving certain word categories in which grapheme-based systems typically perform poorly, and that the problematic categories can be identified prior to system development. An evaluation is offered of when category-based P2G transliteration is beneficial and methods to implement the technique in practice are discussed. Comparative results are obtained for a second language (Vietnamese) in order to determine whether the technique can be generalised. / MSc (Computer Science) North-West University, Vaal Triangle Campus, 2014

Nesta dissertação tomamos por objeto de estudo a romanização, em português, de palavras árabes (e·g, termos, topônimos e antropônimos). Estudamos por que alguns nomes são grafados de maneiras diferentes em um ou outro jornal, por exemplo, analisamos esquemas existentes de romanização e propomos um sistema que seja não apenas conciso e regular, mas também abrangente e minucioso, podendo ser usado tanto por leigos quanto por acadêmicos. Os seis capítulos encontram-se divididos em duas partes. Na primeira parte, abordamos a relação entre transposição ortográfica (i·e, transcrição, transliteração romanização e latinização) e tradução; apresentamos a língua árabe e sua escrita; ressaltamos elementos pertinentes da ortografia do português brasileiro. Na segunda parte, cotejamos oito esquemas de romanização árabeportuguês, apontando suas (des)semelhanças; apresentamos nosso sistema de transposição ortográfica, em seus quatro módulos (leigo simplificado, leigo comum, acadêmico amplo e acadêmico estrito), incluindo exemplos de aplicação em nomes, expressões e um breve texto; concluímos antecipando os desdobramentos da proposta. / This master\'s thesis presents our research concerning the romanization of Arabic words (e·g, technical terms, toponyms and anthroponyms) into Portuguese. We have first studied why the way some nouns and proper names are written varies from source to source, then analysed some of the existing romanization schemes and, finally, proposed a system both concise and regular as well as comprehensive and detailed, one which is suitable for laymen and scholars alike. there are six chapters, distributed in two parts. In the first part, we see four orthographic transposition methods (transcription, transliteration, romanization and latinization) and how they relate to translation; the Arabic language and its writing script; and relevant aspects of the Brazilian Portuguese orthography. In the second part, we compare eight ArabicPortuguese romanization schemes, highlighting their (dis)similarities; introduce all four modules (laymans simplified, laymans common, scholars broad and scholars narrow) of our orthographic transposition system, including samples comprising names, expressions and a small text; and finish off with an assessment on how said proposal may unfold.

Esta tese trata da transposição ortográfica de elementos morfossintáticos do árabe padrão moderno (APM) para o português brasileiro (PB) e constitui uma continua-ção de nossa dissertação de mestrado (Sano, 2011). Nesse trabalho anterior, trata-mos da transposição ortográfica de consoantes no mesmo sentido (APM PB) e desenvolvemos um sistema de romanização com vistas à transcrição e à translitera-ção do APM, ou seja, uma romanização que permitisse recuperar tanto a fala (os fo-nemas) quanto a escrita (os grafemas) da língua de partida. Nesta etapa atual, utili-zamos esse sistema de romanização para analisarmos dois corpora (a gramática de Cowan e o resumo da referida dissertação), com o intuito de identificar os pontos a serem abordados por um sistema de romanização que trabalhe além do nível da pa-lavra. Após esse levantamento, atualizamos o sistema de romanização anterior-mente desenvolvido, agora com o escopo estendido para o nível frasal, e reescreve-mos ambos os corpora aplicando as modificações propostas. / This dissertation deals with the orthographic transposition of morphosintactical elements from Modern Standard Arabic (MSA) into Brazilian Portuguese (BP), being a continuation of our masters thesis (Sano, 2011). In that previous work, we dealt with the orthographic transposition of consonants in the same pair of languages (MSA BP) and developed a romanization system meant to allow for the transcri-ption and transliteration of MSA, i.e., a romanization suitable for recovering both the original speech (the phonemes) and the original writing (the graphemes) of the star-ting language. In the current stage, we have used that system to analyse two corpora (Cowans grammar and our thesis abstract), in order to identify topics to be addres-sed by a romanization system inteded to be used beyond word level. After that, the now extended romanization system was used to rewrite both corpora, applying the proposed modifications encompassing the sentence level.

Ng Mau Kit Michael. / Thesis (M.Phil.)--Chinese University of Hong Kong, 1999. / Includes bibliographical references (leaves 117-120). / Abstract and appendix in English and Chinese. / Abstract --- p.i / Acknowledgments --- p.iv / Chapter 1 --- Introduction --- p.1 / Chapter 2 --- Background Knowledge and Basic Concepts --- p.7 / Chapter 2.1 --- Classification of Natural Languages --- p.7 / Chapter 2.2 --- Chinese Spell Checking --- p.9 / Chapter 2.3 --- Characteristics of Chinese --- p.12 / Chapter 2.3.1 --- Word Frequency and Statistical Information of Chinese Words --- p.12 / Chapter 2.3.2 --- Chinese Grammar --- p.15 / Chapter 2.3.2.1 --- Word Class --- p.15 / Chapter 2.3.2.2 --- Grammar Rules --- p.17 / Chapter 3 --- Problems with Chinese Spell Checking and Related Work --- p.18 / Chapter 3.1 --- Ambiguities --- p.19 / Chapter 3.2 --- Unknown Words --- p.20 / Chapter 3.3 --- Text Errors --- p.21 / Chapter 3.4 --- Combinatory Explosion --- p.23 / Chapter 3.5 --- Related Work --- p.26 / Chapter 4 --- The Chinese Spell Checking System --- p.33 / Chapter 4.1 --- Architecutre of the Chinese Spell Checking System (CSCS) --- p.35 / Chapter 4.2 --- The Segmenter and the Error Detector --- p.39 / Chapter 5 --- The Block-of-Combinations Segmentation Algorithm and Error Detection --- p.42 / Chapter 5.1 --- Single-character-word Function --- p.43 / Chapter 5.2 --- Segmentation Strategy --- p.46 / Chapter 5.3 --- Maximum Number of Combinations of the BOC --- p.51 / Chapter 5.4 --- A Case Study of the BOC --- p.54 / Chapter 5.5 --- Evaluation of the BOC --- p.59 / Chapter 5.5.1 --- Accuracy --- p.59 / Chapter 5.5.2 --- Speed --- p.61 / Chapter 5.5.3 --- Discussion --- p.62 / Chapter 5.6 --- Experiments on Error Detection for the BOC --- p.63 / Chapter 5.6.1 --- Experimental Results of the Error Detection for the BOC --- p.65 / Chapter 6 --- The Genetic Algorithm Segmentation Method --- p.69 / Chapter 6.1 --- Basic Concepts of Genetic Algorithm --- p.69 / Chapter 6.2 --- Genetic Algorithm Model --- p.73 / Chapter 6.2.1 --- Chromosome Representation --- p.75 / Chapter 6.2.2 --- The Flow of the GAS --- p.76 / Chapter 6.2.2.1 --- Crossover --- p.77 / Chapter 6.2.2.2 --- Replacement --- p.78 / Chapter 6.2.2.3 --- Mutation --- p.80 / Chapter 6.2.2.4 --- Termination Criteria --- p.80 / Chapter 6.2.3 --- Fitness Function --- p.81 / Chapter 6.2.3.1 --- Single-character-word Function --- p.82 / Chapter 6.2.3.2 --- Known-word Function and Unknown-word Function --- p.83 / Chapter 6.2.3.3 --- Grammar Rules Scoring Function --- p.83 / Chapter 6.3 --- Maximum Number of Combinations of the GAS --- p.86 / Chapter 6.4 --- Evaluation of the GAS --- p.86 / Chapter 6.5 --- Discussion --- p.88 / Chapter 7 --- The Improved-BOC Algorithm for Handling Unknown Words and Errors --- p.90 / Chapter 7.1 --- Segmentation Principle of the Improved-BOC Method --- p.91 / Chapter 7.2 --- Improvement of the Scoring Function --- p.93 / Chapter 7.2.1 --- The Choice of Grammar Rules --- p.93 / Chapter 7.2.2 --- Phrase-structure Style --- p.96 / Chapter 7.2.3 --- Computer Model of Grammar Rules for Handling Unknown Words --- p.98 / Chapter 7.3 --- Evaluation of Segmentation --- p.102 / Chapter 7.4 --- Error Detection --- p.104 / Chapter 7.4.1 --- Evaluation of Error Detection --- p.106 / Chapter 7.5 --- Discussion --- p.108 / Chapter 7.6 --- "Comparison between the MM, BOC, GA and Improved-BOC" --- p.109 / Chapter 8 --- Conclusion --- p.114 / Bibliography --- p.117 / Appendix A: Sample Result of the Genetic Algorithm Segmentation Method --- p.121 / Appendix B: Set of Grammar Rules --- p.123

Li Hong. / Thesis (Ph.D.)--Chinese University of Hong Kong, 1998. / Includes bibliographical references (p. 133-142). / Electronic reproduction. Hong Kong : Chinese University of Hong Kong, [2012] System requirements: Adobe Acrobat Reader. Available via World Wide Web. / Mode of access: World Wide Web. / Abstracts in English and Chinese.

In Experiments 3 and 4, event-related potentials were recorded to verify the findings of the first two behavioral experiments. Similarly, Experiment 3 employed the lexical decision task and Experiment 4 used the position decision task. First, results revealed that a series of ERP components (P150, P200, and N400) were found to be related to the effect of position-specific radical frequency when the lexical decision task was used (Experiment 3). However, when replacing the lexical decision task with the position decision task but keeping illegal characters as fillers, the effect of position-specific radical frequency became quite weak, which was only associated with P200 (Experiment 4a). Moreover, when using the position decision task but replacing illegal characters with geometric figures as fillers, there were no effects of position-specific radical frequency any more (Experiment 4b). Second, the task was found to influence the time course for effects of position-free radical frequency, which was reflected by P200 in the lexical decision task (Experiment 3) and by N400 in the position decision task (Experiments 4a and 4b). Finally, simple character frequency could play a role in processing compound characters in which simple characters were used as radicals, but such effect was significant only in the position decision task and it was reflected by a change in N400 (Experiments 4a and 4b). / These findings indicate that position-specific radicals could play a role in character recognition, but this effect appears to be constrained by the task and/or the composition of materials. In contrast, effects of position-free radicals keep stable across different tasks and composition of materials. In addition, the findings also suggest that simple character information could exert influence on compound character processing, but only when characters are processed implicitly (e.g., in the position decision task). Implications of this research and future directions are discussed. / This research was designed to examine the processing of different kinds of radical frequency information, i.e. frequency as a position-free radical, frequency as a position-specific radical, and frequency as a stand alone character. Whether the task or the composition of materials could influence the processing of such information was also examined. The first two were behavioral experiments and results indicated that effect of the position-free radical frequency was stable across the lexical decision task (Experiment 1) and the position decision task (Experiment 2). Effect of the position-specific radical frequency emerged only when the lexical decision task was available (Experiment 1). / Wu, Yan. / Adviser: Hsuan Chih Chen. / Source: Dissertation Abstracts International, Volume: 72-04, Section: B, page: . / Thesis (Ph.D.)--Chinese University of Hong Kong, 2010. / Includes bibliographical references (leaves 99-108). / Electronic reproduction. Hong Kong : Chinese University of Hong Kong, [2012] System requirements: Adobe Acrobat Reader. Available via World Wide Web. / Electronic reproduction. Ann Arbor, MI : ProQuest Information and Learning Company, [200-] System requirements: Adobe Acrobat Reader. Available via World Wide Web. / Electronic reproduction. Ann Arbor, MI : ProQuest Information and Learning Company, [200-] System requirements: Adobe Acrobat Reader. Available via World Wide Web. / Abstract also in Chinese; includes Chinese characters in Appendix A-C.

published_or_final_version / abstract / toc / Psychology / Master / Master of Philosophy

Grapheme-based speech recognition systems are faster to develop, but typically do not reach the same level of performance as phoneme-based systems. Using Afrikaans speech recognition as a case study, we first analyse the reasons for the discrepancy in performance, before introducing a technique for improving the performance of standard grapheme-based systems. It is found that by handling a relatively small number of irregular words through phoneme-to-grapheme (P2G) transliteration – transforming the original orthography of irregular words to an ‘idealised’ orthography – grapheme-based accuracy can be improved. An analysis of speech recognition accuracy based on word categories shows that P2G transliteration succeeds in improving certain word categories in which grapheme-based systems typically perform poorly, and that the problematic categories can be identified prior to system development. An evaluation is offered of when category-based P2G transliteration is beneficial and methods to implement the technique in practice are discussed. Comparative results are obtained for a second language (Vietnamese) in order to determine whether the technique can be generalised. / MSc (Computer Science) North-West University, Vaal Triangle Campus, 2014