This thesis investigates the recognition rate improvement strategies for a Mandarin speech recognition system. Both automatic tone recognition and consonant correction schemes are studied and applied to the Mandarin address inputting system and the Mandarin 2, 3, 4-word phrase recognition systems. For automatic tone recognition scheme, the acoustic properties of the four tones in the Mandarin training database are estimated statistically by 4 sets of parameters within 6 minutes. These automatically generated parameters can greatly increase the tone recognition accuracy, and at the same time reduce the amount of time spent in the manual tone parameter adjustment, that is about 8 hours in general. For consonant correction scheme, the sub-syllable models are developed to enhance the consonant recognition accuracy, and hence further improve the overall correct rate for the whole Mandarin phrases. Experimental results indicate that over 90% correct rate can be achieved for the Mandarin address inputting system with 180 thousand place names by applying the above two schemes. Furthermore, the recognition rates for the Mandarin 2, 3, 4-word phrase recognition systems with 116 thousand phrases in total can be improved from 77%, 94% and 97.5%, to 85%, 96% and 98% respectively.

This dissertation is an investigation of the stress system of Squamish (Skwxwú7mesh), one of ten languages that make up the Central division of the Northwest Coast branch of Salishan, a linguistic group indigenous to the Pacific Northwest region of North America. Although other researchers have previously investigated aspects of stress in the language, this work provides the first integrated account of the Squamish stress system as a whole, couched in an Optimality Theoretic framework. The first two chapters are introductory, with Chapter 1 supplying a contextual background for the undertaking within linguistics, and especially within Salishan linguistices, while Chapter 2 provides a thorough grounding in the phonology and phonemics of Squamish in particular. Chapter 3 begins the formal analysis of stress in Squamish by examining the way stress surfaces in free root morphemes,which tend to stress penultimate syllables whenever they contain either a full vowel or a schwa followed by a resonant consonant. Given this outcome, Chapter 4 continues the investigation of basic stress patterns by looking more closely at the interactive roles of schwa, sonority, weight and the structure of syllables and feet in Squamish stress assignment. With the basic stress pattern established, the remaining chapters look at the outcome of stress in morphologically complex Squamish words. Thus, Chapter 5 is an analysis of stress in words involving prefixation, especially those resulting from CVC and CV prefixal reduplication, since non-reduplicative prefixes are unstressable; and Chapters 6 and 7 investigate the occurrence of stress in polymorphemic words resulting from the addition of lexical suffixes and grammatical suffixes, respectively. While stress in roots is generally predictable on the basis of phonological factors alone, that in polymorphemic words may also be influenced by morphological factors, as when a root or suffix has underlying lexical accent, and such factors then take precedence ofer phonological factors. In addition, prosodic domains play an important and interactive work.

In this work, I will be looking at the cross-linguistic restrictions on coda position, especially with regard to sonority and place. I propose that these restrictions can be best captured if we acknowledge two types of constraints; those which restrict the licensing ability of codas, as well as those which set out a relationship between a coda and a following onset. I show that in allowing for this distinction, the contrast between the restrictions on word-internal codas and word-final ones falls out straightforwardly. This study is carried out in the framework of Optimality Theory, but the basic claims are expected to hold in other frameworks, whether rule-based or constraint-based.

Languages differ in their sound patterns, but these differences are, to a large extent, systematic. One goal of Universal Grammar (Chomsky 1957, 1965) is to account for the systematic patterns which are attested across languages. Toward this end, Universal Grammar is considered to contain a set of phonological primitives such as features, and some restrictions on their combination. However, in rule-based phonology, it is assumed that rules are part of the grammar of an individual language. By their very nature, rules describe operations. As such, they are not well-suited to express restrictions on the ways in which segments may combine when no overt operation is involved. To account for such restrictions, Chomsky & Halle (Sound Pattern of English (SPE): 1968) supplemented rules with Morpheme Structure Constraints (MSCs) which define the possible morpheme shapes that a particular language allows (see also Halle 1959). Thus, in SPE, both MSCs and rules played a role in accounting for the phonological patterns observed in languages.

It is widely recognized that differences in both prosodic complexity and position-sensitive contrasts exist both within and across languages. In contemporary phonological theory, these differences are often attributed to differences between heads and non-heads and the asymmetries in licensing potential that exist between such positions. / In this thesis, the consequences of such differences for the second language (L2) acquisition of prosodic complexity and position-sensitive contrasts are explored. It is argued that an explanatorily adequate account of L2 syllabification must include highly-structured representations as well as a theory of licensing, which distinguishes between the licensing of a given position and the licensing of featural content in such a position. Using data drawn primarily from a number of studies that investigate the acquisition of French by native speakers of English and Mandarin, it is demonstrated that the widely-attested interlanguage (IL) syllable-structure-modification processes of deletion, epenthesis, and feature change have a common source. Specifically, all three processes result from the IL grammar's inability to license a syllable position or (some of) the featural content present in such a position in the target representation. Within Optimality theory, the framework adopted, this is formalized through the competition between Faithfulness constraints and Markedness constraints, which evaluate the wellformedness of the licensing relationships. Finally, it is argued that Prosodic Licensing and the principle of Licensing Inheritance from Harris (1997) work together to encode prosodic markedness in representation, as they create a series of head-dependent asymmetries in which heads are strong licensors vis-a-vis their dependents.

Made available in DSpace on 2015-05-14T12:43:20Z (GMT). No. of bitstreams: 1 arquivototal.pdf: 1233471 bytes, checksum: 18383a1198cd8f692689f5bbdff60511 (MD5) Previous issue date: 2009-03-09 / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - CAPES / There has been a lot of discussion about the post-vocalic /s/ and its variable behavior in Brazilian Portugueses, as in Brescancini Brescancini (2002); Callou, Leite, Moraes (2002); Hora (2003); Ribeiro (2006); among others. All of them agree that /s/ occupies the coda position, alternating alveolar (ca[s]ca, me[z]mo, feli[s]), palatal (po[]te, de[]de, mai[]) and glotal (de[h]de, me[h]mo, mai[h]) realizations, and also deletion (me[]mo, mai[]), when the syllabic pattern turns into the less marked one (CV). Although this conception is widespread within the phonological theory, we believe that the particular status the coronal fricative possesses in relation to liquids and nasals (those licensed to occupy the coda position in Brazilian Portuguese) is a fact that deserves to be observed more carefully and one that can lead to another way of analysis. According to Harris & Gussmann (1998) and Ewen & Hulst (2001), some syllables can be formed by onset with a phonetically non-filled nucleus. Accordingly, these syllables satisfy the less marked pattern (CV) in underlying representation, but they do not show, in some cases, the phonetic representation in the surface. The fricatives would be prone to sustain the syllable weigh, since they have more frequency in duration in their production, and as such they do not need, in theory, the phonetic realization of their nuclei. These arguments make us, therefore, propose another classification of the post-vocalic /s/, considering it onset of a phonetically empty nucleus, that, in some dialects, is realized in final position ([fajzi], [majzi]). Therefore, our aim is to analyze the post-vocalic /s/ in Brazilian Portuguese, with the variationist results from Brescancini (2002); Callou, Leite, Moraes (2002); Hora (2003) and Ribeiro (2006), in order to highlight its variable pattern and to propose a ranking for Brazilian Portuguese, using, for such, the Optimality Theory (OT) in Coetzee‟s perspective. In this proposal, the evaluator (EVAL) realizes a harmonic ranking for the set of candidates, in a way that the losers are also ranked, that is, both the constraints and the candidates are ranked. In this way, it is possible to work with non-categorical phenomena, as such related to variation, and to produce a proficuous link between the Sociolinguistic Research and Optimality Theory. / Muito já se discutiu sobre o /s/ pós-vocálico e seu comportamento variável no português brasileiro, a exemplo dos trabalhos de Brescancini (2002); Callou, Leite, Moraes (2002); Hora (2003); Ribeiro (2006); dentre outros. Em todos eles, é consenso que o /s/ pós-vocálico ocupa a posição de coda, alternando-se entre as realizações alveolar (ca[s]ca, me[z]mo, feli[s]), palatal (po[]te, de[]de, mai[]) e glotal (de[h]de, me[h]mo, mai[h]), podendo, ainda, sofrer apagamento (me[]mo, mai[]), encaixando-se, dessa forma, no padrão silábico menos marcado (CV). Embora essa concepção seja bastante difundida na teoria fonológica, acreditamos que o perfil diferenciado que a fricativa coronal pós-vocálica possui em relação às líquidas e às nasais, consoantes licenciadas a ocupar a posição de coda no português brasileiro (PB), seja um fato que merece ser observado com mais cuidado e que pode levar a outro caminho de análise. Segundo Harris e Gussmann (1998) e Ewen e Hulst (2001), algumas sílabas podem ser formadas por onset com núcleo foneticamente não preenchido. De acordo com o proposto, essas sílabas satisfazem o padrão menos marcado (CV) na subjacência, mas não apresentam, em alguns casos, a representação fonética na superfície. As fricativas, por possuírem uma maior freqüência de duração em sua produção, seriam uma das consoantes mais propícias para sustentarem o peso da sílaba, não necessitando, em tese, da realização fonética de seu núcleo. Esses argumentos nos fazem, portanto, propor outra classificação do /s/ pós-vocálico, considerando-o onset de núcleo foneticamente vazio que, em alguns dialetos, chega a se realizar na posição final ([fajzi], [majzi]). Diante desse fato, é nosso objetivo analisar o /s/ pós-vocálico no PB, através dos resultados variacionistas de Brescancini (2002); Callou, Leite, Moraes (2002); Hora (2003) e Ribeiro (2006), a fim de salientar o seu padrão variável e propor uma hierarquização para o PB, utilizando, para tanto, a Teoria da Otimalidade (TO) na perspectiva de Coetzee (2004). Nessa proposta, o avaliador (EVAL) realiza um ordenamento harmônico para o conjunto completo dos candidatos, de forma que os perdedores também são ordenados entre si, ou seja, tanto as restrições são hierarquizadas como os candidatos analisados como variantes são ordenados. Dessa forma é possível trabalhar com fenômenos não-categóricos, a exemplo dos relacionados à variação, e efetivar um casamento profícuo entre a Pesquisa Sociolingüística e a Teoria da Otimalidade.

Esta tese trata do estatuto silábico das consoantes finais /r, l, s, n/ no Português. Partimos, inicialmente, da proposta de Harris e Gussmann (1998), segundo a qual as consoantes finais não ocupam a coda silábica, mas são onsets de sílabas com núcleos vazios. Essa proposta baseia-se em uma concepção de estrutura silábica independente da estrutura segmental e que obedece às condições próprias (condições de licenciamento), as quais determinam a boa formação silábica. Nesse sentido, a estrutura silábica pode ser bem formada independentemente da estrutura segmental que a preenche. Assim se explica a ideia de Harris e Gussmann (1998) de núcleos vazios, isto é, de sílabas bem-formadas em termos de estrutura silábica, mas com núcleo não preenchido. Para procedermos a esta discussão, o trabalho foi subdividido em quatro capítulos. No primeiro, apresentamos a abordagem de que consoantes finais são onsets, buscando entender as implicações dessa abordagem e seus desdobramentos. Fazemos também uma comparação com abordagens alternativas, como a da extraprosodicidade e das semissílabas. Noções como a de licenciamento e outros detalhes da perspectiva adotada por Harris e Gussmann (1998) são explicitadas no capítulo 2, no qual também buscamos abordar os principais pressupostos da tese. Embasamo-nos na perspectiva de que a gramática é regida por princípios e parâmetros, admitindo marginalmente a existência de condições e regras específicas à língua. Adotamos também a perspectiva da existência de mais de um nível representacional. A abordagem considera tanto níveis derivacionais, como proposto pela Fonologia Lexical, Kiparsky (1982), quanto a de representação abstrata das estruturas fonológicas, com vários níveis (tiers) estruturais independentes, hierarquicamente organizados. Dessa forma, o núcleo vazio é criado no Léxico e pode ser preenchido, ainda nesse nível, por vogais introduzidas pela sufixação e pela flexão. Por outro lado, uma sílaba com núcleo vazio pode prosseguir até o nível pós-lexical sem precisar ser preenchido com conteúdo segmental. No capítulo 3, trazemos algumas línguas para as quais foi defendida na literatura a existência de consoante final como onset de um núcleo vazio, que denominamos línguas NV. Diferentes processos ilustram a existência dessas categorias após uma consoante na posição final. A observação das evidências para os núcleos não preenchidos foneticamente em distintos idiomas reforça a ideia de que eles sejam estruturas representacionais possíveis; além disso, traz informações para a construção da argumentação da análise de núcleos vazios para o Português. No capítulo 4, argumentamos pela análise das consoantes finais /l, r, s/ como onsets de sílabas com núcleos vazios. Para essa argumentação, trazemos inicialmente a análise do estatuto silábico das consoantes finais no Português Europeu, com base em Mateus e D’Andrade (2000). Em seguida, analisamos argumentos baseados na literatura sobre o Português Brasileiro para cada um dos segmentos finais, mostrando que a análise como onset final é compatível com o comportamento observando para essas consoantes. A nasal final, por sua vez, não pode ser considerada onset como as outras, pois se trata do processo de nasalização da vogal, em que não há um segmento consonantal, mas um glide nasal, conforme Battisti (1997). Por fim, propomos a representação de uma estrutura CVC] como constituída silabicamente como CV.CØ]. A análise adota, ainda, a existência do Parâmetro da Consoante Final que estaria ativado no Português Brasileiro. / This thesis focuses on the syllabic status of word-final consonants /r, l, s, n/ in Portuguese. We had as initial proposal the one presented by Harris and Gussmann (1998), according to which word-final consonants do not occupy the syllable coda, but the onset of a syllable containing an empty nucleus. Such proposal is based on a conception of syllable structure independent of the segmental structure and obeys constraints (licensing constraints) that determine the syllable well-formedness. Following this assumption, the syllable structure can be well-formed independently of the segmental structure that fills it. This explains the authors’ idea of empty nuclei, i.e. of well-formed syllables in terms of syllable structure despite an unfilled nucleus. To proceed to this discussion, this work was divided into four chapters. On the first chapter, we present the approach of C] as onsets, as a way of understanding the implications of this approach and their unfolding. We also compare it to alternative approaches, such as the extraprosodicity and the semisyllable ones. The notion of licensing, among other notions, and some other details of the perspective adopted by Harris e Gussmann (1998) are presented on Chapter 2, on which we also try to make the main presuppositions of this thesis explicit. We adopted the perspective of a grammar ruled by principles and parameters, assuming marginally the existence of specific conditions and rules to the language. We also adopted the perspective that assumes the existence of more than one representational level. This approach considers both the derivational levels, as proposed by Lexical Phonology, Kiparsky (1982), and the abstract representation of phonological structures, with independent structural tiers, hierarchically organized. Thus the empty nucleus is created on the Lexicon and can be filled, on this same tier, by vowels inserted through suffixation and inflection. On the other hand, an onset of a syllable containing an empty nucleus can proceed up to the pos-lexical level without being necessary to be filled by any segmental content. On chapter 3, we mention some languages to which it was assumed the existence of word-final consonant as the onset of an empty nucleus. These languages are named NV languages. Different processes show the existence of empty nucleus after a word-final consonant. The observation of evidences to empty nuclei in different languages reinforces the idea that these empty nuclei are possible representational structures; besides, it carries information to the argumentation construction of the empty nucleus analysis in Portuguese. On chapter 4, we argue for the word-final consonants /l, r, s/ analysis as onsets of syllables containing empty nuclei. For this argument, we bring up the analysis of the syllabic status of word-final consonants in the European Portuguese, based on Mateus e D’Andrade (2000). Then we analyze arguments based on the literature about Brazilian Portuguese to each word-final segment, showing that the word-final onset analysis is compatible with the behavior observed to these consonants. The word-final nasal, however, cannot be considered syllable onset as the other consonants, because this represents the vowel nasalization process, in which there is no consonantal segment, but a nasal glide instead, according to Battisti (1997). Finally, we propose the representation of a CVC] structure as CV.CØ]. This analysis adopts the existence of Word-final Consonant Parameter that would be active in the Brazilian Portuguese.

Submitted by Suelen Reis (suziy.ellen@gmail.com) on 2013-05-15T13:25:24Z No. of bitstreams: 1 Tese Claudia Moreira.pdf: 2433230 bytes, checksum: 6409815ac6938e00cc2b511ed2808665 (MD5) / Approved for entry into archive by Alda Lima da Silva(sivalda@ufba.br) on 2013-05-16T17:52:34Z (GMT) No. of bitstreams: 1 Tese Claudia Moreira.pdf: 2433230 bytes, checksum: 6409815ac6938e00cc2b511ed2808665 (MD5) / Made available in DSpace on 2013-05-16T17:52:34Z (GMT). No. of bitstreams: 1 Tese Claudia Moreira.pdf: 2433230 bytes, checksum: 6409815ac6938e00cc2b511ed2808665 (MD5) Previous issue date: 2009 / Este trabalho tem como objetivo central investigar a influência da estrutura silábica sobre a aquisição da leitura por crianças e adultos que se encontram em fase de alfabetização escolar. Quatro variáveis relacionadas ao processamento silábico foram consideradas: a) o padrão silábico; b) a acentuação; c) a posição da sílaba na palavra; d) as estratégias de leitura. Os dados foram obtidos através de pesquisa de campo e experimental, desenvolvida ao longo do ano letivo de 2006, em duas escolas públicas da cidade de Salvador. Participaram da pesquisa ao todo 29 informantes, 17 crianças (G1) e 12 adultos (G2). Realizaram-se quatro sessões de leitura de palavras e pseudopalavras com cada sujeito, em encontros periódicos ao longo do ano. Os testes de leitura foram aplicados através de um aplicativo denominado APPTL − Aplicativo para Teste de Leitura, elaborado especialmente para essa tarefa. As leituras foram gravadas em áudio e vídeo. Os resultados mostraram que a estrutura silábica da palavra exerce uma influência significativa sobre o nível de acertos na leitura de palavras e sobre o tipo de estratégia utilizada pelos sujeitos. As palavras formadas pela sílaba CV obtiveram o maior índice de acertos, e, conseqüentemente, a maior ocorrência de estratégia proficiente durante sua leitura; numa posição intermediária, ficaram as palavras constituídas de sílabas CVC; e por fim, as palavras formadas por sílabas CCVC foram as que obtiveram menos acertos, e, em conseqüência, demandaram o uso das estratégias menos proficientes. Quanto à acentuação, as palavras paroxítonas foram lidas com maior número de acertos, seguidas das proparoxítonas e oxítonas. No que diz respeito à posição da sílaba na palavra, as sílabas complexas foram processadas com mais acertos quando se encontravam na posição inicial, seguida da posição interna e final. Os resultados apresentados pelas crianças (G1) foram semelhantes aos ocorridos com os adultos (G2), o que permite concluir que crianças e adultos têm um comportamento similar no que tange ao processamento da sílaba na fase inicial de aquisição da leitura. Tais achados implicam que sejam revistos os métodos de alfabetização oferecidos a crianças e adultos, no sentido de promover um tratamento mais sistemático da sílaba na escola, o que implica uma abordagem que leve mais em consideração a relação entre leitura e oralidade. / Salvador

The present study examines Saudi ESL learner perception of three syllable-final English nasal contrasts: /m/-/n/, /m/-/©¯/, and /n/-/©¯/. It was based primarily on two models, the Perceptual Assimilation Model (PAM) and the Speech Learning Model (SLM), in order to determine how Saudi ESL learners categorized nasal segments in their L2, English. In addition, other models, including the Markedness Differential Hypothesis (MDH), contributed in finding the most difficult contrast. The participants, consisting of 24 Saudis studying in the US, were asked to distinguish between 40 English monosyllabic words grouped into minimal pairs. Regardless of their length of exposure, participants had greater difficulty in categorizing /n/ and /©¯/ as separate phonemes, dealing with these sounds as a Single Category (SC). The /n/-/m/ contrast had the fewest errors among participants with more exposure in the US, who were able to distinguish between the word pairs in this contrast, while those with less exposure had more difficulty in distinguishing /n/ from /m/. This fact shows that the participants, especially after receiving more exposure, were able to categorize the /n/ and /m/ as a Two Category (TC). The contrast /m/-/©¯/ had a number of errors somewhat similar to that found with the /n/-/m/ contrast. Therefore, participants in early stages of learning encountered more difficulties in categorizing /m/ and /©¯/ than those who had spent a longer period of time in the US. This study concluded that the categorization between Arabic and English phonemes was highly important in learning English as an L2. Therefore, the absence of the English nasal /©¯/ from the Arabic phonemic inventory caused difficulties for Saudi ESL learners to categorize /©¯/ as a separate phoneme. Additionally, markedness also played a role since the velar nasal /©¯/ is typologically more marked than /n/ and /m/ making it more difficult to acquire in early stages of L2 learning.

This study provides an Optimality-Theoretic analysis of the syllabification of word-initial biconsonantal clusters in the productions of Faifi Arabic and Asiri Arabic speakers. This study aimed at investigating the role of sonority in the syllabification of onset clusters. Two groups, each made up of 15 participants, were employed in this study to produce English nonwords, which had onsets composed of biconsonantal clusters with different sonority levels. The results of the study showed that the two groups had two different ways of treating the clusters. The Faifi group epenthesized a vowel before the onset clusters, forcing the second consonant in the cluster to become the onset of the following syllable. The Asiri group epenthesized a vowel between the consonants of onset clusters when the sonority slope equaled 2. When the sonority slope equaled 3, the cluster was produced intact. All of these differences were shown to be the result of different rankings of several markedness and faithfulness constraints.