No description available.

This dissertation illustrates the cultural dimension of globalization by examining how the ascendance of South Korean popular culture, i.e., the Korean Wave, reshapes the global imagination and transforms the locality of Korean Americans in Philadelphia. As an ethnographic global media study, I conducted in-depth interviews and participated in Korean cultural events/meetings, as well as visited the sites of living for Korean Americans in the Philadelphia metropolitan area. My research finds that advances in the digitalized media environment made my informants consume copious transnational Korean media every day and individualized their media consumption. Accordingly, their perceptions of Korea/Asia/U.S.’s places in the world are complicated and their ethnic identity has become significant. Their global imaginations also intersect with negotiating gender roles, perceiving attractiveness, and planning future paths. This shift contributes to construction of the in-between identities of Korean Americans by denaturalizing ideas and cultural elements in both Korea and the U.S. Most distinctively, the rise of the Korean Wave stimulates global imagination among young second generation Korean Americans to aspire to and operate their agency in a transnational context that their parents’ generation barely anticipated. Finally, the upsurge of the Korean Wave drives Korean Americans to participate in transforming localities rooted in thickened connectivities and activities centering on Korean popular culture across intra/inter-ethnic groups locally and globally. This conversely facilitates intense engagement and belonging in the local spaces of community among Korean Americans. My study shows how transnational media flow under conditions of globalization positively influences immigrants to embrace their own ethnic identities and local spaces. On the other hand, it implies that there should be further examination of different boundaries of global imagination rooted in gender/class differences as well as race/ethnicity. / Media & Communication

Bacterial taxonomy is the science of classifying, naming, and identifying bacteria. The scope and practice of taxonomy has evolved through history with our understanding of life and our growing and changing needs in research, medicine, and industry. As in animal and plant taxonomy, the species is the fundamental unit of taxonomy, but the genetic and phenotypic diversity that exists within a single bacterial species is substantially higher compared to animal or plant species. Therefore, the current "type"-centered classification scheme that describes a species based on a single type strain is not sufficient to classify bacterial diversity, in particular in regard to human, animal, and plant pathogens, for which it is necessary to trace disease outbreaks back to their source. Here we discuss the current needs and limitations of classic bacterial taxonomy and introduce LINbase, a Web service that not only implements current species-based bacterial taxonomy but complements its limitations by providing a new framework for genome sequence-based classification and identification independently of the type-centric species. LINbase uses a sequence similarity-based framework to cluster bacteria into hierarchical taxa, which we call LINgroups, at multiple levels of relatedness and crowdsources users' expertise by encouraging them to circumscribe these groups as taxa from the genus-level to the intraspecies-level. Circumscribing a group of bacteria as a LINgroup, adding a phenotypic description, and giving the LINgroup a name using the LINbase Web interface allows users to instantly share new taxa and complements the lengthy and laborious process of publishing a named species. Furthermore, unknown isolates can be identified immediately as members of a newly described LINgroup with fast and precise algorithms based on their genome sequences, allowing species- and intraspecies-level identification. The employed algorithms are based on a combination of the alignment-based algorithm BLASTN and the alignment-free method Sourmash, which is based on k-mers, and the MinHash algorithm. The potential of LINbase is shown by using examples of plant pathogenic bacteria. / Doctor of Philosophy / Life is always easier when people talk to each other in the same language. Taxonomy is the language that biologists use to communicate about life by 1. classifying organisms into groups, 2. giving names to these groups, and 3. identifying individuals as members of these named groups. When most scientists and the general public think of taxonomy, they think of the hierarchical structure of “Life”, “Domain”, “Kingdom”, “Phylum”, “Class”, “Order”, “Family”, “Genus” and “Species”. However, the basic goal of taxonomy is to allow the identification of an organism as a member of a group that is predictive of its characteristics and to provide a name to communicate about that group with other scientists and the public. In the world of micro-organism, taxonomy is extremely important since there are an estimated 10,000,000 to 1,000,000,000 different bacteria species. Moreover, microbiologists and pathologists need to consider differences among bacterial isolates even within the same species, a level, that the current taxonomic system does not even cover. Therefore, we developed a Web service, LINbase, which uses genome sequences to classify individual microbial isolates. The database at the backend of LINbase assigns Life Identification Numbers (LINs) that express how individual microbial isolates are related to each other above, at, and below the species level. The LINbase Web service is designed to be an interactive web-based encyclopedia of microorganisms where users can share everything they know about micro-organisms, be it individual isolates or groups of isolates, for professional and scientific purposes. To develop LINbase, efficient computer programs were developed and implemented. To show how LINbase can be used, several groups of bacteria that cause plant diseases were classified and described.

Data-intensive scientific applications often process an enormous amount of data. The scalability of such applications depends critically on how to manage the locality of data. Our study explores two common types of applications that are vastly different in terms of memory access pattern and workload variation. One includes those with multi-stride accesses in regular nested parallel loops. The other is for processing large-scale irregular social network graphs. In the former case, the memory location or the data item accessed in a loop is predictable and the load on processing a unit work (an array element) is relatively uniform with no significant variation. On the other hand, in the latter case, the data access per unit work (a vertex) is highly irregular in terms of the number of accesses and the locations being accessed. This property is further tied to the load and presents significant challenges in the scalability of the application performance. Designing platforms to support extreme performance scaling requires understanding of how application specific information can be used to control the locality and improve the performance. Such insights are necessary to determine which control and which abstraction to provide for interfacing an underlying system and an application as well as for designing a new system. Our goal is to expose common requirements of data-intensive scientific applications for scalability. For the former type of applications, those with regular accesses and uniform workload, we contribute new methods to improve the temporal locality of software-managed local memories, and optimize the critical path of scheduling data transfers for multi-dimensional arrays in nested loops. In particular, we provide a runtime framework allowing transparent optimization by source-to-source compilers or automatic fine tuning by programmers. Finally, we demonstrate the effectiveness of the approach by comparing against a state-of-the-art language-based framework. For the latter type, those with irregular accesses and non-uniform workload, we analyze how the heavy-tailed property of input graphs limits the scalability of the application. Then, we introduce an application-specific workload model as well as a decomposition method that allows us to optimize locality with the custom load balancing constraints of the application. Finally, we demonstrate unprecedented strong scaling of a contagion simulation on two state-of-the-art high performance computing platforms. / Ph. D.

Non-locality and contextuality are key features of quantum mechanics that distinguish it from classical physics. We aim to develop a deeper, more structural understanding of these phenomena, underpinned by robust and elegant mathematical theory with a view to providing clarity and new perspectives on conceptual and foundational issues. A general framework for logical non-locality is introduced and used to prove that 'Hardy's paradox' is complete for logical non-locality in all (2,2,l) and (2,k,2) Bell scenarios, a consequence of which is that Bell states are the only entangled two-qubit states that are not logically non-local, and that Hardy non-locality can be witnessed with certainty in a tripartite quantum system. A number of developments of the unified sheaf-theoretic approach to non-locality and contextuality are considered, including the first application of cohomology as a tool for studying the phenomena: we find cohomological witnesses corresponding to many of the classic no-go results, and completely characterise contextuality for large families of Kochen-Specker-like models. A connection with the problem of the existence of perfect matchings in k-uniform hypergraphs is explored, leading to new results on the complexity of deciding contextuality. A refinement of the sheaf-theoretic approach is found that captures partial approximations to locality/non-contextuality and can allow Bell models to be constructed from models of more general kinds which are equivalent in terms of non-locality/contextuality. Progress is made on bringing recent results on the nature of the wavefunction within the scope of the logical and sheaf-theoretic methods. Computational tools are developed for quantifying contextuality and finding generalised Bell inequalities for any measurement scenario which complement the research programme. This also leads to a proof that local ontological models with `negative probabilities' generate the no-signalling polytopes for all Bell scenarios.

Distributed Real-Time Media Processing refers to classes of highly distributed, delay no-tolerant applications that account for the majority of the data traffic generated in the world today. Real-Time audio/video conferencing and live content streaming are of particular research interests as technology forecasts predict video traffic surpassing every other type of data traffic in the world in the near future. Live streaming refers to applications in which audio/video streams from a source need to be delivered to a set of geo-distributed destinations while maintaining low latency of stream delivery. Real-time conferencing platforms are application platforms that implement many-to-many audio/video real-time communications. Both of these categories exhibit high sensitivity to both network state (latency, jitter, packet loss, bit rate) as well as stream processing backend load profiles (latency and jitter introduced as Cloud processing of media packets). This thesis addresses enhancing real-time media processing both at the network level parameters as well as Cloud optimisations. We provide a novel, bandwidth management algorithm, for cloud services sharing the same network infrastructure, which provides a 2x improvement in system stability. Further examining network impact on cloud services, we provide a novel hybrid Cloud-Network distributed Cloud architecture to enable locality aware, application enhancements. This architecture led to a multi-cloud management overlay algorithm that maintains low management overhead on large scale cloud deployments. On the application level we provide a study of Media Quality parameters for a WebRTC enabled Media Cloud back-end, and provide patterns of quality metrics with respect to back-end stream load and network parameters. Additionally we empirically show that a "minimal load" algorithm for stream allocation, outperforms other Rotational, or Static Threshold based algorithms. / El procesamiento de medios en tiempo real distribuido se refiere a clases de aplicaciones altamente distribuidas,no tolerantes al retardo, que representan la mayoría del tráfico de datos generado en el mundo actual. Las conferenciasde audio y video en tiempo real y la transmisión de contenido en vivo tienen especial interés en investigación, ya quela prospectiva tecnológica estima que el tráfico de video supere a cualquier otro tipo de tráfico de datos en el futurocercano. La transmisión envivo se refiere a aplicaciones en las que flujos de audio/vídeo de una fuente se han de entregara un conjunto de destinos en lugares geográficos diferentes mientras se mantiene baja la latencia de entrega del flujo(como por ejemplo la cobertura de eventos en vivo). Las plataformas de conferencia en tiempo real son plataformasde  aplicación  que  implementan  comunicaciones  de  audio/video  en  tiempo  real  entre  muchos  participantes.  Ambascategorías presentan una alta sensibilidad tanto al estado de la red (latencia, jitter, pérdida de paquetes, velocidad debits) como a los perfiles de carga de la infraestructura de procesamiento de flujo (latencia y jitter introducidos duranteel procesamiento en la nube de paquetes de datos multimedia). Esta tesis trata de mejorar el procesamiento de datosmultimedia en tiempo real tanto en los parámetros de nivel de red como en las optimizaciones en la nube.En este contexto, investigamos si los recursos de la red se podían controlar a nivel de servicio para aumentar laeficiencia y el rendimiento de la red, así como cuantificar el impacto del recurso compartido de la red en la calidad delservicio. Los recursos de red compartidos afectan el rendimiento del servicio en la nube y, por lo tanto, optimizandoo intercambiando recursos de red pueden mejorar el rendimiento del servicio en la nube. Esta posible degradación delrendimiento se debe a la infraestructura de red compartida no regulada (la asignación de recursos de ancho de banda noes consciente de los objetivos del acuerdo de nivel de servicio (SLO) y de comportamiento). Gestionando el ancho debanda de la red a través de control predictivo, permitimos un mejor uso de los recursos de red disponibles y menoresviolaciones de SLO, logrando una mayor estabilidad del sistema por al menos un factor de 2.Las redes de acceso (AN) (extremo, red principal) de los ISP, transportistas y redes comunitarias no tienen unainfraestructura de nube de propósito general, mientras que los proveedores de recursos de Internet proporcionan bajodemanda recursos de la nube. Encontramos una oportunidad para la unificación de los recursos dentro de un AN y fueracon el fin de proporcionar una oferta de nube unificada a través de una federación de nubes y proporcionar movilidad delservicio hacia los usuarios para optimizar la localidad. Este trabajo de investigación proporciona una nueva arquitecturade red híbrida y nube federada que proporciona una infraestructura de red extendida con un despliegue en nube a granescala, incorporándolo directamente a la infraestructura de red. La nueva arquitectura multi-nube permite a los serviciosllegar a un compromiso entre localidad respecto al usuario o el rendimiento en tiempo de ejecución optimizando asípara latencia para conseguir la asignación óptima de recursos de aplicaciones en tiempo real. Para  optimizar  la  latencia  en  las  aplicaciones  de  transmisión  en  vivo  se  propuso  un  nuevo  algoritmo  desuperposición  de  multi-nube  autogestionado  basado  en  una  topología  de  gradiente  en  la  que  cada  nube  de  unaaplicación de transmisión de flujos optimiza la proximidad del cliente a la fuente. El modelo de aplicación se separa enun  diseño  de  dos  capas,  el  back-end  de  entrega  multi-nube  y  los  clientes  de  flujo.  El  backend  de  gradienteautorregulado minimiza la carga de tráfico creando un árbol de expansión mínimo a través de las nubes que se utilizapara  el  enrutamiento  de  cada  flujo.  El  algoritmo  propuesto  tiene  una  tasa  de  convergencia  muy  rápida  en  losdespliegues de nube a gran escala, y no resulta afectado por la rotación de recursos de la nube, así como proporcionauna mayor estabilidad de la transmisión en vivo.En este trabajo ofrecemos un análisis de calidad de los medios de comunicación y mejoras de los emisores deflujo en la nube en tiempo real, así como estrategias de asignación para mejorar el rendimiento de nivel de serviciode las plataformas de comunicación de Web en tiempo real. Los patrones de calidad de los medios están fuertementeinfluenciados por el rendimiento del procesamiento en la nube, y por lo tanto, al ajustar este aspecto, podemos controlarla calidad de los medios. En particular, demostramos empíricamente que a medida que los tamaños de sesión aumentan,la difusión simultánea supera la codificación de capa única. Además, introducimos un algoritmo de asignación de flujopara minimizar los picos de carga en los retransmisores de flujos en la nube y comparamos el comportamiento devarias políticas de asignación de flujos. Con la mínima información y el requisito de asignación de sesión de un únicoservidor, la política de asignación de carga mínima se comporta bastante mejor que otros algoritmos basados en unumbral rotativo o estático. / Distribuerad realtidshantering av mediadata syftar på klasser av starkt distribuerade tillämpningar som inte tolererar fördröjningar, och som utgör majoriteten av datatrafiken som genereras i världen idag. Audio/video-konferenser i realtid och överföring av innehåll "live" är av speciellt intresse för forskningen eftersom teknikprognoser förutser att videotrafiken kommer att kraftigt dominera över all annan datatrafik i den nära framtiden. "Live streaming" syftar på tillämpningar i vilka audio/video strömmar från en källa och behöver distribueras till en mängd av geografisk distribuerade destinationer medan överföringen bibehåller låg latens i leveransen av det strömmade datat (som ett exempel kan nämnas "live"-täckning av händelser). Konferensplattformar för realtidsdata är tillämpningsplattformar som implementerar realtidskommunikation av audio/video-data av typen "många-till-många". Båda dessa kategorier uppvisar hög känslighet för såväl nätverkets tillstånd (latens, jitter, paketförluster, bithastighet) och lastprofiler av ström bearbetning "back-end" (latens och jitter introducerat som Cloud-hantering av mediadatapaket). Denna avhandling adresserar förbättringar inom realtidshantering av mediainnehåll både med avseende på nätverksnivåns parametrar och optimeraringar för molninfrastrukturen. I detta sammanhang har vi undersökt huruvida nätverksresurserna kan kontrolleras på servicenivån i syfte att öka nätverkets effektivitet och prestanda, och även att kvantifiera påverkan av den delade nätverksresursen på servicekvaliteten. Delade nätverksresurser påverkar molntjänstens prestanda och dessa kan genom en optimering eller handel med nätverksresurser förbättra molntjänstens prestanda. Denna potentiella prestandadegradering beror på en oreglerad delad nätverksinfrastruktur (allokeringen av bandbredd är inte medveten om prestanda och mål för servicenivån). Genom att mediera nätverkets bandbredd genom prediktiv kontroll, möjliggör vi ett bättre utnyttjande av de tillgängliga nätverksresurserna och en lägre grad av avvikelser mot SLO, vilket leder till en ökad stabilitet med åtminstone en faktor 2. Accessnätverken (AN) (edge, kärnnätverk) hos ISP, bärare och lokala nätverk har ingen generell molninfrastruktur, medan s.k. "Internet Resource Providers" erbjuder resurser för molntjänster "on demand". Vi ser en möjlighet till ensande av resurserna inuti ett AN och utanför i syfte att erbjuda ett samlat molntjänsterbjudande genom s.k. "Cloud Federation" och erbjuder tjänstemobilitet för användarna för att optimera lokaliteten. Denna forskningansats erbjuder ett nytt hybrid nätverk med Federated Cloud arkitektur vilken ger en utvidgad nätverksinfrastruktur med en storskalig användning av molntjänster, som direkt inkorporerar denna i nätverksinfrastrukturen. Den nyskapande "Multi-Cloud"-arkitekturen möjliggör för tjänster att balansera lokalitet för användaren mot run-time-prestanda och därigenom optimera för latens mot optimal resursallokering för realtidstillämpningar. För att optimera latensen i "live streaming"-tillämpningar föreslås en nyskapande självstyrd "multi-Cloud-overlay"-algorithm baserad på gradienttopologi i vilken varje moln för en tillämpning inom "stream broadcasting" optimerar klientens närhet till källan. Tillämpningsmodellen separeras i en tvålagersdesign, "multi-cloud delivery back-end" och "stream clients". Denna självreglerande gradientbaserade "back-end" minimerar trafiklasten genom att skapa ett minimalt spännande träd genom molnen som används för routing av strömmarna. Den föreslagna algoritmen har en mycket snabb konvergenshastighet vid större moln, och påverkas inte av "churn" hos molnresursen liksom att den erbjuder ökad motståndskraft hos "live"-strömmen. I detta arbete erbjuder vi mediakvalitetsanalys och förstärkning av realtidsmolnets "forwarders", liksom även allokeringsstrategier för att förstärka servicenivåprestanda hos "Web Real-Time Communication"-plattformar. Mediakvalitetsmönster påverkas kraftigt av molnets bearbetsningsprestanda, och således kan vi genom att påverka denna aspekt kontrollera mediakvaliteten. Specifikt demonstrerar vi empiriskt att efterhand som sessionsstorlekarna ökar, så utklassar simulcast enlagersinkodning. Dessutom introducerar vi en strömallokeringsalgoritm för att minimera "load spikes" hos "Cloud stream forwarders" och jämför beteendet hos olika strömallokeringspolicys. Med enbart minimal information och allokeringsbehoven hos en enskild serversession beter sig den minimala lastbalansallokeringspolicyn tydligt bättre än andra "rotational"- eller "static threshold"-baserade algoritmer. / <p>QC 20170425</p>

Este trabalho apresenta um estudo sobre o problema de substituição de linhas de cache em microprocessadores. Inspirado no conceito de Entropia da Informação proposto em 1948 por Claude E. Shannon, este trabalho propõe uma nova heurística de substituição de linhas de cache. Seu objetivo é capturar e explorar melhor a localidade de referência dos programas e diminuir a taxa de miss rate durante a execução dos programas. O algoritmo proposto, Entropy, utiliza a heurística de entropia da informação para estimar as chances de uma linha ou bloco de cache ser referenciado após ter sido carregado na cache. Uma nova função de decaimento de entropia foi introduzida no algoritmo, otimizando seu funcionamento. Dentre os resultados obtidos, o Entropy conseguiu reduzir em até 50,41% o miss rate em relação ao algoritmo LRU. O trabalho propõe, ainda, uma implementação em hardware com complexidade e custo computacional comparáveis aos do algoritmo LRU. Para uma memória cache de segundo nível com 2-Mbytes e 8-way associative, a área adicional requerida é da ordem de 0,61% de bits adicionais. O algoritmo proposto foi simulado no SimpleScalar e comparado com o algoritmo LRU utilizando-se os benchmarks SPEC CPU2000. / This work presents a study about cache line replacement problem for microprocessors. Inspired in the Information Entropy concept stated by Claude E. Shannon in 1948, this work proposes a novel heuristic to replace cache lines in microprocessors. The major goal is to capture the referential locality of programs and to reduce the miss rate for cache access during programs execution. The proposed algorithm, Entropy, employs that new entropy heuristic to estimate the chances of a cache line to be referenced after it has been loaded into cache. A novel decay function has been introduced to optimize its operation. Results show that Entropy could reduce miss rate up to 50.41% in comparison to LRU. This work also proposes a hardware implementation which keeps computation and complexity costs comparable to the most employed algorithm, LRU. To a 2-Mbytes and 8-way associative cache memory, the required storage area is 0.61% of the cache size. The Entropy algorithm was simulated using SimpleScalar ISA simulator and compared to LRU using SPEC CPU2000 benchmark programs.

Im ersten theoretischen Teil der Arbeit werden die Geschichte der Heimatmuseen sowie die Forschungsgeschichte des Heimatbegriffes vorgestellt. Neben historischen und volkskundlichen Forschungen werden hier auch soziologische und ethnologische Theorien berücksichtigt, die vor dem Hintergrund der Globalisierungsdebatte entwickelt wurden. Der Heimatbegriff wird auf diese Weise mit dem Begriff locality gleichgesetzt. Im zweiten empirischen Teil werden die Bezirksmuseen der Berliner Bezirke Köpenick, Prenzlauer Berg, Kreuzberg und Neukölln beschrieben und analysiert; hierbei liegt der Schwerpunkt auf der Untersuchung der Dauer- und Wechselausstellungen und der Analyse der materiellen Kultur im Rahmen der Museumsarbeit. Das bedeutet, dass nicht nur die thematischen Ausrichtungen der Dauer- und Wechselausstellungen analysiert werden, sondern auch die Auswahl der Exponate sowie die Präsentationsweisen. Dabei finden auch Ansätze der Forschung über materielle Kultur Berücksichtigung. Schließlich wird ausgehend von den ersten beiden Teilen ein Museumsmodell entwickelt, das die herausgearbeiteten Charakteristika der Museumsarbeit aufgreift und vor dem Hintergrund des besprochenen Heimatbegriffes erweitert. Heimatmuseen werden hier als ethnographische Museen betrachtet, die aktuelle gesellschaftliche Entwicklungen und Diskussionen widerspiegeln können. / The first part of the dissertation describes the history of local museums and the history of scientific research on the concept of Heimat. Not only historical and ethnographic studies are taken into account, but also sociological and anthropological theories which were formulated against the background of the debate on globalisation. In this way Heimat and locality are treated as equivalent terms. The second empirical part describes and analyses the local museums of the districts Köpenick, Prenzlauer Berg, Kreuzberg and Neukölln in Berlin. The focus lies on the research on main and temporary exhibitions and the analysis of the role of material culture within the framework of the museums. This means the examination of both the thematic structuring of the exhibitions and the range of objects and their presentation. In the course of this material culture studies are taken into consideration. Coming from the first two parts a museum concept is formulated. This concept takes up the characteristics of the museums and adds the considerations on the terms Heimat and locality. In conclusion local museums are regarded as ethnographic museums which are able to reflect current social developments and discussions.

Este trabalho se insere no âmbito dos estudos a respeito da formação de palavras e pretende analisar a estrutura morfossintática de diminutivos e aumentativos do português brasileiro construídos com os formadores -inh/-zinh e -ã/-zã, respectivamente. Mais especificamente, o recorte empírico feito na tese pode ser dividido em duas grandes linhas: uma que aborda a interação entre as marcas de diminutivo e de aumentativo com as marcas de gênero/classe nominal e outra que contempla a (im)possibilidade de que as formações diminutivas e aumentativas sejam não-composicionalmente interpretadas. Para tanto, reanalisamos o estatuto das noções de gênero e classe nominal, propondo que elas ocupam a mesma posição na estrutura sintática. Tal posição é identificada como uma projeção de gênero, que é parte da projeção estendida do nome. Os formadores de diminutivo e aumentativo são analisados com base nas relações que estabelecem com esse núcleo sintático de gênero. A hipótese de base é a de que nuances na relação entre os formadores avaliativos e o núcleo de gênero revelam aspectos significantes da posição estrutural ocupada por cada um deles. Em linhas gerais, propomos que o diminutivo -inh se diferencia dos outros formadores por compartilhar com a raiz o mesmo núcleo de gênero. Essa estrutura é capaz de dar conta, entre outros fatos empíricos, da possibilidade de que a vogal final da forma diminutiva seja idêntica à vogal final da forma não-diminutiva, ainda que tal vogal seja condicionada pela raiz. Por outro lado, a vogal final que completa o aumentativo -ã reflete os padrões gerais da língua, independentemente da raiz presente na derivação. Tomamos esse fato como evidência de que o aumentativo e a raiz possuem núcleos independentes de gênero. No que diz respeito às construções aumentativas e diminutivas encabeçadas pela consoante -z, a presença de núcleos de gênero independentes na estrutura sintática é ainda mais clara, uma vez que tanto a vogal que completa a raiz, como a vogal que completa as formas -zinh e -zã são fonologicamente realizadas. Assim como no caso das formações com -ã, a vogal que completa os formadores de grau encabeçados por -z é completamente independente da raiz que participa da formação, seguindo o padrão mais geral da língua. Esses fatores fazem a análise do aumentativo -ã e a análise das formas encabeçadas por -z bastante similares uma à outra: tais formas possuem, em sua estrutura sintática, um núcleo de gênero que é independente daquele que categoriza a raiz. No entanto, há diferenças no comportamento dessas formas que acabam por separar de um lado o aumentativo -ã e, de outro, as formas -zinh e -zã. Propomos que tais diferenças são derivadas do fato de o primeiro formador se anexar abaixo do núcleo de número, enquanto os dois últimos entram na estrutura depois que ela já possui um núcleo de número. Por fim, dentro de uma visão localista de gramática, em que a atribuição de significado não-composicional deverá ser licenciada dentro de domínios bem definidos, discutimos a composicionalidade das formações partir das posições sintáticas atribuídas a cada um dos formadores em questão. / This work is inserted within the scope of the studies that investigate word formation, and aims to analyze the morphosyntactic structure of diminutives and augmentatives in Brazilian Portuguese built with the formatives -inh/-zinh, and -ã/-za, respectively. More specifically, the empirical path of this thesis can be divided into two main lines: one that addresses the interaction of diminutive and augmentative with the notions of gender/noun classes, and one that addresses the (im)possibility of a non-compositional interpretation being attributed to the structure. In order to do so, it was necessary to review the status of notions like gender and noun class in the grammar, and the formal representation attributed to them. I propose that gender and noun class occupy the very same position in the syntactic structure. This position is identified as a gender projection, which is part of the extended projection of the noun. Diminutive and augmentative markers are, then, analyzed based on the relations they establish with the syntactic gender head. The underlying hypothesis is that differences in the relation established between the evaluative formatives and the gender head reveal important aspects of the structural position that hosts each of them. More specifically, I propose that the diminutive -inh differs from other formatives, because it shares with the root the same gender head. In this sense, -inh is attached to the same gender projection responsible for categorizing the root. This structure is capable of accounting, among other empirical facts, for the possibility that the final vowel of diminutive and nondiminutive forms is identical, even if it is a root-conditioned element. On the other hand, the final vowel that completes -ã augmentatives always reflects the general gender pattern of the language, quite independently of the root. I take this as evidence that, unlike the diminutive, the augmentative and the root have independent gender heads. Thus, the syntactic structure proposed for the augmentative formation has two gender heads: one that attaches to the root, and another one that attaches directly to the augmentative. The distance between the root and the gender head that follows the augmentative is responsible for the default phonological realization of the final vowel of the augmentative. In the same sense, in the diminutive and augmentative formations built with -zinh and -zã, the presence of two independent gender heads is even clearer, since they can be phonologically identified in the output form. It is proposed, then, that, just as in is the case of the -ã forms, the vowel completing -zinh and -zã occupies a gender head that is independent from the one that categorizes the root. Differences in the behavior of these forms point to a split between -ã on one side, and -zinh/-zã on the other. We propose that these differences are derived from the fact that while -ã attaches below a number head, -zinh/-zã, on the other hand, attaches above a number head. Finally, within a localist view of grammar, in which the licensing of noncompositional meaning must be conditioned by local domains of syntactic structure, the possibility and impossibility of non-compositional interpretation being attributed to diminutive and augmentative formations is derived from the syntactic positions assigned to each of the formatives.

O objeto empírico desta tese é um subgrupo de verbos complexos do português brasileiro. Os dados estudados são formações sincronicamente transparentes e composicionais com prefixos a-, eN- e eS- e sufixos -ec-, -iz-, -e- e -ej-, incluindo os chamados verbos parassintéticos, e formações originalmente complexas, porém duvidosas quanto à complexidade atualmente. O corpus contém 380 verbos selecionados a partir de dicionário e organizados por critérios de frequência. O objetivo geral descritivo enfoca questões relativas às propriedades e ao comportamento dos afixos, das raízes e das vogais temáticas. A discussão é organizada em torno dos níveis de estrutura morfológica, morfofonológica, argumental e eventual. O objetivo geral teórico do trabalho consiste em discutir as propostas da Semântica Lexical, da Sintaxe Lexical e da Morfologia Distribuída. Como resultados, o estudo oferece uma primeira classificação em verbos parcialmente transparentes e totalmente transparentes. Aqueles são analisados como fruto de um processo de reanálise histórica comparado ao desaparecimento de preverbos. O estudo mostra que existe um continuum entre formações completamente fossilizadas, reanalisadas como simples, em processo de mudança e completamente transparentes e composicionais. Uma segunda classificação se refere a formações com significado composicional e não-composicional. Os dados não-composicionais são estruturalmente analisados através de uma releitura da restrição de localidade na interpretação das raízes e do uso da noção de polissemia das raízes. Os verbos totalmente transparentes e composicionais são descritivamente classificados em verbos de mudança de estado, de lugar (location), de posse concreta (locatum), de posse abstrata, de reconfiguração e verbos de modificação de v. A característica mais robusta dessa subclasse é a obrigatoriedade de um argumento interno interpretado como objeto afetado (tema ou experienciador, em menor escala) da mudança expressa pelo evento. A investigação aponta que esses prefixos podem ser a realização fonológica de um núcleo misto de natureza lexical e funcional que é responsável por introduzir o argumento interno na estrutura e relacioná-lo à semântica da raiz. Tal núcleo possui minimamente o traço [+r] (relacional) e, em poucos casos, apresenta especificação direcional [+dir]. Com isso, a ideia de que esses prefixos são morfemas direcionais é desmistificada, pois essa informação interna ao verbo complexo é residual e decadente. Em geral, os prefixos se comportam como alomorfes e não há fortes evidências de associação exclusiva de um prefixo a uma determinada estrutura argumental ou classe semântica. Os sufixos são analisados como realizações de núcleos funcionais de tipo v[+voice], v[-voice] e v[+voice, -télico] e também se observa que a ocorrência sufixal em tipos de eventos não se dá de modo tão sistemático como afirma a literatura prévia. A teoria de alomorfia prospota em Embick (2010), baseada em localidade e linearidade, se mostra efetiva para analisar a escolha dos alomorfes dos vi núcleos R (relacionador), v e Th (Vogal temática). O tipo semântico da raiz influencia o tipo de verbo formado, mas pode ser manipulado a fim de sofrer coerção por um processo metonímico ou estrutural. A principal conclusão a partir dos resultados obtidos é que a morfologia verbal do português brasileiro pode revelar tendências em relação à estrutura argumental e a estrutura de eventos, mas não reflete correlações suficientemente regulares ou consistentes. / The empirical object of this dissertation is a subgroup of complex verbs of Brazilian Portuguese. The dataset is composed by synchronically and compositional formations containing the prefixes a-, eN- e eS- and the suffixes -ec-, -iz-, -e- e -ej- and originally complex formations which are dubious in relation to its synchronic complexity. The corpus contains 380 verbs selected from a dictionary and organized by frequency criteria. The general descriptive goal encompasses topics on properties and behavior of affixes, roots and theme vowels and the discussion is guided by the levels of morphological, morphophonological, argument and event structure. The general theoretical goal of this dissertation is to discuss Lexical Semantics, Lexical Syntax and Distributed Morphology proposals. As empirical results, the study offers a primary classification in terms of partially and totally transparent verbs. Partially transparent verbs are treated as resulting from a historical reanalysis process compared to the well known process of disappearance of preverbs. It is assumed that there is a continuum from forms which are: 1) completely fossilized; 2) reanalyzed as simple; 3) forms in process of change; 4) completely compositional and transparent. A secondary classification refers to compositional and noncompositional formations. Non-compositional data are structurally analyzed by means of a new reading on the literature on locality restriction on the interpretation of roots and the use of the notion of root polysemy. Completely compositional and transparent verbs are empirically classified into change of state, change of location, change of abstract and concrete possession, reconfiguration and verbs of modification of v. The strongest characteristic of this subclass is the obligatory presence of an internal argument interpreted as an affected object (theme or experiencer, to a less extent) of the change denoted by the event. The investigation points out that the prefix may be considered as the phonological realization of a head with a mixed lexical functional nature, which is responsible for introducing the internal argument in the structure and relating it to the root semantics. This head has at least the feature [+r] and, in a few cases, it may present directional information [+dir]. Considering this, the assumption that these prefixes are directional morphemes is debunked since this kind of information within a complex verb is residual and decayed. In general, prefixes behave as allomorphs and there are not strong evidences of an exclusive association of a prefix and a certain kind of argument structure or semantic class. The suffixes are analyzed as realizations of three functional heads: v[+voice], v[-voice] and v[+voice, -telic] and it is observed that suffix occurrence in event type is not systematic as previous literature claims. The theory of allomorphy proposed in Embick (2010), which is based on locality and linearity, was efficient in accounting for selection of allomorphs of R, v and Th heads. Finally, semantic type shows influence on verb type but this information can be viii manipulated in order to derive structural or metonymical coercion. The main conclusion to be drawn from the results is the fact that Brazilian Portuguese verbal morphology may reveal certain tendencies in argument and event structure, but it does not reflect sufficiently regular or consistent correlations.