Coordenação de Aperfeiçoamento de Pessoal de Nível Superior / Phonological opacity refers to rules/ processes that should apply in a given context but don‟t or apply even without a visible context. As such processes of underapplication and overapplication are found in the intermediate stages in-between the underlying and surface forms, the Classic Generative Theory (CGT) used to accommodate them without much ado in its rule orderings. Due to its extreme abstraction, new generative models come to light; one of them is the Optimality Theory. In this output-centered model originally just two levels of representation are accepted: the underlying and the surface ones, banning, therefore, intermediate levels between representation and production. Without such levels the theory was obliged to revisit its structure and operation methods so that to take account of opaque generalizations. However, the generative models are still unable to deal with all kinds of opacity with efficacy. This conclusion added to the fact that some scholars point out that there is no unified theory regarding phonological opacity not even in CGT makes us consider that such lack of transparency comes from the paradigm they are immersed into, the rationalism. We think that a change of paradigm would be fruitful to such issue. In this master‟s thesis we contemplate an emergentist view to opacity. During this thesis we come to the conclusion that phonological opacity can be dealt in a transparent manner. Multi-representational and Connectionist models, as they take into consideration a linguistic representation that is coherent with the neurosciences‟ findings, being it flexible, gradual and rich in phonetic details, do narrate the facts considered opaque. However, the first group of models due to being poor in regards of formal characterization becomes a weak model when compared to OT machinery, for example. In this light, we point that a pure connectionist OT (BONILHA, 2004) is an adequate model to handle opacity as it brings transparency to such issue and is also capable to formalize it. / Opacidade fonológica faz referências às regras ou processos que ou deveriam se aplicar em um contexto, mas não se aplicam, ou se aplicam mesmo sem um contexto aparente. Como esses processos de sub-aplicação e superaplicação de regras encontram-se nos estágios intermediários entre a subjacência e a superfície, a Teoria Gerativa Clássica (TCG) os acomodava sem muitos problemas em seus ordenamentos lineares de aplicação. Dada à grande abstração desse modelo, contudo, novos modelos gerativos vêm à tona; um deles é a Teoria da Otimidade (TO). Nesse modelo centrado no output, estão previstos, originalmente, apenas dois níveis de representação: a subjacência e a superfície, banindo, assim, estágios intermediários entre a representação linguística e a produção. Sem tais níveis, a teoria foi obrigada a repensar sua estrutura e funcionamento para tentar acomodar os casos opacos. No entanto, as reestruturações gerativas da TO não conseguem abarcar os fenômenos opacos da língua do homem com eficácia. Abordamos, nessa dissertação, os seguintes modelos gerativos para o tratamento da opacidade: Conjunção Local de Restrições; Teoria da Simpatia; LPM-TO; TO-Serial; e a TO com cadeias de candidatos. Abordamos, também, grande parcela das críticas presentes na literatura para cada um desses modelos. O fato de que a TO Standard e suas variantes não conseguem lidar com a opacidade em todas as suas faces, somado ao fato de que alguns autores apontam que a questão da opacidade não é unificada nem mesmo para a TGC, faz-nos pensar que essa obscuridade é oriunda do paradigma no qual tais modelos se inserem, o paradigma racionalista. Apostamos, então, que a mudança de paradigma seria frutífera para essa problemática. Nesta dissertação de mestrado, desta forma, abrimos mão de representações linguísticas cartesianas e de um mapeamento linguístico complexo e optamos por contemplar uma visão emergentista para a opacidade fonológica. Constatamos ao longo do trabalho que, nesse novo prisma, a opacidade pode ser acomodada de forma transparente. Modelos multirrepresentacionais e modelos conexionistas, ao abarcarem uma representação linguística coerente com os achados da neurociência, sendo esta flexível, gradual e rica de detalhamento fonético, conseguem narrar os fatos taxados como opacos. Contudo, a primeira sorte de modelos, ao carecer de formalização linguística, torna-se enfraquecida ao comparar-se com os mecanismos da TO, por exemplo. Nesta luz, propomos que a TO puramente conexionista, proposta por Bonilha (2004), seja um modelo adequado para acomodar a opacidade, tendo em vista que esse, além de conseguir trazer a transparência para a opacidade fonológica, também consegue, como atestamos, formalizar esses fenômenos linguísticos.

In this thesis English interdental fricative acquisition by Brazilians second language learners has been investigated. Data from oral productions of three students groups has been gathered. The first group is composed by sixth semester undergraduate students of UNIFRA English course. The second and third groups are composed by third and tenth semester students of an extra curricular English course of UFSC. First, a phonetic transcription of selected words that contained interdental fricatives has been performed. Then, an analysis based on Connectionist Optimality Theory proposed by BONILHA (2004) has been carried out. This analysis uses the reranking of constraints, and it reveals that Brazilians second language learners are prone to make substitutions of the segment /θ/ by the segment [t] as well as the substitutions of the segment /ð/ by the segment [d]. Furthermore, there is a strong impact of the Lexical Frequency on the learners behavior. / No presente trabalho, investiga-se a aquisição das fricativas interdentais por aprendizes brasileiros de inglês como segunda língua (L2). Os dados foram colhidos a partir da leitura de palavras contextualizadas, leitura de um texto e narrativas orais produzidas por três grupos de estudantes de inglês. O Grupo 1 é composto por estudantes universitários que estavam cursando, no momento da coleta, o sexto semestre do curso de Letras do Centro Universitário Franciscano. Os Grupos 2 e 3 são formados por estudantes que freqüentavam o projeto extracurricular de ensino de línguas da Universidade Federal de Santa Catarina e estavam no terceiro e décimo semestres, respectivamente. Após a transcrição fonética e seleção das palavras que continham as fricativas interdentais, foi feita a análise dos dados. Essa análise tem como base a Teoria da Otimidade Conexionista proposta por Bonilha (2004) que, através da hierarquia de restrições, demonstra a gramática da interlíngua apresentada pelo aprendiz. Os resultados apontaram para a emergência precoce da fricativa interdental desvozeada, para as substituições do segmento /θ/ pelo [t] e do segmento /ð/ pelo [d] e também a interferência do tipo de coleta realizada nos percentuais de produções corretas. Foi constatada ainda uma grande influência da freqüência lexical no comportamento dos sujeitos estudados.

No description available.

The aim of this thesis is to improve our understanding of the neural systems involved in schizophrenia by suggesting possible avenues for future computational modelling in an attempt to make sense of the vast number of studies relating to the symptoms and cognitive deficits relating to the disorder. This multidisciplinary research has covered three different levels of analysis: abnormalities in the microscopic brain structure, dopamine dysfunction at a neurochemical level, and interactions between cortical and subcortical brain areas, connected by cortico-basal ganglia circuit loops; and has culminated in the production of five models that provide useful clarification in this difficult field. My thesis comprises three major relevant modelling themes. Firstly, in Chapter 3 I looked at an existing neural network model addressing the Neurodevelopmental Hypothesis of Schizophrenia by Hoffman and McGlashan (1997). However, it soon became clear that such models were overly simplistic and brittle when it came to replication. While they focused on hallucinations and connectivity in the frontal lobes they ignored other symptoms and the evidence of reductions in volume of the temporal lobes in schizophrenia. No mention was made of the considerable evidence of dysfunction of the dopamine system and associated areas, such as the basal ganglia. This led to my second line of reasoning: dopamine dysfunction. Initially I helped create a novel model of dopamine neuron firing based on the Computational Substrate for Incentive Salience by McClure, Daw and Montague (2003), incorporating temporal difference (TD) reward prediction errors (Chapter 5). I adapted this model in Chapter 6 to address the ongoing debate as to whether or not dopamine encodes uncertainty in the delay period between presentation of a conditioned stimulus and receipt of a reward, as demonstrated by sustained activation seen in single dopamine neuron recordings (Fiorillo, Tobler & Schultz 2003). An answer to this question could result in a better understanding of the nature of dopamine signaling, with implications for the psychopathology of cognitive disorders, like schizophrenia, for which dopamine is commonly regarded as having a primary role. Computational modelling enabled me to suggest that while sustained activation is common in single trials, there is the possibility that it increases with increasing probability, in which case dopamine may not be encoding uncertainty in this manner. Importantly, these predictions can be tested and verified by experimental data. My third modelling theme arose as a result of the limitations to using TD alone to account for a reinforcement learning account of action control in the brain. In Chapter 8 I introduce a dual weighted artificial neural network, originally designed by Hinton and Plaut (1987) to address the problem of catastrophic forgetting in multilayer artificial neural networks. I suggest an alternative use for a model with fast and slow weights to address the problem of arbitration between two systems of control. This novel approach is capable of combining the benefits of model free and model based learning in one simple model, without need for a homunculus and may have important implications in addressing how both goal directed and stimulus response learning may coexist. Modelling cortical-subcortical loops offers the potential of incorporating both the symptoms and cognitive deficits associated with schizophrenia by taking into account the interactions between midbrain/striatum and cortical areas.

Coordenação de Aperfeiçoamento de Pessoal de Nível Superior / In this study, we investigated the repair strategies applied to English word primary stress in the process of acquisition by native speakers of Brazilian Portuguese (BP). For this purpose, we used Connectionist Optimality Theory (henceforth, COT) by Bonilha (2004) as its theoretical basis. COT, beyond previewing constraint interaction, the interaction of different phonological levels (such as segment, syllable and stress) and constraints reranking in an L2 acquisition process, it also claims that constraints which are specific from the L2 may be acquired. Such restructuring process may be explained by an implemented gradual learning algorithm (BOERSMA & HAYES 2001) which constitutes a great advantage to acquisition data analysis. In the present study, we analyzed English stress acquisition in suffixed words, especially, the production of words whose suffixes carry the primary stress. In this last group, we observed the following suffixes: -oon, -eer, -ee, -ette, -esque, - ese, -ique, -et, -aire, -euse and -eur. Taking into account the process of constraint reranking, we intended to characterize interlanguage hierarchy evidenced by learners from the repair strategies they applied to deviant productions. For this aim, we used oral language data from sixteen informants, academics of Languages - English major from a Brazilian Southern university. Data collection consisted of two recordings of an instrument which contained 135 suffixed and non-suffixed words and 135 carrier-sentences. Afterwards, data were transcribed by auditory method and reviewed by two evaluators. The transcriptions were based on IPA. Data were statistically treated by Chi-Square test via Statistica 7.0 software. We could observe that, among the suffixes that present different behaviors concerning the stress of the primitive words, the ones which receive the primary stress were the least correctly produced, presenting around 50% of errors. Data also confirmed the trends indicated by studies on L2 acquisition that interlanguage systems show the militancy of L1 hierarchy of constraints. This is due to the predominance of the trochaic pattern as a repair strategy - around 50% of deviant productions, and, as argued by Bisol (1992), it is the stress pattern of BP. Concerning the suffixes that carry the primary stress, it seems that L2 vowel length, that may be the main element of stress attribution in words such as refugee and mountaineer, is not easily perceived/produced by BP native speakers, according to the findings by Nobre-Oliveira (2007). Taking into account the results, we understand that some specific constraints to vowel length were not ranked in learners interlanguage hierarchy yet. Also, we relate the phenomenon of the random production of such stress pattern to its low frequency in English lexicon. Thus, frequency seems to be one important factor for learners oral productions. In sum, we believe that such factors as a whole may be conspiring to the fact that informants productions are still deviant from the pattern of the target language. Finally, we argue that COT seems to be an important tool in order to describe and analyze L2 language acquisition data. / Neste estudo, investigamos as estratégias de reparo aplicadas ao padrão de acento primário do Inglês, no processo de aquisição, por falantes nativos de português brasileiro (PB). Para esse propósito, utilizamos a Teoria da Otimidade Conexionista (COT), a partir dos estudos de Bonilha (2004) como sua base teórica. A COT, além de prever a interação de restrições, a interação de diferentes níveis fonológicos (como segmental, silábico e acentual) e o reranqueamento no processo de aquisição de uma L2, também propõe que restrições específicas da L2 devam ser adquiridas. No presente estudo, analisamos a aquisição do acento do inglês em palavras sufixadas, especialmente, na produção de palavras cujo sufixo carrega o acento, como -oon, -eer, -ee, -ette, -esque, -ese, -ique, -et, - aire, -euse e -eur. Levando-se em consideração o processo de reranqueamento, intentamos caracterizar a hierarquia de interlíngua evidenciada pelas produções desviantes dos aprendizes a tal padrão. Com este objetivo, usamos dados de linguagem oral de dezesseis sujeitos, acadêmicos de Letras - habilitação Língua Inglesa de uma universidade do Sul do Brasil. Os dados consistiram de duas coletas, realizadas por meio de um instrumento que contém 135 palavras (entre sufixadas e nãosufixadas) e 135 frases-veículo. Em seguida, houve a transcrição fonética pelo método de oitiva e a revisão por dois avaliadores. As transcrições tiveram por base o Alfabeto Fonético Internacional (IPA). Os dados foram tratados estatisticamente com o teste Qui-quadrado via programa Statistica 7.0. Pudemos observar que, dentre os sufixos que apresentam comportamentos distintos, no que concerne ao acento da palavra primitiva, os sufixos que atraem o acento primário foram os que apresentaram o menor percentual de acertos, aproximadamente 50%. Os dados também confirmam as tendências, apontadas por outros estudos em aquisição de L2, de que os sistemas de interlíngua apresentam a militância da hierarquia de restrições da L1. Evidenciamos isso pelo predomínio do padrão acentual trocaico como estratégia de reparo, que responde por cerca de 50% dos tokens desviantes e, como argumenta Bisol (1992), o padrão trocaico é o predominante do PB. Com relação aos sufixos que atraem o acento primário, há evidências de que o prolongamento da vogal, que é o principal elemento considerado na atribuição de peso da sílaba e na atração do acento em palavras como refugee e mountaineer, não está sendo adquirido pelos falantes nativos de PB, conforme já aponta o estudo de Nobre-Oliveira (2007). Visto os dados de que dispomos nesta pesquisa, compreendemos que algumas restrições ligadas ao prolongamento e fidelidade da vogal acentuada, bem como algumas restrições de marcação, precisam ainda ser ranqueadas na hierarquia de interlíngua dos aprendizes. Também, relacionamos o fenômeno de atribuição flutuante do padrão oxítono sufixado a sua baixa frequência no léxico do Inglês. Assim, observamos que a frequência também parece desempenhar um importante papel nas produções orais dos sujeitos. Em suma, acreditamos que tais fatores devem conspirar para a emergência de produções desviantes ao padrão da língua alvo. Por fim, com este estudo, pudemos comprovar a validade dos presupostos da COT e tomá-la como um modelo linguístico pertinente à descrição e análise de dados de aquisição de L2.

This thesis explores the use of artificial neural networks for modelling cognitive processes. It presents an exposition of the neural network paradigm, and evaluates its viability in relation to the classical, symbolic approach in cognitive science. Classical researchers have approached the description of cognition by concentrating mainly on an abstract, algorithmic level of description in which the information processing properties of cognitive processes are emphasised. The approach is founded on seminal ideas about computation, and about algorithmic description emanating, amongst others, from the work of Alan Turing in mathematical logic. In contrast to the classical conception of cognition, neural network approaches are based on a form of neurocomputation in which the parallel distributed processing mechanisms of the brain are highlighted. Although neural networks are generally accepted to be more neurally plausible than their classical counterparts, some classical researchers have argued that these networks are best viewed as implementation models, and that they are therefore not of much relevance to cognitive researchers because information processing models of cognition can be developed independently of considerations about implementation in physical systems. In the thesis I argue that the descriptions of cognitive phenomena deriving from neural network modelling cannot simply be reduced to classical, symbolic theories. The distributed representational mechanisms underlying some neural network models have interesting properties such as similarity-based representation, content-based retrieval, and coarse coding which do not have straightforward equivalents in classical systems. Moreover, by placing emphasis on how cognitive processes are carried out by brain-like mechanisms, neural network research has not only yielded a new metaphor for conceptualising cognition, but also a new methodology for studying cognitive phenomena. Neural network simulations can be lesioned to study the effect of such damage on the behaviour of the system, and these systems can be used to study the adaptive mechanisms underlying learning processes. For these reasons, neural network modelling is best viewed as a significant theoretical orientation in the cognitive sciences, instead of just an implementational endeavour. / Psychology / D. Litt. et Phil. (Psychology)

In this dissertation a formalisation of nonmonotonic reasoning, namely Default logic, is discussed. A proof theory for default logic and a variant of Default logic - Prioritised Default logic - is presented. We also pursue an investigation into the relationship between default reasoning and making inferences in a neural network. The inference problem shifts from the logical problem in Default logic to the optimisation problem in neural networks, in which maximum consistency is aimed at The inference is realised as an adaptation process that identifies and resolves conflicts between existing knowledge about the relevant world and external information. Knowledge and data are transformed into constraint equations and the nodes in the network represent propositions and constraint equations. The violation of constraints is formulated in terms of an energy function. The Hopfield network is shown to be suitable for modelling optimisation problems and default reasoning. / Computer Science / M.Sc. (Computer Science)

本研究旨在探討如何將漢語中詞彙的產出，建構在過去的語言生成模型(speech production model)之上。 在過去文獻中，語言生成模型主要有兩大流派：序列模型(serial-ordering model)主張各階段的激活過程只能單向進行，而連接模型(connectionist model)則允許各階段之間有雙向互動。此外，兩大模型闡述綜合性詞誤 (同時在語意及語音方面相關的詞誤)的生成也略有不同：序列模型認為綜合性詞誤起因於功能層次(functional level)與位置層次(positional level)同時發生錯誤，但若綜合性語誤的比例顯著高於機率，則較不支持此說法；連接模型則認為綜合性詞誤來自音系層(phonological stage)和詞條層(lemma stage)間的交互作用，倘若大部分的縱向聚合詞誤(paradigmatic errors)皆具有語音相似度，此說法較能得到驗證。 因此，為估算漢語詞彙選擇中語音的相似度，本研究分析了421個在自然語境中收集的單音節縱向聚合詞誤，結果發現綜合性語誤所佔的比例顯著偏高，此外，所有詞誤在字首、語音特徵、韻母、音節結構和聲調各方面的效應都達到顯著，這說明了語音和語意的確會同時影響詞彙的選擇，這也表示連接模型可能較接近中文的語言生成過程。 另外，語料分析的結果也顯示出漢語詞彙產出的特性。第一、在詞彙提取的過程中，字首比語音特徵和韻母扮演了更重要的角色，這同時也支持音節結構應以字首(C)和韻母劃分(CV/CVC)的假設；第二、聲調效應的顯著顯示聲調應存在於字彙的底層結構，因此聲調的促發應先於詞彙結的激化；第三、目標詞彙(target)的音節結構應在音系層的初期就已指定，因此有相同音節結構的詞彙結容易勝出。綜合以上可知，漢語的生成過程較傾向連接模型的架構，唯聲調和音節結構等語言特色宜納入模型討論。 / This study aims to investigate the process of word production in Mandarin, to see how it can be structured in previous models. Speech production models have two primary sects—the serial-ordering model versus the connectionist model—arguing for uni-directional and bi-directional activation respectively. Besides, between these two models, the generation of both semantically- and phonologically-related lexical errors (mixed errors) is different. On the one hand, the serial-ordering model (Garrett, 1988) interpreted mixed errors as malfunctions occur at both functional and positional level. It may not be favored if mixed errors occur more than chance in the corpus. On the other hand, the connectionist model (Dell, 1986) explained it as the feedback activation from the phonological level back to the lemma level. It would be supported if most paradigmatic substitutions show phonological similarities. Therefore, in order to facilitate the phonological similarities in lexical substitutions, 421 single-syllable paradigmatic (non-contextual) lexical errors collected in natural settings are examined. It is found that the percentage of mixed errors is significantly higher. Moreover, in the corpus the initialness, similarity, rhyme, syllable structure and tone effects are all prominent, which proves the joint effect of semantic and phonology in word selection. In other words, the connectionist network might better account for the processing of Mandarin. In addition, the analysis reveals certain issues of word-production in Mandarin. First, initials are primary activation units in word retrieval, rather than phonetic features or rhymes. It also supports the division of hierarchical syllable structure as an onset (C) and rhyme (VC/CVC). Secondly, the significance of tone effect suggests that tone may be stored in the underlying phonological organization of lexicons and thus is prior in word-activation. Thirdly, the syllable structure of the target is assigned at the beginning of phonological stage and thus lexical nodes with the same syllable structure tend to be selected. In conclusion, the analysis proves that the connectionist model could be applied to Mandarin Chinese if the status of tone and syllable structure is included.

Lesen zielt darauf ab, Bedeutung aus geschriebenem Text zu extrahieren. Interessanterweise unterscheiden sich Wörter beträchtlich hinsichtlich der Menge mit ihnen assoziierter Bedeutung, und es wurde kürzlich gezeigt, dass eine hohe Bedeutungshaltigkeit lexikalische und semantische Aufgaben erleichtert. Die vorliegende Dissertation kombiniert ereigniskorrelierte Potentiale (EKPs) und konnektionistische Modellierung, um einige offene Fragen zur Rolle der Bedeutungshaltigkeit bei der Wortverarbeitung anzugehen. Hierbei wurden EKPs verwendet, um den Zeitverlauf unabhängiger Einflüsse der Anzahl semantischer Merkmale und Assoziationen beim Wortlesen zu bestimmen sowie Einflüsse von Bedeutungshaltigkeit auf implizites Wortlernen zu untersuchen. Um die zugrundeliegenden Mechanismen besser zu verstehen, wurden die Ergebnisse anschließend mittels eines semantischen Netzwerk-Modells simuliert. Es zeigten sich keine Einflüsse der Anzahl der Assoziationen, aber eine schnelle Aktivierung semantischer Merkmale, die das EKP bereits ab 190 ms beeinflussten - nur 20 bis 30 ms nach und zeitlich überlappend mit der Aktivierung orthographischer Repräsentationen, die durch N1-Lexikalitätseffekte angezeigt wurden. Im weiteren Verlauf ging eine hohe Merkmalsanzahl mit größeren N400-Amplituden einher. Zudem verstärkten semantische Merkmale Wiederholungseinflüsse auf die Akkuratheit lexikalischer Entscheidungen und N400-Amplituden, was einen ersten Hinweis auf Einflüsse von Bedeutungshaltigkeit auf implizites Wortlernen darstellt. Diese Ergebnisse stehen im Einklang mit merkmalsbasierten semantischen Netzwerk-Modellen. Simulationen legen nahe, dass semantische Aktivierung lexikalische Entscheidungen erleichtert, während Netzwerk-Fehler in engem Zusammenhang mit N400-Amplituden stehen. Da Netzwerk-Fehler psychologisch als implizite Vorhersagefehler interpretiert werden, deuten diese Ergebnisse darauf hin, dass N400-Amplituden implizite Vorhersagefehler im semantischen System widerspiegeln. / Language ultimately aims to convey meaning. Importantly, the amount of associated semantic information varies considerably between words. Recent evidence suggests that the richness of semantic representations facilitates performance in lexical and semantic tasks, but much remains to be learned about semantic richness effects. The present dissertation combined event-related brain potentials (ERPs) and connectionist modeling to address several unresolved issues concerning the role of semantic richness in word processing. Specifically, ERPs were employed to investigate the time course of independent influences of the number of semantic features and associates during word reading (study 1) and influences of semantic richness on implicit word learning (study 2). Aiming at advancing a mechanistic understanding of the obtained results, both studies were subsequently simulated using a network model of semantic cognition (study 3). Results showed no influences of the number of associates, but fast access to semantic features, with influences of feature-based semantic richness starting at about 190 ms - a mere 20 to 30 ms after and temporally overlapping with the activation of orthographic representations as reflected by N1 lexicality effects. Later on, a high number of semantic features induced larger N400 amplitudes. Furthermore, the number of semantic features enhanced repetition priming effects on lexical decision accuracy and N400 amplitudes, providing initial evidence for influences of semantic richness on implicit word learning. These results are in line with feature-based network models of semantic cognition. Simulations with such a model suggest that semantic activation can facilitate lexical decisions, while network error closely corresponds to N400 amplitudes. In psychological terms, network error has been conceptualized as implicit prediction error. Thus, these results are taken to suggest that N400 amplitudes reflect implicit prediction error in semantic memory.

This thesis explores the use of artificial neural networks for modelling cognitive processes. It presents an exposition of the neural network paradigm, and evaluates its viability in relation to the classical, symbolic approach in cognitive science. Classical researchers have approached the description of cognition by concentrating mainly on an abstract, algorithmic level of description in which the information processing properties of cognitive processes are emphasised. The approach is founded on seminal ideas about computation, and about algorithmic description emanating, amongst others, from the work of Alan Turing in mathematical logic. In contrast to the classical conception of cognition, neural network approaches are based on a form of neurocomputation in which the parallel distributed processing mechanisms of the brain are highlighted. Although neural networks are generally accepted to be more neurally plausible than their classical counterparts, some classical researchers have argued that these networks are best viewed as implementation models, and that they are therefore not of much relevance to cognitive researchers because information processing models of cognition can be developed independently of considerations about implementation in physical systems. In the thesis I argue that the descriptions of cognitive phenomena deriving from neural network modelling cannot simply be reduced to classical, symbolic theories. The distributed representational mechanisms underlying some neural network models have interesting properties such as similarity-based representation, content-based retrieval, and coarse coding which do not have straightforward equivalents in classical systems. Moreover, by placing emphasis on how cognitive processes are carried out by brain-like mechanisms, neural network research has not only yielded a new metaphor for conceptualising cognition, but also a new methodology for studying cognitive phenomena. Neural network simulations can be lesioned to study the effect of such damage on the behaviour of the system, and these systems can be used to study the adaptive mechanisms underlying learning processes. For these reasons, neural network modelling is best viewed as a significant theoretical orientation in the cognitive sciences, instead of just an implementational endeavour. / Psychology / D. Litt. et Phil. (Psychology)