Made available in DSpace on 2016-06-02T19:44:02Z (GMT). No. of bitstreams: 1 1686.pdf: 849120 bytes, checksum: 5c60c9efa387d12681a75a1c661ea2cc (MD5) Previous issue date: 2005-03-31 / Universidade Federal de Sao Carlos / The cochlear implant is a prosthetic electric device that is implanted through surgery. I has the propose of substituting the cochlear functions allowing thus, the individuals with pre and post-linguistic neurosensorial deafness to begin to detect environmental sounds. One of the necessary procedures to help someone to develop hearing skills after the implant is to trace electrodes, which is a procedure that basically consist of determining measures of threshold and maximum hearing comfort for each implanted electrode. In order to obtain these within the clinical practice, the oral skills of the individuals with implant must be considered. The precision of these measures become very compromised among implanted children with pre-linguist deafness, due to the deficit of their speech capacity. The purpose of the present study is to propose an evaluation of two procedures that attempt to obtain measures of both, hearing threshold and maximum hearing comfort without using neither, directions or verbal answers. The study consist de two experiments composed by seven pre-linguist children with profoundly neurosensorial deafness, which have acquired the Nucleus 24® cochlear implant system . In one of de procedures, a simple operant discrimination between the presence and absence of de hearing stimulus was established during the training phase. Following, a testing phase was introduced in order to determine the hearing threshold to assess the discrimination loss through a descendant scale, which measures the variation intensity of the electric stimulus (Experiment 1). In the other procedure, it was established that during the electric stimulation of the training phase, an answer could stop the stimulus while no answer produce a positive reinforcement. Next, the intensity of the stimulus was modified during the test phase in ascendant scales with or without an answer, in order to determine whether or not the stimulus was comfortable to the subject (Experiment 2). In both procedures, the electric stimulation was realized through an electrode located in the cochlear medial region. Electro-physiologic measures of the ECAP and ESR were evaluated in an additive form within the same date that the operational procedures were realized. Also, measures of the hearing threshold obtained during clinical approach of the last evaluation, preceding the beginning of the experiments, were registered. All of these measures were compared to results obtained from de operant procedures. The results demonstrated in the Experiment 1, all of the participants learned the taught discrimination and how to use this skill during the variation of the electric stimulation in descendant scales. This experiment was also used to indicate hearing threshold measures with consistence to all of the participants. In the Experiment 2, five of the seven participants mastered the procedure; however, the results indicate that the order two participants learned that the answer would interrupt an electric stimulus and its absence would lead to the reinforcement. Thus, their performance aimed for identification of the consistence level of the maximum comfort in ascendant scales of variation. The training phase was not successful with the other participants, compromising then the consistence of the performance demonstrated during the testing phase; nonetheless, the measures are suggested for them as well. Concerning the relationship among the operant measures and the other measures betaken in the study, the results show that measures of operant threshold are similar to or smaller than the measures obtained through clinical methods. Regarding the maximum comfort values as the most relevant data. It has been emphasized that the two participants that fulfilled the expected performance in the procedure, demonstrated stapedial reflex values (ESR) similar to the obtained from operant form. The data suggest initial evidence that the operant procedure can contribute to the realization of tracing electrodes with larger precision among pre-linguistic children with implants. / O implante Coclear trata-se de uma prótese eletrônica que é implantada cirurgicamente e substitui as funções da coclear, permitindo que indivíduos com surdez neurossensorial pré e pós-lingual passem a detectar som do ambiente. Um dos procedimentos necessários para o desenvolvimento de habilidades auditivas após a realização do implante é o mapeamento de eletrodos, que consiste basicamente, em determinar medidas de limiar e máximo conforto auditivo para cada eletrodo implantado. Dentro da prática clinica a obtenção destas medidas dependem das habilidades orais dos implantados. Entre crianças implantadas, com surdez pré-lingual, a precisão destas medidas torna-se bastante comprometida, pois a capacidade de fala é seriamente limitada. O presente estudo propõe e avalia, em dois experimentos, dois procedimentos operantes para obtenção de medidas de limar auditivo e máximo conforto auditivo sem o uso de instrução ou resposta verbal, com sete crianças com surdes neurossensorial profunda e pré-lingual, usuárias do implante coclear tipo Nucleus 24®. Em um procedimento, na fase de ensino, era estabelecida uma discriminação simples operante entre presença e ausência de estímulos auditivos, em seguida, na fase de testes, a deterioração nesta discriminação ao longo de escalas descendentes da variação na intensidade do estímulo elétrico foi utilizada para avaliar limiar auditivo (Experimento 1). No outro procedimento, na fase de ensino, foi estabelecido que durante a estimulação elétrica, uma resposta interrompia o estímulo e ausência dessa resposta produzia um reforçador positivo, posteriormente, na fase de testes, em escalas ascendentes a intensidade do estímulo era variada e a ausência ou presença de reposta indicava se o estímulo era confortável ou desconfortável (Experimento 2). Em ambos os procedimentos a estimulação elétrica era feita em um eletrodo localizado na região medial da cóclea. Foram avaliadas, em acréscimo, medidas eletrofisiológicas de ECAP e ESR na mesma data em que os procedimentos operantes foram realizados e, também, foram registradas medidas de limiar auditivo obtidas por métodos clínicos, obtidas na ultima avaliação antes do inicio dos experimentos. Todas estas medidas foram comparadas às obtidas pelos procedimentos operantes. Os resultados mostraram que, no Experimento 1, todos os participantes aprenderam a discriminação ensinada e o uso desta habilidade durante a variação da intensidade do estímulo elétrico em escalas descendentes, serviu para indicar medidas de limiar auditivo com consistência também para todos. No Experimento 2, dos sete participantes, cinco realizaram o procedimento e os resultados indicam que dois aprenderam que a resposta interrompia um estímulo elétrico e a ausência desta permitia acesso ao reforçador, posteriormente, o desempenho de ambos em escalas ascendente de variação na intensidade, permitiu identificar nível de máximo conforto com consistência. Com os demais participantes, a fase de ensino não foi eficiente, comprometendo a consistência do desempenho mostrado na fase de teste, mas medidas são sugeridas também para estes participantes. Na relação entre medidas operantes e as outras medidas tomadas neste estudo, os resultados mostram que medidas de limiar operante são semelhantes ou menores que as medidas obtidas por métodos clínicos. Com os valores de máximo conforto, como dado de maior relevância, destaca-se que os dois participantes que apresentaram desempenho esperado no procedimento demonstraram valores de reflexo estapediano (ESR) semelhantes aos obtidos de forma operante. Os dados sugerem evidências iniciais de que procedimentos operantes podem contribuir na realização de mapeamentos de eletrodos com maior precisão entre crianças implantadas pré-linguais. Palavras-chave: Implante coclear; discriminação operante; limiar auditivo; conforto auditivo; criança pré-lingual

Made available in DSpace on 2016-06-02T19:45:54Z (GMT). No. of bitstreams: 1 DissTFG.pdf: 1894668 bytes, checksum: 746e2fb775dd1a84e0583b87b1bd952e (MD5) Previous issue date: 2005-05-29 / Universidade Federal de Sao Carlos / The cochlear implant is a device that allows deaf people to detect surrounding sounds, including the human speech, which are transformed in electric stimulation of the cochlea. This is just the starting point for those who were deprived of audition, especially if the deafness occurred before language acquisition (pre-lingual deafness). After the implant, the person needs to learn (or relearn) to listen, and this is an important research and intervention issue. The goal of this study was to describe the evolution of the listening behavior in pre-lingual implanted children, in a period of time correspondent to the first 26 months after the implant. The analysis consisted of word recognition, figure naming and echoic behavior. In the recognition task, for each dictated word the participants were required to identify the corresponding figure (between three possible figures simultaneously available). On the figure naming task, a figure was shown and the participants were required to pronounce its name. On the echoic behavior required that a dictated word should be repeated by the participants. Twenty words were tested on each evaluation for word recognition, and the same twenty words were tested on the two remaining tasks. A microcomputer presented the participants with the stimuli (so oral-facial clues were avoided), and recorded responses that required selection, spoken words by children were videotaped and transcribed by direct observation. Three successive evaluations were performed with a mean interval of 5 months between them. Pre-training was conducted in order to orient the children on how to do the tasks and use the mouse. The tasks were applied individually, and the data analysis was also performed separately for each participant. The initial evaluation showed intermediate scores on word recognition and low scores or no correct answers on figure naming and echoic behavior. Progressive and substantial increase in word recognition naming and echoic behaviors also improved, but the magnitude of progress was much smaller than that observed in word recognition. In addition to the correct/incorrect criterion of analysis, answers that were close to the vocal emission considered correct (partially correct answers) were also analyzed, and a distinction between wrong answers and no answers at all was made. The evaluation of the partially correct answers was important for the identification of a progressive improvement on the echoic behavior and figure naming performances in which, although the children didn t reach the absolute correct criterion, they showed a systematic progress and attested learning of these important behaviors, which are part of language and whose acquisition is dependent upon audition. A second important observation was the reduced occurrence of no answers in the initial evaluations and the absence of this category of responding in the last evaluation. The improvements observed in the children s behavior are important indictors of learning oral language (comprehension and intelligible speech) after the implant; in the other hand, the gaps observed in their performances strongly suggest the need for educational interventions that aim to accelerate the acquisition of the implant user s repertoires, as well as to refine them, in order to maximize the benefits provided by this device. / O implante coclear é um dispositivo que permite que pessoas surdas passem a detectar sons do ambiente, incluindo a fala humana, que são transformados em estimulação elétrica da cóclea. Este é apenas um ponto de partida para quem ficou privado da audição, especialmente se a surdez teve início antes da aquisição da linguagem (surdez pré-lingual). A partir do implante, a pessoa precisa aprender (ou reaprender) a ouvir e esta é uma questão importante de pesquisa e intervenção. O presente estudo teve por objetivo descrever a evolução do comportamento de ouvir em crianças implantadas com surdez pré-lingual, nos primeiros 26 meses após o implante. Foram analisados o reconhecimento de palavras, a nomeação de figuras e o comportamento ecóico. No reconhecimento de palavras, cada palavra era ditada e o participante deveria identificar a figura correspondente (entre três simultaneamente disponíveis). Na nomeação uma figura era apresentada e a criança era solicitada a dizer seu nome. No comportamento ecóico uma palavra era ditada e a criança deveria repetí-la. No reconhecimento de palavras foram testadas 20 palavras em cada avaliação, e as mesmas 20 palavras foram usadas nas tarefas de ecóico e na nomeação. A apresentação das tarefas foi gerenciada por meio de um microcomputador que apresentava os estímulos (o que evitava pistas oro-faciais) e registrava as respostas de seleção. As falas da criança foram registradas por observação direta e filmadas em videoteipe. Foram realizadas três avaliações sucessivas, com um intervalo médio de cinco meses entre uma e outra. Foi realizado um pré-treino para que as crianças aprendessem a realizar as tarefas e a manusear o mouse. As tarefas eram aplicadas individualmente, o mesmo ocorrendo com a análise de dados. Observou-se, na avaliação inicial, escores intermediários no reconhecimento de palavras e escores baixos ou ausência de acertos na nomeação de figuras e no comportamento ecóico. Ao longo das três avaliações sucessivas, ocorreu um aumento progressivo e bastante substancial no reconhecimento de palavras e aumentos de menor magnitude na nomeação e no ecóico. Além de analisar acertos e erros, foi conduzida também uma análise de respostas que se aproximassem da emissão vocal considerada correta (acertos parciais) e feita a distinção entre respostas erradas e ausência de respostas. A avaliação das respostas parcialmente corretas foi fundamental na identificação de uma melhora progressiva nas tarefas de emissão ecóica e na nomeação que, embora não atingissem o critério absoluto de acerto, mostraram progressos sistemáticos, atestando aprendizagem desses importantes comportamentos que fazem parte da linguagem e cuja aquisição depende da audição. Uma outra observação importante foi que as ausências de respostas foram reduzidas nas avaliações iniciais e praticamente desapareceram na última avaliação. Os progressos apresentados pelas crianças são um importante indicativo de que elas estão aprendendo linguagem oral (tanto a compreensão quanto a emissão de fala inteligível) após o implante; por outro lado, as lacunas apresentadas nesse desempenho sugerem fortemente a necessidade de intervenções educacionais voltadas para acelerar e aprimorar o repertório dos usuários de implantes, maximizando os benefícios possibilitados por este dispositivo.

Discovering the structure of natural language sentences by semi-supervised methods Rudolf Rosa In this thesis, we focus on the problem of automatically syntactically ana- lyzing a language for which there is no syntactically annotated training data. We explore several methods for cross-lingual transfer of syntactic as well as morphological annotation, ultimately based on utilization of bilingual or multi- lingual sentence-aligned corpora and machine translation approaches. We pay particular attention to automatic estimation of the appropriateness of a source language for the analysis of a given target language, devising a novel measure based on the similarity of part-of-speech sequences frequent in the languages. The effectiveness of the presented methods has been confirmed by experiments conducted both by us as well as independently by other respectable researchers. 1

Estudo das superfícies dorsal e ventral da língua e da superfície de interface epitélio-tecido conjuntivo de Bradypus torquatus, foi realizado empregando-se os métodos de microscopia de luz e microscopia eletrônica de varredura. As línguas de Bradypus torquatus foram removidas, dissecadas e seccionadas em 3 partes: rostral, média e caudal e tratadas pelos métodos de microscopia de luz e eletrônica de varredura. Os resultados revelaram que a língua apresenta um ápice arredondado, um corpo e uma raiz. A mucosa dorsal mostra três tipos de papilas linguais: filiformes, fungiformes e vaIadas e uma superfície inferior lisa. Histologicamente, observou-se que a superfície dorsal da língua contém as papilas filiformes, fungiformes e vaIadas, distribuídas nas regiões rostral, média e caudal. A camada epitelial é de espessura variável e possui os estratos basal, espinhosa, granular e córnea, mostrando as células epiteliais. A lâmina própria é caracterizada por um tecido conjuntivo denso formando as projeções papilares longas, curtas e arredondadas. As imagens tridimensionais obtidas pela microscopia eletrônica de varredura, mostraram numerosas papilas filiformes localizadas essencialmente nas regiões rostral e média. Poucas papilas fungiformes estão presentes na superfície dorsal da língua, tendo na maioria localizada nas margens da extremidade rostral da língua, exibindo as formas arredondadas e baixas. Elas são cobertas de células epiteliais queratinizadas. Na região caudal notou-se duas papilas vaIadas, contendo valas profundas e a superfície coberta de epitélio escamoso estratificado onde pode ser identificados poros gustativos. Nas papilas fungiformes identificou-se vários poros gustativos na superfície. As amostras tratadas com a solução de hidróxido de sódio e examinado ao microscópio eletrônico de varredura, revelaram após a remoção total da camada epitelial, o tecido conjuntivo denso em disposição original apresentando papilas conjuntivas de diferentes formas e tamanhos. As regiões rostral e média evidenciaram numerosas papilas correspondentes às filiformes e fungiformes e na região caudal, a papila conjuntiva da papila vaIada apresentou uma forma alongada. Na superfície foram notadas pequenos forames de ductos de glândulas salivares e projeções papilares maiores e menores em aspectos tridimensionais. Na superfície de interface epitélio-tecido conjuntivo, foram notados espessos feixes de fibras colágenas. Cortes corados por Picro-sírius e analisados por meio de microscopia de luz polarizada evidenciaram a coloração de papilas conjuntivas em aspecto vermelho indicando a presença de fibras colágenas do tipo I e fibras colágenas de coloração esverdeada, tipo III. / The study of the dorsal and lingual surfaces and the surface of interface epithelium-connective tissue of Bradypus torquatus\'tongue was performed employing the light and scanning electron microscopy techniques. The tongue of Bradypus torquatus were removed, dissected and sectioned in three parts: rostral, medium and caudal regions and prepared by light and scanning electron microscopy methods. The results revealed that the tongue present a round apex, dorsum and a root. The dorsal mucosae showed three lingual papillae: filiforms, fungiforms and vallate and a ventral smooth surface. Histologically, it was observed that the dorsal surface of tongue contains the filiform, fungiform and vallate papillae located in the rostral, medium and caudal regions.The epitheliallayer is variable in thickness and possess the basal, spinosum, granular and queratinized layers, showing the epithelial cells. The lamina propria is characterized by a dense connective tis sue forming the long, short and rounded papilar projections. The three-dimensional images obtained by scanning electron microscopy, showed numerous filiform papillae located essentially in the rostral and medium regions. Few fungiform papillae are present in the dorsal surface of tongue, usually located in the border of rostral apex of tongue exhibiting the rounded and low in formo They are covered by queratinized epithelial cells. In the caudal region was noted two vallate papillae containing a deep vala and the surface was covered by stratified squamous epithelial cells. The superior third of surface of vallate papillae was covered by stratified squamous epithelial cells where may be identified the gustatory taste pores. In the fungiform papillae, several taste pores are located on the surface. The samples treated with NaOH solution and examined by scanning electron microscopy revealed afier removal of epitheliallayer, the dense connective tissue- in original disposition, presenting different sizes and shapes of conjuntive papillae. The rostral and medium regions envidentiated numerous papillae corresponding to the filiform and fungiform papillae and in the caudal region, the connective tissue papillae of vallate papillae presented a round formo On the surface were noted small foraminae of lingual salivary gland ducts and small and great papillary projections in three-dimensional images. AIso, on the surface of interface epithelium-connective tissue, were noted thick bundles of collagen fibers. The specimens stained with Picrosirius and examined by po larized light microscopy revealed the connective tissue stained in red colour indicating the collagen fibers type I and the type III, those stained in green colour collagen fibers.

Predicting who will benefit from what in the treatment of psychiatric disorders is incremental to future development of psychological treatments. In the current study functional magnetic resonance imaging (fMRI) data from participants with social anxiety disorder (SAD) was used to elucidate whether neural responses to negative evaluation could predict treatment response in SAD. Nine weeks prior to Internet- delivered Cognitive Behaviour Therapy (ICBT) onset, participants viewed negative social stimuli directed either at themselves or an significant other during fMRI scanning. Regression analyses including the differential activations for other-referential criticism in contrast to self-referential criticism in the posterior mid cingulate cortex (pMCC) and the lingual gyrus (LG) predicted 34% of treatment change as measured by residual gain scores on the Liebowitz Social Anxiety Scale Self-Report (LSAS-SR) in our sample. The final regression model, combining these measures with behavioural measures, which by themselves explained 27% of the variance, resulted in a model explaining 50% of the variance regarding treatment response. This lends additional support to the notion that further elucidating the neurobiological underpinnings of core processes in SAD, as well as the neural correlates of treatment response to CBT, would be of great value in predicting treatment outcome.

Rationale: The importance of well developed phonological awareness and the effect of poor phonological awareness skills on reading and spelling have long been recognised. However, a dearth of research exists regarding populations in multi-cultural, multi-lingual contexts both nationally and internationally. This issue is of particular importance in the South African context where many Black learners in the school setting have no other choice than English as the Language of Learning and Teaching (ELoLT). Aim: The purpose of the study was to explore the effect of a multi-cultural, multi-lingual context on the English phonological awareness skills of a group of Black Grade 4 learners in a primary school setting in South Africa. Method: An exploratory, descriptive and contextual research design was implemented, which incorporated both quantitative and qualitative paradigms. An assessment battery consisting of formal and informal assessments was used to determine levels of development. The researcher endeavoured to find statistical correlations between the participants' phonological awareness skills on the one hand, and their phonological abilities, their expressive and receptive language abilities, and their reading and spelling abilities in ELoLT on the other. Participants: Fifteen Black Grade 4 learners, each with a Black language as mother tongue, who attended a mainstream school where English was the language of learning and teaching (ELoLT), participated in the study. The participants' ages ranged between 9 years 11 months, and 10 years 11 months. Results: The participants’ phonological abilities in English as LoLT showed that 47% of the participants produced the target consonants with a consonant approximation, and vowel approximations were produced by all of the participants in single words. All fifteen the participants’ expressive and receptive language abilities in English as their LoLT, were on a below-average level and their expressive language abilities were more advanced than their receptive language abilities. All the participants evidenced problems in terms of their phonological awareness skills. Based on the results obtained for reading decoding abilities, only one third of the participants could be considered to be readers of an average standard. The reading comprehension abilities of all the participants were on a lower level than those of first language speakers of Standard South African English. Furthermore, the participants’ spelling abilities in English as LoLT were not age-appropriate. Even though no significant correlation could be determined between the participants’ reading comprehension and phonological awareness skills (due to the fact that memory probably played a role in their reading comprehension), it was statistically determined that their poor phonological awareness skills could be associated with their below-average phonological, receptive and expressive language, reading decoding, and spelling abilities. Conclusions: The results of this research have implications for the role of speech-language therapists in terms of collaboration, prevention, assessment, and intervention where the development of these learners' phonological awareness skills is concerned. Clinical implications focus on the dissemination of information, therapy planning, and EAL learner support. The need for further research in this field is emphasised. / Dissertation (M (Communication Pathology))--University of Pretoria, 2007. / Speech-Language Pathology and Audiology / M (Communication Pathology) / Unrestricted

Relations that hold between causes and their effects are fundamental for a wide range of different sectors. Automatically finding sentences that express such relations may for example be of great interest for the economy or political institutions. However, for many languages other than English, a lack of training resources for this task needs to be dealt with. In recent years, large, pretrained transformer-based model architectures have proven to be very effective for tasks involving cross-lingual transfer such as cross-lingual language inference, as well as multilingual named entity recognition, POS-tagging and dependency parsing, which may hint at similar potentials for causality detection. In this thesis, we define causality detection as a binary labelling problem and use cross-lingual transfer to alleviate data scarcity for German and Swedish by using three different classifiers that make either use of multilingual sentence embeddings obtained from a pretrained encoder or pretrained multilingual language models. The source languages in most of our experiments will be English, for Swedish we however also use a small German training set and a combination of English and German training data.  We try out zero-shot transfer as well as making use of limited amounts of target language data either as a development set or as additional training data in a few-shot setting. In the latter scenario, we explore the impact of varying sizes of training data. Moreover, the problem of data scarcity in our situation also makes it necessary to work with data from different annotation projects. We also explore how much this would impact our result. For German as a target language, our results in a zero-shot scenario expectedly fall short in comparison with monolingual experiments, but F1-macro scores between 60 and 65 in cases where annotation did not differ drastically still signal that it was possible to transfer at least some knowledge. When introducing only small amounts of target language data, already notable improvements were observed and with the full German training data of about 3,000 sentences combined with the most suitable English data set, the performance for German in some scenarios even almost matches the state of the art for monolingual experiments on English. The best zero-shot performance on the Swedish data was even outperforming the scores achieved for German. However, due to problems with the additional Swedish training data, we were not able to improve upon the zero-shot performance in a few-shot setting in a similar manner as it was the case for German.

Models based around the transformer architecture have become one of the most prominent for solving a multitude of natural language processing (NLP)tasks since its introduction in 2017. However, much research related to the transformer model has focused primarily on achieving high performance and many problems remain unsolved. Two of the most prominent currently are the lack of high performing non-English pre-trained models, and the limited number of words most trained models can incorporate for their context. Solving these problems would make NLP models more suitable for real-world applications, improving information retrieval, reading comprehension, and more. All previous research has focused on incorporating long-context for English language models. This thesis investigates the cross-lingual transferability between languages when only training for long-context in English. Training long-context models in English only could make long-context in low-resource languages, such as Swedish, more accessible since it is hard to find such data in most languages and costly to train for each language. This could become an efficient method for creating long-context models in other languages without the need for such data in all languages or pre-training from scratch. We extend the models’ context using the training scheme of the Longformer architecture and fine-tune on a question-answering task in several languages. Our evaluation could not satisfactorily confirm nor deny if transferring long-term context is possible for low-resource languages. We believe that using datasets that require long-context reasoning, such as a multilingual TriviaQAdataset, could demonstrate our hypothesis’s validity.

Cross-lingual word embeddings are an increasingly important reseource in cross-lingual methods for NLP, particularly for their role in transfer learning and unsupervised machine translation, purportedly opening up the opportunity for NLP applications for low-resource languages.  However, most research in this area implicitly expects the availablility of vast monolingual corpora for training embeddings, a scenario which is not realistic for many of the world's languages.  Moreover, much of the reporting of the performance of cross-lingual word embeddings is based on a fairly narrow set of mostly European language pairs.  Our study examines the performance of cross-lingual alignment across a more diverse set of language pairs; controls for the effect of the corpus size on which the monolingual embedding spaces are trained; and studies the impact of spectral graph properties of the embedding spsace on alignment.  Through our experiments on a more diverse set of language pairs, we find that performance in bilingual lexicon induction is generally poor in heterogeneous pairs, and that even using a gold or heuristically derived dictionary has little impact on the performance on these pairs of languages.  We also find that the performance for these languages only increases slowly with corpus size.  Finally, we find a moderate correlation between the isospectral difference of the source and target embeddings and the performance of bilingual lexicon induction.  We infer that methods other than cross-lingual alignment may be more appropriate in the case of both low resource languages and heterogeneous language pairs.

Named Entity Recognition (NER) is a critical task in Natural Language Processing (NLP), and recent advancements in language model pre-training have significantly improved its performance. However, this improvement is not universally applicable due to a lack of large pre-training datasets or computational budget for smaller languages. This study explores the viability of fine-tuning an English and a multilingual model on a Swedish NER task, compared to a model trained solely on Swedish. Our methods involved training these models and measuring their performance using the F1-score metric. Despite fine-tuning, the Swedish model outperformed both the English and multilingual models by 3.0 and 9.0 percentage points, respectively. The performance gap between the English and Swedish models during fine-tuning decreased from 19.8 to 9.0 percentage points. This suggests that while the Swedish model achieved the best performance, fine-tuning can substantially enhance the performance of English and multilingual models for Swedish NER tasks. / Inom området för Natural Language Processing (NLP) är identifiering av namngivna entiteter (NER) en viktig problemtyp. Tack vare senaste tidens framsteg inom förtränade språkmodeller har modellernas prestanda på problemtypen ökat kraftigt. Denna förbättring kan dock inte tillämpas överallt på grund av en brist på omfattande dataset för förträning eller tillräcklig datorkraft för mindre språk. I denna studie undersöks potentialen av fine-tuning på både en engelsk, en svensk och en flerspråkig modell för en svensk NER-uppgift. Dessa modeller tränades och deras effektivitet bedömdes genom att använda F1-score som mått på prestanda. Även med fine-tuning var den svenska modellen bättre än både den engelska och flerspråkiga modellen, med en skillnad på 3,0 respektive 9,0 procentenheter i F1-score. Skillnaden i prestandan mellan den engelska och svenska modellen minskade från 19,8 till 9,0 procentenheter efter fine-tuning. Detta indikerar att även om den svenska modellen var mest framgångsrik, kan fine-tuning av engelska och flerspråkiga modeller betydligt förbättra prestandan för svenska NER-uppgifter.