Magister Legum - LLM / The detention of juvenile offenders is not encouraged by both the Constitution and a number of international instruments. This right is entrenched in the South African Constitution (section 28(1)(g) ) which provides that every child has the right not to be detained except as a measure of last resort in which case, in addition to the rights a child enjoys under section s12 and 35, the child may be detained only for the shortest appropriate period of time. This Constitutional provision, in clear terms, views the incarceration of juvenile offenders in a serious light as it provides that the detention of juvenile offenders should be a measure of last resort. One of the important international instruments, the United Nations Convention on the Rights of the Child, (Article 37(b) provides that children may be arrested, detained or imprisoned "only as a measure of last resort and for the shortest possible period of time". This thesis examines the impact of the Constitution and some international instruments on the Criminal Law Amendment Act, 105 of 1997 with regard to juvenile offenders. It also sets out the current legal position in South Africa with regard to sentencing of juvenile offenders. / South Africa

No description available.

The presented thesis focuses on garden-path sentence processing in Czech and follows the line of research within the good-enough approach to language comprehension. Some studies within this framework suggest that, after a garden-path sentence is reanalysed, a full and complete syntactic representation is formed, but the system fails to prune the initial misanalysis completely (Slattery et al., 2013; Qian et al., 2017). However, the research of garden-path sentences within the good-enough approach stems from a limited number of syntactic structures. This thesis presents a series of three self-paced reading experiments employing open-ended comprehension questions that aim to test if the conclusions of Slattery et al. (2013) can also be applied to syntactic structures that have not yet been examined within this framework, namely senteces such as "Roztržitý kastelán zamkl průvodkyni na nádvoří koloběžku.". The results are only partially in accordance with the hypothesis. The lingering misanalysis effect was shown only for the sentences with an animate object (these sentences were generally harder to process). Sentences with an inanimate object, that are easier to analyse, show no difference in response accuracy in comparison with similar unambiguous sentences. Keywords: garden-path sentences,...

Background: Currently, the majority of learners within the South African education system speak English as an additional language. Many of these children are therefore learning the language of instruction through the language of instruction. Of particular concern for speech-language therapists (SLTs) are those children who have language impairment. In addition, it is important for SLTs to be able to distinguish between the learners who have language impairment, and those who are merely in the process of acquiring English. Additionally, the identification of language impairment among learners who speak English as an additional language is difficult as there are no overt manifestations of language learning difficulties, and, unfortunately, these learners are easily over-looked during the pre-school and school-age years. Furthermore, specifically within the South African context, there are limited tools available that can be used to screen for, and/or diagnose language impairment among EAL learners. Purpose: The main aim of this study was to explore the use of sentence repetition as a screening tool for the identification of language impairment in learners who speak English as an additional language. Method: The research design of this study was non-experimental, quantitative, descriptive and cross-sectional in nature, with comparative and correlational components. One hundred and seven grade 2 EAL learners from a mainstream school in Gauteng participated in the study. The learners were evaluated on two sentence repetition tests; the Redmond (2005) Sentence Repetition Test and the Recalling Sentences subtest from the Clinical Evaluation of Language Fundamentals-4 (CELF-4) (Semel, Wiig & Secord, 2003). The learners were also evaluated on the Gray Oral Reading Test-4 (GORT-4) (Wiederholt & Bryant, 2001). The results obtained from these measures were correlated in order to determine the internal validity of the two sentence repetition measures, as well as to investigate the extent to which sentence repetition can be used to predict academic literacy. Learners who were identified as being at-risk for language impairment, namely those who fell 1 standard deviation (SD) below the peer group mean on the Redmond (2005) Sentence Repetition test were further evaluated using the Diagnostic Evaluation of Language Variation-Criterion Referenced edition (DELV-CR) (Seymour, Roeper & de Villiers, 2003), to diagnose language impairment, avoid misdiagnosis, and describe the manifestations of language impairment in the second language. Results and Implications: The results of the Redmond (2005) Sentence Repetition test proved to be a valid measure to identify learners who were at-risk for language impairment, provided that the peer group mean was used as a standard of comparison. Twelve of the 107 (11.2%) learners from the study were identified as being at-risk for language impairment. However, after analysis of the results and using a peer group mean from Jordaan’s (2011) study, only 9/12 (7.5%) of the participants were diagnosed with language impairment. This finding highlights the fact that EAL learners are often over-identified as having language impairment and further assessment is necessary to minimise the risk of misdiagnosis of language impairment. The findings from the DELV-CR (Seymour, Roeper & de Villiers, v 2003), in terms of the manifestations of language impairment in the second language, were consistent with the EAL language impaired learners from Jordaan’s (2011) study, as well as the literature. This finding indicates that sentence repetition is a valid screening tool for the identification of language impairment in EAL learners. Furthermore, 11 of the 12 of the at-risk learners scored below the peer group mean on the reading comprehension measure. Thus, it is evident that, in addition to not being able to use language effectively for academic purposes, most children with language impairment are further disadvantaged by poor reading comprehension skills. An implication for future research would be to explore whether the development of a sentence repetition measure in an African language (e.g. Zulu) would yield similar results as the current study. In this way, EAL learners could be identified in their home language. A secondary finding of the current study was that the grade 2 educators were not able to identify learners who required additional language support within the classroom. This finding has implications for the role of SLTs in mainstream education and the need to support educators in their ability to identify learners with SLI and whose academic language development is not on par with that of their peers.

Aspect semantically denotes two concepts: eventualities and their endpoints, or the speaker’s perspective of an event. These aspectual distinctions respectively refer to telicity and perfectively. Telicity specifically denotes the internal temporal organization of events in regards to their endpoint (telic) or lack thereof (atelic). The telic/atelic distinction has been studied in semantics using logical analysis, in particular mereological relations (the relation of parts to parts and parts within wholes) and scalar theory (a representation of measurement composed of degrees that are partially or totally ordered). This thesis presents a psycholinguistic experiment investigating the cognitive nature of one component of aspectual realization, namely telicity, in English. In accordance with the predictions made by mereological and scalar theories for the semantic composition of telicity, it is hypothesized that telic constructions are semantically more complex than atelic constructions. This complexity specifically refers to maximalization; a process which Filip (2008) predicts takes place exclusively in the derivation of telic events. This prediction is made under the assumption that a correlation exists between the number of derivational steps and cognitive complexity. Specifically, constructions which involve a greater number of steps in their derivation are hypothesized to also involve a greater amount of cognitive resources in order to be successfully computed. This correlation is supported by recent research investigating the role of working memory in sentence comprehension. Recent studies have found constructions with greater syntactic derivational complexity are susceptible to interference and processing trade-offs when processed under heavy memory loads. Based on this, if telic constructions involve a greater derivational complexity compared to atelic counterparts, they therefore can be hypothesized to be more susceptible to interference and processing trade-offs from increased memory loads. The results of the experiment provide evidence supporting this prediction, ultimately suggesting telicity is a grammatically more complex process than atelicity. / Thesis / Master of Science (MSc)

Evidence of systematic misreading has been taken to argue that language processing is noisy, and that readers take noise into consideration and therefore sometimes interpret sentences non-literally (rational inference over a noisy channel). The present study investigates one specific misreading phenomenon: failure to notice word transpositions in a sentence. While this phenomenon can be explained by rational inference, it also has been argued to arise due to parallel lexical processing. The study explored these two accounts. Visual, lexical, and syntactic properties of the two transposed words were manipulated in three experiments. Failure to notice the transposition was more likely when both words were short, and when readers' eyes skipped, rather than directly fixated, one of the two words. Failure to notice the transposition also occurred when one word was long. The position of ungrammaticality elicited by transposition (the first vs. second transposed word) influenced tendency to miss the error; the direction of the effect, however, depended on word classes of the transposed words. Failure of detection was not more likely when the second transposed word was easier to recognize than the first transposed word. Finally, readers’ eye movements on the transposed words revealed no disruption in those trials when they ultimately accepted the sentence to be grammatical. We consider the findings to be only partially supportive of parallel lexical processing and instead propose that word recognition is serial, but integration is not perfectly incremental, and that rational inference may take place before an ungrammatical representation is constructed.

Attitydanalys är ett delfält av språkteknologi (NLP) som försöker analysera känslan av skriven text. Detta är ett komplext problem som medför många utmaningar. Av denna anledning har det studerats i stor utsträckning. Under de senaste åren har traditionella maskininlärningsalgoritmer eller handgjord metodik använts och givit utmärkta resultat. Men den senaste renässansen för djupinlärning har växlat om intresse till end to end deep learning-modeller.Å ena sidan resulterar detta i mer kraftfulla modeller men å andra sidansaknas klart matematiskt resonemang eller intuition för dessa modeller. På grund av detta görs ett försök i denna avhandling med att kasta ljus på nyligen föreslagna deep learning-arkitekturer för attitydklassificering. En studie av deras olika skillnader utförs och ger empiriska resultat för hur ändringar i strukturen eller kapacitet hos modellen kan påverka exaktheten och sättet den representerar och ''förstår'' meningarna. / Sentiment analysis is a subfield of natural language processing (NLP) that attempts to analyze the sentiment of written text.It is is a complex problem that entails different challenges. For this reason, it has been studied extensively. In the past years traditional machine learning algorithms or handcrafted methodologies used to provide state of the art results. However, the recent deep learning renaissance shifted interest towards end to end deep learning models. On the one hand this resulted into more powerful models but on the other hand clear mathematical reasoning or intuition behind distinct models is still lacking. As a result, in this thesis, an attempt to shed some light on recently proposed deep learning architectures for sentiment classification is made.A study of their differences is performed as well as provide empirical results on how changes in the structure or capacity of a model can affect its accuracy and the way it represents and ''comprehends'' sentences.

Sentence embedding models play a key role in the field of Natural Language Processing. They can be exploited for the resolution of several tasks like sentence paraphrasing, sentence similarity, and sentence clustering. Fine-tuning pre-trained models for sentence embedding extraction is a common practice that allows it to reach state-of-the-art performance on downstream tasks. Nevertheless, this practice usually requires labeled data sets. This thesis project aims to overcome this issue by introducing a novel technique for the automatic creation of a target set for fine-tuning sentence embedding models for a specific downstream task. The technique is evaluated on three distinct tasks: sentence paraphrasing, sentence similarity, and sentence clustering. The results demonstrate a significant improvement in sentence embedding models when employing the Smooth Inverse Frequency technique for automatic extraction and labeling of sentence pairs. In the paraphrasing task, the proposed technique yields a noteworthy enhancement of 2.3% in terms of F1-score compared to the baseline results. Moreover, it showcases a 0.2% improvement in F1-score when compared to the ideal scenario where real labels are utilized. For the sentence similarity task, the proposed method achieves a Pearson score of 0.71, surpassing the baseline model’s score of 0.476. However, it falls short of the ideal model trained with human annotations, which attains a Pearson score of 0.845. Regarding the clustering task, from a quantitative standpoint, the best model achieves a harmonic mean (calculated using DBCV and cophenetic score) of 0.693, outperforming the baseline score of 0.671. Nevertheless, the qualitative assessment did not demonstrate a substantial improvement for the clustering task, highlighting the need for exploring alternative techniques to enhance performance in this area. / Modeller för inbäddning av meningar spelar en nyckelroll inom området Natural Language språkbehandling. De kan utnyttjas för att lösa flera uppgifter som meningsparafrasering, meningslikhet och meningsklustring. Fin- och finjustering av förtränade modeller för extraktion av meningsinbäddning är en vanlig praxis som gör det möjligt att nå toppmoderna prestanda på nedströmsuppgifter. Denna metod kräver dock vanligtvis märkta datauppsättningar. Detta avhandlingsprojekt syftar till att lösa detta problem genom att introducera en ny teknik för det automatiska skapandet av en måluppsättning för finjustering av meningsinbäddningsmodeller för en specifik nedströmsuppgift. Tekniken utvärderas på tre olika uppgifter uppgifter: meningsparafrasering, meningslikhet och meningsklustring. Resultaten visar en betydande förbättring av modellerna för inbäddning av meningar när Smooth Inverse Frequency-tekniken används för automatisk extraktion och märkning av meningspar. I parafraseringsuppgiften ger den föreslagna tekniken en anmärkningsvärd förbättring på 2,3% när det gäller F1-score jämfört med baslinjens resultat. Dessutom visar den en förbättring på 0,2% i F1-score jämfört med det ideala scenariot där riktiga etiketter används. För meningslikhetsuppgiften uppnår den föreslagna metoden en Pearson-poäng på 0,71, vilket överträffar baslinjemodellens poäng på 0,476. Det faller dock under den ideala modellen som tränats med mänskliga anteckningar, vilket uppnår en Pearson-poäng på 0.845. När det gäller klustringsuppgiften uppnår den bästa modellen ur kvantitativ synvinkel ett harmoniskt medelvärde (beräknat med DBCV och cophenetic score) på 0,693, vilket överträffar baslinjens poäng på 0,671. Den kvalitativa bedömningen visade dock inte på någon väsentlig förbättring för klustringsuppgiften, vilket understryker behovet av att utforska alternativa tekniker för att förbättra prestandan inom detta område. Translated with www.DeepL.com/Translator (free version)

Past studies of grammatical gender have shown that native speakers encounter processing difficulties when encountering a form that does not agree in gender with previous words. However, the specific behavioral and neural responses to these difficulties have not been replicated across studies of the same type. This is in part due to different experimental designs and statistical analyses, but a crucial factor may be the lack of control between nouns of masculine and feminine gender in stimuli creation. Masculine and feminine gender show distinct distributional asymmetries and collapsing them into one condition diminishes the explanatory power of any study examining grammatical gender. We used reading times in a self-paced reading experiment to examine whether masculine and feminine gender violations differentially affect processing speeds. Fifty French speakers read sentences that were well-formed or contained a mismatch in gender between determiner and noun, half of which were masculine and half feminine. Following Beatty-Martínez et al. (2021), we added individual difference measures to determine how participant-specific factors modulate processing. Participants also completed a category verbal fluency task and the AX-CPT, a measure of cognitive control. They found that ERP components were modulated by these components for Spanish speakers and the modulation differed between masculine noun and feminine noun violations. We hypothesized that reading times would be similarly affected in French, a closely related language with the same gender categories. However, no conditions or interactions reached statistical significance. It is unclear whether this is due to the experimental manipulation or lack of control for participants’ language background, as we had a high number of bilingual and multilingual participants. Regardless, elements of the procedure may provide insight on how to design future experiments that lay a groundwork in understanding the most basic elements of gender processing. / Thesis / Master of Science (MSc) / Why is ‘pen’ masculine and ‘car’ feminine? Grammatical gender is a widespread feature of languages that comes naturally to native speakers and perplexes many second language learners. The assignment of gender seems to be random, but upon closer examination, patterns can be established. What do these differences mean for speakers of gendered languages? In the current study, we set out to determine how masculine and feminine grammatical gender is processed in French and how this is influenced by differences between individual speakers. Participants read French sentences that were either grammatical or contained a mismatch in gender between article and noun. Reading times were used to evaluate how speakers react when encountering an ungrammatical form with either masculine or feminine gender. Participants also completed tasks measuring response inhibition and verbal fluency to see how individuals with different cognitive and language skills react differently to unexpected forms.

No description available.