Carleton University Institutional Repository
Not a member yet
20816 research outputs found
Sort by
An Autoethnographic Account of Utilizing ChatGPT as an Oral Language Learning Partner
Get PDFThis paper explores the effectiveness of ChatGPT as an oral language learning partner for a self-regulated learner. Since generative AI tools, including ChatGPT, are used in a wide range of contexts, it is prudent for researchers to understand how language learners, especially self-learners, can utilize ChatGPT successfully. This study employs an autoethnographic research approach to explore ChatGPT as a novel social phenomenon for language learning. As the author and only participant, I reflect on my experiences conversing with ChatGPT over ten weeks to better understand its capabilities, limitations, types of feedback provided, and to explain which second language acquisition (SLA) theories best account for what is going on. Keywords: ChatGPT, autoethnography, self-regulated learning (SRL), corrective feedback, second language acquisition (SLA), oral language partne
The Datafication of Mental Health: Data Brokers in the Biopolitical Economy
Get PDFThis thesis explores how mental health data are quantified, datafied, and commodified by data brokers and assesses whether they are adequately governed in Canada by using Actor-Network Theory and drawing on Foucauldian concepts of biopower and bioeconomics. Methodologically, a lengthy list of mental health data brokers that operate in Canada were identified, and a selection of ten were shortlisted for analysis, focusing on how these actors collect, aggregate, de-identify, repurpose, and co-construct mental health related data through their privacy policies, governance structures, data products, and overall operations. The findings suggest that improved governance, oversight, and regulation mechanisms are required, and that privacy legislation alone is insufficient at mitigating the potential harms of data brokers. Certain aspects of the Artificial Intelligence and Data Act, along with the establishment of a mental health data trust, might better safeguard the wellbeing of individuals affected by the practices of mental health data brokers
A Study on Comparing AI Models for Fair Social Media Sentiment and Engagement Analysis Across Different COVID-19 Stages
No full textAs people share emotions and opinions on social media, training AI models with such data can introduce biases. Ensuring bias-free training models is critical in AI ethics. This study aimed to identify the best-performing unbiased models for sentiment and engagement analysis on social media using data from before, during, and after the COVID-19 pandemic, with language as the bias attribute. Logistic regression, decision tree, LSTM, BERT, and LLAMA models were evaluated using performance metrics such as accuracy, precision, recall, and F1-score to test the data from X. Fairness metrics, including average odds difference, equal opportunity difference, equalized odds ratio, disparate impact ratio, and predictive rate parity, were also employed. The results indicated that LSTM combined with sentiment analysis and BERT applied to engagement analysis were the best-performing unbiased models. This study offers a foundational approach for mitigating bias in AI models and promoting ethical AI practices
Evaluating the Practicability of Wearable IMUs for Clinical Assessments of Spine Movement Quality
Get PDFAssessments of spine movement quality (MQ) and motor control may improve diagnosis and treatment of low back pain (LBP); however, subjective clinical assessments are unreliable, and gold-standard objective optical motion capture (OPT) systems are infeasible for routine clinical use. Inertial measurement units (IMUs) have the potential for clinical integration but are prone to errors that can affect measurement accuracy, validity, and reliability, preventing widespread uptake and acceptance. Through conducting four main studies, the practicability of IMUs for clinical assessments of spine MQ is evaluated. Specifically, the first study evaluates the concurrent validity of the Xsens DOT IMUs to track multiplanar spine movement and range of motion (ROM) relative to OPT equipment by assessing root mean square error (RMSE), mean absolute error (MAE), and intraclass correlation coefficients (ICCs); results confirm concurrent validity within clinical thresholds. The second study demonstrates concurrent validity to measure spine MQ compared to OPT equipment and compares the within- and between-day reliability of both systems by evaluating standard error of measurement (SEM), coefficient of variation (CV), and ICCs. Reliable metrics are identified, and the results suggest that issues related to reliability are attributable to movement variability and sensor placement rather than equipment error; the minimal detectable differences (MDDs) presented may provide thresholds to researchers and clinicians for identifying meaningful changes in ROM and MQ. The third study classifies and critically reviews all potential sources of error during IMU-based assessments of human movement, from which a taxonomic framework is developed. Errors with a high risk of impacting data quality are identified, and a high-level review of methods for error mitigation is provided. The final study refines and validates the taxonomic framework by consulting with expert focus groups. Additionally, an error risk assessment is conducted, recommendations for effective error mitigation techniques are summarized, and suggestions for future development of universal best practice guidelines are provided. The contributions from this thesis will help progress toward establishing best practice methodologies for conducting IMU-based assessments of human movement, which may enable reliable acquisition of substantial amounts of spine MQ and motor control data necessary for improved models of care for LBP
The Measure of a Mind: Coming to Know in an Integrative Cognitive Science
Get PDFNúñez et al. (2019) argue that cognitive science has never managed to produce a coherent research programme. By this, they mean that its subfields (psychology, neuroscience, philosophy of mind, computing, linguistics, anthropology; Boden, 2006, p. 523) are capable of collaboration, but cognitive science as an integrative whole remains a developmentally arrested pre-science. Several conceptual and institutional difficulties lie in the way of producing a Lakatos (1978)ian research programme. Above all, it remains to be clarified how the subdisciplines sit in definite relation to one another, in common revelation of one shared object of study. Accordingly, the field must make ontological commitments that allow it to speak coherently about its object. Such commitments are not without explanatory limits (namely, phenomenal consciousness), but they nevertheless permit strong claims to be made about the formal contents of certain brain states, with broad and deep implications for the study of intelligence and social behaviour
