27716 research outputs found
Sort by
Interpretable computational cetaphor processing
Full text linkMetaphors are a fundamental component of human language, enabling abstract reasoning, nuanced communication, and cultural expression. However, their inherent complexity poses significant challenges for natural language processing (NLP) systems, which must accurately detect, interpret, and translate metaphorical expressions across diverse linguistic and cultural contexts. This thesis addresses these challenges by developing computational methodologies for interpretable metaphor processing, bridging insights from computational linguistics, psycholinguistics, and artificial intelligence.
The work advances three core areas: 1）Enhancing metaphor detection through syntactic pruning (RoPPT), semantic frame integration (FrameBERT), and basic meaning modeling (BasicBERT), achieving state-of-the-art performance across benchmark datasets. 2）Improving cross-linguistic metaphor translation via MMTE – a novel evaluation framework combining human and automatic metrics to assess emotional salience and translation quality in English, Chinese, and Italian. 3）Leveraging interpretability in large language models (LLMs) for metaphor understanding through sparse autoencoders and dictionary learning, decomposing latent representations to extract monosemantic features that improve metaphor transparency.
The findings contribute to the development of more linguistically sophisticated, contextually adaptive, and culturally aware NLP systems. By advancing metaphor processing methodologies and introducing interpretability techniques, this research provides a foundational exploration for applications in machine translation, sentiment analysis, and explainable AI. The proposed models and evaluation frameworks not only improve metaphor understanding in computational settings but also provide a foundation for future work in cross-linguistic and cross-cultural NLP
Essays on Fiscal Sustainability: Assessment and Adjustment
Full text linkThis thesis comprises three chapters on fiscal sustainability, focusing on ageing society. Chapter 1 discusses the assessment of fiscal sustainability, while Chapter 2 and 3 explore fiscal policy measure aimed at improving it.
Chapter 1 evaluates fiscal sustainability using three separate approaches: (1)undertaking an econometric test of the PVBC based on the series of debt, deficit and other macro variables pertinent to fiscal policies as Campbell (1987) test; (2)regressing the primary surplus on debt to estimate a Bohn (2005) rule and (3) measuring the required adjustment in a fiscal tool to achieve a given debt target in the medium-term, the so-called Blanchard (1990) medium-term debt dynamics. Chapter 1 conducts the empirical analysis through the length of a unified VAR model that incorporates demographic change. The combined empirical evidence suggests that fiscal policies in most countries have been on an unsustainable path since 2008.
In Chapter 2, I use a transformed VAR model to analyse the response required by the Bohn (2005) rule to converge the Blanchard (1990) medium-term debt dynamics to zero. The results suggest that major economies should strengthen their responses to rising debt ratios and the responses lies within the range of previous significant response bands. Nonetheless, the reliability of the counterfactual results is challenged by the Lucas (1976) critique.
In Chapter 3, I use a large-scale OLG model calibrated to China and compare two steady states (2020 vs 2100). Exogenous variation in the labour tax rate yields an inverted-U “debt Laffer effect” and model-implied debt limits (fiscal space) in each steady state. Sustainability is assessed by comparing these limits with the sustainable debt ratio implied by the Bohn (2005) rule. The implied sustainable debt ratio is 72.38\%, well below the debt limits in both steady states, indicating sustainability after ageing. However, ageing tightens fiscal space, so feasibility typically requires a higher labour tax rate in 2100
Shaping Muslim Young People's Religious Identities: The Perspectives of Salafi Muslims in Manchester, England, on Diverse Educational Spaces
No full textOver the past few decades, the education of Muslim children in the UK has been at the centre of political and social debate, often linked to broader concerns about integration, national identity, and security. These debates intensified following incidents such as the “Trojan Horse Affair” in 2014 and the implementation of the Prevent strategy, which has disproportionately targeted Muslim communities, fostering a climate of suspicion and surveillance. In this context, Salafi Muslims—often portrayed in policy and media as adhering to a 'hard-line' form of Islam—have been particularly stigmatised. Despite their prominence in public discourse, Salafi communities remain under-researched, especially regarding their educational experiences, perspectives, and school choices. Existing studies on European Salafism tend to focus on extremism and radicalisation; this research shifts the focus to explore the appeal of Salafi religious doctrine and its impact on how Salafi families perceive educational institutions in shaping their children’s Islamic identity. This thesis addresses this gap by exploring the educational preferences, perceptions, and experiences of Salafi Muslim families within a Salafi mosque community in Manchester. Since parental perceptions of social, political, and cultural factors within formal education significantly influence school choices, the Salafi community's educational preferences should preserve a distinct Salafi identity which resists perceived secular or liberal influences in mainstream schooling. This focuses on three key spaces—schools, mosques, and the home—to examine how Salafi parents and young people perceive the role of education in shaping Muslim personhood, a term here defined as a form of identity linked to Islamic moral and religious standards that Muslim parents often want. Given the challenges of accessing a closed and often misunderstood community, this thesis offers fresh insights into how fundamental religious minorities engage with, or resist, England’s educational landscape
Experiences of Online Threats Among Younger Adults in the Kingdom of Saudi Arabia and the United Kingdom
Full text linkThis programme of research explored how younger adults (aged 40 or younger) in the Kingdom of Saudi Arabia (KSA) and the United Kingdom (UK) experience, detect and respond to online security threats. Through online studies, two using an innovative diary method, the research investigated the online threats younger people in the two countries encounter, their concerns about these threats, the cues they use to identify threats and how they respond to them.
The first study involved a scenario-based online survey designed using the MITRE ATT&CK framework and the Cyber Kill Chain model to present 12 realistic online threat scenarios to participants. Participants were asked whether they had encountered threats like these scenarios and to report their concerns, detection strategies, and responses to them. The results highlighted attacks from seemingly trusted sources as a top threat in both countries. However, malware and data theft were more frequently reported among KSA participants compared to those in the UK.
To explore these experiences in greater depth, a 30-day online diary study was conducted with 16 participants from KSA. Participants recorded their encounters with online threats through a short questionnaire sent to them daily. Phishing was the most common threat encountered. A substantial proportion of threats could not be fully classified due to the limited information provided. Detection cues reported by participants were systematically coded using a modified version of the NIST Phish Scale, expanded to include cues applicable to a wider range of threat sources beyond email (e.g. voice calls and social media). Language and content cues were the most frequently reported cues of threats, followed by technical indicators and prior knowledge of encountered threats.
A second diary study was conducted in the UK with 45 participants over 7- and 14-day periods. Insights from the KSA study, particularly regarding the ambiguity of open-ended responses and decreased participant engagement over the long study duration, informed the design of this study. Consequently, the UK study was shorter, had more participants and used a more structured approach, incorporating mainly closed-ended questions. The modified Phish Scale cues were used as a set of options to select from. Phishing and spear phishing were the most frequently reported threats. Technical indicators such as suspicious links and email addresses were the most frequent cues used by British participants.
The studies also investigated the influence of individual differences among participants, such as levels of security knowledge, security behaviour intentions (measured by the SeBIS scale), unrealistic optimism, risk-taking, and thinking orientation.
This thesis contributes new empirical evidence on how younger adults experience, detect, and respond to online threats across two distinct cultural contexts (UK and KSA). Its principal contribution lies in methodological innovation: combining surveys with the novel application of diary methods to capture real-world encounters. This approach generated rich insights into behaviours that are often overlooked in lab-based studies
Multilingual and Multi-Domain Rumour Stance Classification
Full text linkRumour stance classification focuses on a conversation initialised by a rumour-related source post on social media, aiming to determine the stance of each reply’s author towards the rumour. Accurately capturing public stance facilitates assessing the verdict or check-worthiness of the information. Given the diversity and multilingual nature of online rumours, systems must be able to generalise and adapt to diverse domains and languages. This thesis addresses these issues by investigating methods for improving the generalisation and adaptation of rumour stance classification across new rumours, domains, and languages. Firstly, we identify the distinction between rumour stance classification and generic stance classification by analysing the special role of the stance target (i.e, rumour) in model generalisation. We propose a new ensemble-based method to enhance the reasoning with rumours, achieving state-of-the-art performance. Secondly, we assess the generalisability of top-performing models under domain shift, and propose a LLM-assisted self-training framework for effective adaptation without access to both source and target domain labelled data. Thirdly, we reveal how class labels design in prompts affect LLMs’ generalisation in zero-shot in-Context Learning (ICL) (e.g, the lexical choice between “agree” and “support” for positive stance), and introduce an efficient post-hoc method for optimal label selection. Furthermore, this thesis investigates the adaptation of English-centric rumour stance classification models to non-English languages. We create the largest multilingual benchmark dataset with nine diverse high- and medium-resource languages. We then reveal performance inconsistency across languages and further analyse strategies to improve model performance
Water service provision as socio-technical bricolage: understanding the everyday practices of frontline bureaucrats in Malawi
Full text linkOver the last three decades significant progress has been made globally in providing water services to the rural poor. However, sustaining these services remains a challenge. This thesis provides a critical but sympathetic account of the experiences of frontline government staff, whose work often goes unrecognised in the quest to improve policy outcomes. These actors play an important role in supporting communities to manage water services, as per the community-based management (CBM) model. Yet they are not neutral conduits of state policy, they are individuals with their own identities and values, who actively (re)interpret policies in the face of complex local realities and significant resource constraints. This thesis demonstrates how frontline bureaucrats “get their jobs done” through creative and negotiated practices of socio-technical bricolage, advancing critical institutionalist theory and shedding light on how decentralised public services actually function and hence might be improved or transformed.
Employing an actor-orientated ethnographic approach, the research investigated the work-lives of six Water Monitoring Assistants (WMAs) in a rural district of Malawi, revealing how these actors navigated and (re)shaped the state-community interface and water technologies in their everyday work. The analysis shows that, despite facing numerous challenges, these frontline bureaucrats endeavoured to fulfil their mandates and produce outcomes that were acceptable to both state and community. In doing so, they employed a number of different strategies to work around structural constraints and to reconcile multiple interests, values and policy goals. To be successful, they had to exercise both social and technical know-how, skilfully leveraging a variety of material and non-material resources to carry out their tasks and engaging in various forms of institutional, ideational and technological bricolage. These practices helped to ensure that much needed water services functioned. Further research would be valuable to ascertain the longer term implications of such practices, and how they manifest in other contexts
Autistic Mothers' and Birthing Peoples' Experiences of Being a Parent
Full text linkSection One: Autistic Mothers' and Birthing Peoples' Experiences of Parenting: A Systematic Review of Qualitative Evidence and Thematic Synthesis
Objectives
Parenthood is considered transformative and encompasses physiological, social and relational changes. Qualitative research highlights unique strengths and challenges in this experience for autistic parents. The review aimed to collate, critically appraise and thematically synthesise qualitative literature to further understand the lived experience of parenthood for autistic mothers and birthing people.
Design and Method
Four electronic databases were systematically searched using specified search terms. 29 studies between 2016-2025 with 713 participants were included, and data extracted and analysed using thematic synthesis. The quality of included papers was assessed using the Critical Appraisal Skills Programme qualitative research checklist, and a sensitivity analysis was conducted to assess the possible impact on the review findings.
Results
Synthesis revealed three themes and nine subthemes. The three themes were: ‘Being an Autistic Mother/Parent is Overwhelming’, ‘Navigating an Inaccessible System’, and ‘Discovering Identity and Values as an Autistic Parent’.
Conclusions
Findings highlighted both individual and systemic challenges faced during mother/parenthood. Despite this, for many participants, connecting with their children facilitated the discovery of their identity and values as an autistic parent, through which participants identified advocating to support change for themselves and their children.
Section Two: “There’s no real place that’s just for us”: Autistic Mothers’ and Birthing Peoples’ Experiences of Perinatal Mental Healthcare an Interpretative Phenomenological Analysis (IPA)
Objectives
The perinatal period is defined from preconception to two years after birth. Autistic people are more likely to face challenges in navigating sensory, physical and social experiences. This can result in motherhood feeling isolating, placing this population at increased risk of mental illness. The current study sought to understand the experiences of autistic women and birthing people who have a perinatal mental health condition, specifically focusing on their experiences of mental health care.
Design and Method
Five autistic women who experienced challenges with their mental health in the perinatal period completed semi-structured interviews. Interpretative Phenomenological Analysis (IPA) was used to provide an in-depth understanding of this lived experience.
Results
Three group experiential themes (GETs) emerged through the data, each with subthemes: ‘Amplification of Feeling ‘Different’ and Disconnected’, ‘Losing Trust that I will Get Help’ and ‘The Value of Understanding and Connection’.
Conclusions
Findings highlight the individual, systems and societal barriers to receiving perinatal mental health support. Receiving an autism diagnosis, therapeutic and peer support brought understanding, acceptance and facilitated self-compassion. However, there was concern about how much resource participants had left to sustain this journey on their own. Systemic changes are considered to support autistic mothers and birthing people
Classical and machine learning algorithms for analysing complex DNA structures
No full textAtomic force microscopy (AFM) is unique in its ability to image single molecules in liquid with sub-molecular resolution, without the need for labelling or averaging. This enables us to probe biomolecular structures in native-like states and examine conformational changes. For DNA, its innate flexibility enables compaction in the nucleus and processing by essential cellular machinery which drives a large range of these conformational changes and must be regulated to ensure cell survival. However, the large quantities of closed AFM filetypes limit the adoption of open-source tools developed by the image analysis community. The lack of AFM-specific automated analysis tools to process raw data and characterise conformation make high-throughput conformational analyses difficult and laborious.
I have developed AFMReader, an open-source Python file loader for the extraction of AFM images and metadata from proprietary file formats. I also developed TopoStats, a toolbox for; AFM-specific image processing, object identification, and characterisation of individual molecules. Key developments to this pipeline are a new height-biased skeletonisation algorithm, and quantification of overlapping DNA segments, enabling the accurate tracing of branched, crossing, and overlapping DNA structures.
This new automated tracing architecture enables the classification of DNA knots and catenanes produced by the Xer recombination system by extracting a pseudo 3D molecular backbone trace. I characterise DNA replication fork stalling by Lac-repressor protein and the Tur-Ter complex via calculation of the replicated and unreplicated DNA segment contour lengths. I show that this pipeline can be adapted to characterise a possible prebiotic RNA synthesis pathway via molecular backbone height profiles across samples conditions in a fixed location. Finally, I explore the feasibility of a deep learning variational auto-encoder to describe the conformational landscape of supercoiled DNA minicircles. These applications show the versatility of this new pipeline as a toolbox to help quantify and uncover the role of structure in DNA interactions
Artificial Intelligence for Psychological Treatment Selection: Predicting Treatment Outcomes and Understanding Clinician and Patient Perspectives
Full text linkObjectives
Artificial Intelligence (AI) use in clinical decision-making could offer numerous benefits for clinicians, patients, and services. However, AI use in mental healthcare is in its infancy, and little is known about its acceptability to patients and therapists. This qualitative study aimed to better understand perceptions of AI in mental healthcare, and how AI may influence therapy expectations and shared decision-making (SDM).
Methods
Semi-structured interviews were conducted with seven patients and 12 clinicians participating in the TherapyMatch-D (TMD) trial, where patients with depression were randomly allocated to AI-informed psychological treatment selection (TMD) or allocation as usual (AAU). Interviews explored clinical decision-making approaches, experiences using the algorithm, and broader views of AI’s role.
Results
Framework analysis identified nine themes capturing clinicians’ and patients’ experiences of treatment selection. Initial themes outlined factors considered within assessments, usual decision-making approaches, and clinicians’ difficulties. TMD participants felt the AI algorithm improved decision-making quality and confidence in treatment decisions. Clinicians valued the algorithm’s efficiency and usability, whilst patient expectations of therapy appeared improved compared to AAU patients. The algorithm could enhance clinician-patient collaboration, supporting SDM, whilst maintaining patient autonomy. Clinicians noted applicability and integration limitations, recommending better integration with electronic health records and continued human involvement in AI-supported decisions.
Conclusions
AI tools such as the TMD algorithm can enhance the quality, confidence, and collaborative nature of decision-making in mental healthcare, supporting SDM and enhancing positive expectations of therapy. Integrating AI into existing systems and retaining human input could help maximise AI tool effectiveness, efficiency, and acceptability
Modelling and Optimisation of The Continuous Pharmaceutical Manufacturing Process: A New Data-Driven Approach For Right-First-Time Production
No full textPharmaceutical industries, like most industries, are subjected to stringent quality and regulatory requirements to ensure the manufacturing of safe and high-quality medicinal products. Continuous manufacturing has emerged as a transformative approach offering the potential to meet global demands of medicines through efficient and continuous processes. However, its adoption in tablet manufacturing remains constrained by the complex, multivariate behaviour of particulate processes. Moreover, the lack of comprehensive modelling frameworks further hinders understanding and control of the multistage processes.
This thesis aims to develop and evaluate novel predictive modelling frameworks tailored to the continuous manufacturing of pharmaceutical tablets, using data collected from an industrial-scale pilot plant (Consigma-25) encompassing five critical unit operations. An integrated and sequential modelling framework was constructed using ensemble machine learning techniques, including gradient boosting machines and random forests, to predict key quality attributes across stages, with Gaussian mixture models incorporated to reduce uncertainties. To enhance interpretability, a hybrid modelling approach combining artificial neural networks with interval type-2 fuzzy inference system was developed. Additionally, a novel integration of Adaptive Neuro-Fuzzy Inference System with a Genetic Algorithm formed the basis of a model-informed optimisation strategy, enabling the identification of optimal process settings to control the final product quality under “Right-First-Time” manufacturing.
The results demonstrate that proposed frameworks were effective in capturing the non-linearity among process parameters and quality outcomes, achieving values exceeding 0.90 across the frameworks. This represents a predictive capability improvement of 56\% compared with prior studies. The incorporation of interpretable, uncertainty-aware methods ensured model outputs remained effective to illustrate the processes' understanding despite complexity. The model-informed optimisation strategy was validated through practical application within the right-first-time manufacturing concept. These research findings demonstrate the potential of the proposed frameworks to advance pharmaceutical tablet manufacturing by bridging the gap between scientific research innovation and scalable industrial implementation
Data sources:
Access Repository Dashboard
Do you manage Open Research Online? Become a CORE Member to access insider analytics, issue reports and manage access to outputs from your repository in the CORE Repository Dashboard! 👇