DigitalGeorgetown
Not a member yet
425578 research outputs found
Sort by
Advancing Generalization in Coreference Resolution
No full textPh.D.Coreference, as an important component of Natural Language Understanding, is a linguistic phenomenon in which two or more expressions within a text refer to the same entity. This task plays a crucial role in various downstream applications, such as information extraction, dialogue systems, machine translation, and summarization.Recent advances in coreference systems reveal a significant challenge: susceptibility to overfitting during training on standard benchmarks such as OntoNotes, leading to performance degradation when applied to unseen data/genres. Moreover, the absence of singleton spans, i.e. mentions of entities that occur only once in a text, in deep learning coreference models prevents identification of all entities in texts and causes a misalignment with linguistic theories, such as the Centering Theory. To address these issues, this dissertation focuses on the generalization problem in coreference resolution and aims to enhance model adaptability across diverse linguistic contexts while maintaining theoretical coherence with linguistic principles.Firstly, we create OntoGUM -- the largest open, gold standard dataset consistent with OntoNotes -- with 213 documents in 12 genres. We present the details of automatic data conversion from the GUM annotation scheme to the OntoNotes annotation scheme, demonstrate that both state-of-the-art (SOTA) neural models with pre-trained contextualized embeddings and traditional rule-based systems encounter performance degradation, and conduct a genre-by-genre analysis to reveal the strengths and weaknesses of current approaches.Secondly, we build near-gold singleton annotation for OntoNotes by training a mention detection classifier based on syntactic structures. Then, we propose a two-step pipeline-based approach to integrate the extracted singleton information. Experimental results demonstrate performance on par with the SOTA baseline model in-domain and outperform the baseline model out-of-domain. We conduct an analysis to show the effect of singletons on coreference linking.Thirdly, we propose a multi-task learning based coreference model with constrained mention detection, which jointly learns two mention-based tasks, entity type recognition and information status classification. Experimental results show that the proposed model achieves the SOTA performance both in-domain and out-domain. Further error analysis reveals the effectiveness of our proposed approach
Teaching and Learning with AI: A Conversation with Cal Newport and Eddie Maloney
No full textBuilding on the many conversations at Georgetown about Generative AI and how it is transforming teaching and learning, join us for a lunchtime keynote conversation with Georgetown Professor Cal Newport, Provost's Distinguished Associate Professor of Computer Science, and CNDLS Executive Director Edward Maloney. Our conversationalists look forward to a mainly Q&A style keynote responding to audience questions submitted in advance and posed during the session
ScL-2PAM: A Novel Countermeasure That Ameliorates Neuroinflammation and Neuronal Losses in Mice Exposed to an Anticholinesterase Organophosphate
No full textPh.D.Exposure to organophosphate (OP) agents with anticholinesterase activity, even at sublethal levels, can lead to various severe adverse events including neuronal loss and persistent neurological dysfunction. Pralidoxime (2-PAM), the sole FDA-approved acetylcholinesterase reactivator as an antidote for OP poisoning, has very limited effectiveness in the brain due to its inefficient penetration across the blood-brain barrier. To address this, we have previously demonstrated the superior efficacy of a novel brain-penetrating nanocomplex encapsulating 2-PAM (scL-2PAM) compared to unencapsulated free 2-PAM in mitigating cholinergic symptoms and enhancing survival rates in mice with lethal OP exposure. In this study, we evaluate scL-2PAM's potential to mitigate neurological consequences in mice surviving OP exposures using paraoxon, a metabolite of OP pesticide parathion. Exposure of mice to paraoxon led to notable microglial activation, increased expression of inflammatory cytokines, and neuronal losses in various brain regions, including the hypothalamus, piriform cortex, and amygdala. Among the assessed cytokine mRNAs, chemokine ligand 2 (CCL2) exhibited the most significant upregulation after paraoxon exposure, demonstrating a high correlation with the magnitude and duration of the observed cholinergic crisis. Following acute exposures to paraoxon, mice were promptly treated with either scL-2PAM or free 2-PAM, both administered alongside atropine. The effectiveness of two antidote formulations in mitigating neuroinflammation and preventing paraoxon-induced neuronal losses was directly compared by transcriptomic and histological analyses. scL-2PAM nanocomplex outperformed free 2-PAM in reducing CCL2 transcript production and inhibiting microglial activation, as evidenced by histological analysis. The antidote regimen incorporating scL-2PAM also demonstrated greater efficacy in mitigating neuronal losses induced by paraoxon exposures. Our data indicate that scL-2PAM is superior to free 2-PAM in reducing neuroinflammation and mitigating the neurotoxicity that emanates from paraoxon exposures. Thus, scL-2PAM appears to be a new and improved countermeasure for reducing neuroinflammation and mitigating brain damage in survivors of organophosphate exposures. It might be possible to mitigate the longer-term neurocognitive disorders observed in survivors after OP exposure
Carbon Credits as Catalysts: Assessing Their Role in Renewable Energy Investment
No full textM.P.P.This research delves into the complex relationship between carbon credit issuance andrenewable energy investment, exploring how market mechanisms interact with environmental
policy objectives. This study examines data spanning two decades to examine the impact of
carbon credits on the financing of renewable energy projects. Employing both Ordinary Least
Squares (OLS) and Fixed Effects (FE) regression models, the research scrutinizes the correlation
between carbon credit issuance and renewable energy investment, controlling for a range of
economic variables. The findings reveal a nuanced scenario: while the general assumption of a
positive relationship between carbon credits and renewable energy investment is not consistently
supported, the influence of regulatory quality emerges as a critical moderating factor. In
jurisdictions with robust regulatory frameworks, the impact of carbon credit issuance on
renewable energy investment becomes more pronounced, highlighting the essential role of
governance structures in optimizing the efficacy of market-based environmental policies.
The limitations posed by data constraints in the current research context point towards the
need for comprehensive data collection mechanisms, which would facilitate a deeper
understanding of the dynamics at play in carbon credit markets and renewable energy
investment. This study contributes to the environmental economics literature by offering a
sophisticated analysis of carbon credit markets and their role in financing renewable energy. It
provides valuable insights for policymakers, emphasizing the need for harmonized regulatory
environments to maximize the potential of carbon credits as catalysts for renewable energy
development. The thesis also identifies avenues for future research, particularly the necessity of
assessing the long-term impacts of projects financed through carbon credits, to ensure
sustainability and ecological integrity alongside economic growth
Oral Healthcare Utilization in Low-Income Reproductive-Aged Women: The Role of Dental Insurance Coverage in 2018
No full textM.P.P.Oral health is an important aspect of overall health and well-being. Women’s oral health is especially important during the pregnancy and postpartum periods as poor oral health has been connected to higher rates of low-birth weight, preeclampsia, and preterm birth. Maintaining good oral health during pregnancy is thus critical for preventing poor health outcomes for both mother and baby. But previous research shows that low-income populations and women both face inequities in the oral healthcare system. This thesis explores the relationship between dental insurance coverage and oral healthcare utilization among low-income, reproductive age women using cross-sectional, 2018 data from the Centers for Disease Control and Prevention’s National Health Interview Survey. I add new understanding to the growing body of oral health and public policy literature by focusing on this subpopulation. My analysis examines a range of different types of insurance coverage, including private plans that provide coverage for both medical and dental care, private health insurance plans that have no dental coverage, single service dental plans, and Medicaid. I also control for a series of demographic factors. Findings reveal that low-income women of reproductive age with any form of health or dental insurance are significantly more likely to have a dental visit compared to those who have neither health nor dental insurance. The empirical results underscore the pivotal role insurance plays in facilitating access to dental care services, particularly for low-income women
Zhao on Kneese, Death Glitch
No full textReview: Kneese, T. (2023). Death glitch: How techno-solutionism fails us in this life and beyond. Yale University Press. $35.00. ISBN: 97803002482
Enhancing Nigeria’s Health Security: Current Challenges and Solutions Exploring the Intersection of Academia and Global Health Security
No full textMedical and disaster management scholars and professionals have long highlighted Nigeria's healthcare infrastructure and its challenges in supplying quality healthcare and responding to public health emergencies. With the population rising (Statista, 2023), climate change increasing, harsher weather conditions (Campbell, 2018), adverse economic conditions (Bangudu, 2012), political disruption, corruption, and violence (Abah, 2022; Campbell, 2020), and global epidemics and pandemics, the healthcare system in the country is facing multiple challenges. These challenges further exacerbate the existing strain, making it even more difficult to effectively respond to disasters and implement comprehensive risk management strategies (Abah, 2022; Obaseki, 2021). In this study, we target the question, "What can be done to enhance the resilience of the healthcare system in the face of these complex challenges?" By exploring how the US can manage global health emergencies and positively impact the West African region by optimizing Nigeria's health security through collaboration with academic medical institutions (AMIs), this research aims to provide strategic insight and recommendations for policymakers and stakeholders to strengthen healthcare systems in the context of health security, ultimately improving disaster response and risk management strategies in the US and worldwide. By leveraging the expertise and resources of AMIs, we can create a more resilient healthcare system that is better equipped to address global health emergencies and partner with countries like Nigeria to enhance their health security
Book Review: Cartografías cosmopolitas: León de Greiff y la tradición literaria, by Marco Ramírez Rojas. West Lafayette, Indiana: Purdue UP, 2023, 232 p. ISBN 978-1-61249-847-8
No full textReview of - Cartografías cosmopolitas: León de Greiff y la tradición literaria, by Marco Ramírez Rojas. West Lafayette, Indiana: Purdue UP, 2023, 232 p. ISBN 978-1-61249-847-8https://doi.org/10.57928/r6qw-272
Illuminating Inequity: An Analysis of Socioeconomic Disparity in Light Pollution Distribution by U.S. Census Tract
No full textM.P.P.Light pollution is one of the most widely spread and fastest-growing environmental hazards. Despite its negative health, economic, and environmental consequences and disproportionate impact on marginalized communities, light pollution is often overlooked by the environmental justice movement. This critical area of study remains underexplored, with limited research examining light pollution as an intersectional environmental injustice. This study aims to evaluate the injustices associated with light pollution within the context of environmental equity.By harnessing the most recent ground-based skyglow observations from the My Sky at Night Project (2022) and United States (U.S.) Census Bureau and American Community Survey (ACS) 5-year estimate (2017-2021) data, I analyze social inequities in light pollution exposure across the continental U.S. The effect of demographic and socioeconomic factors on skyglow is measured using an OLS regression with the dependent variable, average NELM, and independent variables: gender, race, ethnicity, housing tenure, and median income, controlling for age and population. Interactions between gender and race/ethnicity are included in subsequent models.
Ultimately, the analysis indicated that gender, housing tenure, and ethnicity are correlated with an increase in skyglow. Specifically, for every 10-percentage point increase in renter-occupied units, female population, and Hispanic or Latino population in a census tract, average NELM decreases by 0.14, 0.20, and 0.43, respectively. This suggests that light pollution mitigation policies should targeted in communities with large proportions of renter-occupied units and high percentages of Hispanic/Latino populations. However, the limitations of the available data highlight the need for more ground-based skyglow data collection
The Causes of Existential International Emnity
No full textPh.D.In this dissertation, I undertake three closely linked research projects. First, I make the case for restructuring the literature on international rivalries, or cases of long-lasting conflict between states. Although I find flaws in all existing empirical and theoretical arguments for using ‘rivalry’ as a distinct variable in international relations scholarship, a careful analysis of the statistics and intuitions marshalled in support of those arguments yields compelling grounds for adopting a modified variable. This variable is the degree of inconsistency between states’ preferences, considered in isolation from the strategies chosen by states to realize their preferences. I label this kind of inconsistency ‘enmity,’ refer to its extensiveness as its ‘breadth,’ and describe the broadest enmities as ‘existential’ ones. Second, I propose a research agenda for a rivalry literature reoriented toward investigating the broadening of enmity. I propose three explanatory models of enmity broadening: a stochastic model, drawing largely on structural realism, depicts enmities as accumulations of unrelated disagreements. A spiral model, drawing largely on social constructivism, suggests that initial disagreements spur increasingly hostile states to find additional areas of disagreement. And a philosophical model, drawing on a new theoretical approach that views actors’ political preferences as reflections of their underlying philosophies, explains particularly broad enmities as reflections of deep-seated ideational inconsistencies between states.
Third, I demonstrate the practicality of my proposed research agenda by conducting a case study to test these explanatory models. In the case study, I trace the processes responsible for the emergence of several Arab-Israeli enmities, and conclude that my philosophical model best accounts for these enmities: the stochastic model cannot explain why distinct Israeli and Arab societies emerged at all, and the spiral model cannot explain why low-level rural conflict resulted in broader enmities between urban policymakers. However, the philosophical model fills these gaps in the other models’ historical accounts by explaining Arab leaders’ preferences as the result of their understanding of Zionism as the implementation of a philosophical principle – an unnatural, unspiritual materialism – that threatened their existence and thus inclined them toward the broadest possible enmity with an emerging Israel