33760 research outputs found
Sort by
Lexical categories, (re)categorization, and locality in morphosyntax
No full textThis dissertation is about the nature of syntactic primitives and principles, their status in the grammar, and their interaction with extra-linguistic cognition.
The dissertation has two parts unified by the common goal of streamlining the syntax by asking whether some of its proposed constructs are dispensable, whether the motivation for their existence can be found syntax-externally, and whether they must be assumed to be part of the initial state of the learner. While I discuss a range of phenomena in a number of languages, core empirical evidence throughout comes from adjectival derivation in Bosnian/Croatian/ Serbian (BCS).
In the first part of the dissertation, I consider the status of lexical categories (LCs) in grammar. I argue that LCs noun, verb, and adjective are purely formal, abstract categories which have a distributional role in the syntaxes of individual languages, but which do not have a one-to-one mapping to any interpretive property. I argue against proposals that attribute universal syntactic or semantic properties to the specific LCs. In addition to discussing relevant data from a variety of languages, I provide two detailed case studies on mixed categories: passive and active participles. I show that all participles in the languages under discussion are in fact deverbal adjectives, in every syntactic position they appear in and regardless of their interpretation. While participles may denote (predicates of) properties or eventualities, I argue that these different interpretations are not cross-linguistically associated with more or less verbal or adjectival structure. This reinforces the conclusion that a direct one-to-one correspondence between an item’s LC and its interpretation does not exist. If correct, this proposal has significant con- sequences for our understanding of Universal Grammar. If there are no universal syntactic or semantic properties we can attribute to the LCs, then it becomes superfluous to assume that the individual LCs are part of the initial state of the learner. I propose that the cross-linguistic tendencies we observe around LCs may stem from the way non-linguistic knowledge is organized in the mind/brain.
In the second part of the dissertation, I turn my attention to the formal principles that operate on grammatical primitives, asking specifically what kinds of locality constraints are employed by the grammar. While locality has been extensively studied in generative linguistics, the current offering of locality theories is arbitrary, redundant, baroque, and/or empirically inadequate. There are in essence three competing locality theories currently in circulation within the field: Featural Relativized Minimality (FRM), Phase theory as currently understood in the syntax literature (where it is a successor of Subjacency), and Phase theory as understood in the context of Distributed Morphology (DM). Despite recent attempts to devise a single, unified Phase theory which is responsible for both syntax-internal lo- cality and interface locality, I argue on both conceptual and empirical grounds that the unification is unfeasible. In a detailed empirical study of deadjectival derivation in BCS, I show that adjectivization imposes a DM-locality boundary (for allomorphy and morphological tone assignment), but not a ‘big syntax’-locality boundary (for punctuated movement paths). Nonetheless, I show that the original inventory of locality principles can be reduced if we assume that (i) syntax-internal locality is regulated by FRM, and (ii) interface locality is regulated by Transfer, a modified version of Phase theory which has no syntax-internal effects. I reinterpret the evidence supporting Phase theory through the lens of FRM and demonstrate that the division of labor in (i)-(ii) not only achieves the right empirical cut, but also offers insight into why the grammar may require two distinct locality principles
ORTHOGRAPHIC EFFECTS IN L2 ARABIC WORD PROCESSING: A PRIMING STUDY
No full textLate Arabic learners seem to process words in a similar manner to native speakers, distinguishing their processing pattern from the commonly held assumption that form-relatedness drives lexical access in L2 learners. The morphologically driven facilitation witnessed in L2 Arabic learners inspired the current study to further investigate if orthographic effects can be obtained within the same population. Therefore, the current study investigated if late Arabic learners process orthographically related prime-target pairs in a similar manner to native speakers. The study measured participants' reaction times during a lexical decision task while being primed by orthographic neighbors ( شاب – باب ) and orthographic embedded words ( أسباب - باب ). The results contradicted the prediction and challenge the notion that L2 learners are generally primed byform similarities. L2 Arabic learners seem to differ from L2 speakers of concatenative languages.
The results also confirmed previous claims suggesting a lack of orthographic priming effects in
L2 Arabic learners, indicating that their processing could be morphologically driven
Gamma-Ray Pair Regime Polarimetry with AMEGO-X
No full textMany astrophysical sources such as gamma-ray bursts, active galactic nuclei (AGN), and pulsars emit high-energy radiation in the form of gamma rays. Gamma-ray polarimetry can reveal critical information about their emission mechanisms, which may be otherwise unattainable. The All-sky Medium Energy Gamma-ray Observatory eXplorer (AMEGO-X) is a proposed NASA mission concept designed to detect gamma rays from extreme cosmic sources in the 100 keV to 1 GeV range. Above 10 MeV, gamma rays interacting with the detector primarily undergo pair production, converting into an electron and positron. The azimuthal angle of these two particles carries information about the gamma ray's initial linear polarization. However, multiple scattering of the electron and positron in the detector material is a fundamental limitation and the primary source of polarization sensitivity loss. To evaluate AMEGO-X's ability to measure polarization, Monte Carlo simulations of polarized gamma rays and dedicated data analysis software are used to study the instrument's response. This research presents preliminary simulations that include realistic detector effects to assess the response of AMEGO-X to polarized gamma rays. These results support future advances in high-energy gamma-ray polarimetry
Trust through Transparency: Towards Reliable AI for All
No full textSeemingly performant models can break down in unexpected and uneven ways, from image classifiers failing to recognize an otter out of water, to LLMs being nearly 3 times worse at recalling facts about Somalia than Sweden. In this dissertation, I’ll detail interpretability techniques to scalably illuminate and efficiently intervene on discovered model deficiencies. First, I’ll present evidence for pervasive reliance on spurious correlations by vision models, by way of carefully constructed benchmarks. Then, I’ll automate these approaches, demonstrating the power of leveraging auxiliary models to more efficiently organize data, towards uncovering and articulating subsets where models struggle. Finally, I’ll show how these same techniques can be applied to mitigate instances of real-world geographic disparities and even tackle sociotechnical challenges like artistic copyright infringement. In general, it can be difficult to trust what we do not fully understand, especially when unexpected failures arise. By scalably identifying failure modes before they cause harm, we enhance transparency around model abilities and limitations, thus better informing when models can be trusted to work reliably for all
SHAPING “A WORLD BEYOND A WAR”: THE TECHNOLOGICAL IMAGINARY OF U.S. GEOPOLITICAL DISCOURSE, 1939 – 2016
No full textIn this dissertation, I theorize the technological imaginary to account for how the rhetoric of U.S. presidents has bolstered U.S. supremacy, hegemony, authority, and anxiety. Such presidential rhetoric addresses the destructive forces of rival nations, the threat of nuclear weaponry, and the rise of global threat in World War II, the Cold War, and the U.S. War on Terror in the Middle East. The combination of expanding power and growing anxiety associated with wartime technological advances has created a level of palpable fear that has given presidents considerable authority and deference across these seventy-five years to make decisions on matters of U.S. foreign policy with worldwide implications.
In Chapter 1, I argue that a significant aspect of the U.S. presidential depictions of technology during WWII was the call to increase military strength and defense spending to give the United States a decisive victory in both the European and Asian theaters. As I show, presidents constitute meanings of technology to justify U.S. interventionism, U.S. sovereignty, and U.S. national security expressed through three central topoi: protectionism, securitization, and morality. These three themes are foundational to my theorization of the technological imaginary. Protectionism is grounded in the presumption that nation-states and their leaders have a responsibility to protect their citizens. Securitization helps presidents justify their buildup of the most advanced weapons to protect the people and to deter the enemy from attacking. And presidents turn to U.S. morality—democracy, peace, freedom—to justify their use and authority over military arsenals. I then trace how the themes of protectionism, securitization, and morality circulate across the Cold War and the War on Terror to elaborate the technological imaginary.
In Chapter 2, I focus on how U.S. Cold War presidents articulated a technological discourse that enabled the United States to protect the people and its allies as presidents navigated the treacherous international security environment, projected U.S. technological superiority, and advanced the United States as a moral superpower. As the USSR and the United States built a nuclear arsenal that achieved the level of assured mutual destruction, fears over the weapons designed to protect the people became the source of insecurity. The calls for nuclear buildup to defeat communism clashed with the fears over the weapons themselves, producing what I call a technological paradox. On one side of the paradox is technological essentialism that reinforces the need for the United States to strengthen its defense systems to protect the American people and its allies from enemy threats. On the other side of this paradox is a fear over the nuclear weapons designed to provide such protection. I conceive of this fear as technological nihilism undermining the securitization that such weapons were designed to achieve. Presidents who prioritized technological essentialism and technological nihilism all routinely turn to U.S. morality to justify their technological positioning. Some, such as Presidents Truman and Reagan, justify nuclear buildup in more realist terms to protect the peace and preserve American democracy from the evils of communism. Others justify nuclear treaties in more internationalist terms to reinforce American democratic commitments to diplomacy and to forestall the threat of war. This technological paradox is integral to my conceptualization of the technological imaginary.
Chapter 3 examines the ways in which presidents during the War on Terror manage the technological paradox that integrates arguments of technological essentialism/realism and nihilism/internationalism. I also trace shifts in these arguments from past years as Presidents George W. Bush and Barack Obama assume the authority to determine which nation-states can develop nuclear weapons and those which cannot. While Bush reinforces a commitment to realist principles and technological essentialism, Obama’s rhetoric exhibits the qualities of a technological utopia in places, even as he also affirms the themes of technological essentialism and technological nihilism across his rhetoric.
In the Afterword, I elaborate my theory of the technological imaginary and the role that presidents serve in fusing U.S. national identity with American hegemony through expressions of technology in their foreign policy rhetoric across the three central topoi. Technological essentialism, securitization, and realism have helped presidents and the United States achieve and sustain its technological supremacy, culminating in the U.S. role as imperial leaders of technological weaponry across the globe. Arguments of technological nihilism, morality, and internationalism have provided a counterbalance to such imperialism and mutual-assured destruction. Yet, presidential authority over technological production and determinism reinforces the undemocratic nature of the technological development and the role of economic supremacy in sustaining such power. Presidents assume control to define and use such weaponry and to allow or forbid other political actors from doing so. Presidential reifications of technological essentialism reify a natural order of U.S. technological advancement and supremacy, yet ideological commitments of technological nihilism pose the hope of destabilizing such contrivances
PALS 2025 : Deliverable 3
No full textThis Presentation shows the design process and student involvement in the UMD PALS 2025: Green Isles Project
Measuring Network Dependencies from Node Activations
No full textThis thesis develops a method for inferring dependency networks from binary data by describing and accounting for sources of systematic bias. Node “activation” data is a common source of complex networks, such as scientific collaboration networks from publications, to keyword and tag co-occurrence graphs from documents, epidemiological networks from infection records, and customer product recommendations from buyer carts, to name only a few. However,the way these networks get created often lead to dense edge “hairballs” and heavy post-processing needs. To more accurately reconstruct dependency networks, we propose a flexible technique that generalizes co-occurrence counting, and allows domain-knowledge to inform why nodes activate (and how to use that for inference).
Using our technique, we obtain “Desire Path Density” estimates of latent networks: families of possible networks based on probability distributions over individual edge activations, rather than node co-occurrences. We show that co-occurrence counting is a special case of this estimation technique—and that it assumes node activations arise exclusively from cliques. This assumption leads to “clique-bias” in the recovered graphs, which results in the unwanted hairball effect. Our method, called Forest Pursuit, is a different application of desire path density estimation that attempts to address this bias. It assumes node activations are generated by sampling from distributions over dependency trees.
Forest Pursuit outperforms other algorithms at accurate network recovery, is scalable to very large networks sizes, and can be executed in a streaming or parallel manner over observations. It approximates Steiner trees for each set of activated nodes before aggregating them to estimate the latent global network. We additionally extend Forest Pursuit to be a probabilistic model whose parameters may be inferred through Expectation Maximization. This modification shows improved performance over one-shot Forest Pursuit, at the cost of computation time.
Our introductory chapters provide a unified framework to analyze node activation data and edge (dependency) occurrences in related vector spaces, and provide a taxonomy of network recovery assumption types so that practitioners can find and describe sources of systematic bias from those assumptions. Then, along with derivations for Desire Path Density estimation and Forest Pursuit, we verify our method and compare its performance against others. We created a dataset of synthetic experiments (MENDR), made publicly available as a reference and test-bench for the community. Finally, we use several case studies to demonstrate the impacts of using Forest Pursuit for network analysis in a realistic setting,with insights into the topological differences between networks inferred by other methods and ours. By using Forest Pursuit, practitioners can correct for clique-bias and better use their latent networks to uncover important behavior of complex systems
Cultural Opportunities in Psychotherapy: An Analogue Study of Therapist Perspectives on Religion in Muslim Clients
No full textAlthough considerable efforts have been made to promote multicultural competencies for mental health clinicians, religion/spirituality is still considered a “neglected dimension” of multicultural diversity within the field of counseling (Hage et al., 2006). This may adversely impact religious minorities who express reluctance seeking mental health treatment, underutilize services, and terminate prematurely from therapy. Data was collected from 210 therapists and therapists in training who were randomly assigned to view analogue videoclips in which a therapist either explores religious/spiritual themes or misses cultural opportunities to discuss religion/spirituality in a simulated session with a Muslim client. Results illustrated that when the therapist successfully noticed and responded to cultural opportunities related to the client’s religious/spiritual identity, the therapist was perceived to be more culturally humble and more culturally comfortable. The therapist and client were also perceived to have a stronger working alliance, and the quality of the session was evaluated more favorably by the participant-observers. Recommendations for clinical practice, training, and future research are also discussed
Essays on Development Economics
No full textMy dissertation examines the role of infrastructure, institutions, technology, and environment in the process of economic development.
In Chapter 1, I study distributional impacts of large-scale irrigation in South Africa. Large-scale surface irrigation is essential in keeping agriculture viable in regions prone to drought but it has the potential to exacerbate inequality due to the uneven distribution of its benefits and costs. I investigate this issue in the context of South Africa by estimating heterogeneous effects of irrigation canals on crop productivity and agricultural land expansion by type of farmer. To estimate these effects, I use remote sensing measures of crop yields and a novel land cover classification dataset in a spatial regression discontinuity framework with relative elevation to the nearest canal as the running variable. Areas below the canals serve as the treated group, while areas above serve as the control. The findings show that commercial farmers below canals benefit in terms of higher maize and wheat yields and expand their area under production. Census data further reveal that these expanding commercial farms create employment opportunities for the rural poor. In contrast, subsistence farmers below canals experience lower yields relative to those above and do not expand their cultivated area. Despite the unequal distribution of benefits, a cost-benefit analysis demonstrates that large-scale irrigation infrastructure remains a cost-effective investment.
In Chapter 2, I examine the long-term impacts of the 2004 Indian Ocean Tsunami on human capital accumulation. Natural disasters have devastating immediate impacts but their long-term consequences remain underexplored. Through the disruption of schools in the affected areas they can negatively affect enrollment and school completion rates. I study the impact of the 2004 Indian Ocean tsunami on educational attainment among the affected children and adolescents. Comparing older and younger cohorts within provinces hit by the tsunami and in the rest of the country in a cross-cohort difference-in-differences framework, I find that the tsunami shock increased primary school completion by 1.8 percentage points, while the effect on lower secondary school completion was negative and insignificant and the effect on upper secondary school completion was positive and insignificant. The young adults affected by the tsunami during lower secondary school age or younger were also more likely to perform unpaid family work. A preliminary exploration of mechanisms suggests that the positive effects on human capital accumulation were concentrated among households who did not migrate away from the affected provinces, thus benefiting from relief aid.
In Chapter 3, I study the effects of the account-to-account interoperability between different mobile money providers on mobile money adoption in Tanzania. The introduction of account-to-account interoperability, which allows users to transfer funds between accounts across different MMPs, was expected to foster broader adoption by leveraging network externalities and reducing transaction costs. Using data from the Tanzania National Panel Survey, I develop a differentiated product demand model to estimate the effect of interoperability on mobile money adoption. The model captures how household preferences for interoperability vary by key characteristics such as wealth, education, and urban residence. Results show that households, on average, place a positive value on interoperability, with wealthier, more educated, and urban households exhibiting a stronger preference for interoperable services. This study contributes to the literature on financial inclusion and the importance of regulatory framework in driving mobile money adoption
CLINICAL FEASIBILITY OF THE UNFAMILIAR PEER PARADIGM IN CLINICAL PRACTICE SETTINGS: CLINICAL INSIGHT INTO POSSIBLE IMPLEMENTATION AND EFFECTIVENESS IN CLINICAL PRACTICE
No full textExposure-based treatment for social anxiety disorder (SAD) is an extensively supported evidence-based treatment. However, there are multiple barriers to implementation in routine practice. Furthermore, exposure-based treatments must accurately reflect clients’ social environments for optimal effect. Emulating an adolescent client’s social environment can be particularly challenging, as this often involves interactions with same-age, unfamiliar peers. To this end, the current study explored whether a lab-based model for eliciting social anxiety (i.e., Unfamiliar Peer Paradigm [UPP]) can be modified to leverage task-sharing principles, namely use of lay individuals, to more accurately simulate the social contexts that typify adolescent clients’ symptoms and impairments during in-session exposures. For this dissertation, I reviewed lab-based applications of the UPP and proposed ways to tailor the UPP for therapeutic exposures with adolescent clients. To further expand this proposed model, I sought clinical insight from experts in the field: practicing clinical psychologists. I surveyed clinicians on their current clinical practices of utilizing exposure-based treatment for social anxiety disorders and conducted qualitative interviews to gather feedback on their perceived feasibility and the clinical utility of implementing the UPP in therapeutic settings. Ten practicing psychologists were interviewed. Major themes identified across interviews included generally positive impressions of exposure-based treatments and multiple barriers to implementation. With regards to the UPP, major themes included discussions of feasibility and practicality of implementation, with a generally positive impression of the UPP as a clinically useful approach. With insight from clinical experts, the current study can further the work on utilizing the UPP using lay individuals through task-sharing principles to further enhance the effectiveness of in-person social anxiety-based exposures for adolescents