SEQUENTIAL DECISIONS AND PREDICTIONS IN NATURAL LANGUAGE PROCESSING

Natural language processing has achieved great success in a wide range of ap- plications, producing both commercial language services and open-source language tools. However, most methods take a static or batch approach, assuming that the model has all information it needs and makes a one-time prediction. In this disser- tation, we study dynamic problems where the input comes in a sequence instead of all at once, and the output must be produced while the input is arriving. In these problems, predictions are often made based only on partial information. We see this dynamic setting in many real-time, interactive applications. These problems usually involve a trade-off between the amount of input received (cost) and the quality of the output prediction (accuracy). Therefore, the evaluation considers both objectives (e.g., plotting a Pareto curve). Our goal is to develop a formal understanding of sequential prediction and decision-making problems in natural language processing and to propose efficient solutions. Toward this end, we present meta-algorithms that take an existent batch model and produce a dynamic model to handle sequential inputs and outputs. Webuild our framework upon theories of Markov Decision Process (MDP), which allows learning to trade off competing objectives in a principled way. The main machine learning techniques we use are from imitation learning and reinforcement learning, and we advance current techniques to tackle problems arising in our settings. We evaluate our algorithm on a variety of applications, including dependency parsing, machine translation, and question answering. We show that our approach achieves a better cost-accuracy trade-off than the batch approach and heuristic-based decision- making approaches. We first propose a general framework for cost-sensitive prediction, where dif- ferent parts of the input come at different costs. We formulate a decision-making process that selects pieces of the input sequentially, and the selection is adaptive to each instance. Our approach is evaluated on both standard classification tasks and a structured prediction task (dependency parsing). We show that it achieves similar prediction quality to methods that use all input, while inducing a much smaller cost. Next, we extend the framework to problems where the input is revealed incremen- tally in a fixed order. We study two applications: simultaneous machine translation and quiz bowl (incremental text classification). We discuss challenges in this set- ting and show that adding domain knowledge eases the decision-making problem. A central theme throughout the chapters is an MDP formulation of a challenging problem with sequential input/output and trade-off decisions, accompanied by a learning algorithm that solves the MDP.

Context Models for Understanding Images and Videos

A computer vision system that has to interact in natural language needs to understand the visual appearance of interactions between objects along with the appearance of objects themselves. Relationships between objects are frequently mentioned in queries of tasks like semantic image retrieval, image captioning, visual question answering and natural language object detection. Hence, it is essential to model context between objects for solving these tasks. In the first part of this thesis, we present a technique for detecting an object mentioned in a natural language query. Specifically, we work with referring expressions which are sentences that identify a particular object instance in an image. In many referring expressions, an object is described in relation to another object using prepositions, comparative adjectives, action verbs etc. Our proposed technique can identify both the referred object and the context object mentioned in such expressions. Context is also useful for incrementally understanding scenes and videos. In the second part of this thesis, we propose techniques for searching for objects in an image and events in a video. Our proposed incremental algorithms use the context from previously explored regions to prioritize the regions to explore next. The advantage of incremental understanding is restricting the amount of computation time and/or resources spent for various detection tasks. Our first proposed technique shows how to learn context in indoor scenes in an implicit manner and use it for searching for objects. The second technique shows how explicitly written context rules of one-on-one basketball can be used to sequentially detect events in a game.

A Figured Worlds Perspective on Middle School Learners' Climate Literacy Development

The purpose of this study was to investigate the nature of the relationship between middle school science learners’ conditions and their developing understandings of climate change. I applied the anthropological theoretical perspective of figured worlds (Holland, Lachicotte, Skinner, & Cain, 1998) to examine learners’ views of themselves and their capacities to act in relation to climate change. My overarching research question was: How are middle school science learners’ figured worlds of climate change related to the conditions in which they are embedded? I used a descriptive single-case study design to examine the climate change ideas of eight purposefully selected 6th grade science learners. Data sources included: classroom observations, curriculum documents, interviews, focus groups, and written assessments and artifacts, including learners’ self- generated drawings. I identified six analytic lenses with which to explore the data. Insights from the application of these analytic lenses provided information about the elements of participants’ climate change stories, which I reported through the use of a storytelling heuristic. I then synthesized elements of participants’ collective climate change story, which provided an “entrance” (Kitchell, Hannan, & Kempton, 2000, p. 96) into their figured world of climate change. Aspects of learners’ conditions—such as their worlds of school, technology and media use, and family—appeared to shape their figured world of climate change. Within their figured world of climate change, learners saw themselves—individually and as members of groups—as inhabiting a variety of climate change identities, some of which were in conflict with each other. I posited that learners’ enactment of these identities – or the ways in which they expressed their climate change agency – had the potential to reshape or reinforce their conditions. Thus, learners’ figured worlds of climate change might be considered “spaces of authoring” (Holland et al., 1998, p. 45) with potential for inciting social and environmental change. The nature of such change would hinge on the extent to which these nascent climate change identities become salient for these early adolescent learners through their continued climate change learning experiences. Implications for policy, curriculum and instruction, and science education research related to climate change education are presented.

INFORMATION TECHNOLOGY PROJECTS IN NONPROFIT UNIVERSITY: A PROJECT GOVERNANCE PERSPECTIVE

The universities rely on the Information Technology (IT) projects to support and enhance their core strategic objectives of teaching, research, and administration. The researcher’s literature review found that the level of IT funding and resources in the universities is not adequate to meet the IT demands. The universities received more IT project requests than they could execute. As such, universities must selectively fund the IT projects. The objectives of the IT projects in the universities vary. An IT project which benefits the teaching functions may not benefit the administrative functions. As such, the selection of an IT project is challenging in the universities. To aid with the IT decision making, many universities in the United States of America (USA) have formed the IT Governance (ITG) processes. ITG is an IT decision making and accountability framework whose purpose is to align the IT efforts in an organization with its strategic objectives, realize the value of the IT investments, meet the expected performance criteria, and manage the risks and the resources (Weil & Ross, 2004). ITG in the universities is relatively new, and it is not well known how the ITG processes are aiding the nonprofit universities in selecting the right IT projects, and managing the performance of these IT projects. This research adds to the body of knowledge regarding the IT project selection under the governance structure, the maturity of the IT projects, and the IT project performance in the nonprofit universities. The case study research methodology was chosen for this exploratory research. The convenience sampling was done to choose the cases from two large, research universities with decentralized colleges, and two small, centralized universities. The data were collected on nine IT projects from these four universities using the interviews and the university documents. The multi-case analysis was complemented by the Qualitative Comparative Analysis (QCA) to systematically analyze how the IT conditions lead to an outcome. This research found that the IT projects were selected in the centralized universities in a more informed manner. ITG was more authoritative in the small centralized universities; the ITG committees were formed by including the key decision makers, the decision-making roles, and responsibilities were better defined, and the frequency of ITG communication was higher. In the centralized universities, the business units and colleges brought the IT requests to ITG committees; which in turn prioritized the IT requests and allocated the funds and the resources to the IT projects. ITG committee members in the centralized universities had a higher awareness of the university-wide IT needs, and the IT projects tended to align with the strategic objectives. On the other hand, the decentralized colleges and business units in the large universities were influential and often bypassed the ITG processes. The decentralized units often chose the “pet” IT projects, and executed them within a silo, without bringing them to the attention of the ITG committees. While these IT projects met the departmental objectives, they did not always align with the university’s strategic objectives. This research found that the IT project maturity in the university could be increased by following the project management methodologies. The IT project management maturity was found higher in the IT projects executed by the centralized university, where a full-time project manager was assigned to manage the project, and the project manager had a higher expertise in the project management. The IT project executed under the guidance of the Project Management Office (PMO) has exhibited a higher project management maturity, as the PMO set the standards and controls for the project. The IT projects managed by the decentralized colleges by a part-time project manager with lower project management expertise have exhibited a lower project management maturity. The IT projects in the decentralized colleges were often managed by the business, or technical leads, who often lacked the project management expertise. This research found that higher the IT project management maturity, the better is the project performance. The IT projects with a higher maturity had a lower project delay, lower number of missed requirements, and lower number of IT system errors. This research found that the quality of IT decision in the university could be improved by centralizing the IT decision-making processes. The IT project management maturity could be improved by following the project management methodologies. The stakeholder management and communication were found critical for the success of the IT projects in the university. It is hoped that the findings from this research would help the university leaders make the strategic IT decisions, and the university’s IT project managers make the IT project decisions.

MULTI-UNIT ACCIDENT CONTRIBUTIONS TO U.S. NUCLEAR REGULATORY COMMISSION QUANTITATIVE HEALTH OBJECTIVES: A SAFETY GOAL POLICY ANALYSIS USING MODELS FROM STATE-OF-THE-ART REACTOR CONSEQUENCE ANALYSES

The U.S. Nuclear Regulatory Commission implemented a safety goal policy in response to the 1979 Three Mile Island accident. This policy addresses the question “How safe is safe enough?” by specifying quantitative health objectives (QHOs) for comparison with results from nuclear power plant (NPP) probabilistic risk analyses (PRAs) to determine whether proposed regulatory actions are justified based on potential safety benefit. Lessons learned from recent operating experience—including the 2011 Fukushima accident—indicate that accidents involving multiple units at a shared site can occur with non-negligible frequency. Yet risk contributions from such scenarios are excluded by policy from safety goal evaluations—even for the nearly 60% of U.S. NPP sites that include multiple units. This research develops and applies methods for estimating risk metrics for comparison with safety goal QHOs using models from state-of-the-art consequence analyses to evaluate the effect of including multi-unit accident risk contributions in safety goal evaluations.