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.

DEVELOPMENT OF A DECISION/DEPLOYMENT SUPPORT TOOL FOR VARIABLE SPEED LIMITS IN RECURRENT CONGESTION

Traffic demand increases are pushing aging ground transportation infrastructures to their theoretical capacity. The result of this demand is traffic bottlenecks that are a major cause of delay on urban freeways. In addition, the queues associated with those bottlenecks increase the probability of a crash while adversely affecting environmental measures such as emissions and fuel consumption. With limited resources available for network expansion, traffic professionals have developed active traffic management systems (ATMS) in an attempt to mitigate the negative consequences of traffic bottlenecks. Among these ATMS strategies, variable speed limits (VSL) and ramp metering (RM) have been gaining international interests for their potential to improve safety, mobility, and environmental measures at freeway bottlenecks. Though previous studies have shown the tremendous potential of variable speed limit (VSL) and VSL paired with ramp metering (VSLRM) control, little guidance has been developed to assist decision makers in the planning phase of a congestion mitigation project that is considering VSL or VSLRM control. To address this need, this study has developed a comprehensive decision/deployment support tool for the application of VSL and VSLRM control in recurrently congested environments. The decision tool will assist practitioners in deciding the most appropriate control strategy at a candidate site, which candidate sites have the most potential to benefit from the suggested control strategy, and how to most effectively design the field deployment of the suggested control strategy at each implementation site. To do so, the tool is comprised of three key modules, (1) Decision Module, (2) Benefits Module, and (3) Deployment Guidelines Module. Each module uses commonly known traffic flow and geometric parameters as inputs to statistical models and empirically based procedures to provide guidance on the application of VSL and VSLRM at each candidate site. These models and procedures were developed from the outputs of simulated experiments, calibrated with field data. To demonstrate the application of the tool, a list of real-world candidate sites were selected from the Maryland State Highway Administration Mobility Report. Here, field data from each candidate site was input into the tool to illustrate the step-by-step process required for efficient planning of VSL or VSLRM control. The output of the tool includes the suggested control system at each site, a ranking of the sites based on the expected benefit-to-cost ratio, and guidelines on how to deploy the VSL signs, ramp meters, and detectors at the deployment site(s). This research has the potential to assist traffic engineers in the planning of VSL and VSLRM control, thus enhancing the procedure for allocating limited resources for mobility and safety improvements on highways plagued by recurrent congestion.