Rhetorical Structure Theory (Mann et al. 1988), a popular approach for analyzing discourse coherence, suggests that coherent text can be placed into a hierarchical organization of clauses. Identification of a texts rhetorical structure through automatic discourse analysis is a crucial element for many of todays Natural Language Processing tasks, but no sufficient tool is available. The current state-of -the-art discourse parser, SPADE (Soricut et al. 2003), is limited to parsing discourse within a single sentence. HILDA (Hernault et al. 2010) extends the parsing abilities of SPADE to the document level, but with a decrease in performance. This study achieved document-level discourse parsing without sacrificing performance. Provided text was already segmented into elementary discourse units, the task of discourse parsing was separated into three steps: structuring, nuclearity labeling, and relation labeling. An algorithm was developed for classifying relation existence, nuclearity, and relation label that improved upon previous methods. New features were explored for all three steps to maintain state-of-the-art performance when parsing at the document-level.

Microalgal harvesting strategies for biodiesel production have been a major setback in the industry with high energy estimates of $3400/ton biomass by centrifugation. The present study utilized effective mass and energy balances to reduce these large operating costs. The energy for Stage 1 centrifugation was reduced by 82% when harvesting 28.5% of biomass at 18 L/min as opposed to 95% harvesting at 1 L/min. This strategy was further confirmed using electrocoagulation (EC) with Nannochloris and Dunaliella algae with perforated aluminum and iron electrodes at low (< 6 mg/L) metal ion concentrations. Despite 20% lower harvesting efficiencies, the iron electrodes were more energy and cost efficient with operating costs less than $0.03/L oil when flocculating and settling Nannochloris and Dunaliella cultures. Furthermore, a continuous multistage algae harvester using EC and dissolved air flotation (DAF) for Stage 1 harvesting and centrifugation for Stage 2 dewatering was designed. It was determined throughout the testing that greater EC costs for improved harvesting efficiencies were necessary to offset the large energy requirements of the DAF. The multistage system dewatered a low density (100 mg/L) Nannochloris to 20% solids for a final energy requirement of 1.536 kWh/kg algae ($138/ton). Using the data collected from this research and existing literature, a life cycle analysis was assembled to judge the sustainability of microalgal biofuels in Louisiana. High and low energy estimates for culturing (mixing, CO2, nutrients), harvesting, lipid extraction and energy conversion were compared with the current research. Scaling the EC/DAF system for a full size facility was expected to reduce the harvesting costs to 1.133 kWh/kg algae, resulting as $0.44/L oil for a culture with 20% lipids. Despite this improvement in harvesting costs, the production of algal for the sole purpose of biodiesel was not economically viable. Considering a system with a growth rate of 15 g/m2/day and lipid content of 20%, the energy inputs exceeded the outputs from biodiesel production by 36% under the most ideal conditions. However, incorporating additional revenue through wastewater treatment and biogas production from residual biomass could improve sustainability and profitability of algal biodiesel to an 18.5% energy surplus at its current state.

Coreference resolution is the task of resolving all expressions in a text that refer to the same entity. Such expressions are often used in writing and speech as shortcuts to avoid repetition. The most frequent form of coreference is the anaphor. To resolve anaphora not only grammatical and syntactical strategies are required, but also semantic approaches should be taken into consideration. This dissertation presents a framework for automatically resolving pronominal anaphora by integrating recent findings from the field of linguistics with new semantic features. Commonsense knowledge is the routine knowledge people have of the everyday world. Because such knowledge is widely used it is frequently omitted from social communications such as texts. It is understandable that without this knowledge computers will have difficulty making sense of textual information. In this dissertation a new set of computational and linguistic features are used in a supervised learning approach to resolve the pronominal anaphora in document. Commonsense knowledge sources such as ConceptNet and WordNet are used and similarity measures are extracted to uncover the elaborative information embedded in the words that can help in the process of anaphora resolution. The anaphoric system is tested on 350 Wall Street Journal articles from the BBN corpus. When compared with other systems available such as BART (Versley et al. 2008) and Charniak and Elsner 2009, our system performed better and also resolved a much wider range of anaphora. We were able to achieve a 92% F-measure on the BBN corpus and an average of 85% F-measure when tested on other genres of documents such as children stories and short stories selected from the web.

In this study, an automatic semantic segmentation and object recognition methodology is implemented which bridges the semantic gap between low level features of image content and high level conceptual meaning. Semantically understanding an image is essential in modeling autonomous robots, targeting customers in marketing or reverse engineering of building information modeling in the construction industry. To achieve an understanding of a room from a single image we proposed a new object recognition framework which has four major components: segmentation, scene detection, conceptual cueing and object recognition. The new segmentation methodology developed in this research extends Felzenswalb's cost function to include new surface index and depth features as well as color, texture and normal features to overcome issues of occlusion and shadowing commonly found in images. Adding depth allows capturing new features for object recognition stage to achieve high accuracy compared to the current state of the art. The goal was to develop an approach to capture and label perceptually important regions which often reflect global representation and understanding of the image. We developed a system by using contextual and common sense information for improving object recognition and scene detection, and fused the information from scene and objects to reduce the level of uncertainty. This study in addition to improving segmentation, scene detection and object recognition, can be used in applications that require physical parsing of the image into objects, surfaces and their relations. The applications include robotics, social networking, intelligence and anti-terrorism efforts, criminal investigations and security, marketing, and building information modeling in the construction industry. In this dissertation a structural framework (ontology) is developed that generates text descriptions based on understanding of objects, structures and the attributes of an image.

Photocatalytic concrete pavements are a promising technology for mobile source air pollution remediation, however before widespread application of this technology is realized many unanswered questions remain regarding its overall environmental impact. In response to these questions, the goal of this study was to increase the understanding of the environmental impact of photocatalytic concrete pavement highways. To achieve this goal, the objectives of this study were to (A) construct a model that evaluates the nitrogen oxides (NOx) reduction from photocatalytic pavements, (B) quantify the nitrates released from the photocatalytic degradation of NOx, and (C) identify and characterize pathways for TiO2 nanoparticle exposure. To achieve objective A, a field study was conducted to evaluate the NOx reduction. Results showed evidence of minimal photocatalytic reductions with large variability due to many unknown and known parameters. As a result, this study also investigated the use of laboratory results to better understand the significance of the NOx reduction through the creation of a theoretical mass balance Lavoisier box model. Laboratory results indicated that the nitrogen monoxide (NO) oxidation rate is reaction rate mass transfer controlled following the Langmuir- Hinshelwood (L-H) model. A parametric study was completed to evaluate the L-H constants under different environmental conditions and statistical model was created to describe the NO oxidation rate. Incorporating the resulting NO oxidation rate into a Lavoisier box model the mass transfer mechanisms were compared and objective A was achieved. Objectives B and C of the project deal with evaluating potential unintended consequences resulting from implementation of photocatalytic concretes. To complete objective B, nitrates and TiO2 nanoparticles released to water were quantified. Lastly, TiO2 nanoparticles released to the air during construction activities were quantified and characterized to achieve objective C.

In this research, we focus on the early prediction of whether topics are likely to generate significant controversy (in the form of social media such as comments, blogs, etc.). Controversy trend detection is important to companies, governments, national security agencies, and marketing groups because it can be used to identify which issues the public is having problems with and develop strategies to remedy them. For example, companies can monitor their press release to find out how the public is reacting and to decide if any additional public relations action is required, social media moderators can moderate discussions if the discussions start becoming abusive and getting out of control, and governmental agencies can monitor their public policies and make adjustments to the policies to address any public concerns. An algorithm was developed to predict controversy trends by taking into account sentiment expressed in comments, burstiness of comments, and controversy score. To train and test the algorithm, an annotated corpus was developed consisting of 728 news articles and over 500,000 comments on these articles made by viewers from CNN.com. This study achieved an average F-score of 71.3% across all time spans in detection of controversial versus non-controversial topics. The results suggest that it is possible for early prediction of controversy trends leveraging social media.

Increasing demand on aviation industry calls for more pilots. Thus, pilot training systems and pilot-candidate screening systems are essential for civil and military flying training institutes. Before actual flight training, it is not easy to determine whether a flight trainee will be successful in the training. Due to the high cost of actual flight training, it would be better if there were low cost methods for screening and training candidates prior to the actual flight training. This study intended to determine if subject related factors and flight control stick position have an impact on acquisition of simulated flying skills using a PC-based flight simulator. The experimental model was a factorial design with repeated measures. Sixty-four subjects participated in the experiment and were divided into 8 groups. Experiment consisted of 8 sessions in which performance data, such as heading, altitude and airspeed were collected every 15 seconds. Collected data were analyzed using SAS statistical program. Result of multivariate analysis of variance indicated that the three independent variables: nationality, computer game experience, and flight stick position have significant impact on acquiring simulated flying skill. For nationality, Americans recorded higher scores in general (mean: 81.7) than Koreans (mean: 78.9). The difference in mean scores between Americans and Koreans was 2.8 percent. Regarding computer game experience, the difference between high experience group (82.3) and low experience group (78.3) is significant. For high experience group, American side-stick group recorded the highest (mean: 85.6), and Korean side-stick group (mean: 77.2) scored the lowest. For the low experience group, American center-stick group scored the highest (80.6), and the Korean side-stick group (74.2) scored the lowest points. Therefore, there is a significant difference between high experience group and low experience group. The results also reveal that the center-stick position is easier to learn than side-stick position. The difference in performance score between group of center-stick (mean: 82.1) and side-stick (mean: 76.8) is considerable.

This dissertation presents a novel method of autonomic face recognition based on the recently proposed biologically plausible network of networks (NoN) model of information processing. The NoN model is based on locally parallel and globally coordinated transformations. In the NoN architecture, the neurons or computational units form distributed networks, which themselves link to form larger networks. In the general case, an n-level hierarchy of nested distributed networks is constructed. This models the structures in the cerebral cortex described by Mountcastle and the architecture based on that proposed for information processing by Sutton. In the implementation proposed in the dissertation, the image is processed by a nested family of locally operating networks along with a hierarchically superior network that classifies the information from each of the local networks. The implementation of this approach helps obtain sensitivity to the contrast sensitivity function (CSF) in the middle of the spectrum, as is true for the human vision system. The input images are divided into blocks to define the local regions of processing. The two-dimensional Discrete Cosine Transform (DCT), a spatial frequency transform, is used to transform the data into the frequency domain. Thereafter, statistical operators that calculate various functions of spatial frequency in the block are used to produce a block-level DCT coefficient. The image is now transformed into a variable length vector that is trained with respect to the data set. The classification was done by the use of a backpropagation neural network. The proposed method yields excellent results on a benchmark database. The results of the experiments yielded a maximum of 98.5% recognition accuracy and an average of 97.4% recognition accuracy. An advanced version of the method where the local processing is done on offset blocks has also been developed. This has validated the NoN approach and further research using local processing as well as more advanced global operators is likely to yield even better results.

Much evidence suggests that the cause of lower back pain (LBP) and injury is frequently related to the posture of lifting, the load, muscle fatigue, and other factors. The purpose of this study was to evaluate the effect of progressive fatigue on factors that have previously been associated with increased risk of LBP in various occupational settings, during a repetitive lifting task where freestyle lifting technique was utilized. A laboratory experiment was conducted to evaluate several fatigue analysis, electromyography amplitude, kinematic, and kinetic parameters of repetitive freestyle lifting during a 2-hour lifting period. Each of ten (10) young adult male participants lifted a load from floor height to a lowering platform of 76cm height at a rate of 4 lifts/minute. The mass lifted was determined utilizing the psychophysical approach. The task consisted of 8 consecutive 15-minute periods of lifting, before, between, and after which subjective fatigue rating and strength measurements were taken, and during which kinematic variables were recorded. Effect of time, at α=0.05 level, was observed on subjective fatigue rating (p<0.0001) and on static strength (p=0.0184). Subjective fatigue rating increased over time, indicating that the participant "felt" increasingly fatigued as the experiment progressed. Static composite strength decreased an average of 20% from the beginning to the end of the experiment. Effect of lifting posture (semi-squat, semi-stoop, and stoop) was observed on peak trunk flexion angle (p=0.0122), trunk flexion angle at initiation of the lift (p=0.009), and knee angle at initiation of the lift (p=0.0007), indicating that in freestyle lifting, participants assume quantitatively different lifting techniques. A significant effect of the time-posture interaction was observed on the dynamic leg lift floor to knuckle height strength (0.0237), indicating that dynamic strength may change depending upon lifting posture selected. No generalizable effect of the independent variables was observed on the remaining parameters for all participants. Indicators of general physical fatigue, particularly dynamic floor to knuckle height leglift strength and subjective fatigue rating, were observed to possess some significant predictive capability in variation of a number of kinematic and force parameters.

Hollow particle filled polymers known as syntactic foams are lightweight and highly damage tolerant. Syntactic foams are used as core materials in sandwich composites. The use of such materials in aeronautical and space structures make it necessary to understand their characteristics for various environmental and loading conditions. The first part of the present work takes modeling and finite element analysis approach to understand and predict the deformation behavior of syntactic foams. Contact analysis is performed on single particle models by the finite element analysis approach. In the second part extensive experiments are carried out to characterize syntactic foams for hygrothermal and compressive properties, and syntactic foam core sandwich composites for compressive and flexural properties. Flexural tests are carried out in three and four point bending and short beam shear configurations. Syntactic foams are tested in three different specimen sizes and orientations to characterize them as per the recommendations of various ASTM standards. Effect of specimen aspect ratio on the measured mechanical properties can be determined by such an approach. The effect of change in the internal radius of hollow particles, called cenospheres, on mechanical properties is studied for all these loading conditions. Five different types of cenospheres are selected for the study of the internal radius dependence of mechanical properties of syntactic foams and their sandwich composites. All selected types of cenospheres have the same outer radius, however, their internal radius is different. Hence, difference in mechanical properties of syntactic foams is caused due to a difference in only one parameter, the cenosphere internal radius. Such unique approach made it possible to identify the individual contribution of matrix resin and cenospheres in the mechanical properties of syntactic foams. Specimen deformation behavior and fracture features are correlated to deformation curves obtained during the testing.