Doctor of Philosophy, The Ohio State University, 2023, Linguistics

Given a pair of sentences, a premise and a hypothesis, the task of natural language inference (NLI) consists of identifying whether the hypothesis is true (Entailment), false (Contradiction), or neither (Neutral), assuming that the premise is true. NLI is arguably one of the most important tasks for natural language understanding. Datasets have been collected in which pairs of sentences are annotated by multiple annotators with one of the three labels. However, it has been shown that annotation disagreement, or human label variation (Plank, 2022), is prevalent and systematic for NLI – human annotators sometimes do not give the same label for the same pair of sentences (Pavlick and Kwiatkowski, 2019, i.a.). Label variation questions the widespread assumption in natural language processing that each item has a single ground truth label and casts doubt on the validity of measuring models' ability to produce such ground truth labels. In this dissertation, I investigate the question of why there is label variation in NLI and how to build models to capture it. First I analyze the reasons for label variation from the perspective of linguists, by developing a taxonomy of reasons for label variation. I found that NLI label variation can arise out of a wide range of reasons: some are due to uncertainty in the sentence meaning, while others are inherent to the NLI task definition. However, it is unclear how well the perspective of linguists reflect that of linguistically-uninformed annotators. Therefore, I collect annotators' explanations for the NLI labels they chose, creating the LiveNLI dataset containing ecologically valid explanations. I found that the annotators' reasons for label variation are similar to the taxonomy across the board, but some other reasons also emerged. Explanations also reveal that there exists within-label variation: annotators can choose the same label for different reasons. There is thus a wide range of variation that NLI models should capture. (open full item for complete abstract)

Committee: Marie-Catherine de Marneffe (Advisor); Michael White (Committee Member); Chenhao Tan (Committee Member); Micha Elsner (Committee Member) Subjects: Computer Science; Linguistics

Master of Arts, The Ohio State University, 2022, Anthropology

The Inhibitory Cascade Model (ICM) is a model of molar development positing that molar tooth germs influence one another's growth by a balance of activator and inhibitor molecules as they form in series. Previous research on mammals has generally validated the model, but conformation of human molars to ICM predictions has been questioned. Previous work on humans has also not assessed the extent to which population variation in molar proportions conforms to ICM predictions. Here, two different human population samples are assessed based on expectations of the ICM. Both are expected to exhibit a pattern of decreasing size across molar series, as has been shown to be the most frequent pattern in previous studies for humans. However, the two populations are predicted to differ in terms of molar size ratios given that one population sample (Seminole Nation) derives from an ancestry group that exhibits high frequencies of third molar agenesis, while the other (Gullah-Geechee) does not. If the ICM influences differences in third molar agenesis between peoples of Native American vs. African ancestry, then it is expected that molar sizes will decrease across the molar row to a greater extent in the Seminole than they do in the Gullah. Such a difference, though not definitively demonstrating a role for the ICM in third molar agenesis, would be consistent with the possibility that in the Seminole, there is greater inhibition of growth across the molar row than there is in the Gullah. Buccolingual (BL), mesiodistal (MD), and area (BL x MD) dimensions of the teeth were analyzed. Frequencies of size proportion relationships for molar areas were analyzed, with attention to patterns consistent with ICM expectations. Molar size ratios for the BL and MD dimensions as well as for areas were compared for the two population samples in simple and multiple linear regression models. Across the entire sample set, there is a clear pattern aligned with expectations of the ICM, with greatest (open full item for complete abstract)

Committee: Debbie Guatelli-Steinberg Dr. (Advisor); John Hunter Dr. (Committee Member); Hubbe Mark Dr. (Committee Member); Barbara Piperata Dr. (Committee Member) Subjects: Physical Anthropology

Master of Science, The Ohio State University, 2021, Health and Rehabilitation Sciences

Manuscript 1: Methodology Dependent Variation in Volumetric Bone Mineral Density Calculation throughout the Body Introduction: Bone quality assessment using quantitative computed tomography (QCT) may provide a more in-depth and accurate assessment of osteoporosis and fracture risk than dual-energy x-ray absorptiometry (DXA). However, QCT methodologies utilizing single-scan calibration curves may not account for differential x-ray attenuation caused by the patient which may influence calculated volumetric bone mineral density (vBMD) and skew bone quality and fracture risk assessment. Methodology: Clinical CT scans were conducted on 50 male post-mortem human subjects with phantom calibration rods throughout the scan. Height and weight were collected to determine subject BMI. Hounsfield units (HU) from skeletal volumes of interest (VOIs) were collected from the lumbar spine and left femoral neck, humerus, radius, tibia, and calcaneus. The femoral neck was segmented into trabecular (Tb), cortical (Ct) and Total (Tb and Ct) VOI's, the lumbar spine and the calcaneus consisted of Tb and Total VOIs, and the humerus, radius, and tibia were assessed for Ct bone. HU from each VOI was converted to vBMD using both a general scan specific (Gen.) calibration curve constructed from phantom rods within the CT slices of the lumbar region and location specific (LS) calibration curves constructed from phantom rods in slices for each of the skeletal VOIs. Results: Significant variation in vBMD calculated from Gen. and LS calibration curves was observed in the femoral neck, calcaneus, and tibia in all skeletal compartments 15 (p<0.01). However, no significant differences were observed in any of the lumbar spine, humerus, or radius VOIs (p>0.01). Additionally, BMI was not able to explain variation in vBMD values at any site (p>0.01). Conclusions: Using a single calibration curve to calculate vBMD in other anatomical locations, may skew bone quality and differential fracture risk ass (open full item for complete abstract)

Committee: Randee Hunter PhD (Advisor); Amanda Agnew PhD (Committee Member); Jun Zhang PhD (Committee Member) Subjects: Anatomy and Physiology; Biomedical Research; Health; Health Care

Doctor of Philosophy, The Ohio State University, 2014, Anthropology

Senescence varies substantially within and among populations. Data examined here extend knowledge on modern human variation by analyzing elders from Nagasaki Prefecture, Japan, enhance our understanding of relationships between senescence and human biology, and provide more information concerning the use of allostatic load (AL) as a measure of senescent decline among a non-Western population. Developing a valid method for assessing physiological variation due to senescence will benefit those studying health outcomes and survival of elders. It also will aid in focusing healthcare funds and interventions by targeting those most likely to experience unwanted outcomes. Understanding how Japan's elders are surviving and adapting to old age, life-long stress, and developing dysfunction with increasing age provides a model of how others may slow senescence in other settings. Background: 96 elderly residents of Sakiyama City, Nagasaki Prefecture (ages 55-89) and 27 elderly residents of Hizen-Oshima, Nagasaki Prefecture, Japan (ages 51-82) were sampled for components of allostatic load (AL) and other aspects of physical and physiological variation. Surveys were conducted by local health care nursing staff and members of a joint American-Japanese research team during participants' yearly physical examinations. Japan was selected as the study site because Japanese men and women rank among the longest-lived people in the world and the population is relatively genetically homogenous. Methods: AL is a summary measure of physiological activity across multiple regulatory systems pertinent to disease risks. AL incorporates data on ten components: systolic and diastolic blood pressure, high density lipoproteins, total cholesterol, glycosylated hemoglobin, dihydroepiandrosterone-sulfate, cortisol, noradrenaline, adrenaline, and waist:hip ratio. It is calculated by summing the number of components for which an individual's values are in the highest risk quartile. Two altern (open full item for complete abstract)

Committee: Douglas E. Crews PhD (Advisor); Jeffrey Cohen PhD (Committee Member); Randy Nelson PhD (Committee Member) Subjects: Aging

Doctor of Philosophy, The Ohio State University, 2007, Anthropology

This dissertation reconstructs behavioral and biological variation among prehistoric Jomon foragers and Yayoi agriculturalists using bioarchaeological data. The Jomon were a group of foragers from Japan dating to approximately 13,000 until 2500 BP. The Yayoi were the first agriculturally dependent people on the Japanese islands, dating from 2500 until 1700 BP. Data collected from human skeletal remains were used to test the following hypotheses about these groups: 1) Patterns of systemic stress among prehistoric Jomon foragers was variable; 2) Systemic stress increased following the transition to agriculture; 3) Systemic stress patterns among Yayoi agriculturalists was variable; 4) Differences in body size and proportions will be observed among and between Jomon foragers and Yayoi agriculturalists; 5) Systemic stress experienced during the agricultural transition in prehistoric Japan was greater than other East Asian agriculturalists and improved compared with North American agriculturalists. Non-specific indicators of stress suggest the Jomon from western Japan experienced greater disease loads than those from eastern Japan. This trend is associated with plant dependent diets and resource scarcity in western Japan. Stature variation is, however, not recorded between these two groups indicating that systemic stress severity was not different between the eastern and western Jomon. Oral health declined following the transition to agriculture in prehistoric Japan in association with increased consumption of carbohydrates. Non-specific indicators of stress reduced in frequency following the transition to agriculture. These trends indicate that the quality of life for prehistoric Japanese was generally improved compared to prehistoric North American agriculturalists and similar to that observed in prehistoric East Asian agriculturalists. In addition, patterns of health observed among Yayoi agriculturalists, specifically enamel hypoplasia and stature variation, indicates (open full item for complete abstract)

Committee: Clark Larsen (Advisor) Subjects: Anthropology, Physical

Doctor of Philosophy, Case Western Reserve University, 2010, EECS - Computer and Information Sciences

Haplotype is an important representation of human genetic variation and is thus valuable for investigating the genetics behind diseases.However, humans are diploid and in practice, genotype data instead of haplotype data are collected directly. Consequently, there are great demands for efficient and accurate computational methods to reconstruct haplotypes from genotype data. Our project started with the development of a rule-based haplotyping method for pedigree data with tightly linked markers. We formulate Mendelian constraints as a linear system of inheritance variables and solve the linear system using disjoint-set data structures. Our algorithm achieved the lowest time complexity among all existing methods. Comparisons with two popular algorithms showed that this algorithm made 10 to 105-fold improvements on a variety of parameter settings. Based on the zero-recombinant haplotype inference, we went on to construct a general framework for haplotyping population and pedigree mixed data that consist of many families with unrelated founders, by combining novel techniques of recombination event detection and maximum likelihood optimization. This method makes it possible to do the genome-wide haplotype inference on pedigree and population mixed data.

Committee: Jing Li (Committee Chair); Xiaofeng Zhu (Committee Member); Mehmet Koyuturk (Committee Member); Soumya Ray (Committee Member) Subjects: Bioinformatics; Computer Science