• Designed and implemented a language model with a novel architecture consisting of multi-level interactions in human language (word structure/morphology and pronunciation/phonology)• Wrote the code to train models with this architecture, including optimizing weights via the Expectation-Maximization algorithm and using SciPy’s optimize function• Determined the general shape of the solution space (asymptotic) given the interplay of the structure of the training data with the model architecture• Proposed and executed a novel approach (random sampling and filtering of randomly initialized trained language models) to manage language modeling in an asymptotic solution space• Trained ~800 models on 8 morpho-phonological phenomena• Improved model performance by curating training data, identifying new features to annotate, and adjusting the learning rate • Designed test data to evaluate the trained models’ abilities to generalize to novel data in a human-like manner (425 models on 3 morpho-phonological phenomena)• Analyzed parameters to understand why 425 trained models generalized in the way they did, and to classify models into generalizability families• Presented research at multiple scientific conferences and published a paper on my language model

I collaborated with 2 language scientists to launch a package in R, which (1) trained language models, (2) automated cross-validation for a set of trained models, (3) compared models according to the AIC, BIC and the likelihood ratio test, and (4) predicted probabilities for test set items (including an option to adjust temperature).• Reviewed code, tested, debugged and wrote documentation• Designed 3 data sets and multiple edge cases to test the package's functions• Derived and implemented the gradient function for the objective function, leading to a 4-fold increase in optimization speed• Utilized Git for version control and GitHub for collaborative work

• Managed a team of 8 teaching assistants for a course with ~480 students, and resolved student issues• Trained ~90 students to use audio analysis software to annotate and analyze speech data• Identified weekly learning objectives, designed materials and communicated goals to align student experience across 8 tutors• Supervised, advised and guided undergraduate projects• Established and disseminated procedures on how to classify, annotate, interpret, analyze, and report textual and speech data• Created and implemented criteria to evaluate student performance -- including quality of annotations, linguistic analysis• Taught ~20 UCLA undergraduate weekly discussion sections ranging in size from 10-25 students each. Topics included: phonetics, phonology, morphology, general linguistics• Motivated students, connected them with resources and explored strategies alongside them to perform to their potential under stress-inducing circumstances (e.g. practical phonetics assessments and presentations)

• Annotated, cleaned and spliced ~1070 speech files for use as training and test data• Designed stimuli and wrote script to randomly generate ~1,350 datapoints that were informed by corpus frequencies for human language experiments• Scraped pronunciations and 26 other attributes of 456,890 entries via the Oxford English Dictionary API• Extracted and combined data across multi-lingual corpora and databases (CELEX, Lexique 3, CMU Pronouncing Dictionary, Oxford English Dictionary) to perform statistical analysisAnalyzed complex language data to gain modeling insights:• Synthesized and distilled complex conflicting language data from French, Turkish and Danish to propose 2 novel unifying features that interact to generate data from the three languages• Re-analyzed language data to eliminate unneeded complexity from the features of the CiV Lengthening language model• Designed 2 finite-state transducers (FST) for long-distance dependencies (One was a 2-way deterministic FST, the other was a cascade of two 1-way deterministic FSTs). Phenomenon: unbounded metathesis.Experimental design and data analysis:• Designed experiments to ascertain whether reported linguistic phenomena were currently used by speakers • Collected, cleaned and annotated human data from experiments• Performed statistical analysis on data sets using R, Python and Excel• Prepared visualizations to effectively communicate experimental resultsPresented work at conferences and published a paper describing research findings