Seminar in Computational Linguistics

In this seminar, I will outline recent studies orbiting three popular topics in natural language processing: multilinguality, dependency parsing, and interpretability. First, I will present work on extending state-of-the-art methods in bilingual word embedding mapping to the multilingual case (read: more than 2 languages), and whether or not such alignments are a viable option for cross-lingual learning. Second, I will discuss a multilingual case study in dependency parsing in the age of Sesame Street (e.g. ELMo, BERT), analyzing the performance of neural parsers when augmented with newly-proposed contextualized embedding models. Lastly, I will present work that relates the aforementioned embedding models to human processing data via the Representational Similarity Analysis (RSA) framework. Ultimately, I hope to establish a common thread across these (seemingly) disparate themes and present related directions for future work.