The rapid advancements in digital and information technology have inadvertently barred older, and technologically disadvantaged populations from acquiring information and getting necessary tasks done. This digital divide handicaps marginalized people and exacerbates pre-existing inequities. I intend to tackle this inequality by utilizing NLP and IR for human-computer interaction and collaboration to improve accessibility, with a particular focus on providing everyone with equal access to information. This will not only benefit individuals, but also help introduce more informed, active, and diverse populations into society.

So far, my work has focused on efficient and controllable language analysis, IR, and IE, for more accessible and robust question answering and task-oriented dialogue systems.

I have spent considerable time working on language analyzers, including rule-based POS taggers, NER models, and speech-act analyzers whose high throughput and controllability make them suitable for search engines and in high-stakes domains. I also worked to build more robust ML-based models and a hybrid model that kept the customizability and robustness of either method.

In IR and IE, I have worked on MRC, summarization, and recommendation systems. For many of these projects, having to process millions of documents in a short time often required me to make optimizations on all fronts, including data processing, and the use of system resources and databases. I find this a useful experience because making small and efficient systems can be as important as making them bigger and more powerful, especially in making more accessible systems.

My papers were focused on improving the performance of models using linguistic features. The first paper was on classifying speech-acts using neural networks and rule-extracted features, and for my master's thesis, I worked on MRC which utilizes features like POS tags and wh-words. I expected these reliable, low-overhead features would help improve accuracy while maintaining efficiency.

Building upon my past work, I plan to work on processing spoken language and providing target knowledge in a useful way for the user. In pursuing these objectives, I aim to emphasize efficiency to build more deployable systems and provide reliable, grounded information. I also intend to explore linguistic theories and statistical methods for more efficient modeling and gain a deeper understanding of the latest and traditional NLP techniques.