Pranav A

I am Pranav and I am currently working as a PhD researcher at the University of Hamburg.

My email is cs.pranav.a (at) gmail.com.

My full name is Pranav Agrawal, but I prefer to be called by 'Pranav A'. For academic publications, I use the name 'A Pranav'. I use either he/him or they/them pronouns.

Github Link  /  LinkedIn Link  /  CV (Link to PDF)

Research Interests and Publications

Broadly, my interests are in NLP (Natural Language Processing) and AI ethics.

Regarding AI ethics, my main areas of interest include inclusivity, data sovereignty and fair policymaking. Presently, my research efforts are concentrated on the following topics:

• How can community-based approaches enhance inclusivity at conferences?
• In cases where control over digital identity, as manifested in datasets, models, and algorithms, has been taken away from queer individuals, what tools can be devised to counteract this?
• Why do deadnames occur in citations of publications, and what tools can be developed to prevent the circulation of deadnames and outdated information?

We introduce Glitter, an English-German benchmark featuring extended passages with professional translations implementing three gender-fair alternatives. Our experiments reveal significant limitations in state-of-the-art language models, which default to masculine generics and rarely produce gender-fair translations even when explicitly instructed.

We expand Yann LeCun's 'cake that is intelligence' analogy to the full life-cycle of AI systems, from sourcing data to evaluation and distribution. We describe each step's social ramifications and provide actionable recommendations for increased participation in AI discourse.

Emperical comparisions on character based models against word based models on common Chinese semantic benchmarks.

Community-led participatory design case study of Queer in AI contributed lessons on decentralization, building community aid, empowering marginalized groups, and critiquing poor participatory practices.

Queer in AI provides a tutorial for diversity & inclusion organizers on making virtual conferences more queer-friendly through inclusivity based on their community's experiences with marginalization.

The paper contributes a contextual subword segmentation method along with benchmark datasets that outperformed previous Chinese character conversion approaches by 6 points in accuracy, especially for code-mixing and names entities.

The paper contributes information retrieval ranking functions with heuristics like positional tokenization and graphical error modelling to the problem of cognate detection.