about me

Hi, I'm Nick.

I’m a researcher in AI code generation at Microsoft Research, Cambridge. I’m interested in code as an output modality for generative AI due to its executability, a property distinct from free textual generation. Executable code is excellent for tasks involving calculation, planning, tool use, and verifiable reasoning, which traditional text-based approaches in LLMs struggle with. Further, generated code can exercise deep control of the computing environment, which I believe will enable a future of more powerful computing experiences for end-users.

My past research centers on the formal semantics of natural language and computer modeling of meaning, including with LLMs, Knowledge Graphs, Entailment Graphs, and bespoke neural architectures. I have done award-winning research showing that even though LLMs lack the understanding of language inherent in humans, they learn a latent space of concepts that is highly organized in a way which can approximate linguistic reasoning.

I hold a Ph.D. from the University of Edinburgh and the Institute for Language, Cognition, and Computation, where I was advised by Mark Steedman. I also hold a B.Sc. in Computer Science from Brown University.


Recent News:
Feb '24 I started a postdoc position with Microsoft Research, Cambridge.
Nov '23 Our paper, Smoothing Entailment Graphs with Language Models, won the "Best Paper" award at AACL 2023.
Oct '23 Our paper, Sources of Hallucination by Large Language Models on Inference Tasks, was accepted to EMNLP Findings 2023.
Oct '23 I defended my Ph.D. thesis, Inference of Natural Language Predicates in the Open Domain.
selected publications

Nick McKenna*, Tianyi Li*, Liang Cheng, Mohammad Javad Hosseini, Mark Johnson, and Mark Steedman. Sources of Hallucination by Large Language Models on Inference Tasks. Findings of EMNLP, 2023. Paper Link.

Nick McKenna, Tianyi Li, Mark Johnson, and Mark Steedman. Smoothing Entailment Graphs with Language Models. AACL, 2023. *Best Paper Award!* Paper Link.

Nick McKenna, Liane Guillou, Mohammad Javad Hosseini, Sander Bijl de Vroe, Mark Johnson, and Mark Steedman. Multivalent Entailment Graphs for Question Answering. EMNLP, 2021. Paper Link.


For a full list of publications, refer to my CV.

contact

Reach me by email or find me on X/Twitter and LinkedIn.

© 2024 Nick McKenna