Kelly Marchisio
kmarc@jhu.edu
Welcome!
I am a Member of Technical Staff at Cohere (https://cohere.ai/) specializing in multilingual natural language processing. I started the multilingual NLP group at Cohere in June 2023. Our work resulted in Cohere’s first public launch of a multilingual model in March 2024 (Command-R).
Previously, I completed my PhD in Computer Science at the Center for Language and Speech Processing at Johns Hopkins University, where I was advised by Dr. Philipp Koehn. My research focuses on unsupervised and semi-supervised machine translation, bilingual lexicon induction, multilingual representation learning, and efficient natural language processing. During my time at JHU, I interned with Meta AI London (FAIR Labs, Summer 2022, Advisor: Mikel Artetxe) and Google Translate (Summer 2020).
Prior to JHU, I completed an MPhil in Advanced Computer Science at Christ’s College, University of Cambridge. My thesis advisor was Helen Yannakoudakis. I was employed at Google, Inc. in Mountain View, CA from 2013-2017 where I was a Web Solutions Engineer (2015-2017) and Support Specialist for gTech Ads, where I led automation efforts for North America (2013-2015). Previously, I completed an Ed.M. in Mind, Brain, and Education at Harvard University, and a B.A. in Psychology/Sociology and French (minor) at Franklin & Marshall College in Lancaster, PA.
News
Summer 2024: It’s been a busy summer at Cohere! (1) I am co-supervising the internships of Nate Robinson and Piotr Nawrot, (2) We have released four pre-prints since May:
How Does Quantization Affect Multilingual LLMs?
Kelly Marchisio, Saurabh Dash, Hongyu Chen, Dennis Aumiller, Ahmet Üstün, Sara Hooker, Sebastian Ruder
Understanding and Mitigating Language Confusion in LLMs
Kelly Marchisio, Wei-Yin Ko, Alexandre Bérard, Théo Dehaze, Sebastian Ruder*
RLHF Can Speak Many Languages: Unlocking Multilingual Preference
Optimization for LLMs
John Dang, Arash Ahmadian, Kelly Marchisio, Julia Kreutzer, Ahmet Üstün, Sara Hooker
Aya 23: Open Weight Releases to Further Multilingual Progress
Cohere & Cohere For AI
March 2024: Our work since June contributed to the launch of Command-R, Cohere’s first public multilingual model!
December 2023: Presented at NeurIPS 2023 in New Orleans:
Improving Language Plasticity via Pretraining with Active Forgetting
Yihong Chen, Kelly Marchisio, et al. 2023
July 2023: Presented at ACL 2023 in Toronto:
Mini-Model Adaptation: Efficiently Extending Pretrained Models
to New Languages via Aligned Shallow Training (Findings)
Kelly Marchisio, Patrick Lewis, Yihong Chen, Mikel Artetxe
12 June 2023: I had my first day as Member of Technical Staff at Cohere!
7 June 2023: My PhD defense will be held at Johns Hopkins University at 2pm ET (UTC-4:00). Please email me if you’d like a Zoom link! UPDATE: I successfully defended! Title: “Multilinguality from Static Embedding Spaces: Algorithmic, Geometric, and Data Considerations”
Feb 2023: Talks at George Mason University (8 Feb) and the Artificial Intelligence Society at Johns Hopkins (22 Feb)
December 2022: Two papers to-appear at EMNLP in Abu Dhabi:
IsoVec: Controlling the Relative Isomorphism of
Word Embedding Spaces
Kelly Marchisio, Neha Verma, Kevin Duh, Philipp Koehn
Bilingual Lexicon Induction for Low-Resource Languages using
Graph Matching via Optimal Transport
Kelly Marchisio, Ali Saad-Eldin, Kevin Duh, Carey Priebe, Philipp Koehn
October 2022: I am honored to have been named an Amazon Fellow! https://twitter.com/AmazonScience/status/1577765575207460865
Summer 2022: I will be interning at FAIR in London, UK with Mikel Artetxe.
More about me
In my non-CS life, I’m a classically-trained soprano and was a Choral Scholar at Christ’s College, Cambridge. I’ve had fun singing acappella through the years – with Google MTV’s Googapella (2014-2017), the Common Chords at the Harvard Graduate School of Education, and Sweet Ophelia at Franklin & Marshall College.
Apart from singing, I enjoy vegetable gardening – and if “trying new foods” can be considered a hobby, then it’s certainly one of favorites!