Suchir Salhan

About me

click to read more

Suchir Salhan is a PhD candidate at the University of Cambridge +

My work is guided by a single central question: how do intelligent systems learn? I ask it across minds and machines at once, tracing the same puzzle through human and machine learning, language and intelligence, cognition and computation. Language is where these threads meet: it offers rich structure (syntax, semantics, morphology, phonology) alongside decades of developmental theory, cross-linguistic variation, and behavioural evidence against which learning can be measured.

I’ve had a long fascination with the intersection of language and computation—how humans have developed the capability to acquire natural language to communicate, learn, and reason, despite the diversity of linguistic systems; and how we might build machines that can do the same. I arrived in Cambridge in 2020 to pursue a BA and MEng in Computer Science & Linguistics at Gonville & Caius College, Cambridge, where I earned a “starred First” and a Distinction.

I got into research early, and slowly the two worlds stopped competing. This is what my work is about — minds and machines: cognition, language and learning on one side; AI, agents and small models on the other; and, between them, questions of interpretability, representation and behaviour. My master’s thesis pulled me into BabyLM, and the BabyLM cinematic universe is expanding. I co-organise BabyVLM. At first glance, a small language model looks like a weaker version of a large one. I think that misses the point. So rather than asking only “How good is the model?”, I became interested in a different question.

Research Interests

click to read more

My research sits at the intersection of Machine Learning, Cognitive Science, and Linguistics, with a particular focus on developing Small Language Models. +

My Research Agenda

How do models learn?

How does language shape learning?+

What representations emerge?+

Can we trust small AI?+

Addressing these questions are directly beneficial for the AI community and stakeholders using AI in their systems. My work helps stakeholders build mergeable, efficient small models and agentic pipelines that are cheaper to deploy than frontier systems. More generally, I aim to contribute to the AI community by understanding and building a foundation for multilingual, fairer AI that reaches beyond English into under-served languages and build more human-centered AI that is culturally-adaptive, safe, fair and trustworthy.