About the role

Moleculyst is building a reasoning checker that works directly in English. When an LLM draws conclusions from contracts, annexures or laws, our system checks every step for soundness and completeness against the source documents. We don’t formalize in Lean. Each step is an English sentence, read through a dictionary that can be corrected and checked against a small first-order logic core. We’re hiring a Computational Linguist to own the language side of this system.

What you’ll work on
Parse English with the English Resource Grammar (ERG) and build our MRS-to-first-order-logic mapping
Build DRS (Discourse Representation Theory) processing for multi-sentence text, covering anaphora, scope and reading pruning, kept domain-independent with no hand-written rules
Design our controlled-English fragment, in the spirit of Attempto Controlled English. Many phrasings can map to one logical step, but one phrasing never maps to two.
Build the WordNet-based sense dictionary,



with LLM-driven sense selection and recent entries, and people correcting wrong readings
Turn pragmatics into a working spec: the questions an LLM must ask to resolve ambiguity, vagueness and missing context in queries, source documents and its own output, and when to ask a person
Test the checker on real contracts and laws, using Z3 as a test oracle

What we’re looking for
Solid formal semantics: first-order logic, DRT (Kamp & Reyle), MRS, quantifier scope, anaphora
Grammar engineering experience, ideally with HPSG/ERG and DELPH-IN tools
Lexical semantics and word sense disambiguation, with working knowledge of WordNet
Grounding in pragmatics: presupposition, implicature, reference resolution
Solid Python, and comfort working with LLMs, including locally hosted models

📌 Computational Linguist New Delhi
🏢 Moleculyst
📍 New Delhi