Irina Saparina

Member of Technical Staff, Cohere (CodeGen)

Irina Saparina

I work on agentic code generation at Cohere, where my focus is on making models better at building applications from nothing but a natural language description (“vibe coding”).

I finished my Ph.D. in Informatics at the University of Edinburgh, supervised by Prof. Mirella Lapata. My research focused on semantic parsing and code generation tasks that require advanced reasoning to handle real-world complexity: recognizing ambiguity, generating diverse interpretations, and adapting to language variation.

Publications

  1. 2026

    Reasoning about Intent for Ambiguous Requests

    Irina Saparina, Mirella Lapata

    COLM 2026

  2. 2026

    Long-Context Reasoning Through Proxy-Based Chain-of-Thought Tuning

    Miao Li, Irina Saparina, Alexander Gurung, Mirella Lapata

    ACL 2026 Oral

  3. 2025

    Who Gets Cited Most? Benchmarking Long-Context Language Models on Scientific Articles

    Miao Li, Alexander Gurung, Irina Saparina, Mirella Lapata

    Preprint

  4. 2025

    Asking a Language Model for Diverse Responses

    Sergey Troshin, Irina Saparina, Antske Fokkens, Vlad Niculae

    UncertaiNLP @ EMNLP 2025

  5. 2025

    Disambiguate First, Parse Later: Generating Interpretations for Ambiguity Resolution in Semantic Parsing

    Irina Saparina, Mirella Lapata

    Findings of ACL 2025

  6. 2024

    AMBROSIA: A Benchmark for Parsing Ambiguous Questions into Database Queries

    Irina Saparina, Mirella Lapata

    NeurIPS 2024, Datasets & Benchmarks Spotlight

  7. 2024

    Improving Generalization in Semantic Parsing by Increasing Natural Language Variation

    Irina Saparina, Mirella Lapata

    EACL 2024 Oral

  8. 2023

    Searching for Better Database Queries in the Outputs of Semantic Parsers

    Anton Osokin, Irina Saparina, Ramil Yarullin

    Findings of EACL 2023

  9. 2021

    SPARQLing Database Queries from Intermediate Question Decompositions

    Irina Saparina, Anton Osokin

    EMNLP 2021

  10. 2019

    Cost-Sensitive Training for Autoregressive Models

    Irina Saparina, Anton Osokin

    WiML @ NeurIPS 2019