PeopleofSuperintelligence

Researcher

Ryan McDonald

Also known as Ryan T McDonald

Ryan McDonald is an AI researcher whose work connects dependency parsing, multilingual language tools, and AI agents. He completed a PhD at the University of Pennsylvania in 2006, spent 15 years as a research scientist in Google's Language Team, and joined ASAPP as chief scientist in 2021. At ASAPP, he described multimodal models that use agent actions to predict what an agent should say or do; later, he coauthored large language model systems for web tasks and simulated tool use. His 2013 multilingual parsing paper received the ACL Test-of-Time Paper Award in 2023. As of October 2026, he is an AI consultant and investor.

Now at
ASAPP
Works
116
Citations
16,071
h-index
48

Publication numbers from OpenAlex, as of 15 October 2026.

Timeline

  1. 2006

    A PhD in dependency parsing

    He completes a University of Pennsylvania PhD in 2006 with a dissertation on discriminative training and spanning-tree algorithms for dependency parsing.

  2. 2013

    Universal dependency annotation

    He coauthors Universal Dependency Annotation for Multilingual Parsing in 2013, contributing to a framework for consistent grammatical annotation across human languages.

  3. 2021

    Joining ASAPP as chief scientist

    After 15 years as a research scientist in Google's Language Team, he joins ASAPP as chief scientist in 2021.

  4. March 2023

    Agent models for customer service

    In March 2023, he describes ASAPP multimodal models that use streams of agent actions to predict what an agent should say or do next.

  5. 2023

    An ACL Test-of-Time award

    At ACL 2023, he and his coauthors win the Test-of-Time Paper Award for Universal Dependency Annotation for Multilingual Parsing, their 2013 paper.

  6. 2024

    Stacked policies for web tasks

    In 2024, he coauthors SteP, a system that dynamically composes large language model policies for web tasks and tracks policy control as a stack of calls.

  7. July 2025

    Self-training dialogue agents

    In July 2025, he coauthors JOSH, a self-alignment method that trains dialogue agents from their own outputs in sparse-reward simulations, and ToolWOZ, a tool-calling simulation environment.

  8. As of October 2026

    AI consulting and investment

    As of October 2026, he is an AI consultant and investor who consults on autonomous-agent and scientific-simulation work.

Sources and credits

Photo: We have no published photo of Ryan McDonald yet. Send us a free-licensed photo.

  1. ryan mcdonald ryanmcd.github.io. Checked 15 October 2026.
  2. Universal Dependency Annotation for Multilingual Parsing - ACL Anthology aclanthology.org. Checked 15 October 2026.
  3. Universal Dependencies universaldependencies.org. Checked 15 October 2026.
  4. Why I joined ASAPP: Taking AI to new levels in enterprise solutions - ASAPP asapp.com. Checked 15 October 2026.
  5. Athens Natural Language Processing Summer School 2026 athnlp.github.io. Checked 15 October 2026.
  6. Generative AI for Agent Augmentation: Agents Models not Language Models - ASAPP asapp.com. Checked 15 October 2026.
  7. Announcement of the 2023 ACL Test-of-Time Paper Award aclweb.org. Checked 15 October 2026.
  8. SteP: Stacked LLM Policies for Web Actions asappresearch.github.io. Checked 15 October 2026.
  9. Sparse Rewards Can Self-Train Dialogue Agents - ACL Anthology aclanthology.org. Checked 15 October 2026.