Overview
The Computation research cluster uses machine learning and computer simulations to study how idiosyncratic patterns of language variation develop structure over time and how exposure leads to the acquisition of variable patterns. We also develop machine-learning tools that support research across the lab.
Active Workstreams
- Phontrast, a toolkit for comparing measures of sound-category distinctness, with two related papers in submission and review.
- Simulating phonetic drift: can computers learn to speak with a foreign accent?
- LingViz, a lab-built tool in early development, beginning with its Perception module.
- Vowelchemy, a lab-built project in early development.
- Improving an AI tutor through a VITAL minigrant project, now wrapping up.
- Modeling and evaluating how talkers and artificial agents adapt to conversation partners.
Methods and Toolchain
- Statistical learning models and agent-based simulations.
- Information-theoretic and statistical measures of phonological separation.
- Feature engineering from acoustic, behavioral, and conversational datasets.
- Model comparison, validation, and reproducible analysis workflows.
Recent Software Releases
- phontrast 2.4.0 - Contrast and separation metrics for phonological categories, also available on CRAN. Related development is available in the phonJSD repository.
- eeg-pipeline - EEG processing tools used in lab analyses.
- LingViz - An early-stage lab-built visualization tool, beginning with a Perception module.
- Vowelchemy - A lab-built project in early development.
Integration Across Lab Clusters
- Uses community and cognition datasets to test model generalizability.
- Provides quantitative tools for measuring category structure and adaptation.
- Feeds model and evaluation results back into experimental design.
Project Snapshot
- Cluster: Computation
- Current tools: Phontrast, LingViz, and Vowelchemy
- Current models: phonetic drift and conversational adaptation
- Inputs: acoustic, behavioral, conversational, and sociolinguistic data
- Outputs: simulations, evaluation frameworks, and reusable research tools
Current Activity
- Phontrast: Two related papers are moving through journal submission and review.
- LingViz: The initial Perception module is in early development.
- Vowelchemy: Early development is underway.
- AI tutor: The VITAL minigrant project is wrapping up.
- Adaptation modeling: Current work evaluates adjustment by human talkers and artificial agents.
Last Updated
September 2026