Name: Haley Shea Barfield

School of Psychology – Master’s Thesis Proposal Meeting

Date: Tuesday, May 26, 2026

Time: 3:30 – 5:00pm

Location: Virtual

Teams Link: Available upon request (reach out to hbarfield6@gatech.edu if interested in attending) 

Thesis Chair/Advisor:

Bruce Walker, Ph.D. (Georgia Tech)

Thesis Committee Members:

Mengyao Li, Ph.D. (Georgia Tech)

Richard Catrambone, Ph.D. (Georgia Tech)

Title: Investigating neural correlates of thematic ambiguity resolution and ditransitive sentence processing with electroencephalography (EEG) and the event-related potential (ERP) technique

Abstract: In psycholinguistics, "thematic relation" refers to semantic information about "who is doing what (to or for whom)" in a given sentence. Within Friederici’s neurocognitive model of language processing, the subprocess of thematic relations has been shown in both Phase 2 and Phase 3, meaning the precise timing of this aspect of sentence processing is not entirely clear. Part 1 of this project (Master’s thesis) aims to contribute evidence toward constraining the relative timing of subprocesses within Friederici’s model by examining neural activity associated with thematic role assignment, thematic reanalysis, syntactic-semantic conflict, and resolution of ambiguity.

To do so, this study focuses on comprehension of stimulus sentences with three ditransitive conditions (independent variable; attributive, non-attributive/double-object dative, and ambiguous), all following the same part-of-speech word order. Stimulus sentences are displayed using rapid serial visual presentation (RSVP) during a novel classification task, following a within-subjects (repeated measures) experimental design. To achieve the objectives of Part 1 and establish the model training and testing datasets for Part 2, electroencephalographic (EEG) data (a neurophysiological dependent variable) will be collected from participants during the experimental sentence classification task and analyzed using the event-related potential (ERP) technique, testing for N400, P600, or combined N400-P600 component effects. Behavioral data including response times, classification judgments, and confidence responses (dependent variables collected via E-Prime) will be analyzed using linear mixed models (LMMs) and generalized linear mixed models (GLMMs) for classification accuracy (congruence) of non-ambiguous trials. Hypotheses for the ambiguous condition include longer response times and greater amplitude for the semantic-associated N400 and/or syntactic-associated P600 components, in comparison to non-ambiguous conditions. Longer response times may reflect increased cognitive workload during thematic reanalysis of ambiguous sentences. Approximate timing of significantly greater ERP component amplitude (i.e. in the 400ms or 600ms post-stimulus windows) may help to narrow down the positioning of thematic role assignment to either Phase 2 or Phase 3 of Friederici’s model.

Looking ahead, Part 2 of this project (dissertation) will explore the intersections of neurolinguistics and emerging technologies (e.g. artificial intelligence (AI) and brain-computer interface (BCI)) by training a machine learning (ML) model to classify a subset of Part 1’s collected neural data by sentence type with respect to human interpretations of structure (reserving further ML details for the dissertation proposal). Regarding future directions and broader impacts at the career scope, to provide context of long-term goals: This work (including a series of follow-up studies covering other basic sentence structures in English, anticipated to begin in the post-doctoral phase) provides an initial foundation before neural data can become useful for accurately disambiguating human intentions during language production (e.g. improving semantic alignment and accuracy of response when ambiguous prompts are given to AI agents) and could influence development of brain-based benchmarks for metalinguistic evaluation of large language models (LLMs). As neurotechnologies continue to emerge, human sentence processing research using EEG/ERP methods could potentially help establish groundwork for future improvement of voiceless, hands-free human-computer interaction by drawing on findings from neurolinguistics to better account for the unresolved problem of ambiguous language pervasive in existing subvocalization-based BCI paradigms. Thus, the eventual development of a silent language BCI paradigm that is robust to the problem of linguistic ambiguity is a main motivator for this research and continuation of its trajectory beyond the graduate level.