Dissertations, Theses, and Capstone Projects

Date of Degree

9-2025

Document Type

Master's Thesis

Degree Name

Master of Arts

Program

Linguistics

Advisor

Valerie Shafer

Subject Categories

Psycholinguistics and Neurolinguistics

Keywords

Event-related potentials (ERPs), Anterior Negativity (AN), Mismatch Negativity (MMN), Implicit learning, Statistical learning, Function Words (FWs)

Abstract

Research has indicated that children and adults can implicitly learn language patterns; however, questions still remain regarding the role and nature of implicit learning, particularly of sequential patterns. Implicit statistical learning is a type of unconscious learning of sequential patterns, which may be especially relevant to phrase structure learning. Prior studies have highlighted the differential processing of function words (FWs), such as determiners, relative to content and nonsense words and have also pointed to particular neural response patterns (specifically, Anterior Negativity, or AN) to early syntactic violations. To date, there is relatively little work directly investigating the role of FWs in implicit statistical learning of linguistic patterns.

The present study used electrophysiological, Event-Related Potential (ERP) measures to test adults’ ability to implicitly learn grammatical patterns. We hypothesized that: 1) adults would demonstrate implicit learning of the statistical linguistic patterns; 2) violations of the patterns would elicit an AN; 3) the inclusion of real FWs would elicit a more robust neural response, and 4) if the violations were processed at the phonetic/phonological, rather than the syntactic level, this would be seen as increased negativity at the FCz site, consistent with Mismatch Negativity (MMN) to phonetic and phonological prediction errors.

Sixteen participants were included: half who were native English-dominant (ED) and half who were Turkish-English bilinguals, but dominant in Turkish. All were between the ages of 18 and 35 years. In two experiments, ERPs were recorded from 17 electrode sites while participants watched a muted movie while listening to speech stimuli. The stimuli were designed to be analogous to elements of English Noun Phrases (NPs). Both experiments included nonsense word forms (e.g., /ua/, /ia/ /ea/). In Experiment 1 (E1), two of these nonsense forms were placed in the initial noun phrase position. In Experiment 2 (E2), two of the six stimuli were real English FWs, “the” and “an”. Experiments were separated into two blocks; during the first block (B1) of each experiment, the six stimuli were presented in a pattern mimicking the phrase structure Determiner (Adjective)(Adjective) Noun using 16 possible combinations (NP analog patterns). Each of the 16 patterns was presented 13 times in the block. During the second block (B2) 20% of these patterns were reversed, violating the learned pattern.

ERP amplitudes to the grammatical (learned) versus deviant (violation) stimuli over left (F7), right (F8) and midline (FCz) sites were compared in an early (50–200 ms) and late (300–500 ms) time windows for both experiments. Three-way repeated measures ANOVAs revealed a significant three-way interaction of grammaticality (learned versus violation) x time window (early, late) x site (F7, F8, FCc) for E2. No significant findings including condition were observed in E1. Post-hoc pairwise t-tests following up the three-way interaction in E2 revealed site differences only for the learned condition. In the early window, mean amplitudes were significantly more negative at the left compared to the right site. In the late window, greater negativity was observed at midline as compared to the right site.

These findings suggest that learning of the patterns was more successful in E2, where the topography of processing the stimuli was different for learned patterns. This suggests that the FWs facilitated implicit learning of the statistical patterns. Within the grammatical condition, there appear to be topographical differences which are not present for the ungrammatical condition. This may suggest differences in the neural sources engaged in processing grammatical versus ungrammatical patterns; however, further research is required to elucidate these differences.

The lack of evidence for learning of the patterns in E1 is currently insufficient to conclude that learning did not take place because the language experience of the participants was not considered, due to lack of power. Further data are being collected with a proposed number of 40 participants (half English and half Turkish-English), which will provide sufficient power to observe a medium effect size.

Download

Share

COinS