Modeling the mechanisms of verb bias learning

During language production, speakers acquire statistical information about linguistic units. One example of this occurs in verb bias learning, in which speakers update statistics about how likely a particular verb is to occur in a specific syntactic structure. This dissertation explores the mechanisms that support verb bias learning using a combination of empirical work and cognitive models of empirical results. Chapter 2 implements two cognitive models that reproduce previous empirical results. The models in this section behave similarly to humans when asked to learn and unlearn rules, and when they learn expected and unexpected verb-structure combinations. Chapters 3 and 4 detail two experimental investigations into how learning a new bias for one verb can transfer to another, semantically-related verb. Chapter 3 details the initial investigation, and shows that verb bias learning can transfer to similar dative verbs, but that transitive verbs show neither training nor transfer. Chapter 4 explains two follow-up norming studies to select new verb pairs, and then a behavioral replication of the study in Chapter 3. This new study shows no transfer for the new set of dative verbs, and replicates the finding of no training and no transfer for the transitive verbs. Chapter 5 uses a cognitive model to generate the results of the two dative transfer studies, and shows that the finding of transfer in Chapter 3 but not in Chapter 4 can be explained by a difference in how unexpected a structure is for each verb. Chapter 6 models the transitive results found in Chapters 3 and 4, and shows that a predictive object-first cue effectively blocks verb bias learning, consistent with the possibility that transitive learning is blocked by these or related cues. Finally, Chapter 7 addresses findings from Lin (2020) and Thothathiri and Braiuca (2021), which explore how humans switch from using verbs to predict sentence structure to using other cues. However, the models struggle to replicate these results, which suggests a need for a more complex model. This dissertation shows that many of the findings in the verb bias learning literature can be accounted for using the proposed cognitive model, and contributes new findings that also fit within this framework. Additionally, it lays out specific reasons why the model fails to account for cue-switching results, and how improvements to the model could guide further empirical research.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Psychology

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2022

Author dc:creator

• Kelley, Amanda

Contributors dc:contributor

• Dell, Gary S.
• Fisher, Cynthia L.
• Federmeier, Kara D.
• Willits, Jon
• Montag, Jessica L.

• cognitive modeling
• verb bias
• implicit learning
• distributional learning

Statement dc:rights

• Copyright 2022 Amanda Kelley

Language dc:language

en, eng

Handle dc:identifier

https://hdl.handle.net/2142/116091

OAI record oai:identifier

https://www.ideals.illinois.edu/oai-pmh?verb=GetRecord&metadataPrefix=etdms&identifier=oai%3Awww.ideals.illinois.edu%3A2142%2F116091