Structured semantic knowledge mediates variability in episodic memory
Abstract/Contents
- Abstract
- This thesis builds on previous empirical evidence and theoretical work to further investigate the variability in recognition memory behavior. Specifically, the global match-ing models provide mechanistic explanations of how similarities of memory representa-tions contribute to the memory signals that guide recognition memory decisions. In addi-tion, it is commonly believed that long-term semantic knowledge has an impact on the recognition memory related to that knowledge, and the relationship between semantic knowledge and recognition memory is mediated, in part, by interactions between the hip-pocampus (important for aspects of recognition memory) and distributed cortical regions (important for semantic knowledge). While this framework has garnered empirical and theoretical support, the detailed mechanisms underlying the influence of semantic knowledge on recognition memory remains unclear. My research explores how similarity among semantic knowledge influences the gradient of the memory signals that give rise to false and accurate recognition judgements. I ask 1) whether it is possible to systematical-ly modulate false and accurate recognition decisions with objective similarity measure-ments of semantic knowledge, and 2) what is the computation that translates similarity of semantic knowledge to recognition memory signals. I answer these questions with two experiments: Experiment 1 investigated how model-based semantic similarity measure-ments modulate false and accurate memory behavior by leveraging tools from computa-tional models (i.e., Natural Language Processing (NLP)) to systematically quantify se-mantic similarity of the stimuli. Building on the findings from Experiment 1, Experiment 2 used the model-based semantic similarity measurements to further examine the compu-tation that transforms semantic similarity to recognition memory signals. In Experiment 1, I used NLP-derived measurements of semantic similarity to generate word lists with varying degrees of semantic similarities as experimental stimuli. These word lists were used in a memory task to allow for precise control of the semantic similarities. The primary findings from Experiment 1 demonstrated that recognition memory behavior is modulated by semantic similarities. More specifically, NLP-derived semantic similarity predicted both false recognition memory to lures and accurate recog-nition memory to old words: false and accurate recognition judgments increased as a function of NLP-derived semantic similarity. The results from Experiment 1 highlight the fundamental role global similarity computations play in generating recognition memory signals. In addition, our demonstration of the relationship between semantic similarity structure and recognition memory is largely compatible with the findings reported in the literature. Building on the findings in Experiment 1, one outstanding question is how se-mantic similarities are transformed to memory strength. To answer this question, in Ex-periment 2, we expanded our present findings to investigate how the dynamic of within-list distribution of semantic similarity affected recognition memory behavior. In Experiment 2, building on the results from Experiment 1, I explored the nature of the transformation between semantic similarities and recognition memory signals. In-spired by computational models of recognition memory, there could be a linear or a non-linear transformation between semantic similarities and recognition memory strengths. To differentiate between these two possibilities, I examined how skewed and uniform distri-butions of semantic similarities affect recognition memory. The main findings from Ex-periment 2 showed that there was no significant difference between skewed and uniform distribution of semantic similarities on false and accurate recognition memory. The null results indicate that encountering a few semantically highly similar items (i.e., the skewed condition) is not enough to significantly increase false memory for similar lures beyond a linear similarity computation (i.e., the uniform condition). One possible explanation for our null findings is that the memory representations used in our study are defined only by the semantic features of the stimuli; however, the proposed memory representations in global matching models and the neural representations in the fMRI studies both consist of various features beyond the semantic features. Another possible explanation for the null findings is that the process of extracting word embeddings from word co-occurrences already encompasses nonlinear transformations of the semantic space. Building on our findings, a potential research direction is to further explore the computa-tion between NLP-derived semantic similarity and memory signals with more complex models.
Description
| Type of resource | text |
|---|---|
| Form | electronic resource; remote; computer; online resource |
| Extent | 1 online resource. |
| Place | California |
| Place | [Stanford, California] |
| Publisher | [Stanford University] |
| Copyright date | 2021; ©2021 |
| Publication date | 2021; 2021 |
| Issuance | monographic |
| Language | English |
Creators/Contributors
| Author | Wang, Shao-Fang |
|---|---|
| Degree supervisor | Wagner, Anthony David |
| Thesis advisor | Wagner, Anthony David |
| Thesis advisor | Grill-Spector, Kalanit |
| Thesis advisor | McClelland, James L |
| Degree committee member | Grill-Spector, Kalanit |
| Degree committee member | McClelland, James L |
| Associated with | Stanford University, Department of Psychology |
Subjects
| Genre | Theses |
|---|---|
| Genre | Text |
Bibliographic information
| Statement of responsibility | Shao-Fang Wang. |
|---|---|
| Note | Submitted to the Department of Psychology. |
| Thesis | Thesis Ph.D. Stanford University 2021. |
| Location | https://purl.stanford.edu/yn404hh9734 |
Access conditions
- Copyright
- © 2021 by Shao-Fang Wang
- License
- This work is licensed under a Creative Commons Attribution Non Commercial 3.0 Unported license (CC BY-NC).
Also listed in
Loading usage metrics...