Natural language processing for computing the influence of language on perception and behavior
Abstract/Contents
- Abstract
- Modern natural langauge processing (NLP) systems have achieved outstanding improvements over the last ten years, due in part to the rise of highly expressive multi-layer neural networks and massive datasets. Despite this progress, however, large gaps remain between the language capabilities of NLP systems and human beings. First, these systems operate as engines of correlation which specialize heavily to the data on which they are trained and evaluated. Second, human understanding of language is embedded in social context, with abstract and non-literal cues like subjectivity and identity that are difficult to integrate with traditional supervised and unsupervised machine learning frameworks. This dissertation seeks to make two steps forward with respect to these limitations of correlational machine learning and abstract social understanding. Both steps involve building NLP systems with the ability to reason about language in terms of how readers might respond to that language. First, I propose algorithms for discovering which parts of a text are causally implicated in behavioral responses among readers, and for estimating the causal effect of linguistic properties on behavior. In the second step, I propose generative algorithms for automatically manipulating the presence of abstract social concepts in text. In particular, I focus on the case of subjective bias, developing a text editing system which is the fusion of a discriminative bias identification module and generative text editing module. Overall, this dissertation argues that despite recent breakthroughs in NLP, the limits of language technology remain behind that of human beings. It attempts to close this gap by developing systems which can reason in new ways about how readers respond to text.
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 | Pryzant, Reid |
|---|---|
| Degree supervisor | Jurafsky, Dan, 1962- |
| Thesis advisor | Jurafsky, Dan, 1962- |
| Thesis advisor | Liang, Percy |
| Thesis advisor | Wager, Stefan |
| Degree committee member | Liang, Percy |
| Degree committee member | Wager, Stefan |
| Associated with | Stanford University, Computer Science Department |
Subjects
| Genre | Theses |
|---|---|
| Genre | Text |
Bibliographic information
| Statement of responsibility | Reid Pryzant. |
|---|---|
| Note | Submitted to the Computer Science Department. |
| Thesis | Thesis Ph.D. Stanford University 2021. |
| Location | https://purl.stanford.edu/vs342rg2015 |
Access conditions
- Copyright
- © 2021 by Reid Pryzant
- License
- This work is licensed under a Creative Commons Attribution 3.0 Unported license (CC BY).
Also listed in
Loading usage metrics...