Leveraging Natural Language Processing to Uncover Comprehension Processes in Constructed Responses (Poster 2)

• In Event: Leveraging Constructed Responses to Understand Text Comprehension

Abstract

1. Objectives. Recent advances in data science techniques, availability of big data sources, and computational power have provided a means to examine discourse in novel ways. Natural Language Processing (NLP) techniques in particular allow researchers to examine properties of discourse (e.g., spoken language, written texts, essays) along multiple dimensions, such as the complexity of the words that students use in their writing to the emotional themes present in a narrative. The purpose of this poster is to describe how our research team has leveraged theoretically-motivated NLP techniques to reveal insights into the ways in which individuals construct meaning during reading.

2. Perspectives. Our project relies on the assumption that successful text comprehension requires readers to establish a coherent representation of a text they are reading. This representation is constructed through readers’ activation of knowledge (either from their prior experiences or from something they have read earlier in the text) and the generation of connections between this knowledge and what they are reading. Our argument is that we can glean insights into these coherence-building processes by leveraging NLP techniques and applying them to the responses that students generate while reading.

3. Method. Across multiple projects, we have relied on theoretically-motivated NLP techniques to examine the properties of CRs at multiple levels. Much of our research has focused on examining the cohesion of the responses – in other words, we model the ways in which readers generate connections across the responses they generate during reading.

4. Data Sources. We examined CRs produced by students while reading. These CRs varied in the type of text that was being read (e.g., history, science), as well as the nature of the task (e.g., self-explanation, multiple text comprehension). We additionally examined how NLP metrics relate to individual differences amongst students to better understand the role of specific knowledge and skills in coherence-building processes.

5. Results. Our results indicate that NLP techniques can be used to tap into coherence-building processes of readers. Specifically, analyses of cohesion may serve as a proxy for the coherence of readers’ mental representations.

6. Significance. Insights from our work have both theoretical and applied implications. First, NLP analyses allow researchers to better understand the multiple-dimensions at play during comprehension. Thus, we can better understand how readers establish connections but also the situations when they may struggle to make connections readily. These techniques may also be used in the future to drive adaptivity in educational technologies focused on reading strategy training.

Authors

• Laura K. Allen, University of Minnesota
  Presenting Author

• Lauren Flynn, University of Minnesota
  Non-Presenting Author