Using Latent-Semantic Analysis and Network Analysis for Monitoring Conceptual Development
Fridolin Wild, Debra Haley, Katja Bülow · LDV-Forum/Journal for language technology and computational linguistics · 2011
This paper describes and evaluates CONSPECT (from concept inspection), an application that analyses states in a learner's conceptual development.It was designed to help online learners and their tutors monitor conceptual development and also to help reduce the workload of tutors monitoring a learner's conceptual development. CONSPECT combines two technologies -Latent Semantic Analysis (LSA) and Network Analysis (NA) into a technique called Meaningful Interaction Analysis (MIA).LSA analyses the meaning in the textual digital traces left behind by learners in their learning journey; NA provides the analytic instrument to investigate (visually) the semantic structures identified by LSA.This paper describes the validation activities undertaken to show how well LSA matches first year medical students in 1) grouping similar concepts and 2) annotating text. Theoretical JustificationThis section mentions two related Cognitive Linguistic theories that support the approach taken in CONSPECT: Fauconnier's Mental Spaces Theory and Conceptual Blending Theory (Evans and Green 2006).These theories hold that the meaning of a sentence cannot be determined without considering the context.Meaning construction results from the development of mental spaces, also known as conceptual structures (Saeed 2009), and the mapping between these spaces.Mental spaces and their relationships are what LSA tries to quantify.LSA uses words in their contexts to calculate associative closeness among terms, among documents, and among terms and documents.This use of context is consistent with Fauconnier's claim that context is crucial to construct meaning.Various researchers use network analysis to analyse conceptual structures: Schvaneveldt et al (1989), Goldsmith et al (1991) and Clariana & Wallace (2007) are among the researchers who use a particular class of networks called Pathfinder, which are derived from proximity data (Schvaneveldt, Durso et al. 1989).These researchers assume that "concepts and their relationships can be represented by a structure consisting of nodes (concepts) and links (relations)."The strength of the relationships can be measured by the link weights.The networks of novices and experts are compared to gauge the learning of the novices.JLCL 2011 -Band 26 (1) -9-21 10 JLCL