How few is too few? Determining the minimum acceptable number of LSA dimensions to visualise text cohesion with Lex

How few is too few? Determining the minimum acceptable number of LSA dimensions to visualise text cohesion with Lex Abstract Building comprehensive language models us- ing latent semantic analysis (LSA) requires substantial processing power. At the ideal pa- rameters suggested in the literature (for an overview, see Bradford, 2008) it can take up to several hours, or even days, to complete. For linguistic researchers, this extensive pro- cessing time is inconvenient but tolerated— but when LSA is deployed in commercial software targeted at non-specialists, these processing times become untenable. One way to reduce processing time is to reduce the number of dimensions used to build the mod- el. While the existing research has found that the model’s reliability starts to degrade as dimensions are reduced, the point at which reliability becomes unacceptably poor varies greatly depending on the application. There- fore, in this paper, we set out to determine the lowest number of LSA dimensions that can still produce an acceptably reliable language model for our particular application: Lex, a visual cohesion analysis tool. We found that, across all three texts that we analysed, the cohesion-relevant visual motifs created by Lex start to become apparent and consistent at 50 retained dimensions. 1 Introduction Latent Semantic Analysis (LSA) is a well-established method for describing the se- mantic content in textual data as a set of vectors in a high dimensional semantic space (Wade- Stein & Kintsch, 2004).!It is used for a range of applications across a range of fields, including linguistics, cognitive science, education, infor- mation science and text analysis (Evangelopou- los, Zhang, & Prybutok, 2012), and it has been verified as an effective method in the majority of these fields both practically and theoretically (Evangelopoulos et al., 2012; Wade-Stein & Kintsch, 2004). 2 Lex: an overview The application of LSA we are focusing on, pioneered by Foltz, Kintsch, & Landauer (1998), is its use in predicting the coherence of a piece of text by identifying and measuring its lexical cohesive ties. Building on this work, we have designed an LSA-based tool, which we have called Lex, to allow writers and editors to visually analyse the cohesion—and, by exten- sion, coherence—of their own text. Users upload their text, and the tool derives the relatedness of meaning that occurs in each sentence throughout the text using a LSA language model by investi- gating word usage patterns in a large text corpus Caroline McKinnon School of Communication and Arts The University of Queensland Brisbane, Australia c.mckinnon@uq.edu.au Ibtehal Baazeem School of Information Technology and Electrical Engineering The University of Queensland Brisbane, Australia ibtehal.baazeem@uq.net.au National Center for Computer Technol- ogy and Applied Mathematics King Abdulaziz City for Science and Technology Riyadh, Saudi Arabia ibaazeem@kacst.edu.sa Daniel Angus School of Communication and Arts The University of Queensland Brisbane, Australia d.angus@uq.edu.au Caroline Mckinnon, Ibtehal Baazeem and Daniel Angus. 2015. How few is too few? Determining the minimum acceptable number of LSA dimensions to visualise text cohesion with Lex . In Proceedings of Australasian Language Technology Association Workshop, pages …

Content truncated to 500 words