Update: Although there was a major earthquake last night in northern Japan, we will hold the seminars as scheduled. Please stay safe, and we are looking forward to your participation. [March 17, 6:45 AM]

Talk #1. March 17 (Thu) 12:00-13:00 (JST)

The Contribution of Neuroscience to Validity in the Assessment of Second Language Listening

Vahid Aryadoust (Nanyang Technological University, Singapore)

[Registration Form]

Abstract: Listening is key to language learning, communication, and interaction. In second language (L2) listening, listeners encounter unique challenges that they do not usually face in their first language. L2 researchers use a variety of assessment methods to examine and quantify these challenges. The methods consist of the application of assessment techniques that are intended to elicit the cognitive processes of the listeners. However, this method is only able to provide raw scores, which are the very end product of the listening process. Using quantitative methods (e.g., item response theory, factor analysis, structural equation modeling, classification and regression trees etc.), researchers attempt to mine the test scores in the hope of reverse-engineering the cognitive and response processes of the test takers to infer whether the assessment tool measures what it is supposed to measure. In this talk, I review two cutting-edge technologies that allow the measurement and quantification of the neurocognitive processes of L2 listening test takers without relying on test scores: eye tracking and functional near-infrared spectroscopy (fNIRS). I review several studies that we have conducted at our lab to achieve different research goals such as measuring cognitive load in listening and equating listening tests. I will also discuss the challenges of this fledgling stream of research and implications for broader educational and language measurement fields.

Talk #2. March 18 (Fri) 17:00-18:00 (JST)

Analysis of Non-compensatory Choice Structures using the Bayesian Viewpoints Shift Model

Tomoya Okubo (OECD, France)

[Registration Form]

※ This talk will be given in Japanese (本講演は日本語で行われます)

Abstract: Describing human choice behaviour is the focus of psychology and related fields such as behavioural economics and marketing science. A straightforward approach to understanding the relationship between stimuli (products) and respondents (consumers) in terms of choice behaviour involves analyzing choice data using statistical models such as multidimensional scaling (MDS) models. MDS models have been used widely in various fields, especially in marketing science and the unfolding (ideal-point) model represent utilities of respondents as a function of the parameters of stimuli and respondents based on choice (preference) data. Several MDS models have been proposed in order to analyze the choice behaviour; however, they do not necessarily provide us sufficient information for real data analysis. A possible reason for this is that respondents' choices are not always made in accordance with the models assumed by researchers. However, classical MDS models, such as the vector and unfolding models, assume a transitive preference order in respondents' choices. In these models, irrational choice behaviours are attributed to measurement error. A critical issue in describing choice behaviour using MDS models is their inability to consider inconsistency in choices. One approach for allowing intransitive preference choices is using the viewpoints shift model. This MDS model makes it possible to deal with complex preference situations. It allows the presence of multiple preference orders for a respondent, whereas classical MDS models allow only one preference order. In this model, respondents have multiple viewpoints and select one based on the pairs to be compared. There has been modest, but growing, literature on Bayesian methods for MDS modelling. This study contributes to this line of research by proposing a novel Bayesian approach for the viewpoints shift model. The Bayesian framework can also provide the benefit of dealing with missing data. In reality, it is difficult to create perfect experimental situations when collecting data using Thurstone's paired comparison judgments, even though this approach provides detailed information on both stimuli and respondents. This presentation is structured as follows. First, we introduce utility structures of the viewpoints shift model. Then, we propose a Bayesian solution of the viewpoints shift models (vector and unfolding) and follow post-processing procedures to obtain the appropriate posterior distribution of the parameters. Then, we present our data analysis of lexicographic choice data using the Bayesian viewpoints shift models. Finally, we overview our models and summarize this presentation.

Talk #3. March 22 (Tue) 12:00-13:00 (JST)

Cognitive Models and Bayesian Methods for the Mnemonic Similarity Task

Michael D Lee (University of California, Irvine, USA)

[Registration Form]

Co-authors: Manuel Villarreal, Craig E.L. Stark (University of California Irvine)

Abstract: The Mnemonic Similarity Task (Stark, Kirwan, & Stark, 2019) is a modified recognition memory task designed to place strong demand on pattern separation. The sensitivity and reliability of the MST make it an extremely valuable tool in clinical settings. There are several versions of the MST task, using different approaches to presenting stimuli, and allowing different sets of responses from participants. We develop cognitive models, based on the multinomial processing tree framework, for a number of MST tasks. The models are implemented as generative probabilistic models and applied to behavioral data using Bayesian graphical modeling methods. We demonstrate how the combinations of cognitive modeling and Bayesian methods allows for flexible and powerful inferences about performance on the MST. These demonstrations include identifying individual differences in decision strategies, hierarchical extensions of the basic model that measure differences in memory in a fine-grained way, and assessing the agreement between the memory measures produced by different versions of the task.

Sponsor

JSPS KAKENHI Grant Number JP18HP2006, The Behaviormetric Society.