All good readers are the same, but every low-skilled reader is different: an eye-tracking study using PISA data

An Approach of Query Audience's Attention in Virtual Speech

Virtual speeches are a very popular way for remote multi-user communication, but it has the disadvantage of the lack of eye contact. This paper proposes the evaluation of an online audience attention based on gaze tracking. Our research only uses webcams to capture the audience's head posture, gaze time, and other features, providing a low-cost method for attention monitoring with reference values across multiple domains. Meantime, we also propose a set of indexes which can be used to evaluate the audience's degree of attention, making up for the fact that the speaker cannot gauge the audience's concentration through eye contact during online speeches. We selected 96 students for a 20 min group simulation session and used Spearman's correlation coefficient to analyze the correlation between our evaluation indicators and concentration. The result showed that each evaluation index has a significant correlation with the degree of attention (p = 0.01), and all the students in the focused group met the thresholds set by each of our evaluation indicators, while the students in the non-focused group failed to reach the standard. During the simulation, eye movement data and EEG signals were measured synchronously for the second group of students. The EEG results of the students were consistent with the systematic evaluation. The performance of the measured EEG signals confirmed the accuracy of the systematic evaluation.

Predicting the next sentence (not word) in large language models: What model-brain alignment tells us about discourse comprehension

Current large language models (LLMs) rely on word prediction as their backbone pretraining task. Although word prediction is an important mechanism underlying language processing, human language comprehension occurs at multiple levels, involving the integration of words and sentences to achieve a full understanding of discourse. This study models language comprehension by using the next sentence prediction (NSP) task to investigate mechanisms of discourse-level comprehension. We show that NSP pretraining enhanced a model's alignment with brain data especially in the right hemisphere and in the multiple demand network, highlighting the contributions of nonclassical language regions to high-level language understanding. Our results also suggest that NSP can enable the model to better capture human comprehension performance and to better encode contextual information. Our study demonstrates that the inclusion of diverse learning objectives in a model leads to more human-like representations, and investigating the neurocognitive plausibility of pretraining tasks in LLMs can shed light on outstanding questions in language neuroscience.

What Went Wrong for Bad Solvers during Thematic Map Analysis? Lessons Learned from an Eye-Tracking Study

Thematic map analysis is a complex and challenging task that might result in map user failure for many reasons. In the study reported here, we wanted to search for differences between successful and unsuccessful map users, focusing—unlike many similar studies—on strategies applied by users who give incorrect answers. In the eye-tracking study, followed by a questionnaire survey, we collected data from 39 participants. The eye-tracking data were analyzed both qualitatively and quantitatively to compare participants’ strategies from various perspectives. Unlike the results of some other studies, it turned out that unsuccessful participants show some similarities that are consistent across most analyzed tasks. The main issues that characterize bad solvers relate to improper use of the thematic legend, the inability to focus on relevant map layout elements, as well as on adequate map content. Moreover, they differed in the general problem-solving approach used as they, for example, tended to choose fast, less cautious, strategies. Based on the collected results, we developed tips that could help prevent unsuccessful participants ending with an incorrect answer and therefore be beneficial in map use education.

Active photonic sensing for super-resolved reading performance in simulated prosthetic vision

In this work, we study the enhancement of simulated prosthetic reading performance through "active photonic sensing" in normally sighted subjects. Three sensing paradigms were implemented: active sensing, in which the subject actively scanned the presented words using the computer mouse, with an option to control text size; passive scanning produced by software-initiated horizontal movements of words; and no scanning. Our findings reveal a 30% increase in word recognition rate with active scanning as compared to no or passive scanning and up to 14-fold increase with zooming. These results highlight the importance of a patient interactive interface and shed light on techniques that can greatly enhance prosthetic vision quality.