Yoav Meiri, Omer Shubi, Cfir Avraham Hadar, Ariel Kreisberg Nitzav, Yevgeni Berzak. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Readers can have different goals with respect to the text that they are reading.Can these goals be decoded from their eye movements over the text?In this work, we examine for the first time whether it is possible to distinguish between two types of common reading goals: information seeking and ordinary reading for comprehension.Using large-scale eye tracking data, we address this task with a wide range of models that cover different architectural and data representation strategies, and further introduce a new model ensemble.We find that transformer-based models with scanpath representations coupled with language modeling solve it most successfully, and that accurate predictions can be made in real time, shortly after the participant started reading the text.We further introduce a new method for model performance analysis based on mixed effect modeling.Combining this method with rich textual annotations reveals key properties of textual items and participants that contribute to the difficulty of the task, and improves our understanding of the variability in eye movement patterns across the two reading regimes. 1

We present EyeBench, the first benchmark designed to evaluate machine learning models that decode cognitive and linguistic information from eye movements during reading. EyeBench offers an accessible entry point to the challenging and underexplored domain of modeling eye tracking data paired with text, aiming to foster innovation at the intersection of multimodal AI and cognitive science. The benchmark provides a standardized evaluation framework for predictive models, covering a diverse set of datasets and tasks, ranging from assessment of reading comprehension to detection of developmental dyslexia. Progress on the EyeBench challenge will pave the way for both practical real-world applications, such as adaptive user interfaces and personalized education, and scientific advances in understanding human language processing. The benchmark is released as an open-source software package which includes data downloading and harmonization scripts, baselines and state-of-the-art models, as well as evaluation code, publicly available at https://github.com/EyeBench/eyebench.

Spaceborne observations serve as input to models and retrieval algorithms of various atmospheric properties. Specifically, we seek 3D volumetric scattering tomography of clouds. Towards this, cloud-fields are to be imaged simultaneously from multiple directions. The CloudCT project aims to demonstrate this by a formation of ten nanosatellites. Based on this data, scattering tomography will seek the 3D volumetric distribution of cloud microphysical properties. We present constraints and fundamental considerations for spaceborne formation-based cloud tomography. We quantitatively compare visible polarized imagers, visible unpolarized imagers, and short-wave infra-red unpolarized imagers. Each possibility is considered using a large eddy simulation of clouds. We study tomographic quality in the presence of sensor and photon noise, calibration errors and stray light. We find that a polarized imager of a red waveband is preferable.

Can human reading comprehension be assessed from eye movements in reading?In this work, we address this longstanding question using large-scale eyetracking data.We focus on a cardinal and largely unaddressed variant of this question: predicting reading comprehension of a single participant for a single question from their eye movements over a single paragraph.We tackle this task using a battery of recent models from the literature, and three new multimodal language models.We evaluate the models in two different reading regimes: ordinary reading and information seeking, and examine their generalization to new textual items, new participants, and the combination of both.The evaluations suggest that the task is highly challenging, and highlight the importance of benchmarking against a strong text-only baseline.While in some cases eye movements provide improvements over such a baseline, they tend to be small.This could be due to limitations of current modelling approaches, limitations of the data, or because eye movement behavior does not sufficiently pertain to finegrained aspects of reading comprehension processes.Our study provides an infrastructure for making further progress on this question.1

Calibration of an atmospheric lidar is often required due to variations in the electro-optical system. Rayleigh fitting commonly performed may fail under various conditions. Temporal and spatial variations both affect lidar signals. We hence opt for spatiotemporal analysis. We present a novel deep-learning (DL) lidar calibration model based on convolutional neural networks (CNN). We demonstrate our method on simulated data that mimics natural ground-based pulsed time-of-flight lidar signals. Such an approach can better address measurements with a poor signal-to-noise ratio (SNR) and provide a more frequent calibration.

We introduce a comprehensive method for space-borne 3D volumetric scattering-tomography of cloud micro-physics, developed for the CloudCT mission. The retrieved micro-physical properties are the liquid-water-content and effective droplet radius within a cloud. We include a model for a perspective polarization imager, and an assumption of 3D variation of there. Elements of our work include computed tomography initialization by a parametric horizontally-uniform micro-physical model. This results in smaller errors than the prior art. The mean absolute errors of the retrieved liquid-water-content and effective-radius are reduced from 62% and 28% to 40% and 9%, respectively. The parameters of this initialization are determined by a grid search of a cost function. Furthermore, we add viewpoints in the cloudbow region, to better sample the polarized scattering phase function. The suggested advances are evaluated by retrieval of a set of clouds generated by large-eddy simulations.

Methods based on statistical learning have become prevalent in various signal processing disciplines and have recently gained traction in atmospheric lidar studies. Nonetheless, such methods often require large quantities of annotated or resolved data. Such data is rare and requires effort, especially when exploring evolving phenomena. Existing simulators and databases primarily focus on atmospheric vertical profiles. We propose the Atmospheric Lidar Data Augmentation (ALiDAn) framework to fill this gap. ALiDAn serves as an end-to-end generation and augmentation framework of spatiotemporal and multi-wavelength resolved lidar simulated data. ALiDAn employs a hybrid approach of physical models, data statistics, and sampling processes. Additionally, it takes into account geographical and seasonal characteristics of aerosols, meteorological conditions, along with short- and long-term phenomena that affect lidar measurements. This approach can provide diversified data and robust benchmarks to assist in developing and validating new lidar processing algorithms. We demonstrate simulations compatible with a pulsed time-of-flight lidar. Our approach leverages a broader use of existing databases and can inspire similar data augmentation to other types of lidars and active sensors.