Eye tracking offers unique insights into cognitive processes, making it a promising tool for evaluating machine translation (MT). This study explores the feasibility of using an iPhone 12 camera-based eye tracker with a 14-inch laptop display for conducting translation evaluation in personal workspaces, offering a more accessible and cost-effective alternative to traditional setups. Participants evaluated source sentences, selected translations, and identified problematic words while their gaze metrics were recorded and analyzed. Our findings reveal statistically significant correlations between gaze patterns and preferred translations, as well as increased visual attention to problematic words. These results demonstrate that home-based eye tracking systems are technically sufficient for capturing gaze behavior accurately enough for MT evaluation purposes. A potential practical application is to speed up translation proof-reading using eye tracking technique to automatically mark portions of text that should be attended to and improved based on the gaze pattern during a quick reading.

Eye-tracking corpora have become valuable resources for understanding human reading behavior and developing cognitively-informed NLP models. However, existing resources predominantly focus on left-to-right Latin script languages, leaving a significant gap for morphologically rich, right-to-left languages like Arabic. This paper presents a cross-linguistic analysis of eye movement patterns using the AraEyebility corpus, the first Arabic eye-tracking corpus comprising 57,617 words read by 15 native speakers. We systematically compare gaze metrics across Arabic and established English corpora. Our analysis reveals distinct patterns in fixation duration, saccade length, and regression frequency that reflect Arabic’s unique orthographic properties: cursive script, diacritization, bidirectional reading (text right-to-left, numbers left-to-right), and morphological complexity. The findings demonstrate that Arabic readers exhibit longer mean fixation durations and more frequent regressions compared to English readers, suggesting higher cognitive processing demands. We discuss implications for developing cognitively-aligned NLP models and provide recommendations for future multilingual eye-tracking research. The AraEyebility corpus is publicly available to support Arabic NLP research.

Automatic text simplification has mostly relied on human judgments when it comes to what is considered easy or difficult to read. Eye movements while reading can offer a more direct and objective signal of processing effort and reading ease. In this paper, we explore gaze-guided text generation (GGTG), an approach to control reading ease in generated texts, and assess its use for sentence simplification. GGTG employs a gaze model that is trained to predict eye-tracking measures such as reading times or regression rates, which are then used to rerank next-token probabilities generated by a language model. We evaluated the approach on an English sentence simplification benchmark and found gains in automatic evaluation metrics, although the simplification operations are mostly limited to the lexical level. Its modular nature also allows GGTG to be combined with other simplification techniques such as prompting or fine-tuning.

Understanding how text readability affects reading behaviour is crucial for improving accessibility and health communication. We analyse sentence-level eye-tracking data from the French Eye-TrAcking (FETA) corpus, which includes original and manually simplified texts from three domains: general, medical, and clinical. Using clustering of fixation-based features, we identify recurrent processing patterns and examine how these patterns change under text simplification. Cluster quality is evaluated using silhouette scores and participant-level bootstrap stability. Simplification does not uniformly reduce reading effort but reorganises processing in domain-dependent ways. Medical texts show strong diversification, general texts moderate diversification, and clinical texts show a reduction in the number of distinct reading profiles. Hence, rather than uniformly facilitating reading, simplification redistributes effort across sentences, underscoring the need for domain-sensitive readability approaches.

This study uses regression analysis of Brazilian Portuguese eye-tracking data to examine variability in reading times across grammatical categories. Mixed-effects models reveal distinct patterns: numerals elicit high individual variability in early-stage reading, while function words (e.g., adpositions, determiners) drive differences in late-stage integration. In contrast, nouns show stable effects. These findings demonstrate that individual differences in reading are systematically linked to specific parts of speech, with numerals and function words as key loci of variability.

Consistent alignment of multi-modal experimental data—such as verbal utterances in elicitation tasks, (static) visual stimuli, and gaze data—presents a challenge in linguistic research. These elicitations often encode information about the visual perception strategies or cognitive processing of the scene. Thus, it is helpful to transform them into a structured, visually grounded format which captures the visual nature of the data, ideally able to be aligned with the corresponding gaze data. To achieve this, the present paper conceptually proposes the annotation framework for verbal elicitation paths as a data type and presents the first release of the associated newly developed CoordiMap annotation tool. The tool enables structured mapping of verbal elicitation data from experimental studies onto the corresponding visual stimuli. Independent of specific paradigms, the tool supports the annotation of verbal utterances in a linearized form based on coordinates directly marked on the image of the stimulus. The format is conceptually inspired by eye-tracking data formats, in which gaze behavior is represented as temporally linearized paths overlaid on the stimulus. The paper motivates the development of the tool and its annotation methodology by theoretical and experimental considerations regarding the relationship between visual perception and language production. As this a work in progress, the functionality of the annotation tool is demonstrated through an exemplary use case.

Understanding human language processing via eye-tracking (ET) is precise but limited by scalability. Mouse-Tracking (MoTR) offers a cost-effective alternative, yet its viability for long-form reading in languages like Romanian remains underexplored. The primary challenge lies in the motor-induced noise and biomechanical discrepancies between hand and eye movements. Here we show that combining targeted technical enhancements with a Hertz-based velocity transformation allows MoTR to serve as a robust proxy for ET. We evaluate this by training a BERT-enhanced Fusion Model that integrates semantic context to bridge the mechanical gap, achieving an internal consistency of ρ ≈ 0.58 and a cross-modal correlation of ρ ≈ 0.22 in the velocity domain. These results indicate that when properly normalized, manual tracking captures similar cognitive constraints as gaze, with predictive accuracy approaching the empirical bounds of human behavioral variance.

This paper described the use of eye-contact and facial expression tracking as part of a comprehensive approach to assertiveness training in a VR-based anti-bullying simulation environment. We briefly discuss the psychological foundations of assertiveness and then focus on our approach to tracking the facial expressions and eye-contact that a user maintains while communicating with the virtual bully in the simulation. We also outline additional non-verbal indicators tracked by the software and discuss the dialog system, which drives the simulation. Finally, we outline some ethical considerations, discuss the limitations of our current software prototype, and list future directions for enhancing assertiveness training in anti bullying education.

The task of identifying the location that a user looks at, commonly known as gaze estimation, has various HCI and NLP applications. Traditional gaze estimation methods use special hardware such as eye-trackers or ordinary cameras such as webcams to perform this. However, they are not applicable to the majority of web users either because the user does not have them or does not want to use them due to privacy reasons. In this paper, we propose the idea of using multimodal LLMs to analyze the content of the user’s screen along with mouse location to estimate the gaze location. It primarily uses the results of studies that extract common reading patterns such as the F-pattern and Z-pattern. Our experimental results on The Eye Of The Typer (EOTT) dataset provide promising results for estimating gaze location.

This study presents a survey of research integrating eye-tracking (gaze) data into Language Models (LMs) as a means of cognitively grounding NLP models and applications in human reading behavior. Although contemporary LMs excel at learning statistical patterns from text, they fundamentally lack human-like reading and comprehension capabilities. Incorporating gaze data may offer a window into cognitive processing, yet its impact on LMs remains underexplored. Addressing a persistent bottleneck, namely, the high cost and limited scale of laboratory eye-tracking, we propose a roadmap consisting of three streams of research for advancing this novel research domain: (1) developing cognitive multimodal corpora, (2) leveraging generative models for gaze synthesis to overcome the data bottleneck caused by the high costs of human eye-tracking, and (3) training LMs with gaze-guided attention mechanisms and input augmentation. Furthermore, we illustrate practical applications in readability assessment, educational analytics, and assistive communication, demonstrating how gaze-informed models can enable adaptive technologies. Finally, we critically examine ongoing challenges, including the lack of data standardization, the misalignment between human and machine language processing, and the urgent ethical imperative for privacy-preserving architectures to protect sensitive biometric gaze data, motivating privacy-aware data practices and model designs for scalable deployment.