This paper investigates how the distribution of attention during problem-solving tasks can reveal cognitive processes in programmers of varying abilities, especially in program comprehension where the impact of abilities on the dynamic change of eye movement behavior over time is less understood. Through a quantitative analysis of eye gaze behavior, categorized into focused behavior (concentrating on a few positions) and overview behavior (scanning many positions), based on AoI sequences, the study analyzes data from 66 novice programmers tackling program comprehension tasks across selection and iteration constructs. The research aims to determine differences in attention allocation between high- and low-ability participants during programming problem-solving. Utilizing the TS-AOI method, findings show that high-ability participants maintain more focused attention and demonstrate shifts in attention from the early to the late phases of tasks, unlike their low-ability counterparts. These insights could guide strategies to support students in enhancing their programming comprehension skills.
Large language models continue to face challenges in translating low-resource languages with scarce parallel data. This study investigates how to fine-tune them effectively using target-side monolingual data. Existing approaches–dominated by back-translation and recent LLM-based rewriting–remain limited by noisy synthetic sources, unguided simplification, and the absence of a principled mechanism for integrating monolingual sentences into the training objective. To address this, we developed a semi-supervised framework that integrates marginal distribution estimation and curriculum-guided rewriting to exploit monolingual data for low-resource translation. Experiments in four low-resource directions demonstrated substantial gains, averaging +8 spBLEU and +10 COMET over strong baselines, while three additional mid-resource directions showed stable improvements and consistent trends. Reference-free metrics further validated robust gains in fluency and adequacy. The findings establish a scalable paradigm for low-resource translation, revealing that the principled integration of marginal likelihood estimation and generative rewriting enables large language models to achieve superior performance under extreme data scarcity.
3 September 2026 in Biomedical Signal Processing and Control