Adopting a mixed experimental approach, this study compares the effects of L1-glossed keyword subtitles, L2-glossed keyword subtitles and non-glossed keyword subtitles on incidental L2 vocabulary acquisition. Thirty CET-4 undergraduates were randomly assigned to three groups. Data were collected through immediate vocabulary tests, three-week delayed retention tests, eye-tracking experiments and Likert questionnaires. One-way ANOVA reveals a clear hierarchical facilitating effect: L1-glossed subtitles rank first, followed by L2-glossed subtitles, and non-glossed subtitles perform worst, with significant gaps found in orthographic recognition, meaning recognition and meaning recall. Eye-tracking results indicate L1 glosses greatly reduce fixation duration and counts on target words and subtitle areas to optimize visual attention allocation. Correlation analysis proves appropriate focus on target words improves vocabulary gains, while excessive scanning of subtitle regions hinders learning. Delayed tests show L1-glossed subtitles produce the lowest forgetting rate and best long-term retention. Questionnaire data also demonstrate learners hold the highest recognition, usability evaluation and willingness to use L1-glossed subtitles. This study verifies that L1-glossed keyword subtitles reduce cognitive load, deepen form-meaning connections, and promote both short-term acquisition and long-term retention of L2 vocabulary.