Emotionally salient stimuli preferentially capture attention, a phenomenon termed emotional attentional bias. This effect occurs not only with emotional faces and images but also in language processing. Unlike emotional faces and images, which provide explicit visual cues, emotional attentional bias in language relies on semantic processing of emotional words. Specifically, it involves extracting emotional information from words via semantic access. Research indicates that this process is closely linked to the representation of emotional semantics in words, that is, how words convey emotional meanings. Based on their semantic representation, researchers classify emotional words into two types: emotion-label words (e.g., "joy," "anger"), which directly denote emotions, and emotion-laden words (e.g., "success," "failure"), which convey emotions indirectly. These two types of words differ significantly in semantic processing, especially regarding the activation and extraction of emotional information. However, these findings are primarily based on studies of isolated word processing. In natural language processing, especially in Chinese reading, it remains unclear whether these two categories of emotional words show semantic differences and whether these representational differences affect reading processes. Moreover, it is unclear whether emotional semantic activation during reading depends on the integration of lexical contextual information. In this study, we employed natural reading tasks combined with eye-tracking technology, embedding emotional words within sentences to investigate the mechanisms underlying differences in emotional semantic processing and the influence of contextual diversity.

To investigate differences in emotional semantic processing across valences during natural reading, participants were first recruited to rate the emotional prototypicality of two types of emotional words, defined as the degree to which a word represents a specific emotion. The study comprised two experiments. Experiment 1 targeted positive valence, whereas Experiment 2 targeted negative valence. Both experiments used a 2 (emotion word type: emotion-label vs. emotion-laden) × 2 (contextual diversity: high vs. low) within-subjects design for natural sentence reading. Forty-eight participants took part, and their reading was recorded using an EyeLink 1000 Plus eye-tracker. Participants read one set of four experimental materials comprising 128 sentences: 80 experimental, 40 fillers, and 8 practice sentences. Approximately 30% of the sentences were followed by comprehension questions, to which participants responded with "yes" or "no."

The rating results revealed significant differences in emotional prototypicality between the two types of emotional words. Emotion-label words exhibited higher emotional prototypicality and more accurately represented emotional meanings. In positive valence (Experiment 1), emotion-label words showed longer fixation times, indicating greater attentional capture, whereas words with high contextual diversity had shorter fixation times, suggesting faster processing. No interaction effect was observed. In negative valence (Experiment 2), the main effect of emotion word type persisted, with emotion-label words showing longer fixation times during both early and late stages. Notably, the processing difference between emotion word types in negative valence was moderated by contextual diversity, affecting the entire processing span from early to late stages. Specifically, emotion-label words exhibited longer fixation times only under low contextual diversity, whereas no fixation time differences were observed between emotion-label and emotion-laden words under high contextual diversity.

The results reveal significant differences in the semantic representation of emotional words. Compared with emotion-laden words, emotion-label words more typically represent emotional meanings and exhibit higher emotional prototypicality. Furthermore, these findings suggest stable differences in emotional semantic processing during Chinese reading, with emotion-label words demonstrating a semantic advantage over emotion-laden words. Emotion-label words more effectively convey emotional information, capturing and sustaining readers' attention, emphasizing their ecological validity and supporting the embodied theory of semantic representation. These findings further demonstrate cross-linguistic consistency in emotional semantic differences and activation patterns. Additionally, contextual diversity moderated the semantic advantage of emotion-label words differently across valences, indicating that emotional semantics encompass both depth and breadth, and revealing a dissociation mechanism for activating positive versus negative emotional semantics during reading.