The "Affective Text" task aimed to classify emotions and valence (positive/negative) in news headlines, exploring the connection between emotions and lexical semantics. The task involved evaluating automatic approaches to emotion recognition. News headlines were chosen because they typically contain strong emotional content and are written to provoke emotions, making them suitable for automatic emotion recognition. The task included two main components: emotion classification and valence classification. The data set consisted of news headlines from major newspapers and search engines, with 250 annotated headlines for development and 1,000 for testing. Annotators used a web-based interface to rate each headline on a scale for emotions and valence. Inter-annotator agreement was measured using Pearson correlation. Five teams participated, developing systems for emotion and valence classification. Systems included rule-based, knowledge-based, and machine learning approaches. Results showed that emotion annotation is challenging, with systems performing well on fine-grained measures but showing room for improvement. The task highlighted the importance of lexical semantics in emotion recognition and the difficulty of accurately capturing emotional content in text.The "Affective Text" task aimed to classify emotions and valence (positive/negative) in news headlines, exploring the connection between emotions and lexical semantics. The task involved evaluating automatic approaches to emotion recognition. News headlines were chosen because they typically contain strong emotional content and are written to provoke emotions, making them suitable for automatic emotion recognition. The task included two main components: emotion classification and valence classification. The data set consisted of news headlines from major newspapers and search engines, with 250 annotated headlines for development and 1,000 for testing. Annotators used a web-based interface to rate each headline on a scale for emotions and valence. Inter-annotator agreement was measured using Pearson correlation. Five teams participated, developing systems for emotion and valence classification. Systems included rule-based, knowledge-based, and machine learning approaches. Results showed that emotion annotation is challenging, with systems performing well on fine-grained measures but showing room for improvement. The task highlighted the importance of lexical semantics in emotion recognition and the difficulty of accurately capturing emotional content in text.