This paper explores text-based emotion prediction using supervised machine learning with the SNoW learning architecture. The goal is to classify the emotional affinity of sentences in children's fairy tales for use in expressive text-to-speech synthesis. Initial experiments on a dataset of 22 fairy tales show promising results over a naive baseline and BOW approach, with some dependency on parameter tuning. The study also discusses results for a tripartite model covering emotional valence and feature set variations. Plans include developing a more cognitively sound sequential model that considers a larger set of basic emotions. The application area involves text-to-speech synthesis, where emotional content in narrative text is crucial. Emotional text-to-speech synthesis requires identifying the appropriate emotional meaning of text passages and rendering prosodic contours to convey emotional content. The paper discusses previous work in affective computing, including studies on sentence-level emotional affinity and emotion classification in speech and human-computer dialogues. The empirical study involves a machine learning model, corpus, features, and parameter tuning. The model classifies sentences as either emotional or non-emotional, with results showing improved accuracy over naive baselines. The study also explores feature sets, including content BOW and other features derived from text. The feature set includes various linguistic and narrative features, such as sentence length, punctuation, and story progress. Parameter tuning was conducted using two conditions: sep-tune-eval and same-tune-eval. Results showed that the model achieved higher accuracy with the sep-tune-eval condition. The study also discusses the challenges of emotion classification, including the subjective nature of the task and the difficulty in separating emotional and non-emotional content. The results indicate that the model can correctly classify emotional content in children's stories, with some cases correctly identified by both annotators. The study plans to use a larger dataset and a hierarchical sequential model to better capture emotional categorization. Future work includes refining the feature set and exploring emotional intensity in a learning scenario. The paper concludes that further research is needed to comprehensively address the text-based emotion prediction problem.This paper explores text-based emotion prediction using supervised machine learning with the SNoW learning architecture. The goal is to classify the emotional affinity of sentences in children's fairy tales for use in expressive text-to-speech synthesis. Initial experiments on a dataset of 22 fairy tales show promising results over a naive baseline and BOW approach, with some dependency on parameter tuning. The study also discusses results for a tripartite model covering emotional valence and feature set variations. Plans include developing a more cognitively sound sequential model that considers a larger set of basic emotions. The application area involves text-to-speech synthesis, where emotional content in narrative text is crucial. Emotional text-to-speech synthesis requires identifying the appropriate emotional meaning of text passages and rendering prosodic contours to convey emotional content. The paper discusses previous work in affective computing, including studies on sentence-level emotional affinity and emotion classification in speech and human-computer dialogues. The empirical study involves a machine learning model, corpus, features, and parameter tuning. The model classifies sentences as either emotional or non-emotional, with results showing improved accuracy over naive baselines. The study also explores feature sets, including content BOW and other features derived from text. The feature set includes various linguistic and narrative features, such as sentence length, punctuation, and story progress. Parameter tuning was conducted using two conditions: sep-tune-eval and same-tune-eval. Results showed that the model achieved higher accuracy with the sep-tune-eval condition. The study also discusses the challenges of emotion classification, including the subjective nature of the task and the difficulty in separating emotional and non-emotional content. The results indicate that the model can correctly classify emotional content in children's stories, with some cases correctly identified by both annotators. The study plans to use a larger dataset and a hierarchical sequential model to better capture emotional categorization. Future work includes refining the feature set and exploring emotional intensity in a learning scenario. The paper concludes that further research is needed to comprehensively address the text-based emotion prediction problem.