Create a Naive Bayes text classifier from scratch in Python. The model should be trained on a small dataset of positive and negative movie reviews. Implement Laplace smoothing to handle zero-frequency words and provide methods to predict the sentiment of new, unseen sentences.