Joint Blending is a modeling technique in computational linguistics that integrates syntactic and semantic representations at the token level to enhance the coherence of multi-task language models by aligning their hidden states through differentiable alignment functions.