ULTra: Unveiling Latent Token Interpretability in Transformer Based Understanding