Quantum-inspired fuzzy matching network with morphology-enhanced word embeddings

Quantum Natural Language Processing (QNLP) has demonstrated significant advantages in addressing the uncertainty and vagueness in language understanding, with many related works relying on quantum-inspired neural networks. However, most existing quantum-inspired neural networks are based on word-level embeddings, which fail to effectively capture the complexity of language and simulate real human cognitive processes. Recognizing that the introduction of morphological information can inject prior semantic and syntactic knowledge, thereby enhancing the quality of the embeddings, and that the incorporation of fuzzy logic can further alleviate the inherent vagueness in language, we propose a quantum-inspired fuzzy matching network with morphology-enhanced word embeddings (QFNMWE). Our proposed model leverages morpheme-level and word-level embeddings to learn a richer multilevel semantic representation, followed by a fuzzy fusion and fuzzy measurement. Experimental results on various benchmarking datasets demonstrate that our QFNMWE outperforms a wide range of state-of-the-art baselines in different downstream tasks.