Quantum Language Model With Entanglement Embedding for Question Answering.

Area of Science:

• Natural Language Processing
• Quantum Computing
• Artificial Intelligence

Background:

• Quantum language models (QLMs) offer transparency and interpretability.
• Current QLMs do not fully leverage quantum probabilistic descriptions for word sequences.
• There is a need for neural network modules to capture nonclassical correlations in text.

Purpose of the Study:

• To propose a novel neural network module for quantum language models.
• To explicitly capture nonclassical correlations within word sequences using quantum entanglement.
• To enhance the performance and interpretability of quantum language models.

Main Methods:

• Developed a novel entanglement embedding (EE) module for neural networks.
• Transformed word sequences into entangled pure state representations.
• Implemented a quantum-inspired neural network structure (QLM-EE) on classical computing devices.

Main Results:

• Observed strong quantum entanglement within word sequences, indicating parallelized correlations.
• QLM-EE demonstrated superior performance on question answering datasets compared to classical deep neural networks and other QLMs.
• Quantifying entanglement improved the post-hoc interpretability of the model.

Conclusions:

• The proposed QLM-EE effectively captures nonclassical correlations in word sequences.
• Entanglement embedding enhances both performance and interpretability in quantum language models.
• This quantum-inspired approach offers a promising direction for advancing natural language processing.