University of Padova 

Abstract: Quantum Theory (QT) has been widely adopted by many research areas other than physics. In the field of natural language processing (NLP), exploratory research has been conducted to unveil the quantum phenomena in human language understanding from a cognitive aspect, but there is a lack of applicable models for concrete NLP tasks inspired by QT. We theorize the quantum interpretation of language understanding by mapping linguistic units of various levels to quantum states on a unified Hilbert Space, which inherently tackles the word polysemy and compositionality issues. On the implementation phase, we build complex-valued neural networks to turn this quantum theoretical framework into a textual representation, and address the text classification and question answering tasks on its basis. The quantum-inspired models have comparable performances to the state-of-the-art, and the numerical constraints on the complex-valued components increase the robustness and allow us to understand how the models work from a quantum perspective. 

Contact: Lei Wang, 9853