LLM preprocessing can make quantum NLP systems practical for real financial text by simplifying sentences before processing, though the relationship between training data size and performance is counterintuitive.
This paper explores using LLMs to rewrite financial sentences to make them compatible with DisCoCat, a quantum natural language processing framework. The researchers found that LLM-assisted rewriting can reduce computational complexity by over 70% while maintaining sentiment meaning, achieving modest accuracy improvements on financial sentiment analysis tasks.