Quantinuum has announced the release of λambeq Gen II, the latest version of its open-source quantum natural language processing (QNLP) software. λambeq enables users to convert sentences into quantum circuits, which can then be executed on real quantum hardware. The update reflects five years of development since Quantinuum’s initial demonstration of QNLP and incorporates recent progress in both quantum hardware and linguistic formalism. Whereas earlier versions were based on DisCoCat (Distributional Compositional Categorical) semantics, Gen II introduces a new formalism called DisCoCirc that enables modeling of larger text structures and more expressive compositionality across language.
DisCoCirc improves upon its predecessor by capturing more of the compositional structure inherent in natural language, supporting scalability to full texts rather than isolated sentences. The new framework also introduces language neutrality, which makes it adaptable across linguistic domains, and ensures that the resulting quantum models are canonical—well-defined with respect to both compositional structure and learnability. These properties collectively mitigate several pain points of earlier QNLP research, such as trainability of quantum models and interpretability of results. The Gen II release is accompanied by evidence of improved quantum performance and demonstrations of compositional generalization and explainable AI (XAI) functionality.
The λambeq package remains open source and integrates with Quantinuum’s quantum stack. Users can convert syntactic structures directly into quantum circuits using Python and VQC (variational quantum classifier) methods. The release includes updated tutorials and documentation to support research and experimentation in scalable QNLP applications.
For more details, readers can access the full announcement here, the technical paper here, the main documentation here, and a detailed tutorial on DisCoCirc here.
May 22, 2025
Leave A Comment