Meaning Representations as Variational Quantum Circuits

Tilen Gaetano Limbäck-Stokin, Tanishka Birdavade, Kin Ian Lo, Mehrnoosh Sadrzadeh

Keywords: Quantum Machine Learning, Syntax, Semantics, Vision-Language Models, Machine Learning, ICML

TL;DR: Parameter-efficient variational quantum circuits grounded in linguistic composition for contrastive vision-language captioning.

Abstract: Large vision-language models struggle with a "compositionality gap" due to their reliance on unstructured statistical approximations to capture complex relations. We propose CCG-VQC, a quantum framework that explicitly maps linguistic syntax into parametrised quantum circuits. To separate the impact of architectural design from scale, we introduce MicroCLIP, a parameter-matched classical transformer. While CCG-VQC achieved 71.19 % on ARO-Attribution, MicroCLIP struggled to surpass random chance (50.85 %), and even standard CLIP lags behind at 61.00 %. This demonstrates that in low-parameter regimes, explicit linguistic structure outperforms statistical approximation.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 199