Quantum language models have shown competitive performance on sequential tasks, yet whether trained quantum circuits exploit genuinely quantum resources -- or merely embed classical computation in quantum hardware -- remains unknown. Prior work has evaluated these models through endpoint metrics alone, without examining the memory strategies they actually learn internally. We introduce the first mechanistic interpretability study of quantum language models, combining causal gate ablation, entanglement tracking, and density-matrix interchange interventions on a controlled long-range dependency task. We find that single-qubit models are exactly classically simulable and converge to the same geometric strategy as matched classical baselines, while two-qubit models with entangling gates learn a representationally distinct strategy that encodes context in inter-qubit entanglement -- confirmed by three independent causal tests (p < 0.0001, d = 0.89). On real quantum hardware, only the classical geometric strategy survives device noise; the entanglement strategy degrades to chance. These findings open mechanistic interpretability as a tool for the science of quantum language models and reveal a noise-expressivity tradeoff governing which learned strategies survive deployment.
Quantum language models have shown competitive performance on sequential tasks, yet whether trained quantum circuits exploit genuinely quantum resources -- or merely embed classical computation in quantum hardware -- remains unknown. Prior work has evaluated these models through endpoint metrics alone, without examining the memory strategies they actually learn internally. We introduce the first mechanistic interpretability study of quantum language models, combining causal gate ablation, entanglement tracking, and density-matrix interchange interventions on a controlled long-range dependency task. We find that single-qubit models are exactly classically simulable and converge to the same geometric strategy as matched classical baselines, while two-qubit models with entangling gates learn a representationally distinct strategy that encodes context in inter-qubit entanglement -- confirmed by three independent causal tests (p < 0.0001, d = 0.89). On real quantum hardware, only the classical geometric strategy survives device noise; the entanglement strategy degrades to chance. These findings open mechanistic interpretability as a tool for the science of quantum language models and reveal a noise-expressivity tradeoff governing which learned strategies survive deployment.
Hybrid quantum-classical models represent a crucial step toward leveraging near-term quantum devices for sequential data processing. We present Quantum Recurrent Neural Networks (QRNNs) and Quantum Convolutional Neural Networks (QCNNs) as hybrid quantum language models, reporting the first empirical demonstration of generative language modeling trained and evaluated end-to-end on real quantum hardware. Our architecture combines hardware-optimized parametric quantum circuits with a lightweight classical projection layer, utilizing a multi-sample SPSA strategy to efficiently train quantum parameters despite hardware noise. To characterize the capabilities of these models, we introduce a synthetic dataset designed to isolate syntactic dependencies in a controlled, low-resource environment. Experiments on IBM Quantum processors reveal the critical trade-offs between circuit depth and trainability, demonstrating that while noise remains a significant factor, observable-based readout enables the successful learning of sequential patterns on NISQ devices. These results establish a rigorous engineering baseline for generative quantum natural language processing, validating the feasibility of training complex sequence models on current quantum hardware.
Recently, a Quantum Language Model (QLM) was proposed to model term dependencies upon Quantum Theory (QT) framework and successively applied in Information Retrieval (IR). Nevertheless, QLM's dependency is based on co-occurrences of terms and has not yet taken into account the Quantum Entanglement (QE), which is a key quantum concept and has a significant cognitive implication. In QT, an entangled state can provide a more complete description for the nature of realities, and determine intrinsic correlations of considered objects globally, rather than those co-occurrences on the surface. It is, however, a real challenge to decide and measure QE using the classical statistics of texts in a post-measurement configuration. In order to circumvent this problem, we theoretically prove the connection between QE and statistically Unconditional Pure Dependence (UPD). Since UPD has an implementable deciding algorithm, we can in turn characterize QE by extracting the UPD patterns from texts. This leads to a measurable QE, based on which we further advance the existing QLM framework. We empirically compare our model with related models, and the results demonstrate the effectiveness of our model.
Quantum entanglement is a key resource in quantum computing and quantum information processing tasks. However, its quantification remains a major challenge since it cannot be directly extracted from physical observables. To address this issue, we study a few machine-learning based models to estimate the amount of entanglement in two-qubit as well as three-qubit systems. We use measurement outcomes as the input features and entanglement measures as the training labels. Our models predict entanglement without requiring the full state information. This demonstrates the potential of machine learning as an effcient and powerful tool for characterizing quantum entanglement
Stock market forecasting remains one of the most computationally demanding tasks in financial engineering, requiring sophisticated models capable of processing nonlinear patterns, high-frequency volatility, and large volumes of time-series data. Recent advancements in quantum computing offer the potential to surpass classical machine learning (ML) by leveraging principles such as superposition and entanglement to accelerate learning and optimization. This paper conducts a comprehensive comparison between quantum machine learning (QML) models and classical deep learning architectures for stock price prediction, emphasizing the technological improvements and optimization of computational methods. Our analysis integrates insights from prior studies, including the application of quantum-enhanced Gramian Angular Fields with CNNs [2], shallow parameterized quantum circuits for market indicators [4], and hybrid quantum-classical algorithms tailored to finance [1]. Experimental evaluation (using IBM Qiskit's quantum simulators and real hardware) demonstrates competitive performance of quantum neural networks (QNNs) against state-of-the-art recurrent neural networks (RNNs) and long short-term memory (LSTM) architectures [3]. Across AAPL/MSFT and short horizons, compact QNNs match strong LSTM baselines while using an order of magnitude fewer parameters, and—critically—transfer from calibrated simulators to IBM hardware with marginal degradation after mitigation. This device-aware training/evaluation loop exemplifies a deployable path for financial QML under realistic noise and resource constraints and motivates standardized, time-series-aware testing for the community. Results highlight the engineering value of quantum acceleration in feature encoding, the research novelty of hybrid ensemble strategies, and the future scope of benchmarking across leading quantum computing machines such as IBM, Google Sycamore, Rigetti, IonQ, and Xanadu.
Deep learning models are powerful but opaque. As quantum machine learning matures, the field faces a defining choice: build quantum models that are equally opaque, or exploit the mathematical structure of quantum mechanics to make them inherently interpretable. We show that the latter is possible. By tracking quantum mutual information~(MI), entanglement entropy, and state fidelity through the layers of a Quantum Transformer Block (\qtb{}), a fully-coherent variational circuit with quantum analogues of both attention and feedforward, we gain direct insight into how the model processes information: which tokens it attends to, when correlations form, and why predictions fail. On four tasks with known dependency structure we show that (i)~learned MI matrices align with ground-truth task structure (AUC=0.69 on lookup), (ii)~disabling entangling gates collapses accuracy from 100\% to 15\% while MI→0, proving entanglement is the mechanism, (iii)~accuracy and MI co-evolve during training (ρ=0.92 on lookup), and (iv)~per-sample MI predicts prediction correctness on the conditional task with ROC AUC=0.84. All results are validated on IBM Quantum hardware (ibm\_kingston, Heron~r2): the circuit's reasoning process, from product state through structured entanglement, is directly observable on a superconducting processor. These proof-of-concept results, obtained on small synthetic tasks, suggest that the physics of quantum computation can provide intrinsic interpretability signals with no direct classical counterpart, motivating study of whether this advantage persists at scale.
We present the first application of pregroup grammar-based quantum compositional natural language processing (QNLP) to Arabic — a morphologically rich, free-word-order language whose structural complexity provides a uniquely demanding testbed for theories of meaning composition in quantum circuits. Our system converts Arabic sentences into quantum circuits whose topology mirrors grammatical structure: subjects, verbs, and objects become quantum gates, and the typed dependencies between them — the pregroup grammar — determine how those gates are wired together. We conduct three controlled experiments spanning word order, morphological tense, and verb sense disambiguation, comparing quantum circuit methods against classical baselines including AraVec (Arabic word embeddings) and AraBERT (a pre-trained Arabic transformer). The central finding is a clean causal ablation: quantum circuits encoding only grammar topology, with no parameterised components, achieve exactly 50% on a matched-pair word order task where classical bag-of-words models fail (12.8%); adding a single layer of parameterised entangling gates raises performance to 64.9% (95% CI [62.8, 66.3]). This 15-percentage-point gain — with zero variance at the ablation baseline across all seeds and folds — is entirely attributable to parameterised entanglement, establishing a clean causal claim. We additionally introduce the first vocabulary-controlled Arabic word sense disambiguation dataset, using matched sentence pairs and shared lexical pools to isolate structural from lexical disambiguation signals, and characterise a SPSA label inversion phenomenon whose rate is measurably reduced by ancilla qubit encoding. Arabic's Semitic root-and-pattern morphology has a formal correspondence to quantum tensor products — identified independently in the computational linguistics literature — that no other language in the existing QNLP literature shares, positioning Arabic as a theoretically motivated and practically significant target for quantum compositional methods.