[Paper Review] Toward Quantum Machine Translation of Syntactically Distinct Languages

Mina Abbaszade, Mariam Zomorodi‐Moghadam+2 more|arXiv (Cornell University)|Jul 31, 2023

Quantum Computing Algorithms and ArchitectureComputer Science3 citations

TL;DR

This study explores quantum machine translation using noisy intermediate-scale quantum (NISQ) devices by integrating parametrized quantum circuits with a classical LSTM-based encoder-decoder model. It demonstrates that rotation gate angles—optimized via Shannon entropy—serve as effective communication channels between quantum circuits of English and Persian, achieving a mean absolute error of 0.03 and loss of 0.016 with the Adam optimizer on a 160-sample dataset.

ABSTRACT

The present study aims to explore the feasibility of language translation using quantum natural language processing algorithms on noisy intermediate-scale quantum (NISQ) devices. Classical methods in natural language processing (NLP) struggle with handling large-scale computations required for complex language tasks, but quantum NLP on NISQ devices holds promise in harnessing quantum parallelism and entanglement to efficiently process and analyze vast amounts of linguistic data, potentially revolutionizing NLP applications. Our research endeavors to pave the way for quantum neural machine translation, which could potentially offer advantages over classical methods in the future. We employ Shannon entropy to demonstrate the significant role of some appropriate angles of rotation gates in the performance of parametrized quantum circuits. In particular, we utilize these angles (parameters) as a means of communication between quantum circuits of different languages. To achieve our objective, we adopt the encoder-decoder model of classical neural networks and implement the translation task using long short-term memory (LSTM). Our experiments involved 160 samples comprising English sentences and their Persian translations. We trained the models with different optimisers implementing stochastic gradient descent (SGD) as primary and subsequently incorporating two additional optimizers in conjunction with SGD. Notably, we achieved optimal results-with mean absolute error of 0.03, mean squared error of 0.002, and 0.016 loss-by training the best model, consisting of two LSTM layers and using the Adam optimiser. Our small dataset, though consisting of simple synonymous sentences with word-to-word mappings, points to the utility of Shannon entropy as a figure of merit in more complex machine translation models for intricate sentence structures.

Motivation & Objective

• To investigate the feasibility of quantum natural language processing (QNLP) for machine translation on NISQ devices.
• To address the computational limitations of classical NLP in handling large-scale linguistic data by exploring quantum parallelism and entanglement.
• To establish a quantum-classical interface where rotation gate parameters act as semantic carriers between language-specific quantum circuits.
• To evaluate the performance of quantum-enhanced translation models using classical optimizers and standard NLP metrics on a small but representative dataset.
• To demonstrate the utility of Shannon entropy as a figure of merit for tuning quantum circuit parameters in translation tasks.

Proposed method

• Employed a hybrid quantum-classical encoder-decoder architecture based on long short-term memory (LSTM) networks for sequence-to-sequence translation.
• Mapped English and Persian sentence pairs into parametrized quantum circuits (PQCs) using the DisCoCat model and ZX-calculus for circuit translation.
• Used rotation gate angles as trainable parameters to encode linguistic meaning and enable inter-language communication between quantum circuits.
• Applied Shannon entropy to quantify and optimize the information content of quantum circuit outputs, guiding parameter tuning.
• Trained the model using stochastic gradient descent (SGD), Adam, and RMSprop optimizers over 1000 epochs, with early convergence observed after ~200 epochs.
• Evaluated performance using mean absolute error (MAE), mean squared error (MSE), and loss, with results reported per optimizer and model configuration.

Experimental results

Research questions

• RQ1 Can parametrized quantum circuits on NISQ devices effectively model cross-lingual sentence translation with minimal classical data?
• RQ2 How do rotation gate angles in quantum circuits influence the performance of machine translation tasks?
• RQ3 To what extent can Shannon entropy serve as a reliable metric for optimizing quantum circuit parameters in NLP applications?
• RQ4 What is the comparative performance of different optimizers (SGD, Adam, RMSprop) in training quantum-enhanced LSTM models for translation?
• RQ5 Can quantum circuits with limited qubit counts (e.g., 8–12 qubits) encode and process meaningful linguistic correlations comparable to classical word embeddings?

Key findings

• The best-performing model, using two LSTM layers and the Adam optimizer, achieved a mean absolute error (MAE) of 0.03, mean squared error (MSE) of 0.002, and a loss of 0.016.
• Validation loss converged after approximately 200 epochs when using the Adam optimizer, indicating efficient training dynamics.
• The Adam optimizer outperformed SGD and RMSprop, achieving the lowest loss (0.01574) and MAE (0.0324) among all optimizers tested.
• Shannon entropy was found to be a valuable figure of merit for evaluating and tuning quantum circuit parameters, especially in relation to rotation gate angles.
• The study demonstrated that quantum circuits with 8–12 qubits can represent high-dimensional vector spaces (e.g., 256–4096 dimensions), suggesting potential for compact semantic encoding.
• The results indicate that quantum-enhanced models can achieve low error rates on small, syntactically simple datasets, paving the way for future scaling to complex sentence structures.