Cross-domain neural machine translation often suffers from semantic drift because fluent target sentences can still deviate from the meaning of the source sentence when domain distributions differ or parallel data are limited. Although large language models provide strong multilingual semantic priors, most existing translation frameworks use them mainly as encoders or generators and do not explicitly constrain source and target representations to remain semantically consistent during training. To address this problem, this paper proposes a semantic alignment framework for cross-domain neural machine translation with large language models. The core idea is to use a pretrained multilingual LLM encoder to build a unified sentence-level semantic space and to explicitly align translation pairs within this space. On this basis, a dual-view semantic alignment mechanism is introduced, combining representation-level distance minimization with contrastive semantic discrimination so that matched translation pairs become closer while mismatched pairs remain separable. An adaptive alignment strategy further adjusts the alignment strength according to the current semantic residual, allowing the model to emphasize translation fitting at the beginning of training and to impose stronger semantic consistency constraints after the representation space becomes more stable. In this way, the proposed framework treats semantic alignment as the central mechanism and uses the LLM encoder and adaptive weighting strategy as supporting components for robust cross-domain translation. Experimental results on general-domain, low-resource, and domain-shifted benchmarks show that the proposed method yields consistent improvements in BLEU, ChrF, METEOR, and COMET, with especially clear gains on datasets with stronger domain mismatch. These results suggest that explicit semantic alignment is a promising way to improve semantic fidelity and cross-domain robustness in neural machine translation. Conventional methods often struggle to accurately detect and address quality issues in such dynamic environments. This paper proposes a novel anomaly detection and control method based on image feature modeling to enhance process monitoring and decision-making in supply chain operations. By leveraging advanced image processing techniques, key features of production processes are extracted and modeled, enabling accurate identification of deviations and anomalies. Experimental results demonstrate that the proposed method significantly improves detection accuracy and response time compared to traditional approaches. This study contributes to the development of intelligent quality control solutions, offering scalability and robustness for real-world supply chain applications.
Zhang et al. (Sat,) studied this question.