Multimodal machine translation (MMT) aims to enhance translation quality by integrating visual information. However, existing methods often extract visual features using pre-trained models while learning text features from scratch, leading to representation imbalance. These methods are also prone to being misled by redundant visual information, which results in suboptimal performance. To address these challenges, we propose CAMT, a novel cross-modal VQA-augmented MMT method. CAMT aligns image-source text pairs and image-question text pairs through dual-text contrastive learning, thereby improving semantic consistency across modalities. Additionally, we design an effective strategy for generating question–answer pairs to enhance fine-grained alignment and filter out irrelevant visual noise, while also addressing the scarcity of VQA annotations. Extensive experiments on multiple benchmark datasets demonstrate the effectiveness of the proposed CAMT framework, which consistently outperforms state-of-the-art MMT methods across multiple evaluation metrics.

Recently proposed dialogue state tracking (DST) approaches predict the dialogue state of a target turn sequentially based on the previous dialogue state. During the training time, the ground-truth previous dialogue state is utilized as the historical context. However, only the previously predicted dialogue state can be used in inference. This discrepancy might lead to error propagation, i.e., mistakes made by the model in the current turn are likely to be carried over to the following turns.To solve this problem, we propose Correctable Dialogue State Tracking (Correctable-DST). Specifically, it consists of three stages: (1) a Predictive State Simulator is exploited to generate a previously “predicted” dialogue state based on the ground-truth previous dialogue state during training; (2) a Slot Detector is proposed to determine the slots with an incorrect value in the previously “predicted” state and the slots whose values are to be updated in the current turn; (3) a State Generator takes the name of the above-selected slots as a prompt to generate the current state.Empirical results show that our approach achieves 67.51%, 68.24%, 70.30%, 71.38%, and 81.27% joint goal accuracy on MultiWOZ 2.0-2.4 datasets, respectively, and achieves a new state-of-the-art performance with significant improvements.