The field of natural language processing is moving towards more efficient and scalable solutions for multimodal multilingual translation, with a focus on mitigating cross-lingual interference and improving language-specific representations. Recent research has explored the use of layer selection mechanisms and neuron-level adaptation strategies to achieve state-of-the-art results in multimodal translation. Additionally, there is a growing interest in understanding the internal language processing of large language models, including how they handle code-mixed inputs and the role of language-specific neurons in shaping model behavior. Noteworthy papers include: LLaVA-NeuMT, which proposes a novel multimodal multilingual translation framework that achieves SOTA results on two datasets. Language Arithmetics, which introduces a method for systematic language neuron identification and manipulation, allowing for more effective language steering and improved downstream performance.