Possible Affordances and Constraints Brought by the Integration of LLMs and LMMs into Chinese-English Video Game Localization

At present, Large Language Model-Based Translation (LLM-BT) is still a novel, but promising, automatic translation approach. This paper explores the possibilities that Large Language Models (LLMs) and Large Multimodal Models (LMMs), capable of processing and generating multimedia (i.e., text, audio, image, video), can bring to specialised translators, specifically video game localisers. The study resorts to the concept of affordances, originally proposed by J.J. Gibson (1966, 1979), to explore the possibilities of action that derive from the unique relations between LLMs-LMMs and translators. Prompted through text-based or audio-based interfaces, these models could work as unified, one-stop solutions to offer multilingual assistance, adaptable, interactive automated translation and multimedia processing, contextualisation and (re-)creation, in one central location. Notwithstanding such a potential, as of today, these models suffer from limitations, including biases, hallucinations and subpar processing of specialised, culture-specific or multimodal inputs. Despite such and other challenges, LLMs and LMMs have the potential to offer unique affordances in an era of ever-increasing human-machine collaboration in translation processes.