Researchers have developed a framework called UniLang that allows large language models to work with both natural language and machine-native symbols. This is done by expanding the model's vocabulary and embedding space to include grounded machine-native representations. The goal is to enable these models to directly operate on structured information without needing to convert it into natural language. The authors tested UniLang on two tasks: sequential recommendation and legal precedent prediction, and found that it outperformed strong baselines. --- Why it matters: This work matters because it could allow large language models to be used in a wider range of applications, such as working with structured data like databases or spreadsheets, without needing significant retraining or modification. Source: https://arxiv.org/abs/2608.19529