Machine translation has come a long way, but getting truly accurate and nuanced translations, especially for less common languages, remains a challenge. Large language models (LLMs) like those powering popular translation tools, excel with high-resource languages like English or Spanish. But they falter when dealing with languages that have less available data. New research suggests a clever workaround: telling the LLM a little about the grammar of the language it's translating. This technique, called GrammaMT, uses what linguists call “interlinear glossed text” (IGT). Think of IGT as a cheat sheet. It breaks down a sentence in the source language, explaining the role of each word (noun, verb, adjective, etc.) and its underlying meaning. GrammaMT feeds this IGT as part of the prompt to the LLM, essentially giving it a grammar lesson on the fly. The research team tested three variations of this approach: 'gloss-shot', 'chain-gloss', and 'model-gloss.' Gloss-shot simply includes the glossed text alongside translation examples. Chain-gloss goes a step further, asking the LLM to first generate the gloss itself before attempting the translation. Finally, model-gloss uses a separate, specialized model to produce the gloss, which is then fed to the translation LLM. Testing GrammaMT on a range of languages, from common ones like Portuguese and Russian to endangered languages like Tsez and Gitksan, the results were impressive. Across the board, GrammaMT significantly boosted translation accuracy. The most effective method often depended on the specific language, but the key takeaway is clear: even a little grammatical context makes a big difference. This approach could be a game-changer for low-resource languages, making it easier to preserve and share information in these often-marginalized communities. While the research primarily focused on translating into English, the team also explored reverse translations, with promising initial findings. Future research could investigate how to optimize these techniques for different language pairs and improve the automatic generation of glosses. The GrammaMT research offers a powerful new tool in our quest to break down language barriers and connect people across cultures.