This study aims to construct a framework of linguistic properties of mathematical tasks that can be used to compare versions of mathematics test tasks in different natural languages. The framework will be useful when trying to explain statistical differences between different language versions of mathematical tasks, for example, differences in item functioning (DIF) that are due to inherent properties of different languages. Earlier research suggests that different languages might have different inherent properties when it comes to expressing mathematics. We have begun with a list of linguistic properties for which there are indications that they might affect the difficulty of a task. We are conducting a structured literature review looking for evidence of connections between linguistic properties and difficulty. The framework should include information about each property including methods used to measure the property, empirical and/or theoretical connections to aspects of difficulty, and relevance for mathematical tasks.