This paper presents the second place system for the R2VQ: competence-based multimodal question answering shared task. The task consisted in building question answering systems that could process procedural recipes involving both text and image, and enriched with semantic and cooking roles. We tackled the task by using a text-to-text generative model based on the transformer architecture, with the aim of generalising across different question types. Our proposed architecture incorporates a novel approach for enriching input texts by incorporating semantic and cooking role labels through what we call Label-Enclosed Generative Question Answering (LEG-QA). Our model achieves a score of 91.3, with a significant improvement over the baseline (65.34) and close to the top-ranked system ((92.5). After describing the submitted system, we analyse the impact of the different components of LEG-QA as well as perform an error analysis.