Agents act on information. At a given point in time, the organisation of information into a structure like a vector, matrix, tensor, or graph defines the agent's state. The superset formed from assigning all possible values to the features forming a state constitutes the state space. The position of an agent in a state space is a state tensor representing the information influencing an agent's decision. When a swarm of agents operate simultaneously, the swarm state tensor formed from all their individual state tensors carries information on both the individuals and the swarm. Real-time interpretation of this complex mathematical construct into a natural language makes the operation of this complex system transparent to humans. This paper introduces a novel methodology to interpret the swarm state tensors. We bring together our prior work on swarm ontology and the Jingulu swarm language (JSwarm) to design a data-generation pipeline to train the Text-To-Text Transfer Transformer (T5). The data pipeline methodology creates a dataset with 0.5 billion data points, which is then used to fine-tune T5. We used existing metrics and designed two new situationawareness-based metrics to assess the ability of the refined T5 to transform the swarm state tensors in real-time into plain English statements. Results show superior performance of the proposed methodology when compared to baseline T5.