Text this: Integrating Subsymbolic and Symbolic Processing in Artificial Vision