Automation of any kind of problem solving usually requires finding suitable symbolic representations for the problems we intend to solve. Such encodings should be expressive enough to encode any imaginable problem we might have to deal with while imposing as little representational complexity as necessary, both for comprehensibility by humans and straightforward symbol manipulation by machines.
In this work we present a practical example that uncovers limitations of a currently widely-used encoding in the field of machine learning, namely attribute-value representations. The extensions we will apply thereupon will lead us to a uniform data representation, which will hopefully provide a suitable basis for the development of more capable general-purpose machine learning systems.