The Symbol Grounding Problem re-framed as concreteness-abstractness learned through spoken interaction

The Symbol Grounding Problem points out that the underlying mechanisms of computation are symbolic and, therefore, missing crucial information when they are used for processing natural language until they are somehow able to perceive the world directly. Our goal in this paper is twofold: First, we review some of the recent literature that claims to address (even if just to a small degree) the Symbol Grounding Problem, and explain why it is still yet a problem partially due to a misinterpretation of the problem and that there are more modalilties that symbols need to ground into beyond just pictures, including emotion. Second, we re-frame the problem as a problem of handling concreteness and abstractness because (perhaps surprisingly) computational models of distributional meaning seem to capture abstractness more directly than they do concreteness. We take inspiration from child development and offer a toy example of how one could approach modeling concrete and progressively more abstract words. We conclude by posing some open questions and offering paths for future work.