When we hear something like – “…and he hit the ball”, commonsense brings to mind something like, say, (there being) a bat (for hitting the ball with) since hitting a ball with a hand (which is the hypothetical literally-default connotation) is odd and uncommonsensical.
Let’s see a broad explanation for the above.
Firstly, the explanation rests upon an intuitive argument that the commonsense part of the natural language understanding of the mind rests upon phrasal processing – that is picking up words as chunks, to the extent possible, instead of isolated words. We read fast, and there are “auto-complete bubbles” that arise in the mind when we read something like “beat him to”, containing “death”. That is, when we read something till “beat him to”, there is an auto-complete bubble of “death” that arises in the mind.
Now, there are 2 aspects to the commonsensical part of this phrasal chunk processing. One is the one described above – auto-completing the words of the chunks. This has to be clearly syntactic in nature, and is based upon memory (networks of memory). It depends upon having repeatedly heard the phrases (like “beat him to death” or “won the prize”).
Coming to the explanation – allied to this syntactic aspect of common sense, to the phrasal chunk processing, has to be a (commonsensical) semantic one too. This is the one we are referring to here in this discussion, wherein commonsense-MEANING-completion too has to happen in the mind (like the commonsense syntactic word completion). This is what leads to the arousal of the entity ‘bat’ upon reading something like “hit the ball”. This is commonsensical semantic auto-completion. This also proves that there have to be commonsense knowledge-base semantic networks in the brain, (as are used in tools like MIT’s ConceptNet), to enable the process.
So, to summarize, the natural language understanding in the mind is phrasal in nature, to the extent possible. This mechanism has 2 commonsense aspects – 1) Syntactic, memory-based auto-completion of words, and 2) Semantic ‘commonsense knowledge-base semantic networks’-based auto-completion of “meaning”. It is the latter which leads to the arousal of things which aren’t actually present in the given data.