A tiny principle regarding Commonsense in Linguistics

When there are 2 processes/actions A and B, such that : 

A is an enabler for B, and A wouldn't have any meaning if it wasn’t succeeded by B, then A becomes commonsense.

2 illustrative scenarios - 

a) We say “Switch on the light”, and NOT “press the button of the light”. Pressing the button of the light is process A, and ‘switch on the lights’ is B. Pressing the button is an enabler for switching on the lights, and just pressing the button which is, say, not connected to the lights would be meaningless.

Similarly, we say - "turn to the right", and not "turn the steering to the right".

b) When we say “Go for a swim”, there are 2 actions here - the actual act of swimming which is the one thats really referred to, and the “going” (for the swim) which isn't important. That is why, when we say ‘if you go for a swim, your body will become wet’, it is meant implicitly, by ‘going for the swim’, the actual act of swimming and not the act of ‘going’ (which won't make you wet).

The otherwise case : 

A = having food B = washing hands.

Here, A doesn't satisfy the 2 conditions. That is 1) one does not have food so that he can wash hands and 2) One can have food and not wash hands; that won't make the having of food meaningless. So A isn't commonsense when one talks of B. That is, washing hands doesn't imply that one was having food, commonsensically.

General principle : When there are 2 processes/actions A and B and A does not meet the 2 requirements of the condition statement above, then A is commonsense when B is mentioned. Otherwise, it isn't so.

6 kinds of inferences from a sentence.

6 kinds of inferences from a sentence - 

Sentence : A gave a ball to B. 

Story : Why did he give a ball to B? Because B had lost his ball and was crying.

Story-commonsensical inference about A - That means A is a nice guy.

Story-logical-cum-commonsensical inference about A - A gave the ball after B started crying

Story-logical inference about A - some smart connection between a few data points of the story (not sufficiently long enough here) leading to some inference about A.

Commonsense inference about A - A gave the ball with his hands, with his hands pointing towards B etc.

Logical-cum-commonsensical inference about A - A had a ball in his ownership/control for some time, 

Logical inference about A - Say some quantum mechanical or relativistic inference about the event of A giving a ball to B! 

The first 2 in each set, leading to a total of 4 inferences, constitute the meaning of a sentence in a story. Some specific story can destroy the 2 in the latter.

Understanding Commonsense Knowledge

 

UNDERSTANDING COMMONSENSE KNOWLEDGE

We all have pieces of commonsense knowledge within us. But we also have an understanding of them - we know, in crude terms, the why and how - the working - of the processes related to them. This constitutes commonsense reasoning behind the piece of commonsense knowledge, and which (the reasoning) is invisible/unsaid too. Every piece of commonsense knowledge is accompanied by a commonsense understanding of it, which lies in a piece of commonsense reasoning behind it. 

Machines don't have an understanding of the commonsense knowledge pieces in the KB. They don't know why and how ovens bake or scissors cut or pulling tears or eating puts something in the stomach - in crude terms. This is required for commonsense reasoning. 

We need to thus represent the mechanism and thus the crude steps in the workings of eating, cutting, baking, tearing etc.

This will substitute writing hundreds of pieces of varied knowledge.

Consider this example : Mary had a sore mouth. So Jack didn't give her the chicken. 

The CYC system can infer here, but in terms like : if you have a sore mouth, then you avoid eating food. But it doesn't have a “real understanding” of why what it is outputting is true.

Systems which understand KB pieces can “explain” this too (not just in discrete knowledge steps) but by pointing out the stage at which this problem lies, in the relevant process. 

If you have a commonsensically laid-out process in the KB, prepared by filling in the requisite boxes of ‘STATE -> ACTION -> STATE -> ACTION -> …..’ you will have, for say, eating, something like - 

(S)           (A)       (S)         (A)           (S)        (A)             (S)           (A)        (S)        (A)    (S)

Hold food in -> Open -> mouth open -> put held       -> food inside ->close mouth-> mouthclosed  -> apply force->food enters->none->food enters

Hand                Mouth                           food in mouth    mouth                                 with food inside    to swallow     food-pipe                  stomach

Now, the system can diagnose the knowledge piece (sore mouth implies avoid food) by pointing out the stage (4) - sore mouth closed with food inside - as the reason or problem-space, which gives an understanding and “feel” of the commonsense knowledge piece, and which also comprises a kid’s understanding. 

This also automatically creates varied commonsense knowledge pieces like - the relative order of the actions, states and ‘actions and states mixed’, and internal cause-effect pairs in the process. This provides a rich substrate for the system to do commonsense reasoning of various kinds on the situation.

The meaning of 'MEANING'

What is meaning?

Consider this sentence - A gave a ball to B.

Now, one thing that immediately comes to mind (commonsense inference) is that A gave the ball with his hands and they pointing towards B.

Another inference one can draw from this is that - the ball was first in A's hands and then in B's hands. Now this inference is not "reversible" i.e. the ball being in A's hands first and then in B's hands doesnt necessarily imply that A GAVE the ball to B. But the commonsense inference is reversible i.e. A transferring something with his hands and they pointing towards B implies the original statement - amounts to A giving it to B. 

Commonsense inferences are REVERSIBLE; inferences in general could be REVERSIBLE or IRREVERSIBLE.

Perhaps, from the above, commonsense inferences can be called MEANING, and the rest as just INFERENCES. Or in other words we could define meaning as 'COMMONSENSE INFERENCE'.