"Search Space" of commonsense knowledge.

The reservoir of the commonsense related to most said knowledge (BTW : all knowledge in the world is said BY someone, and TO someone) is the 'speaker' and the 'listener'. Or in broader words - the 'source' of the knowledge and the 'destination' of the knowledge.

 

Consider this written outside a temple - Please remove your shoes before entering the temple.

Now, like every piece of knowledge, this too is "incomplete" and has connected bits to it, whose answer lies in commonsense. For example, who is 'your'? Remove shoes from where? 'the temple' means which temple? etc. The answers to most of these questions will lie in the "space" of the source and the destination of the data/knowledge. First question's answer - the reader's shoes. Hence, destination. Second - From the listener's (/ destination's) legs. Third - the temple to which this sign board is attached to OR the temple from which this instruction is coming from. So this is the source. 

 

This would easen the process of getting to the answer of the commonsense questions to a given piece of knowledge by narrowing the "search space (which is the 'source' and 'destination')".

 

Take another small example - Suppose a textbook says - let us delve into a further chapter. Who is 'us'? Answer to this question is the source plus the destination. (Source coming from the book - the author. And the destination being the reader). Chapter "further" to what? Answer - current chapter. This is in the "source space".

Why Commonsense KBs are not "human" -

A note on the "Cognitive-compatibility" of commonsense KBs

Commonsense KBs contain statements like - Soaps are materials used for cleaning things. (Or, Cars are large, external things used for transportation; or 'Trucks are large, heavy surface level vehicles which can't control their altitude).

This is an effective, technical summarization of one's knowledge about soaps, cars and trucks. This is constructed with logical articulation. This is "thought-out" (is after applying thought).

What we actually know about soaps (and also the ways in which we know it) is as follows - 

Soaps are mostly solid and come in different colours. Sometimes they are liquid.

We require soap to have a bath.

We rub soaps quite hard, to and fro, on our bodies (and other bodies).

Soaps need to be applied with water.

When we rub soap, we get lather. The lather has to be washed off later.

When we apply soap to something, it gets cleaned.

The above are the actual (so to speak), knowledge-bits stored in our minds, about soaps. They are fragmented, fuzzy, practical, kluge-like, reflective of our real-life experiences with soaps, etc. Whereas the entries in the present KBs are technical aggregations of the properties of entities/concepts. That is certainly not how knowledge is stored in our brains. Ask a kid about a soap, or see the occurrences of the word 'soap' in his linguistic-expressions and his thoughts and they will resemble the list above. This list is reminiscent and indicative of the real ways and forms in which we have cognized the knowledge about soaps, in our minds, in life.

BUILDING COMMONSENSE KNOWLEDGE BASES - 3 WAYS.

 BUILDING COMMONSENSE KNOWLEDGE BASES - 3 WAYS.

Here are few examples of commonsense knowledge –

Soaps are materials used to clean things.

When children are given new things, they become happy.

Cars are large, external things used for transportation.

Cars contain people.


There are 2 kinds of items in the world - General and Specific. For example, ‘House’ (general) and ‘Pat Hayes’ (specific). Commonsense Knowledge is typically a relationship amongst general items (general concepts).

The above examples of commonsense knowledge contain general items (concepts) like - Material, cleaning, thing, external, transportation, containment, people, houses

Generate triplets, or quartets, or pentets, or.....(the more the better) of concepts purely randomly. For example, from the above set of concepts, we have one triplet as –

House, new, cleaning

OR, a quartet – house, transportation, containment, people

Now, form a sentence with these general items / concepts.

3 ways, (in increasing order of obviousness of the ways)-

1) Google search – search strictly for complete English sentences containing ALL those words.

Now, the web is written for people with commonsense. So a lot of general, basic knowledge won’t be in such worded, expressed form on the web. But still, quite a lot will be.

2) Ask the general public to form sentences from the word-sets presented to them on the web. This is a lot easier and more useful and feasible than asking the general public to feed in commonsense knowledge like the Open Mind Common Sense (OMCS) project of MIT Media Lab did.

3) First enlist ALL possible kinds of sentences in English language on the basis of the permutations of N, V, Adj, Adv,.....Now, given a random, say, quartet of words, form valid grammatical sentences from them FALLING INTO the possible types of valid Part-Of-Speech-permutations enlisted before. Put up the list on the web. Now children/people just have to tick or cross on each if it MEANS something sensible or not. For example, one combination of the quartet example taken above, which falls into a valid grammatical format is – People contain transportation houses. This will get a cross.

Why it is practically impossible to build Commonsense KBs -

Consider any 'incident'/'event' in everyday life. From it, at least one inference can be drawn. For example, consider the incident 'A child got a new bicycle today'. From this, one inference that can be drawn is - So he must be happy today! Now, this inference-drawing is an exhibition of common-sense. This is based on a piece of commonsense knowledge - when children get new things, they become happy. 

So, corresponding to every possible KIND OF event/incident in real life, there will be at least one inference possible (exhibition of common-sense), which in turn is CORRESPONDINGLY BASED UPON A PIECE OF COMMONSENSE KNOWLEDGE. So, corresponding to every possible kind of event/incident in real life, there will be atleast one piece of commonsense knowledge.

Now, the number of different possible kinds of incidents/events in real life is practically infinite. Just imagine - 'I deleted an email'. 'I painted a chair'. 'I arrived at the party'......  And rarely any piece of commonsense knowledge will repeat. So the number of pieces of commonsense knowledge responsible for the exhibition of all instances of common-sense is an impractical number!

A very obvious query -

At the heart of reason is nothing but 'DEFINITION' (or definitive substitution). 

Firstly, let's see one obvious point. Given an entity X, if there is any sentence involving the definition/synonym of X, I can cut-paste that definition/synonym into that sentence and the meaning of the sentence remains the same.

Now, a simple point is that given any sentence with entities E1, E2, E3,...., En in it, if I give ANY INFORMATION about an Ei, that information will behave as a "definition" of Ei. And, it will give an inference. (I bought this lamp from 'Laxmi stores'. 'Laxmi stores' was established in 1950. So, I bought this lamp from a store established in 1950).

Question - Can we represent all 'Reason' using hyper-text (or nested hyper-text) pages, since, as we see above, 'REASON' itself fundamentally functions/jumps in that way?

Analogy between 2 kinds of "meanings" -

1) Meaning of Words -

Suppose you are given this sentence - The book is on the table. You understand "what it MEANS". It means that there is a book resting on a table. Simple. 

2) Meaning of Reality-data - 

Suppose you see someone doing repetitions of a 100 kg dumbbell bar at the speed/frequency of squatting flies. You would immediately say - "this MEANS that the dumbells are fake (say, hollow)".

Now, in both the cases you used the word 'MEANS'. You used the concept of 'meaning'. 

Let's draw analogies. In place of words composed of letters and arranged in a certain order, in the first type, you have some real entities arranged in a certain way (configuration), in the second type. In both the cases you applied a set of knowledge-rules - in the first case, the rules of language, and in the second case, the commonsense (or otherwise) rules of the world. And in both the cases you drew an inference from these which was the 'MEANING'. This gives us a general definition of meaning - The inference drawn by applying thought to a given configuration of data, using certain knowledge-rules. 

But this means that meaning is nothing but an inference. Note that the meaning of 'X' is what X STANDS FOR and not what X IMPLIES (which is a further step via thinking more, on X). What X 'stands for' is also an 'implication' of X and hence meaning is a subset of implication. In a sense, MEANING IS THE "COMMONSENSE" OF IMPLICATION, since it is the first, immediate thing implied by any configuration of data. 

This leads to a point - Coming back to a previous part of this write-up, if the knowledge-rules of the world, applied in the second type were uncommonsensical, the inference would still have been the 'meaning', as long as the thinking involved in the drawing of the inference from them,applied on the given data, was not "more" or a "complex", or in other words, was obvious. 

Now, this previous sentence raises a question - what if there is a sentence, understanding whose "meaning" is tough. In that case, would you call the 'meaning' of that sentence, (inferred from the rules of language), as an 'implication' (because it's tough and not commonsense)? No, we would still call it the 'meaning' of the sentence, because the meaning of X is what X stands for, and the understood tough meaning of the given sentence would still be what the sentence (given in words) stands for (in reality). Implication always comes AFTER 'standing for', however difficult the drawing of the inference of the 'standing-for"' is.

3 kinds of commonsense -

1) Embedded in the structure of language itself - 

This has been explained in the case of the "write a letter"-example sent before (a few days ago). A letter is a physical paper body. You can cut it, throw it, fold it. But how can you "write it"? How can you "write a physical body?" But this shows that language itself has taken shape so as to include this commonsense, wherein writing a letter means writing text/contents on a piece of paper.

2) Nothing to do with any Linguistic theory / (Not embedded in the structure of language) - 

This is the conventional commonsense in AI literature. "The police threw tear gas at the demonstrators because they feared violence". Who is 'they' here? Identifying this has nothing to do with any Linguistic theory, but your commonsense knowledge of the world about policemen, demonstrations and demonstrators etc.

3) Having both of the above aspects - 

Consider this sentence - "This is that shoe's sock". 

When we say this, why are we using the " apostrophe 's' "? It is because the commonsense that A particular shoe belongs to A particular sock and the other to another (i.e. the right one to the right and the left one to the left) is absorbed in the way language is formed - apostrophe 's' indicates the association of a shoe with its right sock. This is one kind of commonsense (type 1).

The other kind of commonsense here is that some real information that is assumed and well-known is hidden. In the literal utterance "shoe's sock" it is meant "the sock worn in the leg that very shoe is worn in". This is the conventional (type 2) commonsense involved here.

An idea to define 'Meaning' -

Meaning of a word is how it relates other words to each other, with those other words taken in a general sense.

For example, what is the meaning of a table? It is how it relates 'four vertical legs' and 'a top flat surface' to each other. The meaning of the 'table' lies in this relationship.

 

Let me explain this in an even stronger and simpler way. What is the meaning of a hyphen ('-')? As in, when we say : The golden rule - "leave it", (this is in the context of Psychological tricks/tips to counter your problems/stress etc., what does the '-' mean? It means that 'leave it' IS THE 'golden rule'. What have I done here? In defining or explaining the meaning of '-', I have related, I have expressed the relationship between, 2 other things ('the golden rule' and 'leave it'). And also, this is to be done in a general sense i.e. we should say "when there is 'X -Y' the '-' means that Y IS X". Similarly, the word 'go' means the relationship between 'something (some entity/concept)' and 'displacement away' with that relationship being that 'that something's displacement away'.

"NEWTON'S 3RD LAW" IN LINGUISTICS -

Consider these few typical sentences - 

1) John kicked the ball.

2)The ice melted.

3) The flower is beautiful

4) The phone is on the table.

Whatever you utter in language, it will be in one of these 2 'forms' - 1) A-outward, denoted by A-> OR 2) A-inward, denoted by A<-

What do these forms mean? The first means that from 'A', something is going outwards, coming out. When John kicks the ball, the kick, as an action, a force, comes "out" from John. The second means that something is coming upon / happening to 'A'. When ice melts, there is an effect "upon" the ice. Hence A-inward. When I say the flower is beautiful, the flower "hosts"/"bears" the beauty. So it is A-inward. When it is said that the phone is on the table, the phone is bestowing (or one could say "acting") ON the table. So it is A-outward.

In the case of 'A->' there will always be an effect (A'<-) simultaneously. John will receive a force exerted by the ball on his leg (by literally Newton's 3rd law). In case of 'A<-', there will be a corresponding action or effect done by A (A'->). The ice will suck in the heat from the surroundings (while melting) and make it cooler. In the case of the flower being beautiful, the corresponding A'-> is that the beauty is spread into the surroundings by the flower. In case of the phone being on the table, the A'<- is that the table supports/holds the phone (it exerts a reaction force on the phone, in Mechanics' terms). 

INTERFERENCE - Semantic spatial overlap

Consider 2 chunks of words, following each other, in a sentence.

E.g. - 1) (Break the tree) + (with an axe)

2) (Drive the car) + (on the road)

3) (Watch the match) and (cheer for the team)

The bracketed sets of words are the chunks.

In some cases, reversing the order of the chunks (or changing the order of multiple chunks in a sentence) preserves the meaning of the sentence. In some cases, it doesnt. In examples (1) and (2), the meaning is preserved. 'Break the tree with an axe' and 'With an axe, break the tree' mean the same. Similarly, in the 2nd example. In example (3) it isnt. 'Watch the match and cheer for the team' doesnt quite mean the same as 'Cheer for the team and watch the match' since the order of events changes the semantics in the 2 cases. Whats the difference? When there is "INTERFERENCE" (semantic spatial overlap) of the chunks with each other, the meaning of the sentence isnt preserved; it is affected by the order. When there is none the meaning is preserved. 

What is this "interference" of? It is of the "semantic VIDEOS" of the chunks. 

Explanation : Lets see it all in the context of the examples above.

1) What is the semantic video of 'break the tree'? A tree breaking. While constructing the video of a chunk, do not factor in or involve the other chunk. What is the semantic video of 'with an axe'? Hands holding an axe with a slight one-second movement of the axe and the hands in the air, away and besides from the tree. (The word 'with' is responsible for the one-second movement). These 2 videos are at DIFFERENT places in space. So there is no spatial overlap. Hence there is no interference. Hence the meaning is preserved upon changing the order - 'Break the tree with an axe' or 'With an axe, break the tree'.

2) What is the semantic video of 'drive the car'? It is the sheer literal video of a car being driven (even the road is missing). Only a moving car, being driven by someone. What is the semantic video of 'on the road'? A hypothetical finger being pointed (with a slight jerk!) to the top surface of a road signifying "on the road". These 2 videos are "disjoint" in space (though touching each other; but not OVERLAPPING). Hence no interference and no change in meaning upon change of order.

3) Now see the 3rd example. The video of 'Watch the match' is 'eyes staring at the TV screen'. The semantic video of 'cheer for the team' is 'jumping up and down while looking at the TV. These 2 clearly spatially overlap. Hence there is a sort of an interference in the 2 videos playing. Hence the meaning changes in the cases of the 2 different orders of the chunks.

This theory sheds light on how there being a spatial overlap between the 'visuals' of the chunks, which relates to the way the brain processes the event as a whole upon being exposed to the 2 chunks in combination, relates to the Linguistics (Semantics here) of the sentence!

Commonsense embedded in the structure of Language -

Consider this sentence - John is writing a letter.

This means that John is composing a letter and penning down the contents on a paper.

But the question is - how can you "WRITE" a letter? A letter is a physical paper body with text on it. You can cut it, fold it, throw it, but how can you "write" it? How can you "write a physical body?" But the written content of a letter is so central and significant to what a letter means and stands for, that we say that the "letter is written" or "has been written", instead of "the letter has it/something written on/in it" (which would be the explicit meaning), for a finished written letter. So commonsense has made its way into shaping language itself. This is one way. Commonsense has made its way into the very construction of the structure of language machinery. We are all familiar with sentences whose explicit and implicit meanings are different, but that difference arises at the meaning or semantics-stage, not at the level of the very infrastructure of language itself, as is the case here.

The specific piece of commonsense involved here is that 'if X is majorly signified, explained or propertized by Y, then X is (as good as) Y'. That is, you can, say, refer to or call X as Y only, or say, make a form of X to be Y only (directly) etc. 

Classes' as regards to knowledge -

There are 4 kinds of knowledge, based on the 2 kinds of distinctions - 'Commonly known' v/s 'Rare', and 'General' v/s 'Specific'.

Commonly known + General - All cars run on wheels.

Commonly known + Specific - Einstein's first name was Albert.

Rare + Specific - Rwanda's capital is so and so.

Now comes Rare + General. The trick here is as follows. If I say 'cricketers wear black bands on their arms while playing a match if someone from the fraternity has passed away recently', then you will say that yes, this is rarely known but not quite general, because it pertains (specifically) to cricket only. So it should be in the Specific + Rare category. And if you say - "wait, is it so that all sportspersons wear black bands in such cases?", to make it fit in the Rare + General category - it would occur to you - "oh, how come I didn't know such a prevalent/common general fact (prevalent across all sports)?" So it should then be in the General + Common category. In either case, there doesn't remain anything in the 'Rare + General' category. This is a property of commonsense knowledge - there is nothing easily fittable in the 'Rare + General' category of knowledge. 

MEANING at 4 LEVELS

 Consider this simple sentence  - John was qualified.

One representative meaning at each of the "4 LEVELS" - 

Order of the words - It was John who was qualified. The very arrangement of the words, serially, from left to right, in the given order tells that it was JOhn who was qualified.

Definitions - John passed exams.  The definition which the word 'qualified' stands for.                              

Semantics of the sentence -  John was educated. The "general", "overall" meaning of the sentence.            Commonsense Semantics -  John had a degree   . Incorporating worldly commonsense knowledge.

Do we need a dictionary like this?

 Why does a word exist? Every word exists to describe a special reality. 'Fever' exists for an ill-health temperature condition, which is beyond 98.4F temperature. But there is no word for a cut laptop. 

The first part portrays the Commonsense Meaning (ill bodily temperature condition). The second part portrays the technical meaning (T > 98.4F)

What if we have a dictionary like this?

WORD                              COMMONSENSE MEANING                                      TECHNICAL                                                                                                                                                     MEANING

                      (why does the word exist? what special reality does it describe?)  (Meaning in scientifically-                                                                                                                                    studied technical terms)

Fastest                                   Maximum speed, rank 1 speed                                highest speed

Table              man made thing for keeping things at a certain height in space      3D-Geometrical description - flat surface supported by perpendicular legs

Continuation of previous post

The first and foremost fact about or aspect of anything is that it exists. That's the first thing the mind processes upon coming across it. 

The "first" commonsense (the SENSE) about anything is related to its purpose of existence. (Why a table? Why a computer? Why the 'comment' option in Facebook? - why do all these things exist? i.e. what do they exist for?) Thats the first inception of a "thought" after the detection of anything. Hence it (the "first" commonsense about anything) is linked to its existence. The SENSE about a table mentioned in the earlier mail (a table is meant to keep things at a certain height from the ground, in space), clearly relates to the existence of the table. 

Commonsense is NOT about FACTS.

 Commonsense is not about facts. It, being a 'SENSE', is not captured by the commonsense facts or commonsense knowledge.

Consider a table. Firstly, why do you keep a table standing on its legs, and not on its surface? Well, that commonsense. Meaning? Meaning, a table is meant to keep things at a certain height from the ground, in space. Hence a table is kept that way. This is the common SENSE (or the rationale) about a table. This is not captured by, say, the 3 commonsense facts or pieces of commonsense knowledge about a table that - 1. "A table stands on legs", 2. "There is a flat surface on top of the legs of a table" and 3. Things are kept (/a table is used to keep things) on top of it, on its surface.

Why computers don't have language.

 This is a small view of mine.

Can you have 2 entities such that there is no common feature between them? No, impossible. Because both of them will be things. This is the inherent similarity between things, associated with the phenomenon of existence itself - that which exists will be a thing and that will be the similar, common feature to anything that exists.

Now, the way I experience / perceive / deal with /...something in the universe decides my (Linguistic) naming of it. So I can call anything as a thing because there is similarity in my experience of those “anythings”. What is that similarity? The very (possible) phenomenon of mental representation of anything, is itself the commonality/similarity among the various things that exist in the universe. Plus there is similarity in some aspects of all those particular mental representations themselves.

Hence, the fact that I have a commonality in the mental representations of everything that exists in this universe, is the primary enabler/source/facilitator of ‘language’. That is why something like ‘language’ is possible.

Hence, language is impossible without a MENTAL representation. So for computers to have language, we need to first give them requisite “mental” features.

A Commonsense Knowledge Principle.

Consider Commonsense Knowledge - 

- Cars have numbers.

Everyone knows that vehicles have numbers. This is commonsense knowledge. But what is this piece of knowledge doing or talking about? Identifying cars. So in my view that's not much different (from the point of view of commonsense thinking, or from that of using this knowledge in practice) than knowing that human beings have names. Or let me extend this to a general principle of knowledge, wherein all such knowledge falls into - "Whenever there is a multitude of samples of the same thing, they are identified". So animals have names - tiger, lion, giraffe etc. Companies have names, flats have numbers, tickets have numbers etc.

- Consider another piece of commonsense knowledge - Guns have triggers. What is this doing or talking about? The operational cause of a gun. So this is no different (at the level of USING THIS KNOWLEDGE COGNITIVELY) than knowing that cars have keys or ovens have buttons.

This should be used in building commonsense knowledge bases. Knowledge principles should be identified and all relevant knowledge be clubbed in relevant categories.