Robot Language 

 

 

 

Class notes:

 

This is an outline for a minimalist language that a robot may be able to use. In this language, there would have to be a small amount of verbs, maybe 5-10. There could only be about 10 times as many nouns as verbs. There would also be prepositions and definite articles.

 

This language would be mostly speach acts performed by the robot and it would have to have some kind of perspective talking information of 'I' or 'he' in order to facilitate communication between the user and robot and inter robot communication.

 

An interesting phenomena is that of the double property of the self. Namely, regarding the self as a body and as a mind. This disticntion parallels the double pane hypothesis of consciousness that the subject is simultaneously experiencing raw sensory data that contains an angle of viewing and a mental representation of the space the subject is in. This distinction aligns with the first person (subjective) vs. third person (objective)  perspectives.

 

An idea to start creating this language would be to think of 10 tasks to convey to the robot and think of which words would be necessary in order to communicate this idea.

 

Class 4/10:

 

Notes and Examples from 4/10

 

The case of the embedded clause:

 

Sentence: Go to the box that is red

    Node 7 can be nominalized in the sentence the red box. The 2 'red' comes right under the box in circumstance that does not have go. A kind of horizontal transformation.

 

Sentence: The red and blue and green cat with a blue bell and.....

 

cascade of 8's.

 

It is like a long check list that looks to see if it has the certain features. Sometimes teh specification comes in an embedded clause. This embedded clause can be un-embedded to come directlybelow that which it describes like in the case of 'go to the box that is red'-> 'go to the red box'

 

Sentence: Give me the ball that I gave you.

 

Sentence:

Horizontal transformation: the cat that is red and that is blue and this is......

    -This situations is a cascade of 7's.

 

Class 4/15:

 

We reviewed Searles take on speech acts. We discussed that a robot should have an understanding of speech acts in order to be able to sucessfully interact with a human user and other agents in its environment. Searle has five categories of speech acts. The categories are:

 

Assertive	Statements,descriptions
Directive	the hearer is involved, assuming a certain duty to you, dvising, admonishing, asking, begging, dismissing, excusing, forbidding, instructing, ordering, permitting, requesting, requiring, suggesting, urging, warning
Commissive	The speaker is involved, agreeing, guaranteeing, inviting, offering, promising, swearing, volunteering
Expressive	Apologies, thanks, congrats
Declarative	You are fired.

 

 

Class discussion acknowledged that in order to have a sucessful robot language, it would need some understanding of the speech act system. There has to be some way to threaten, or make the robot comply. Let the agent know that if it does not do this, then something will result. In order for this kind of reasoning to occur the agent must have a self representational system. It must know that it is a thing that is obligated to do some act. Also, a temporal understanding is implicit in this kind of knowledge. It must understand that things happen sequentially and that if it makes a promise, at some point in the future, it must fulfill its promise.

 

Self representation is not yet implemented in the agent, but that is the next step in the progression of the programming aspect.

 

We also discussed the talking bots that one can chat with on the internet. We came to the conclusion that they utilize a key word search and implement a series of scripted responses that are programmed to be executed upon certain situations. The robot does not have any knowledge and if you asked it "what are we talking about" it would not have a meaningful response. Here is an example of ELIZA (talking bot) talking about a movie with a human.

 

Chomskian linguistics was brought up and we discussed the difference between that system and stemmatic syntax. Chomsky looks at relations between words and puts a label on certain patterns that occur such as a verb phrase or a noun phrase, whereas stemmatic syntax nodes are cognitively motivated. For example, stemma looks at certain domains that are utilized in communication and groups them based on how each word is used in a context. This allows the rules of this system to be programmed into a computer resulting in a schematic understanding of the sentence based on which nodes are used. If one were to program in Chomksy's system, a tree would result, but the program would have no meaning of the sentence whatsoever, it would only be guided by patterns that it has seen the past and applying those patterns to the current text.

 

A few task examples from Huston's presentation:

 

Push this ball to the arch.

Give me the red object.

Non-give(take) the thing from him.

    -We could negate certain verbs in order to cut down on the verbs we have to program into the agent.

Report (how are you) feeling.

Give the non-clean ball to me.

    -We could also try negative adjectives.

Push the non-red non-square thing.

    -Mabe even negating nouns to specify others.

Take the he ball to she.

    -Pronouns to specify possesion.

    -Or it could be:

Take the ball of him to she.

    -Some languages work like this. The house of Joe. Casa de Joe. Or House mine in french: Chez moi.

 

Some ideas for verbs are:

 

-to do

-to be

-go

-take

-push

-put

-come

-give

 

Some ideas for prepositions are:

 

-next

-under

-over

-to

-from

-between

-in

-on

-with

-before

-after