[Korrelan]

http://www.theverge.com/2016/12/5/13846396/lexus-led-lit-is-colors-dua-lipa-vevo

 

http://newatlas.com/vitaly-bulgarov-korean-robot-mech-suit/47057

Ok what I need for the morning commute is this with the led light coverage of the Lexus... so I can display appropriate messages has I stomp on the bonnets of idi*ts who cut me off.

[Art]

Ah yes...'bonnets'...for the English speaking folks across the big pond (aka - the USA), those would Not be the hats of little old ladies but rather the Hoods of vehicles, to which he was referring! Whew!
terminology.

hmm...wonder how many chatbots or A.I. have differentiated or the ability to differentiate between country expressions / dialect, thematics, colloquial usage, origins, slangs, etc.?

Others are lift, flat, widdershins, trainers, jumper, waistcoat, braces...and so on. Whereas here in the States they would be: elevator, apartment, counter-clockwise, sneakers, sweater, vest, suspenders....

Quite a difference for humans not to mention an A.I.

Thoughts?

[infurl]

hmm...wonder how many chatbots or A.I. have differentiated or the ability to differentiate between country expressions / dialect, thematics, colloquial usage, origins, slangs, etc.?

The ability to recognize different dialects is something I've built into my parsing software from the beginning.

Here's a trivial example from the lexicon.

Code

Noun_COLOR := "color" #common_noun #count #american #plain_case #singular
Noun_COLOR := "color's" #common_noun #count #american #genitive
Noun_COLOR := "colors" #common_noun #count #american #plain_case #plural
Noun_COLOR := "colors'" #common_noun #count #american #genitive
Noun_COLOR := "colour" #common_noun #count #british #plain_case #singular
Noun_COLOR := "colour's" #common_noun #count #british #genitive
Noun_COLOR := "colours" #common_noun #count #british #plain_case #plural
Noun_COLOR := "colours'" #common_noun #count #british #genitive

[Korrelan]

Nope… a lot of those cars are driven by little old ladies in bonnets…

Ain’t English a colourful… sorry… colorful language.

 

[infurl]

Ain’t English a colourful… sorry… colorful language.

You can add this to your collection of useful data sources if you haven't already.

http://aschmann.net/AmEng/

[Korrelan]

Holy cr*p... that's a lot of info... lol

I'll work my way through it... to see if anything is useful to me.

Cheers.

[Art]

@ Korrelan...Yep that's pretty much exactly what I said when I first looked at the page! Yikes!!

But what about handling Context, usage? How does your bot (a bot) know when and how a certain word might be used?

Take Boot for example. In England a boot is an enclosed space (often used for storage) in a car whereas we Americans call that a Trunk.

But the word boot, might mean trunk, or in America, footwear as in a work boot or hunting boot. Then there is the Fred got the boot from work today (meaning he was redacted / fired / let go / employment terminated). I going to boot him right in his butt if he doesn't behave!
==========
Trunk is also an elephant's long, frontal appendage above it's mouth. A trunk is a storage place in an American Auto. A trunk is a mid to moderately sized box-like contained with four sides, bottom and top which can be fitted with hardware for locking the top/lid. A tree also has a trunk used for transporting sap and nutrients to the branches and leaves.
==========

So how do we best fit an A.I. with the "understanding" or intention of these types of words, of which there are many?

[LOCKSUIT]

It's the only page guys..........every link links to just a different area of the page. Some link to websites.

Is that like the works of a lifetime of someone's interest on geographical language?

pffffffffffffffffffff

kill it with fire

[Korrelan]

@Art

But what about handling Context, usage? How does your bot (a bot) know when and how a certain word might be used?

Chatbots are definitely not my forte; but I have written a couple of very simple engines...

http://aidreams.co.uk/forum/index.php?topic=6458.0#.WF2gttSLRhE

I’ll have a go at answering this one… don’t laugh lol.

Give the bot a short term memory that stores the major points of the current conversation.

After you have parsed the input sentence and referenced each word/ word pair against the knowledge base; place the results in a short term memory stack and assign a count down timer to each item.  Just store an index to the knowledge base entry.

When adding knowledge/ words to the main knowledge base concatenate/ include all the indexes of the items in the short term memory with currently running timers.  This stores that word/ knowledge within the context of the proceeding conversation.

When searching the context to a word; take into consideration the items in the list with the counter running. There will be several entries in the knowledge base for each word or word pair; each with a separate prefix/ suffix/ variable that donates its use within a specific context.

As the conversation moves forward an outline of the current topic will be represented by the short term memory and word context can be selected based on what came before.  Base your output/ reply on the current contents of the short term memory.

When a timer runs down to zero remove it from the short term memory list.

I think I also had specific sentences that cleared the short term memory… new subject, etc.

 

[infurl]

@korrelan Thanks for posting that, it's an inspiring coincidence.

Yesterday I made up my mind to have a go at creating a SHRDLU (BlockWorld) of my own using the libraries that I've developed. If I can persevere it will be interesting to see how it turns out when developed using modern software engineering and algorithms.

Almost all the original source code is available from the SHRDLU resurrection website for anyone who is interested.

http://www.semaphorecorp.com/misc/shrdlu.html


@art When handled properly there isn't really a lot of ambiguity in English. Most English words have a number of different senses but unless the writer/speaker is making a concerted effort to be ambiguous (e.g. jokes, double entendre, puns, they're an *ssh*le) they are easily decoded.

It's almost always necessary to look beyond the word to determine which sense is intended, and often, even to distinguish noun from verb, but there are a finite number of patterns (called verb frames) in the English language (less than 200) and they are well documented in resources like VerbNet or the English Resource Grammar for example. Semantic parsers make use of these frames to easily distinguish the precise meaning (sense) of a word in a particular context.

Anaphora (what does "it" refer to?) can be a bit more difficult. Many cases can be resolved correctly using an algorithm like the one that korrelan described, many more can be resolved by paying attention to prepositions and semantic properties in verb frames. Some require a knowledge base and reasoning capabilities to resolve correctly (Winograd Schemas)

http://www.cs.nyu.edu/faculty/davise/papers/WinogradSchemas/WSCollection.xml

[Korrelan]

Cool. I've never seen the original Blockworld  code.

 

[Art]

Terry Winograd - wrote the SHRDLU program for his PHD thesis while a student at MIT.

The name SHRDLU was derived from ETAOIN SHRDLU, the arrangement of the alpha keys on a Linotype machine, arranged in descending order of usage frequency in English.

https://en.wikipedia.org/wiki/SHRDLU

I spoke to Terry many, many years ago about his program and found him to be quite an interesting and pleasant person.
I think his program was actually ported over to a somewhat similar version that would run on the Windows platform.

The 3D version along with spoken output would be cool to mess with.