My Harry Workman intelligent agent project explores various aspect of the human mind. Some of the work is similar to other projects described in the forum here. However, I’m wondering if anybody else working on these:
inductive reasoning
abductive reasoning
temporal/spatial reasoning
semantic memory conflict detection and resolution
episodic memory
procedural memory
belief justification
metacognition

Toborman - Aug 11, 2011:

My Harry Workman intelligent agent project explores various aspect of the human mind. Some of the work is similar to other projects described in the forum here. However, I’m wondering if anybody else working on these:
inductive reasoning
abductive reasoning
temporal/spatial reasoning
semantic memory conflict detection and resolution
episodic memory
procedural memory
belief justification
metacognition

Yes I am working on mostly all of this, but it’s very difficult,
I found only some poor material or none at all (but found lots of written unusable nonsenses!)

inductive reasoning is possible with first order logic (FOL), an some logical reasoner (OWL engine) the results are poor, and also the conversion from natural language to FOL is complex and prone to ambiguous results. This can be enhanced using Fuzzy Reasoning, but there is no framework available and to build this from scratch may be a lifetime, also the theories are not perfectly right and mature.

temporal reasoning is rather complex, and I attacked it with some minor success: I can recognize almost any syntagmatic time period or reference as a whole, and operate in a limited fashion, this time info is then ready to be used to generate natural language and make the correct inflection on some answer, also. This works now at my framework in Spanish, and still needs some work to be accomplished for English.

spatial reasoning   is more complex, I did a “conversation position solver” me-he-you-this-that-here-there, etc. and solve simple things, also did a more sophisticated GIS-solver to find GPS location of cities, for Weather forecasting (working with SMS on my country) the GIS Database has almost all cities in the world, common places, airport names, country/county/area names and solves ambiguity with a proximity common sense algorithm (there are a LOT of same-named cities). Actually I did not add this modules to the bot’s framework, but it’s easy to add it as an external web-service function, and this will be it soon.
My current NLP-GLR parser also recognizes actually ‘geographic-place’ named entities like an address or a place description like this: “Rockville 235 3rd stage dept. A, Chicago W, IL” this may check against a GIS database to see if the address may be valid or not.

episodic memory Actually the system wears this kind of memory, in a simple way, only temporal (conversation time) is taken into account, also has a smart/silly ‘forget’ function imitating a medium-aged human mind. Also using this as referential, for example “yesterday he came” is tightly tied to temporal reasoning + parsing, and as soon as I can have the time-model and system perfectly running, I’ll tie all together!

procedural memory is not implemented, have to see the available models for this kinda things!

belief justification & metacognition last stage..!, for a FOL you can explain the way the resolution of the constrain has been done (is unique) or show the simplest one, this involves Natural language Generation and planning, I built a planner and a NLG subsystem, but my Fuzzy-FOL reasoner is still not operational (sorry!)

hope this clarify your query!

AndyHo - Aug 12, 2011:

Yes I am working on mostly all of this, but it’s very difficult,
I found only some poor material or none at all (but found lots of written unusable nonsenses!)

I’m happy to hear that someone else is working on these. I also have found many of these areas ambiguously defined, which is why I’m attempting to create better definitions in the Human Mind Map to help me eliminate the ambiguity.

Thanks for your explanations of each area. I’ll be sharing my experiences in each area in subsequent posts.

Thanks again, and good fortune on your project.

Toborman - Aug 13, 2011:

AndyHo - Aug 12, 2011:

Yes I am working on mostly all of this, but it’s very difficult,
I found only some poor material or none at all (but found lots of written unusable nonsenses!)

I’m happy to hear that someone else is working on these. I also have found many of these areas ambiguously defined, which is why I’m attempting to create better definitions in the Human Mind Map to help me eliminate the ambiguity.

Thanks for your explanations of each area. I’ll be sharing my experiences in each area in subsequent posts.

Thanks again, and good fortune on your project.

I’ll need good fortune (and a small fortune) to en up with something working as I want to!

Today’s Advance:  My system can conjugate (flexion) Spanish verbs in all the 99 ways you can express a verb, like “love” (amar) even in compound ways, also the flexioner is attendant to special issued grabbed from the semantics, if you give him the verb, and a set of attributes, many of them you grabbed from semantic ontology from other ‘part of speech’ chunks. then you feed them to the engine, and voilá, you get the verb perfectly inflected, even deducting the tense, case and modal features, filling in lacking parameters and correcting coherence-mistakes (not all of them but is a start..) for example this is used at my ‘common sense section’ where you can tell the system thing like:

“mañana comer el gato tal vez” (tomorrow eat the cat may be)

This sentence has a bad grammar constuction (both Spanish and English) and in Spanish it is even worse, you have the verb in inifinitive which is a timeless, caseless, modeless inflection

So at the verb you don’t have a clue to figure out and must use semantics grabbed from the other parts, so the output is (trying to infer and seeing that the verb is transitive and there is a lack of a direct object:

¿que comería el gato mañana? (What would the cat eat tomorrow?)

You see here that the verb is correctly inflected in future (because of tomorrow) + subjunctive (because the doubt ‘may be’) and the sentence is well constructed grammatically (due to a well-formed template)

This is the kinda things I am solving (slowly and a step at a time)

regards, and thanks for the wish!

PD: I am also modeling emotions, using Plutchik’s theories, they were very useful and I built a framework to include a “emotional human soul” which can be assessed with several kind of situations and the resulting state is similar to the emotion a human may show in similar situations, it has been very funny to make this!!

I enjoyed it because as I tested the beast, the beast got angry, then frightned, then happy, then inloved, and as the output was modeled using my Linguistic Fuzzy Set the outcome in words under certain situations, were really awesome, funny and sometimes astonishing!  some of them were

- I feel unhappy / I feel angry / I feel a little confused, etc.

There is a whole math behind, and the stimulus is also complicated, for example you must take something from him, so he feels the loss and becomes sad, or give him something valuable, so he becomes happy! it may be a good shrink-game for an Eliza bot.. haha!

author=“AndyHo” date=“1313192215

Today’s Advance:  My system can conjugate (flexion) Spanish verbs in all the 99 ways you can express a verb, like “love” (amar) even in compound ways, also the flexioner is attendant to special issued grabbed from the semantics, if you give him the verb, and a set of attributes, many of them you grabbed from semantic ontology from other ‘part of speech’ chunks. then you feed them to the engine, and voilá, you get the verb perfectly inflected, even deducting the tense, case and modal features, filling in lacking parameters and correcting coherence-mistakes (not all of them but is a start..) for example this is used at my ‘common sense section’ where you can tell the system thing like: