You're shadow boxing, Jim. It's refreshing to see that someone else besides myself has noticed the difference between the truth (what we confidently believe) vs. The Truth (which is something we can never be sure we've discovered). I built this into my system through soft, non-Boolean truth values.
As for recognizing a definite set of prepositions, you act as though I claimed the same preposition is treated the same way, regardless of context. If "in" means something different when talking about sets (containership, as in "it's in the box") than it does when talking about money (possession, as in "we're in the money"), why not recognize the different meanings of "in" based on what it connects, instead of assuming a universal set of inference rules for "in" relationships? I see no reason to duplicate that information and complicate the implementation by making the information explicit in the form of additional, more specialized link types. Aside from that, I used the same mechanism for prepositions as I did for "kind" words like nouns and verbs: the word node is related to multiple kind/class nodes, and the preposition itself is an instance of one of these kinds. This means that I can easily differentiate between different meanings of "in", by preferring one kind over another in the network's structure. Thus prepositions are not among the list of hardcoded links in my system. Hardcoded links consist primarily of grammatical relations such as the link types corresponding to "is a modifier of" and "is the complement of" with which I string together the 3 part object/verb-preposition-complement prepositional relation, as I tried to demonstrate with my ascii diagram. -- Sent from my Palm Pre On Oct 22, 2012 7:19 AM, Jim Bromer <[email protected]> wrote: [email protected] <[email protected]> wrote: there are certain grammatical categories that have a very small set of words that doesn't expand, and other categories where new words can be easily created. Prepositions, determiners, and particles belong to the first group... -------- This is a mistake that you should not have made. When I talk about conceptual relativity I don't mean that there are no fundamental truths. I mean that we have such limited cpacities that we cannot walk around thinking that we can declare something to be a fundamental truth that we completely understand. Look a prepositions...A preposition (usually) refers to relative position of objects (objects of thought.) So anytime you mention position a preposition might be inferred. Since many things from our discourse about the real world-universe exist or take place somewhere that means that position may be inferred or even explicitly detailed in much of our comments. Many ideas which take place in our minds are stories that deal with place. And (this is where the trap door opens underneath your authoritative declaration) since we can use metaphor to describe ideas that means that most of our ideas contain references that can be inferred or which are explicitly implied to refer to relative positioning. So while the list of propositions may be conveniently finite, the number of possible inferences or implications of the character (or category) of relative position are uncountable. (According to my metaphor your authoritative declaration about the small unexpanding set of words that constitute the category of prepositions has been de-elevated. Even if it is technically true that there are only a few prepositions the number of phrases and sentences that can refer to the character that a preposition defines are uncountable.) Jim Bromer On Sat, Oct 20, 2012 at 3:28 PM, [email protected] <[email protected]> wrote: Please expand a bit on how "a word-object can also become an abstraction of a relation or part of the definition of a process of abstraction". I'm not sure I follow you. As for the simplification process, I don't see why that's necessary. Using 3 link base-level types -- "source of link", "sink of link", and "type of link" -- I can easily represent an expanding set of link labels at a higher level of abstraction. The link and it's label/type become nodes of their own. (Call them "meta links" if you like.) This means I can even relate the links or link types to each other. Forgive the awful ascii art, but here's a visual representation: (source node) | [source of link] | V (link node)<--[type of link]--(link type node) ^ | [sink of link] | (sink node) Nodes are in parens, (), and base level link labels are in square brackets, [], in case my diagramming skills make that less than obvious. Looking at human language, there are certain grammatical categories that have a very small set of words that doesn't expand, and other categories where new words can be easily created. Prepositions, determiners, and particles belong to the first group, and nouns, verbs, adjectives, and adverbs belong to the second. I would argue that prepositions represent built in relationships that the human mind recognizes, which correspond to standard link types in the semantic net. Determiners and particles become predefined properties or labels that get applied to nodes. And the nouns, verbs, etc. correspond to "kind" nodes, to which "instance" nodes can be connected. These instance nodes are then labeled and linked according to the limited set that the human mind recognizes. In my implementation, I use meta links for prepositions, but base level links for direct grammatical relationships like "is the subject of", "is the complement of", etc. There's no reason (aside from efficiency) why I couldn't switch to strictly using meta links. -- Sent from my Palm Pre On Oct 20, 2012 8:25 AM, Jim Bromer <[email protected]> wrote: Aaron Hosford wrote: I do think that reasoning and learning should always be running in parallel to the behavioral and perceptual processes, and should be able to step in and make adjustments when appropriate. That's the reason for going with a universal format for all information processed by the system, namely semantic nets. I think that the data representation has to be simple because AGI is going to be so complicated. However, I totally disagree with the -conventional notion of a semantic net-. The idea of a semantic net is that of a network based on a simplification of the categorization of relations of the word-objects of the network using a few 'kinds' of abstractions to characterize those relations. Now you might say that the idea of a semantic net could be improved on to make it capable of representing potentially more profound insights, but my view is that it cannot be made to fully accommodate the full extent of the meaning of words because if it did it would not be what we typically think of when we think of a semantic net. I don't think you are *just* talking about a "universal format", but of a heavy simplification process. So whereas I do think that a simplifying process is necessary and I do think that a universal format and something like a semantic net is a good way to go, I am not talking about a traditional kind of semantic net in which the relationship between words is found by a single abstraction or by a handful of abstractions of the relations between words and referential objects of the words and sentences. This kind of semantic net was based on a superficial analysis that indicated that the relations between word-objects might be simplified using an concise list of abstractions. I am thinking of a relativistic semantic net where a word-object can also become an abstraction of a relation or part of the definition of a process of abstraction. Jim Bromer On Thu, Oct 18, 2012 at 10:27 AM, [email protected] <[email protected]> wrote: Co-occurrence was really the wrong word. I forget it has the bag-of-words connotation. I imagine an efficient lookup could be designed by using a hash table with hash values based on a bag-of-words approach, but actual recognition would have to be based on the structure of the sentence, as you say. Anaphora resolution is designed into the system. The system doesn't pick a single object that can be matched by a pronoun. It picks a list of them based on recency of use, and links the pronoun to each of them via links with strength based on recency. It then performs higher-level analysis based on the object attributes indicated by the pronoun and the context in which the pronoun is used. Reasoning, which is as yet unimplemented, will be able to step in and further modify these link strengths based on additional information garnered from inference. This approach does produce some combinatorics, but with a reasonable upper bound dictated by the size of the recency list, which can be set to something comparable to the limits of human pronomial references and still be well within the computational constraints of the system. Interesting that you mention higher-level structure to the conversation being important to understanding. I recently read an article about a research team building a system that does exactly that, using a template-based approach. I am probably wildly wrong, but I *think* it was a fellow named Wilson and the system was named GENESYS. I'll look it back up and get you something definite here in a bit. I do think that reasoning and learning should always be running in parallel to the behavioral and perceptual processes, and should be able to step in and make adjustments when appropriate. That's the reason for going with a universal format for all information processed by the system, namely semantic nets. AGI | Archives | Modify Your Subscription AGI | Archives | Modify Your Subscription AGI | Archives | Modify Your Subscription ------------------------------------------- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-c97d2393 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-2484a968 Powered by Listbox: http://www.listbox.com
