Jim and Boris, I think things are done backwards from what you are presuming here...
I think we use placement words (things we already know or can reasonably presume) for context, latch onto things that sound interesting, then "look around" for adjectives, adverbs, object phrases, etc., to tell us about the interesting things. This "solves" the disambiguation problem of what modifies what by presuming that adjectives and adverbs modify the most "interesting" things around them. Sure, this often ends up assigning modifiers differently than a linguist might, but if you start contriving examples I think you will see that people typically work this way, even if linguists don't. For example, "*the flashy red truck...*" does "*flashy*" modify "*red*" or "*truck*"? Almost anything can modify a color, as a trip to a paint store will quickly confirm. Hence, a linguist would decide that " *flashy*" is an adverb modifying the adjective "*red*". However, reflecting on when you first read this, you probably thought that "*flashy*" was referring to the truck. Of course, a paint store owner would probably read "*the flashy red truck*" quite differently than you would, and you are probably NOT going to solve THAT problem through human-level disambiguation, because it bears upon a fundamental defect in language structure. Sure a computer might use some fancy algorithm to apply knowledge about a person's background for disambiguation, but that process would introduce LOTS of errors, and false-positive errors cause great havoc in most logical processes. As I learned when programming DrEliza, there comes a point when the probability of error rises high enough to abandon analysis of a sentence, as the best of several bad options. Once you accept analytical abandonment and start looking at what you SHOULD be abandoning, most of the fancy disambiguation methods that people talk about in papers and on forums goes away. This bottom-up method of parsing also seems likely more "neurological" because it does NOT involve any lengthy search or solution process to resolve. Certainly if a computer is slowed down by a process, we would also be slowed down if we did the same thing. Hence, I tend to dismiss theories built on complex and time-consuming methods of trying many possible solutions until one is found that best fits, as we just couldn't be doing such things in real time as people speak. Hence, my placement/payload approach leads to simple methods of directly parsing complex text that appears to be close to "human", albeit according to rules you won't find in any linguistics text. Steve =================== On Sat, Mar 30, 2013 at 9:06 AM, Jim Bromer <[email protected]> wrote: > No. The idea that they a referent is a kind of search and compare process > does seem obvious. However, the conclusion that sensory inputs is at the > bottom of a hierarchy (of search and compare or something) is not a valid > conclusion. For example, there is the theory of perception where sensory > inputs are believed to be 'defined' (or something) or tagged with higher > level analysis shows that there are solid foundations of non-hierarchal > reasoning. Furthermore, the idea that a system of hierarchy is both > complete and sound is contra-indicated by the evidence of the dismal > results of AGI so far (which is also foundational to your conclusion that > there is a snowball's chance in hell (scih) of the need for sensory > inputs in order for a necessarily hierarchal search and compare process). > If a single hierarchy was sound and complete then traditional logic would > be sound and complete. I think I was saying that an effort to create a > linguistic theory which was able to include a method that determined some > of the referents of a statement that was being analyzed was a search and > compare method which meant it would be complex. Finally, a keyboard (for > example) is a sensory device. It would not make any sense to talk about an > AI program that was not capable of reacting to input in some way (except > maybe as pretty far flung theoretical mathematical thing where the "input" > could be contrived to be derived from the initial input.) > Jim Bromer > > > On Sat, Mar 30, 2013 at 10:52 AM, Boris Kazachenko <[email protected]>wrote: > >> ** >> >> > It is quite possible that the progressive discovery of referents is >> just a search and compare operation... >> >> Isn't this tautological? Isn't it also tautological that, to be >> selective, search ( comparison must be hierarchical? And that at the bottom >> of this hierarchy of complexity are sensory inputs? And that if your >> algorithm can't start from these inputs, then it has a snowball's chance in >> the hell at starting anywhere higher? >> Hello? >> ** >> ** >> *From:* Jim Bromer <[email protected]> >> *Sent:* Saturday, March 30, 2013 9:11 AM >> *To:* AGI <[email protected]> >> *Subject:* Re: [agi] Steve's placement/payload theory of language >> >> So anyway, I think that linguistics has to be involved with the >> progressive determination of referents and how these referents can be used >> to define the meaning of the other parts of an expression. This is so open >> that formal linguistics may not be able to define this well but it can be >> defined in general ways or for general common meanings. Because we can >> provide encodings that, in turn, means that we can use terms in a >> specialized way. (Like when I said that I use the term referent to refer >> to a real world object or a real world event, or to a mental object or >> mental event). People in these groups sometimes become annoyed because we >> can't figure out what they are talking about even though they have talked >> about their ideas numerous times. Part of the problem is that we can't >> recall all the specialized definitions that individuals use. I believe >> that this problem is aggravated because specialization is an important part >> of communicating and even the best of writers rely on this even when they >> are using conventional terminology. >> >> So I think that the major obstacle confronting AGI linguistics right now >> is the discovery of referents. Yes this could sometimes be alleviated with >> multi-modal sensors, but there is no evidence that multi-modal sensory >> methods that would allow an object to be seen and heard or sensed in other >> ways would resolve this problem of deducing what is being referred to. It >> is quite possible that the progressive discovery of referents is just a >> search and compare operation which seems to be major slowdown in computer >> science today. >> >> Jim Bromer >> >> >> On Fri, Mar 29, 2013 at 5:40 PM, Jim Bromer <[email protected]> wrote: >> >>> It is a little difficult for me to answer this question so I will >>> start with one part before I forget. I had to look up sign (linguistics) >>> and agent (linguistics) to get some idea about what he was talking about. >>> I would say that my approach is (or would be/will be) both. In reading >>> about signs I noticed that the idea of the referent is considered to be >>> distinct from the idea of the signified. When I refer to a referent I am >>> referring to a real world thing or event or a mental idea. I believe that >>> the one thing that is missing in modern AI Linguistics is a way to follow >>> what is being referred to. It probably is just too complicated for a >>> computer program to figure out efficiently. My gmail is malfunctioning so >>> I will try to continue this later. >>> Jim Bromer >>> >>> >>> On Fri, Mar 29, 2013 at 1:09 PM, Piaget Modeler < >>> [email protected]> wrote: >>> >>>> Steve and Jim, >>>> >>>> Kindly respond... >>>> >>>> ------------------------------ >>>> Date: Thu, 28 Mar 2013 09:34:07 +0100 >>>> Subject: Re: FW: [agi] Steve's placement/payload theory of language >>>> >>>> From: Roland Hausser >>>> To: [email protected] >>>> >>>> >>>> Hello Michael, >>>> >>>> Thank you very much for your email. I read >>>> >>>> the comments by Jim Bromer and Steve Richfield >>>> with great interest. They lead me to the >>>> following questions: >>>> >>>> * Are their respective approaches sign-oriented >>>> or agent-oriented? >>>> >>>> * What do they think about defining basic concepts >>>> as types of the recognition and action procedures >>>> of an agent? >>>> >>>> * How about reusing these basic concepts as the >>>> literal meanings of a language? >>>> >>>> Happy Easter to you! >>>> >>>> Looking forward to be reading from you, >>>> >>>> Best regards, >>>> >>>> Roland Hausser >>>> >>>> *AGI* | Archives <https://www.listbox.com/member/archive/303/=now> >>>> <https://www.listbox.com/member/archive/rss/303/10561250-470149cf> | >>>> Modify <https://www.listbox.com/member/?&;> Your Subscription >>>> <http://www.listbox.com> >>>> >>> >>> >> *AGI* | Archives <https://www.listbox.com/member/archive/303/=now> >> <https://www.listbox.com/member/archive/rss/303/18407320-d9907b69> | >> Modify <https://www.listbox.com/member/?&;> Your Subscription >> <http://www.listbox.com> >> *AGI* | Archives <https://www.listbox.com/member/archive/303/=now> >> <https://www.listbox.com/member/archive/rss/303/10561250-470149cf> | >> Modify <https://www.listbox.com/member/?&;> Your Subscription >> <http://www.listbox.com> >> > > *AGI* | Archives <https://www.listbox.com/member/archive/303/=now> > <https://www.listbox.com/member/archive/rss/303/10443978-6f4c28ac> | > Modify<https://www.listbox.com/member/?&;>Your Subscription > <http://www.listbox.com> > -- Full employment can be had with the stoke of a pen. Simply institute a six hour workday. That will easily create enough new jobs to bring back full employment. ------------------------------------------- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657 Powered by Listbox: http://www.listbox.com
