Steve, I think you need to have a lot of related information about a subject to 'know' one simple thing about it. And I believe that is the only possible explanation for discerning the meaning (or meanings) of a phrase or a sentence. So I do agree with you up to a point. I finally realized that statistical determination of meaning has the utility that it represents a more limited search method since it is only looking at a sampling of possible usages. So I think that over reliance on a sampling is part of the contemporary problem. But I would still say that the goal is to make the most likely cases stand out in order to narrow in on the best meanings.
Incidentally, some words and phrases explicitly denote a broadening of the possible meanings so not all phrases are intended to narrow in on a more precise target. The meaning of a sentence is not the same as the target of the meaning. So a precise meaning might, for example, refer to a broad generalization. - Jim Bromer On Mon, Aug 12, 2013 at 11:45 AM, Steve Richfield <steve.richfi...@gmail.com > wrote: > Supposing for a moment that we think in the multiverse, and that language > carries multiversal meanings between disparate world models, the following > train of logic says that statistical disambiguation is a really bad idea, > except maybe in some odd applications like low-quality translation w/o > footnotes... > > Everyone has been presuming that language states specific meanings, but > lets turn this around and presume that a given statement simultaneously > means EVERYTHING that it could conceivably mean, complete with the usual > Bayesian weightings. Other surrounding statements would also simultaneously > mean everything that they could conceivably mean, but most of those > meanings would be incongruous with the meanings of surrounding statements, > so that only a few, and maybe/hopefully only one meaning for an entire > passage would emerge. Where several/many meanings survive, live with it, > because that is the way of the multiverse. > > If only one of the meanings proves helpful in guiding subsequent action, > and disambiguation were perfect, then this would work the same as a > statistical disambiguation approach. In the vast majority of cases parallel > erroneous interpretations would be no problem, because their erroneous > qualifiers would specify non-existent leaves on the world-model tree, and > so would affect nothing. > > HOWEVER, where multiple applicable meanings survive overall analysis, the > presence of competing meanings would greatly reduce the reliability of any > particular meaning, which is EXACTLY what is needed and what statistical > disambiguation does NOT do. > > As evidence of this process, consider the quote "Try not. Do, or do not. > There is no 'try'." from *The Empire Strikes Back*. This refers to all of > a person's endeavors. Without multiversal meanings, this would be quite > difficult to explain. With multiversal meaning, the intended meaning is > quite obvious without further explanation. > > So, a blank page would simultaneously be simultaneously saying everything > that could conceivably be said - a sort of selection of your entire > world-model. Then, you put some words on the page, and in the process you > zero in on a particular branch in your world-model tree. You put more words > on the page, and in the process select particular leaves in your world > model. > > Your world model is probably as different from mine as two different trees > can be different from one another. So, of course a given set of words > doesn't lead to the "same" branch in our two world models, because there is > no "same branch". This is no problem, because your words will find ALL of > the branches that "fit", and **NOT** just a particular branch that seems to > be the "best fit". > > This suggests an ENTIRELY different approach to programming to > "understand" natural language. > > Any thoughts? > > Steve > > *AGI* | Archives <https://www.listbox.com/member/archive/303/=now> > <https://www.listbox.com/member/archive/rss/303/24379807-f5817f28> | > Modify<https://www.listbox.com/member/?&;>Your Subscription > <http://www.listbox.com> > ------------------------------------------- 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
