Jim, English has a basic SVO structure. However, there is a lot that doesn't fit into SVO, or does fit into SVO but doesn't mean what it says.
The challenge is to tease this information out, without having it shredded by the parser. With BNF or other similar way of defining syntax, you can write syntax equations to recognize anything you can imagine. I add flags, the ability to select scope for any test, etc., to completely unlock people's syntax analyzing imaginations. OTOH, once you wed yourself to a particular narrow structural model (like LA grammar), everything MUST be shoehorned into that grammar. My own structural model is much looser and pretty well fits everyday speech. In short, everyday English is NOT an LA grammar. With narrower models, everything MUST be (one or more) recognized parts of speech, and must hang together with other things that are also recognized parts of speech. This is GREAT for people who write grammatically correct English, but this applies to only around half of the sentences found in the wild. Getting beyond "grammatically correct" requires bottom-up recognition of aberrant structures, sometimes substitution of plain English to replace idiomatic English, etc. There is no logic to many idioms, so these must be handles in an ad hoc manner, with a special rule for each illogically structured idiom. Continuing On Sun, Mar 24, 2013 at 4:47 PM, Jim Bromer <[email protected]> wrote: > Steve said, > Beneath a lot of words I have made a really simple statement - that for a > computer to grok the semantics of what was said, it must do SOME sort of > pattern matching, be it with triggered rules as I have outlined, or in a > database as Roland has outlined. Triggered rules have NO complexity > restraints, whereas databases are restrained by the limits of their > indexing systems. > Oops, "triggered" was a red herring. This applies to rules in general, not just triggered rules. > > Can you explain this to me? Triggered rules do the pattern matching in > your model? > Sure. Rules will be stated in some sort of BNF-like syntax, that allows them to recognize just about anything. The triggering mechanism will be completely hidden from view, waiting to cause each syntax equation to be evaluated if its least frequently used word or other element is present, instead of triggering its evaluation whenever another equation wants its result as is now standard practice. > What does that mean? > When you let linguists write syntax equations, there are NO LIMITS in the complexity of the analysis. Models like SVO can guide programming, but programming can deviate as far from such models as is necessary to process the language. As I understand LA, it starts coming unglued when the language deviates from the LA model. Models should guide programming, not restrict it. > How does that work? > Very well I hope. Steve ==================== > On Sun, Mar 24, 2013 at 5:35 PM, Steve Richfield < > [email protected]> wrote: > >> PM, >> >> Here's the challenge: >> >> Everyone has a theory of NL, including me, which I outlined in the patent >> application. My theory, which involves the interchangeable placement of >> locating and payload words, is MUCH more directly programmable than other >> theories I have seen, and further, is consistent with the human ability to >> without difficulty understand fractured statements. Hence, before digging >> too deeply into new theories, I tend to first look at their advantages >> relative to my own theory. >> >> We should be discussing things, not throwing books and papers back and >> forth, especially in our fast moving world where right now in the spring, >> publication deadlines are passing almost on a daily basis. >> >> Beneath a lot of words I have made a really simple statement - that for a >> computer to grok the semantics of what was said, it must do SOME sort of >> pattern matching, be it with triggered rules as I have outlined, or in a >> database as Roland has outlined. Triggered rules have NO complexity >> restraints, whereas databases are restrained by the limits of their >> indexing systems. >> >> Note that DrEliza.com is built in VB, which is a pretty thin wrapper on a >> JET SQL database. IMHO with lots of experience, databases are TOO SLOW for >> practical application. I got past some of the limitations in DrEliza.com by >> having an initialization step that reads EVERYTHING in the DB. It sits >> there for a full minute doing nothing, and then it is ready for action once >> everything has been read into the DB cache. However, DrEliza.com is just a >> small demo of a MUCH bigger future implementation, where this approach >> wouldn't be practical. >> >> Looking for some way of picking apart various approaches, I saw that the >> handling of idioms was key, because they are a problem for ALL approaches, >> and further, they were SO common in everyday speech and writing. Hence, I >> keep asking about idioms. >> >> This paper seems to be yet another paper about ontologies, but this >> doesn't bear on my stated concerns - that everyday English (and other >> languages) are only semi-grammatical in a way that short-circuits most >> proposed parsing methods, with or without ontological analysis piled on top >> of them. >> >> And, yes, I am also guilty of throwing books and papers at people, >> usually when I don't have any good answers. Hence, I have learned to >> interpret this action in others. >> >> So, either Roland is going to discuss this, or I will set his methods >> aside for having the defects that they appear to have. >> >> Steve >> ============== >> > *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
