On Tue, Oct 11, 2016 at 9:19 AM, vishnu <[email protected]> wrote:
> Hello all, > > Say, I have the following example sentences > > > - apple is rich in vitamins. > - apple makes the doctor away. > - apple is healthy. > - apple is red in color. > - eva eats apple. > - Steve Jobs invented apple. > - apple iphone is usually costly. > - headquaters of apple inc. is in california. > - apple products are robust. > > > i just parsed these sentences and got R2L outputs. > Some of these sentences are not grammatical, and will fail to parse. or will parse incorrectly, or lead to other strange errors. For example: "The apple is rich in vitamins" or "Apples are rich in vitamins" but *Apple is rich in vitamins. where * is the standard convention in linguistics, meaning "this sentence is not grammatical" The issue is that, for English, some of the meaning of a sentence is conveyed in the word-order, and thus, sentences with bad word order or bad syntax become hard to understand. The use has to think harder -- mybe they are not grammatical, because a joke is being said? Maybe one of the words is mis-spelled? Maybe one of teh words don't mean what I think it means? Grammtical failure leads to mis-understanding and mis-comprehension. For opencog, grammatical failure results in mis-comprehension at the parser level. There is no second pass at a higher layer that attempts to fix such failures (although Amir is kind-of working on this) > > > 1. When you look at the example sentences, you will know that i have > framed all sentences with the word "apple" that comes in two different > contexts. Apple as a fruit and and as a company. I think, Atoms have > multiple truth values depending upon the context. > Only if you use the context link. > (I assume Atoms with the same Context will have similar Truth Values.?!) . > No, they can have any truth value at all. > Is it possible to filter atoms based on some TV depending upon particular > context? > Sure, but you have to write your own code to filter as you wish. > > 2. Like STV, i am unable to see STI in the R2L output. > There is no concept of STI in R2L. STI is a concept used to control the inference engine. It is not used anywhere else. R2L does set the STV, based on the difficulty of finding a correct parse for a sentence. If the sentence was easy to parse, then the parse is given a high confidence. If the sentences was hard to parse, or if one of the parses seems very unlikely or maybe wrong, then it is given a low confidence. If ungrammatical sentences parse at all, then they are given low confidence. > I assume the atom with the highest number of links gets automatically > high STI value. > No. The STI assigned to an atom depends on the algorithm that is doing that assignment. Different algos .. are different. > So is it possible to retrieve top most important atoms based on STI ? > Yes. > ( also may be top atom's related atoms in a similar context). If so, how > can i acheive that? > In C++, the "AttentionBank" stores atoms, ranked by importance. It provides some methods to ask for those. I'm not sure what the scheme and python API's are for this. I assume they're there, somewhere. > 3. The process of forgetting and recovering from the disk is possible only > when i save atoms in postgres. So everytime when i have huge text, i should > parse each line using (nlp-parse ""), convert into R2L output using > (parse-get-r2l-outputs.....) and in turn should store the obtained results > in postgres. > Yes. > Am i missing anything here?. > No, not really, other than that the postgres interfaces remain hard to use. Also, there currently is no way to delete an atom, once it has been put into the database. Attention values are not stored in the database. Attention is used only during reasoning, and the code that does that remains experimental. Different algos for assigning attention are being explored; it is not clear what will work best. This remains a difficult task, so far. > > > > Thanks in advance > Vishnu > > > -- > You received this message because you are subscribed to the Google Groups > "opencog" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to [email protected]. > To post to this group, send email to [email protected]. > Visit this group at https://groups.google.com/group/opencog. > To view this discussion on the web visit https://groups.google.com/d/ > msgid/opencog/f641c4f3-0a91-40bb-bb65-1ca0daaf6689%40googlegroups.com > <https://groups.google.com/d/msgid/opencog/f641c4f3-0a91-40bb-bb65-1ca0daaf6689%40googlegroups.com?utm_medium=email&utm_source=footer> > . > For more options, visit https://groups.google.com/d/optout. > -- You received this message because you are subscribed to the Google Groups "opencog" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To post to this group, send email to [email protected]. Visit this group at https://groups.google.com/group/opencog. To view this discussion on the web visit https://groups.google.com/d/msgid/opencog/CAHrUA37vaSMGBKHM6_KENEtgpxYDZmeKYa734riHpakoe8qEEQ%40mail.gmail.com. For more options, visit https://groups.google.com/d/optout.
