Nil, I talked to Man Hin and Amen about this a bit...
What they said is roughly as follows. Suppose that you feed in the sentence "Pigs are very cute" into the Atomspace, retaining all the outputs of nlp-parse Next, suppose that you feed in the sentences "Dogs are slightly ugly" and "Fred is a dog" retaining all the outputs of nlp-parse Suppose that PLN draws the conclusion corresponding to "Fred is slightly ugly" (expressed in Atomese, not in English... i.e. PLN derives the R2L output corresponding to "Fred is slightly ugly" (let's call this C) ) ... Then I believe that, if you feed C to the microplanner, it should invoke sureal in a way that leads to the sentence "Fred is slightly ugly" getting produced... ON THE OTHER HAND ... if when you feed in "Dogs are slightly ugly" and "Fred is a dog" you do NOT retain all the outputs of nlp-parse in the Atomspace, but only retain the output of R2L on these input sentences -- THEN I think the microplanner/sureal will not succeed at surface-realizing from C ... The point is that sureal needs the links between e.g. the ConceptNode for "dog" and the WordNode for "dog" (which then has links to the link-parser dictionary entry for "dog" ...).... These Atoms are produced when parsing a sentence involving "dog" such as "Dogs are slightly ugly" or "Fred is a dog" ... Later on we will improve the microplanner to do word selection as a separate process ... so that if the ConceptNode that PLN comes up with is not already linked to some WordNode with a link grammar dictionary entry, then the word-selection process will find a set of Atoms that represent that ConceptNode to a reasonable approximation... -- Ben On Mon, Jul 4, 2016 at 3:49 PM, ngeiswei <notificati...@github.com> wrote: > Sorry to slam you with messages. I'm reporting my progress in case someone > comes by and happens to be able to help. > > According to opencog/nlp/sureal/surface-realization.scmIf the SetLink > contains an interpretation it will just output the corresponding existing > sentence. Otherwise it will call create-sentence which ultimately calls > sureal-match which call C++ function do_sureal_match. > > The question then is, what are the requirements of do_sureal_matchin > order to generate a sentence. Looking at the sureal unit tests I need no > test using instances"likes@..."instead of"likes"`, so maybe that's why it > doesn't work. I guess I could generate structure without instances. Yet the > fuzzy pattern matcher will try to select as answers the syntactically most > similar structures to the query, and so I don't see how my inferred > structure could be handed to sureal then. Anyway, more to investigate... > > — > You are receiving this because you are subscribed to this thread. > Reply to this email directly, view it on GitHub > <https://github.com/opencog/opencog/issues/2238#issuecomment-230227008>, > or mute the thread > <https://github.com/notifications/unsubscribe/AFolXFUY1_FVucKLwaxgCEZTwnT89Tf-ks5qSLsNgaJpZM4JEC7x> > . > -- Ben Goertzel, PhD http://goertzel.org Super-benevolent super-intelligence is the thought the Global Brain is currently struggling to form... -- 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 opencog+unsubscr...@googlegroups.com. To post to this group, send email to opencog@googlegroups.com. 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/CACYTDBcsEkKo1z52DkF_sQ-CkRLAS6OHxZwjakgwkoPVJoKSKQ%40mail.gmail.com. For more options, visit https://groups.google.com/d/optout.
