On converting natural language text to atomese.

46 views
Skip to first unread message

Mindey I.

unread,
Dec 20, 2017, 11:53:19 PM12/20/17
to opencog
Hello Everyone,


say I'd like to convert natural language specification of goals into atomese, so as to be able to apply the opencog's solvers to ground them, and possibly extend opencog with other types of solvers.

For something specific, let's say we have descriptions of UN sustainable development goals ( http://www.un.org/sustainabledevelopment/sustainable-development-goals/ , https://en.wikipedia.org/wiki/Sustainable_Development_Goals ), which are phrased in human language.

And, let's say we want to convert them into atomese, in such a way, as to precisely capture their meaning.

In other words, we have some texts, rather than spoken language. What would be the current best way of converting plain text in a natural human language to a semantic representation in atomese, assuming global/general context?


Thanks
Mindey

Linas Vepstas

unread,
Dec 21, 2017, 12:13:17 AM12/21/17
to opencog
The current best-way is to just dump the link-grammar parses into the atomspace.  There's an atom for that; you just invoke it and the parses magically appear in the atomspace.

The link-grammar parses are fairly low-level, but accurate. higher-level is relex, but it is less accurate, its fuzzier. You need to run the relex server for that. You pipe it a sentence, and the atoms just show up in the atomspace.

Higher-up is relex2logic aka R2L which attempts to create logical forms for English sentences. Its got issues; the biggest issue is that the PLN solver doesn't work very well on the output that R2L is generating.

There's nothing wrong with starting with the low-level parses, and trying to do something with those.  The higher-level layers, although more abstract, are .. messy.

--linas



--
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+unsubscribe@googlegroups.com.
To post to this group, send email to ope...@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/8a8e4329-5de3-480b-8846-da976b9589b4%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.



--
"The problem is not that artificial intelligence will get too smart and take over the world," computer scientist Pedro Domingos writes, "the problem is that it's too stupid and already has."
Reply all
Reply to author
Forward
0 new messages