Announce: LinkGrammar can read dicts from AtomSpace

[Linas Vepstas]

Link Grammar can now access language dictionaries located in the AtomSpace. That means that if you have language data (and it's in the AtomSpace), you no longer have to dump it to a file or an SQL DB before LG can use it. In my case, I'm trying to learn dictionaries; I can now use them "live", even as they are being learned.

Going in the other direction has always been possible. The ability to get pares from LG, into the AtomSpace, has been around for a maybe a decade, now. The new work closes the loop, making things bidirectional. ... bidirectional, and live, dynamic -- no more batch-processing. The buzzwords are "life-long learning".

The code is alpha: there's 3 or 5 more features and functions that are still needed. Might take some weeks or a month to knock these out. There is a demo, but its stunningly underwhelming, mostly it just proves that it works. I'll try to get a better dataset set up in .. a few weeks, month.

One neat thing about the implementation: in the demo, the dictionary data is living in a cogserver in a docker container (that you too can build "trivially" by following some short instruction). LG contacts this cogserver over the net, and fishes out the data that it needs. So there are *two* atomspaces: the local one directly compiled into LG, and a network-remote one, holding all the data.  I think this is kind of neat, because after all the talk of a "distributed atomspace", here's an actual "production" example that is more than  just another tutorial.

More info here:

https://github.com/opencog/link-grammar/tree/master/link-grammar/dict-atomese

-- Linas

--

Patrick: Are they laughing at us?

Sponge Bob: No, Patrick, they are laughing next to us.