Linas Vepštas

Hmm. Confusing. For link-grammar, I am trying to interpret cost as if it was minus the log likelihood (i.e. the entropy or mutual information of something). Viz, cost = -log...

Ooops. Clarification/correction to above: 4. If costs are log of probabilities, then I am mis-stating something in bullet 3. The conclusion is still right, because it is "well-known" that the...

The way I see it,, both `[[A+]] & [[B+]]` and also `[A+] & [[[B+]]]` should be the same as `[[[[ A+ & B+]]]]` and all of these variants should be...

BTW -- a related issued in the sql-back dicts -- the ones that I sent you will have exactly the same disjunct appearing multiple times, with different costs, each in...

> the concept of maxcost may be bogus I agree (I looked at the file) it is bogus, precisely because the cost has to be distributive. The 'Denis' variant kind-of-ish...

I would like to get rid of max-cost, and argue that the original article was faulty. The original max-cost is an idea that almost seems to kind-of make sense, but...

I posted above comments before I saw your last; I'm now confused; hang on.

Oh .. > using cost instead of maxcost How did you change line 176 ? I haven't quite figured it out, but it seems all of `build_clause` suddenly gets a...

> Error: Different number of errors (384 != 398) > Error: Different number of errors (82 != 83) What are these?

Thanks. I'll have to look at that. Clearly, "goes" has a problem, I guess.