Re: [bpfk] Improvements to fragments in ilmentufa parser

2015-03-28 20:43 GMT+03:00 Jorge Llambías <jjllambias@gmail.com>:

On Sat, Mar 28, 2015 at 4:05 AM, Gleki Arxokuna <gleki.is.my.name@gmail.com> wrote:
2015-03-28 1:21 GMT+03:00 Jorge Llambías <jjllambias@gmail.com>:

I missed the part where "noi mo cu co'e" became grammatical.

It has two possible explanations:
a). It means {zo'e noi mo cu co'e}
b). It is continues by one speaker a sumti said by another speaker

OK, but (b) seems to be part of a pre-parser interpretation strategy, that has to decide whether a given string is to be fed as is to the parser or needs to be first concatenated with some other string before being fed to the parser. I don't see how PEG can handle that pre-parsing stage. (b) seems to be about what to feed the parser as input rather than about what the parser should output given some input.

And that's what I'd like to have. That's why I'm against fragments in PEG itself. {mi penmi lo pendo - be ma} - this {be ma} fragment isn't able to restore the full tree.

sumti_4 = expr:(sumti_5 / relative_clauses / gek sumti gik sumti_4) {return _node("sumti_4", expr);}

This could be dangerous, as it makes "ta prenu poi do sisku" grammatical, but not with the expected meaning.

I don't know what is the expected meaning here.

Something like "ta me lo prenu poi do sisku", a relatively frequent error, I think.

It can be restored to {ta prenu zo'e poi do sisku ke'a} or instead to something like {fasnu fa lo nu ta prenu poi do sisku ke'a} or instead as {lo ta prenu poi do sisku ke'a cu co'e}.

Did you mean "fasnu fa lo nu ta prenu _kei_ poi do sisku ke'a"?

It doesn't matter. This phrase (without {kei}) officially (in yacc) parses as

(fasnu {<fa [lo ({nu <ta [prenu VAU]> KEI} {poi <do [sisku VAU]> KU'O}) KU ]> VAU})

Maybe you meant it should be restored to "fasnu fa lo nu ta prenu _ku_ poi do sisku ke'a"? This is another option indeed.

Otherwise you just have the same original "pa prenu poi do sisku" inside a "nu".

But I'm confused how you would get either of those two from "ta prenu poi do sisku" where "poi do sisku" occupies an argument slot of "prenu". Is the stand-alone relative clause to be taken as an ordinary sumti or not? If yes, isn't it just filling the (potential) second slot of "prenu"? Wouldn't "mi citka poi plise" mean that I eat an apple?

I don't provide answers here. I ask myself how we should deal with it and should we deal with it at all.

My personal preference is that it's {fasnu fa lonu ta prenu poi do sisku ke'a} whatever that means. Although, I'm bad at distinguishing inner and outer relative clauses so maybe I'd prefer {fasnu fa lonu ta prenu ku poi do sisku ke'a}.

I started this discussion not to issue any decrees but because few people seem to be working on PEG when it's obvious that it can and should be improved.

Also things lika {da poi prenu ku'o noi melbi".

How is this supposed to parse according to you? I can see it again either as restoring {zo'e} or {fasnu fa lo nu} or as {lo da poi prenu ku'o noi ke'a melbi cu co'e}.

With your proposed sumti rule, I would understand it as two sumti. I'm saying that's confusing, because it looks like a single sumti with two relative clauses.

I don't insist on restoring {noi melbi} into {zo'e noi melbi}, but I would never guess that it is {zi'e} that is omitted here.

If no one including me (oh well, only two people here) like {broda noi melbi} as {broda zo'e noi mo} then let's think this option is rejected.

However, I would like it to be restored to {lo broda noi melbi}. If people make mistakes it's because how they learnt or assumed something incorrectly from existing books.

An English phrase "Does which is beautiful" is {[fasnu fa lo nu] zukte noi melbi} or {[fasnu fa lo nu] zukte ku noi melbi}.

If relying on English (and some other European languages) doesn't matter here at all then I can see only one possible explanation of this common mistake: people have been learning gismu from their glosswords incorrectly assuming that gismu may mean nouns whereas they never mean nouns but predicates or (very roughly) verbs. {i mlatu} never means "a cat" but "It is a cat".

I think it's safer to require a "lo" for bare relative clauses: "lo noi pendo cu melbi" (this was also discussed as a good alternative to "poi'i" in many cases).
This is related to split sumti like {fa lo gerku fa noi pendo mi cu bajra} where instead of {fa lo gerku} you have {fa zo'e}.

I guess what I'm saying is that relative-clauses feel more like selbri than like sumti to me, so that using them bare as sumti feels weird.

ca'e di'u assumed as an axiom until someone else raises their head.

I would expect "fa lo gerku fa lo noi pendo mi cu bajra", not "fa lo gerku fa noi pendo mi cu bajra".

As for you suggestion it's a continuation of the official {lo noi pendo ja'a melbi cu co'e} so it requires separate commits (do you have ideas where to patch the peg, btw?)

Maybe something like:
  sumti-tail-1 <- quantifier sumti / quantifier? selbri? relative-clauses?

Are you able to test this suggestion? I tried, it didn't work for me.

I haven't thought out all the ramifications, but this should allow things like "ro poi broda gi'e brode cu brodi" and "lo ci cu broda".

la xalbo suggested having a list of test sentences so that we can quickly test whether some change affects any reference test sentences (it shouldn't affect to preserve or even increase backward compatibility).