Re: [bpfk] te sumti detection using PEG

2015-04-08 18:15 GMT+03:00 Alex Burka <durka42@gmail.com>:

I'm impressed that you got it this far, but as I've said before I really don't see the PEG as the place for this. First of all, it's mixing two separate steps in the interpretation of a sentence (namely parsing and sumti place resolution).

I think this is the same step separated only in our minds. If it shouldn't be called "parsing" then okay.

What concerns me is that one can create any arbitrary system of te sumti resolution especially when you put ZAMs after FAMs.

Programmers would then say {i'asai} to such BPFK decision and implement this system on top of PEG no matter how unnatural it would be for human brain.

I used only PEG to study the core of the language itself to see what system would require as few conceptually new rules as possible (copy-pasting existing rules doesn't count).
I already explained my vision of some results of this analysis in FA-autorestoration thread.

And as you said, this basic functionality requires 60 new rules ... and it has severe limitations as a te sumti detector, since it gives up after the first explicit FA, doesn't interact with SE/GIhE/BE/JAI, etc.

No longer 60 rules due to optimizations and on the opposite due to supporting new features.
Currently, everything excluding putting ZAMs after FAMs is supported from x1 to x5.
Here are some more examples:
mi do ti bai do ta gau mi fe do gau ti klama gau tu mi =>
([{FA mi} {FE do} {FI ti} {bai do} {FO ta} {gau mi} {fe do} {gau ti}] [CU {klama SF} {<gau tu> <FIhA mi>} VAU])

cusku zo coi fi mi =>
([FA ZOhE] [CU {cusku SF} {FE <zo coi>} {fi mi} VAU])

mo mi ti do tu gi'e co'e ta tu =>
([FA ZOhE] [CU {mo SF} {FE mi} {FI ti} {FO do} {FU tu} VAU] [gi'e {CU <co'e SF> <FE ta> <FI tu> VAU} VAU])

Also since CLL asserts that bridi head can never be empty this is implemented:
i carvi =>
(i [FA ZOhE] [CU {carvi SF} VAU])
This ensures x1 exists.
In English this carvi1 is "It" (it rains). In other languages it is zero-marked.

Like I said, good proof of concept, but I'd be surprised if this is the route to a general te sumti detector.

PEG.js can indeed be the route to it since here PEG is coupled with _javascript_, however, I didn't use any _javascript_ except output of strings and nodes as in the original camxes.js.

This is the proof of concept that having a variable memory is not necessary for the language to work up to some point.

For detecting more places higher than x5 it'd be desirable to get memory by storing what's needed in _javascript_ arrays however I doubt very much the language needs more than 5 arguments in functions. If they are needed in e.g. emulation of programming languages I suggest that FAMs are used instead of ZAMs.

There've been different requests to allow empty {lo ... ku} sumti. This all results in some funny long outputs from very short inputs:

lo =>
([FA {lo <COhE SF> KU}] [CU {COhE SF} VAU])
lo lo =>
([FA {lo <(¹lo [COhE SF] KU¹) (¹COhE SF¹)> KU}] [CU {COhE SF} VAU])

lonunoi =>
([FA {lo <(¹[nu {<FA ZOhE> <CU (²COhE SF²) VAU>} KEI] SF¹) (¹noi [{FA ZOhE} {CU <COhE SF> VAU}] KUhO¹)> KU}] [CU {COhE SF} VAU])

I have no idea whether we need it. In some cases {fa zo'e} and selbri autorestoration required changing the choice order of subrules to test.
I also removed one rule treating {sa} in linkargs. If the community thinks {sa} is important I will work on it.

At this point my ToDo list of altatufa "parser" is empty so my work is done unless new features are requested, unnoticed bugs discovered or optimizations or prettifications of the code are envisioned.

On Wednesday, April 8, 2015 at 8:50 AM, Gleki Arxokuna wrote:

Terminology:
*FAM - a term taking a FA-position with FA explicitely filled with {faxiveimo'eko'a} where mo'eko'a is a precise number (e.g. not {xo'e}
* ZAM - a bare term taking a FA-position with FA omitted. Positioning rules can restore exact value of ko'a in {faxiveimo'eko'a}
* BAM - all other terms, e.g. prefixed with BAI or PU etc.

So the issue of te sumti detection is to turn all ZAMs into FAMs in the syntax tree.
Can we do that using PEG? I'm not that sure because
some brivla have infinite number of places like e.g. {jutsi}. {du} is a special case since every te sumti of it can just take {faxixo'e} position.

Currently I'm unaware of any possibilities for remembering values of variables (te sumti numbers) in PEG.js thus we cant increment to any given number of te sumti without first hardcoding all of them in PEG itself.

However, if we limit ourselves to just 5 places and basic cases of omitting FA then we can do that using PEG.
The current version of my fork of camxes.js produces these outputs:

1. ([FAXIPA mi] [CU {prami <FAXIRE do> VAU}])
2. ([FAXIPA mi] [CU {djuno <fi do> VAU}])
3. ([FAXIPA mi] [CU {djica <FAXIRE (¹lo [nu {<FAXIPA (²lo plise KU²)> <cu (²farlu [FAXIRE mi] [FAXICI {lo tricu KU}] VAU²)>} KEI] KU¹)> VAU}])

FAXIPA, FAXIRE, FAXICI are restored FA.

This is how sentence looks now in my PEG:
sentence = expr:(
&(terms_1ZAM CU_elidible selbri terms_1ZAM terms_1ZAM !terms_1ZAM) (termsfa bridi_tail_t1fefi) / /* mi klama do ti*/
&(terms_1ZAM CU_elidible selbri terms_1ZAM) (termsfa bridi_tail_t1fe) / /* mi klama do*/
&(terms_1ZAM terms_1ZAM CU_elidible selbri !terms_1ZAM) (termsfa termsfe bridi_tail_t1) / /* mi do klama*/
&(terms_1ZAM terms_1ZAM CU_elidible selbri terms_1ZAM !terms_1ZAM) (termsfa termsfe bridi_tail_t1fi) / /* mi do klama ti*/
&(terms_1ZAM terms_1ZAM terms_1ZAM CU_elidible selbri !terms_1ZAM) (termsfa termsfe termsfi bridi_tail_t1) / /* mi do ti klama*/
&(terms_1ZAM CU_elidible selbri) (termsfa bridi_tail_t1) / /* mi klama*/
terms? bridi_tail_t1 (joik_jek bridi_tail / joik_jek stag? KE_clause free* bridi_tail KEhE_elidible free*)*) {return _node("sentence", expr);}

Examples for each case is shown in comments.
This addition to PEG required hardcoding terms for fa,fe,fi separately and every case of bridi tail like "selbri x2 x3", "selbri x2", "selbri x3". To hardcode them one needs to also hardcode all inner rules until you meet selbri at bridi_tail_3 level and till tense_modal for sumti (a modification of tense_modal where {fa} or {fe} or {fi} are hardcoded correspondingly).

This required adding 60 new lines to PEG.
Even for supporting these basic cases the work isn't done yet, optimizations of those copy-pasted strings might be possible.

Anyway this is just a proof of concept.

To test the current state of alta parser say "alta: mi prami do" on La Naxle page.

--
You received this message because you are subscribed to the Google Groups "BPFK" group.
To unsubscribe from this group and stop receiving emails from it, send an email to bpfk-list+unsubscribe@googlegroups.com.
To post to this group, send email to bpfk-list@googlegroups.com.
Visit this group at http://groups.google.com/group/bpfk-list.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "BPFK" group.
To unsubscribe from this group and stop receiving emails from it, send an email to bpfk-list+unsubscribe@googlegroups.com.
To post to this group, send email to bpfk-list@googlegroups.com.
Visit this group at http://groups.google.com/group/bpfk-list.
For more options, visit https://groups.google.com/d/optout.