[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [bpfk-announce] Next Up: Morphology



At 12:33 PM 2/11/05 -0800, Robin wrote:
xorxes has put together a lovely morphology algorithm for us, and he
and Pierre have beaten on it quite a bit.  So, let's take a look at
it.

I don't necessarily expect you all to actually read the thing; I
expect xorxes to provide a list of things that he thinks might be
controversial and for us to go over that.  If it turns out in the
future that someone's beliefs about what the algorithm does are
violated by the actual algorithm, we can always come back and fix
it.

Initial voting time is set at 2 weeks; closure on 25 Feb 2005.

-Robin

Various comments in no particular order.

If we vote in the Morphology Algorithm, we are agreeing to the setting of what the valid word-forms are. It is obvious that fu'ivla forms and cmavo forms have changed, but nowhere is there a reasonbly concise statement of what the new forms are.

The "algorithm" is program code. One usually checks a program against the desired results to see if it works. We have no statement of the desired results. I can't read the code without more explanation. For example : "& indicates that the element to the right must follow" [follow what? it's at the beginning - example: BU <- &cmavo (b u ) &postword] "but the marked element itself" [the thing to the right of the &, I presume] "does not absorb anything" [what does this mean?].

Correct me if I'm wrong, but have we just added things like "sia" /sya/ and "rua" /rwa/ to the CVV space? In making Lojban, we made a deliberate decision to exclude these (which were present in Loglan) for 2 reasons. Some of them are too hard to say ("rui"), especially as a single syllable. Some, like "sia", tend to degenerate into other sound sequences (as English "sion" endings are now pronounced "shun").

What are the valid fu'ivla forms? It looks to me like "jritata" is valid (initial cluster OK, passes slinku'i test, doesn't break up into gismu and/or rafsi). So, given "All fu'ivla can be used as final rafsi", "prijritata" is no longer a valid fu'ivla because it is "pri"+jritata", right? Thus we can add to the "[to]slinku'i" test the "prijritata" test to eliminate invalid fu'ivla. This seems too complex to me.

Also, we need to revise "slinku'i" itself a bit if r-hyphens are allowed when not required. We have a valid lujvo "li'erla'i" (li'e + r + la'i). If we allow someone to add an affix on the front without removing the "r" ("tos + li'e + r + la'i), we then need to disallow a potential fu'ivla "sli'erla'i" (originally OK - good initial cluster, doesn't break down into rafsi and/or gismu, used to pass to+sli'erla'i "slinku'i" test but now will not). Making fu'ivla is a mine-field.

In the presentation on Controversial topics, it looks like fu'ivla initial clusters disallow sx (good!) - but not cx? "jl", "jr" "zl", "zr" are allowed? "jk"?

If "non-y cmavo can have non-final rafsi by adding 'y", then are we removing the restriction that there must be a consonant cluster in the first 5 [non-'-non-y] letters? "oi'ycai'ylujvo". How do they attach on the left if they're vowel-initial - eg: klama + o + sutra = klamy[what]sutra?

"cmene can have inital rafsi by adding iy". Can names have "iy" internally, or is that disallowed like la/lai/la'i/doi? Must the name be entirely unstressed if it doesn't have the penultimate syllable? I don't think it causes a problem, but I'm not yet sure.

I notice there are 2 things that are being allowed to have rafsi by following them by 'y: cmavo and fu'ivla. Does this cause a problem? Take for example stura [gismu] + o'a [cmavo] + klama [gismu]. This would be, stura'o'a'yklama (note: here, I'm presuming that the vowel-initial cmavo will attach to the left using ', as vowel-initial fu'ivla do). But, might "stura'o'a'yklama" instead break down as stura'o'a [fu'ivla] + klama? Or is stura'o'a not a valid fu'ivla because it could break down into stura + ' + o'a when in front of something else (oh boy - another test for fu'ivla)? Which has precedence?



--
mi'e noras                                             noras@cox.net
Nora LeChevalier