Re: Arabic and BACK TO Self-segregating morphology

On Wed, 21 Dec 2005 11:37:32 -0500, Jim Henry <jimhenry1973@...>
>wrote:
>On 12/20/05, Patrick Littell <puchitao@...> wrote:
>> [snip]
>
>> ........for things like (prefix)i(word)u(word)o(suffix) there are at
>> least
>> five ways to parse it.  We can figure out which morphemes are
>> prefixes, which are roots, and which are suffixes, but not their
>> hierarchical structure.
>
>In other words, it's sometimes unclear whether a prefix
>modifies the first element of a compound, or modifies
>the whole compound -- & ditto with suffxes w.r.t. the
>final element of the compound or the whole compound word.
>
>> Anyway, there are answers; I just want to see if anyone has any
>> interesting solutions that don't require infixation, circumfixation,
>> or the equivalent.
>
>What about having high- and low-precedence binder morphemes?
I was just about to mention that, when I read your post!

>That would require a larger inventory of vowels, probably.  But suppose
>i, o and u are the high-precedence binder morphemes and order of
>compounding precedence is left to right, then
See below for varying order-of-binding.

>kainalijaupalitaosa
>
>could only be parsed as
>(((ka + nalija) + palita) + sa)
>
>But if y, e, and a are the low-precedence binder morphemes,
>you could also have, e.g.:
>
>kaynalijaepalitaosa
>= (( ka + nalija) + ( palita  + sa ))
>
>This scheme may be too complex to speak and hear in real time;
>we would have to experiement for a while and see how
>it works.
>
>--
>Jim Henry
Operator-precedence grammars are studied in Compiler Theory; and here's a
summary of what you'd need to know.

1) Each (binary infix) operator is either Left-Associative or Right-
Associative or Non-Associative.

So, if /a/ were Left-Associative, /bacad/ could only be parsed as (bac)ad;
and if /e/ were Right-Associative, /beced/ could only be parsed as be(ced);
and if /u/ were Non-Associative, /bucud/ would be ungrammatical and make no
sense; /buc/ would be ok, and /cud/ would be ok, but not /bucud/.

*1*) As a variant, you could allow multiple non-associative operators of
the same precedence to be grammatical, as long as there were no way to
assign any difference to the groupings.  In other words, /bucud/ in the
example above, might be OK, provided it could not be interpreted as
(buc)ud nor as bu(cud).

2) Given any two (binary infix) operators, either the first takes
precedence over the second, or the second takes precedence over the first,
or they have equal precedence.

3) Precedence is a totally-ordering relationship; if /a/ takes precedence
over /e/, and /e/ takes precedence over /i/, then /a/ takes precedence
over /i/.

4) Operators of equal precedence _must_ have the same kind of
associativity; they are either all Left-Associative, or all Right-
Associative, or all Non-Associative.  To put it another way, operators with
differing associativity must also have differing precedence.

-----

OK, imagine a language with seven vowels /a/ /e/ /i/ /o/ /u/ /w/ /y/ and 19
consonants.

Imagine all lexical, semantically non-empty, morphemes are consonant-
clusters.

Imagine the vowels all represent connectors -- makes no difference whether
they are compounding-connectors, or affixing-connectors, or whatever.

Imagine their precedence order is:
/w/ /y/ Left-associative, equal precedence at first;
/o/ /u/ Right-associative, equal precedence at second;
/i/ Left-associative (third highest precedence);
/e/ Right-associative (fourth highest precedence);
/a/ Non-associative (lowest precedence);

Now consider the pattern /bVcVdVf/.
If each V could be any one of /a/ /e/ /i/ /o/ /u/ /w/ /y/, this could lead
to 7*7*7=343 different "words"; however, no more than one of the Vs can be
replaced by /a/, so there are actually
6*6*6 + 3*6*6 = 216+108 = 324 possible words.

/becedef/ is ((bec)ed)ef
/bacedef/ is ba((ced)ef)
/becadef/ is (bec)a(def)
/becedaf/ is ((bec)ed)af

/bicidif/ is bi(ci(dif))
/bacidif/ is ba(ci(dif))
/bicadif/ is (bic)a(dif)
/bicidaf/ is (bi(cid))af

/becedef/ is ((bec)ed)ef
/becedif/ is (bec)e(dif)
/becidef/ is (be(cid))ef
/becidif/ is be(ci(dif))
/bicedef/ is ((bic)ed)ef
/bicedif/ is (bic)e(dif)
/bicidef/ is (bi(cid))ef
/bicidif/ is bi(ci(dif))

/bicidif/ is bi(ci(dif))
/bicidof/ is bi(ci(dof))
/bicodif/ is bi((cod)if)
/bicodof/ is bi((cod)of)
/bocidif/ is (boc)i(dif)
/bocidof/ is (boc)i(dof)
/bocodif/ is ((boc)od)if
/bocodof/ is ((boc)od)of

/becedef/ is ((bec)ed)ef
/becedof/ is (bec)e(dof)
/becodef/ is (be(cod))ef
/becodof/ is be((cod)of)
/bocedef/ is ((boc)ed)ef
/bocedof/ is (boc)e(dof)
/bocodef/ is ((boc)od)ef
/bocodof/ is ((boc)o)dof

/bicidif/ is bi(ci(dif))
/bicidwf/ is bi(ci(dwf))
/bicwdif/ is bi((cwd)if)
/bicwdwf/ is bi(cw(dwf))
/bwcidif/ is (bwc)i(dif)
/bwcidwf/ is (bwc)i(dwf)
/bwcwdif/ is (bw(cwd))if
/bwcwdwf/ is bw(cw(dwf))

-----

You could vary the number of levels of precedence, and the associativity or
lack of it at each level, as well as the direction of associativity at each
level where there is any, to solve most of your problems.

I'd probably recommend a lowest-level non-associative one for compounding
words, a highest-level left-associative one for introducing a bias in favor
of suffixes as opposed to prefixes,
and run through the whole "Left, Right, Non", cycle twice
to allow each kind of associativity to have precedence over each kind.

(That would make six levels, for those keeping count.  With only one binder
per level, you would only need six binders; however you might want make a
semantic or grammatical difference in two or more different binders at the
same level.)

This could completely solve the problem of whether a prefix or suffix on a
compound word modifies just the component, or just the whole compound.

-----

Well, that may be more than you wanted to read.  I hope it helped, instead.

Tom H.C. in MI