Lexical Relatedness Morphology (was: [Conlangs-Conf] Conference Overview)

Hi all,

On Sat, 6 May 2006 "David J. Peterson" wrote:
>
> And wrote:
> <<
> And after reading John Q's summary of the talks, I was wondering
> whether yours, David, was just espousing a Word-and-Paradigm model of
> inflection, or whether it went further and somehow advocated radical
> suppletion... Perhaps you'd give a short summary?
>  >>
>
> First, it's a cohesive model for inflection and derivaton (and, with
> appropriate tweaking, I think the model could handle phonology
> *and* syntax.  Unfortunately, I can count on two fingers the number
> of people seriously working with Bochner's Lexical Relatedness
> Morphology, and neither of them have left the morphological
> realm, yet).  And, second, no--at least, by the definition that
> suppletion > is the following:
>
> -Where X and Y are in a systematic morphological relationship
> such that by looking at either X or Y, one can predict the meaning
> and form of the member, ...
the member?


> ... and this systematic morphological
> relationship holds in general for the language as a whole, suppletion
> is where the systematic relationship between a pair X and Y is
> systematic in meaning, but totally unpredictable in form (i.e.,
> the form X is in no way related to the form Y).
Now, there's a hard definition to follow ...


> So, in English, "go" is the present tense, and "went" is the past tense.
> Verbs having a present and past tense form is something morphologically
> encoded by English, and there are various systematic patterns which
> relate present and past tense forms ("miss/missed", "raze/razed",
> "eat/ate", "hit/hit", etc.).  The pair "go/went", however, are in no
> way related to each other in form.  This is different from, say,
> "ride/rode"
> because there is still a phonological similarity between "ride" and
> "rode", and the pair belongs to a larger pattern of words that work
> the same way (e.g., "write/wrote").
So "suppletion" describes ... what?  The simple past /
simple present relationship?  As an abstract noun
ending in -tion, "suppletion" hints at the possible
(probable?) existence of a verb "to supplete".
What suppletes, and what does it supplete?  Your
definition hints at these questions but does not
answer them.  Here's another definition, with
examples, from SIL at http://tinyurl.com/m7fzc:

"Definition
 Suppletion is the replacement of one stem with another, resulting in an
allomorph of a morpheme which has no phonological similarity to the other
allomorphs.

Example (English)
  The following table illustrates stem suppletion:

Morphological process | Regular, nonsuppletive stem | Suppletive stem
Addition of past tense suffix | walk—walked | go—went
Addition of comparative or superlative suffix | big—bigger— biggest | good—better—best

The following table illustrates affix suppletion:

Morphological process | Regular, nonsuppletive affix | Suppletive affix
Addition of plural suffix | cat—cats | cherub—cherubim ox —oxen"

I found this a little easier to follow.  However,
being mindful of your point that a morpheme-
based analaysis may not always be possible or
desirable, I believe it should be possible to give
a clear definition of suppletion that exposes
the pattern-disrupting properties of go/went
and of good/better/best, without assuming that
the explanation must lie in a change of (a possibly
unattested) stem.  In short, the SIL definition
begs the question of the nature of suppletion.
(It also fails to include the affixation case given
in the examples!)  But what is fundamental to
suppletion is the observation that a pattern has
been broken.


> I *think* where there may be confusion is the idea of dispensing
> with morphemes.  I don't mean to dispense with affixes.  Affixes
> are great, and do a lot of work for a lot of languages.  What I was
> talking about was dispensing with the formal notion of a morpheme--
> that is, the notion that a morpheme is the smallest unit of meaning
> in language, and that a language is simply composed of thousands
> and thousands of morphemes.
>
> Without going into it too much (I plan to do a whole write-up and
> put it on my website, and when it's done, I'll post here), the very
> notion of the morpheme is problematic.  Morphemes *look* pretty
> good when you have pairs like this:
>
> cat = CAT (a feline entity), singular
> cats = CAT, plural
>
> Looking at those two, it's easy to say that cat = CAT, and -s = plural.
> It becomes somewhat more complicated with something like this:
>
> fish = FISH, singular
> fish = FISH, plural
>
> And also:
>
> man = MAN, singular
> men = MAN, plural
>
> These examples have been explained away by morpheme-based
> theories by positing zero morphs--morphemes that have no
> phonological form but mean something.  So fish is "fish" plus a
> null suffix that means plural, and this null suffix is an allomorph
> of /-s/.  There's a second null suffix which changes the vowel in
> [&] to [E] to form a plural, and that's another allomorph of /-s/.
Excellent examples!  And the immediate problem with
allowing morphemes to have allomorphs is simply this:
Knowing one allomorph no longer has the same pre-
dictive power that knowing *the* morpheme does.  At
best, one has a probablility tree for derivations.
Though the "X/allo-X" meme for any concept X is
seductive, there is a price to pay in scientific terms
when we use it: loss of predictive power.  And we test
our scientific theories by how good their predictions
are.  In a given domain, it may be that the best pre-
dictions possible *are* no more than probabilities.
Still, one prefers simpler predictions where possible.


> In other words, what a morpheme-based theory of morphology
> does with language is it treats *everything* like affixation.  And
> yet, at other times, it has a lot of problems with what looks like
> simple affixation.
>
> So in English we have a bunch of "berry" words.  "Berry" is mono-
> morphemic (though note that /-y/ is a common suffix, whether
> it has meaning or not).  In a word like "blueberry", you have a
> compound with "blue" and "berry", rendering "a berry that's
> blue" (whether a blueberry is or not.  I think of them as blue).
> Then, however, there are words like "cranberry", "boysenberry",
> "huckleberry" and "raspberry" which look to be built in the same
> manner, but which, when broken down, appear to be a meaningless
> prefix (or word) attached to a meaningful base.  ...
I'd like to think that a "raspberry" was a berry
whose canes were rasp-like; the ones in my garden
certainly act that way! ;-)  For the most part, I
agree that your base terms appear presently mean-
ingless; however, that does not mean that they were
so when the present words were first formed.  The
association of strawberries with straw is well-
known, and well-founded too.  I might add "young-
berry", "mulberry", "dewberry", "elderberry" and
"loganberry" to your list of berries; the association
of these with "young", "mull", "dew", "elder" and
"logan" varies from dubious to certain.


> ... In a morpheme-
> based theory, all of these words would be necessarily mono-
> morphemic.  In a sense, this is accurate: If a person has never
> seen or tasted a huckleberry, if they hear the word for the first
> time, they're not going to have any idea what color the berry is,
> what it tastes like, etc.  *However*, they're probably going to be
> able to guess it's a berry--more easily than if the word word "blick".
> Yet, a morpheme-based theory would treat "huckleberry" and
> "blick" identically.
>
> The general point is that morpheme-based theories treat instances
> like the "berry" in "huckleberry" as accidents, and treat word-internal
> morphological processes like prefixation or suffixation.  Further,
> by positing all these different allomorphs, it treats systematic
> relationships like accidents.  ...
I think you're right.


> ... Here's my favorite example to illustrate
> that point.
I hope you have a better example than this
one ...


> In Spanish, you can take a name ending in masculine /-o/ or
> feminine /-a/ (crucially treated as suffixes by a morpheme-based
> theory) and add the suffix /-it/ before it to get a nickname.  Here
> are some examples:
>
> Nacho > Nachito
> Dora > Dorita
> Carla > Carlita
> Carlo > Carlito
... because the derivation clearly proceeds by
inserting an infix /-it-/, not a suffix /-it/ as
you assert.  If it did, we'd have instead:
> Nacho > Nachoit
> Dora > Dorait
> Carla > Carlait
> Carlo > Carloit
which are clearly not Spanish words.


> Then you get this pair:
>
> Carlos > Carlitos
>
> Under a strict morpheme-based analysis, "Carlitos" is a huge problem.
> You have to posit one of the following solutions:
>
> (1) /-os/ is an allomorph of /-o/ (only used with the stem "Carl-")
>
> (2) /-it-/ is an infix which is an allomorph of /-it/ that is only used
> with the name "Carlos"
>
> (3) /-s/ has some sort of meaning
>
> It seems to me like (1) is the best solution of the three.  However,
> it's positing a single morpheme that only works with one word.
> (2) does the same thing, but positing that Spanish has exactly one
> infix that accidentally looks like the suffix /-it/ and only is used
> with one name seems utterly preposterous.  And with (3), what
> meaning could be associated with this /-s/?  A general "this is a
> name" suffix?  If so, why does it only show up on one name?
No problem(o)!  The derivation
Carlos > Carlitos
uses exactly the same infix /-it-/ as all
the others do.


> So, keeping with this Carlitos example, the solution I proposed is
> to abandon the formal notion of a morpheme.  By doing so, you
> can recognize that there's a general pattern between names in
> Spanish that looks like this:
>
> (a) Xo  <->  Xito
Or, I'd rather say, a general pattern:
 (1) XO  <->  XitO
where X is the stem of the name, of
almost arbitrary phonological form, and
O is the ending, of form V[C], eg -a/-o/-os.

> On the left is a form for a name (X is a variable that stands in for
> the rest of the name), and on the right is a nickname associated
> with it.  This is just stating a relationship between forms and
> meaning; it's not suggesting that /-o/ or /-it/ means anything.
> Further, there's nothing in the rule that would prevent it from
> applying to Carlos.  Carlos can have it's own rule:
>
> (b) Xos  <->  Xitos
>
> But this rule is simply a more specific instantiation of the rule in
> (a), and isn't necessary.  What this kind of rule does for a speaker
> is it models (or is an attempt to model) the human capacity for
> linguistic analogy.  Thus, if a speaker has the notion of the
> relationship
> in (a), and hears a name like "Carlos", if there's a phonological
> relation between the first item in (a) and the name, then the
> speaker can analogize that there may also be a member like that
> on the right to get "Carlitos".
Yep.  It's essential to get the power of
patterns that speakers can use to form
new, analogic words; or to explain newly
encountered words, as needs must.


> I personally think this type of theory lends is a pretty good model
> of acquisition, because when you encounter a word for the first
> time (say a noun), it's not always going to be in the nominative
> singular.  You hear words in context, and never in citation form.
> So say for some reason a new verb pops up in English:
>
> "Yesterday, I was at the store, and this guy in a cart came by and
> blope me, and I was like, 'Dude!  Watch where you're going!'"
>
> So a hearer can take a pretty good guess at what the word means
> (i.e., either to "hit", or maybe "to hit with a shopping cart"), they
> know it's in the past tense, and the form they're given is "blope".
> Immediately, two patterns should come to mind:
>
> X  <->  X
> XiY  <->  XoY  (using English orthography)
>
> The first is the pattern for "hit/hit", the second the pattern for
> "ride/rode".  At this point, the hearer can take a guess at how
> the rest of the paradigm (in this case, the present tense) is formed.
> Given that the second pattern up there is more common (or
> maybe...  "Hit", "put"...  I can't think of anymore), they might
> guess the second, but who knows.  And they could be wrong.
> However, you'd predict that they'd fit that word into an extant
> pattern of English, and wouldn't say something like, "Why did
> he bloip you?"
>
> This is a nice feature of Lexical Relatedness Morphology, because
> in a morpheme-based theory, if you hear "wrote", there's no
> formal way to predict the singular would be "write".  Rather
> than having a list of fully inflected words with patterns of
> relationship between those words stored in the lexicon, you
> have a list of morphemes, with, at most, a series of diacritics
> for what attaches where, for example...
>
> -er (suf.) comparative [attaches to class A adjectives only]
> big (adj.) class A
> large (adj.) class A
> hilarious (adj.) class B
> perpendicular (adj.) class B
> magnificent (adj.) class B
> morphological (adj.) class B
> funny (adj.) class A
>
> Then when a new adjective comes in, you have to decide which
> class it belongs to, with no formal basis for deciding as much, unless
> you specify which words can fall into class A, and which can't.
> And then if someone comes up with a novel form (possibly to
> be funny), "Yeah, well my car, is more, like...magnificenter,
> so...yeah!",
> the question is, how on earth could they have ever done that, if
> "magnificent" is class B?  And how do we understand it?
>
> Darn, I went on much longer than I intended to.  And what's even
> more perplexing, is that I don't even know if I answered your
> question.  Is this close?
David, I don't know whether you answered
And's question, but you certainly gave me
some food for thought - thanks!

Regards,
Yahya


--
No virus found in this outgoing message.
Checked by AVG Free Edition.
Version: 7.1.385 / Virus Database: 268.5.2/329 - Release Date: 2/5/06