Upper limits to morpheme length - formal vs. functional account.

[Everett, Daniel]

Dear Lise,

No disagreement necessarily. 

All accounts need to be formalized. For me the question is whether they are formalizations over structures or over functional, cultural, or other considerations e.g. cognition, climate, altitude, etc. There will always be some fundamental computational residue that needs its own account. These may or may not represent distinct components of the overall formalization. Grammars are composites of computational and other strategies. Divide and conquer may not be the best strategy in the sense that neither cognition, structure, or computation has privileged status. One may in one context but not another. All are not always needed. 

Dan

Sent from my iPhone

> On Dec 22, 2013, at 16:05, "Lise Menn" <lise.menn at Colorado.EDU> wrote:
> 
> Thanks, Dan, but at the risk of opening another topic, I'd say that a psycholinguistic/diachronic account doesn't obviate the need for a formal account (or vice versa), because formal accounts and functional accounts are different kinds of entities. They serve different purposes and both are useful.
> Formal constraints/accounts, like macro-level physical laws, look elegant and abstract away from particulars (think about the simple, elegant laws relating pressure, temperature, and volume of gases).  Functional accounts are clunkier, concrete, but provide explanations for why things are the way they are (cf. the statistics of how molecules behave when they bump into each other, which is way beyond me but is what underlies the simple gas laws).  Both kinds of accounts should yield testable predictions, and it seems that different folks prefer to work more with the abstract formulations or more with the concrete mechanics. And both kinds of people are needed to keep a science both accessible and empirical.
> 
> Lise
> 
> Lise Menn
> Home Office: 303-444-4274
> 1625 Mariposa Ave
> Boulder CO 80302
> http://spot.colorado.edu/~menn/index.html
> 
> Professor Emerita of Linguistics
> Fellow, Institute of Cognitive Science
> University of  Colorado
> ________________________________________
> From: Everett, Daniel [DEVERETT at bentley.edu]
> Sent: Saturday, December 21, 2013 3:15 PM
> To: Lise Menn
> Cc: Funknet List; LINGTYP at LISTSERV.LINGUISTLIST.ORG
> Subject: Re: [FUNKNET] Upper limits to morpheme length
> 
> Lise,
> 
> Great comments. These remarks likely obviate the need for a more formal account. But now we have a bit of both.
> 
> Thanks,
> 
> Dan
> 
> Sent from my iPhone
> 
>> On Dec 21, 2013, at 17:09, "Lise Menn" <lise.menn at Colorado.EDU> wrote:
>> 
>> The cognitive underpinnings of the Hayes and Wilson constraint (and of Zipf's law, of course) would come from several sources that I can think of (there might well be others):
>> 
>> 1) the difficulty of catching all the phonemes when you a long, unfamiliar, unanalyzed word (hard to imagine Chaugoggagoggmanchagaugagoochaubungungamogg surviving in English without having been written down)
>> 
>> 2) The very sparse neighborhoods of long monomorphemic words (that is, the rarity of pairs of long monomorphemic words that differ by only one phoneme) means that they are identifiable by listeners even when some of the sounds are inaudible, so misunderstandings won't offer any barrier to elision of the sounds
>> 
>> 3) Speakers will abbreviate long words because - other things being equal - they are more work to produce.
>> 
>> Lise
>> Lise Menn
>> Home Office: 303-444-4274
>> 1625 Mariposa Ave
>> Boulder CO 80302
>> http://spot.colorado.edu/~menn/index.html
>> 
>> Professor Emerita of Linguistics
>> Fellow, Institute of Cognitive Science
>> University of  Colorado
>> ________________________________________
>> From: funknet-bounces at mailman.rice.edu [funknet-bounces at mailman.rice.edu] On Behalf Of Everett, Daniel [DEVERETT at bentley.edu]
>> Sent: Saturday, December 21, 2013 12:39 PM
>> To: Funknet List; LINGTYP at LISTSERV.LINGUISTLIST.ORG
>> Subject: Re: [FUNKNET] Upper limits to morpheme length
>> 
>> Folks,
>> 
>> Thanks for the suggestions. I just received the following from Bruce Hayes which exactly answers my question. With Bruce's permission, I pass this along here.
>> 
>> All the best for the end of one year and the beginning of another. I hope you have all finished posting your grades and are now able to relax a bit.
>> 
>> -- Dan
>> 
>> 
>>> 1) The main way to get really long morphemes, I suspect, is to borrow from languages with which you have little contact, so you can't parse their long polymorphemic words. Hence English Okaloacoochee, Hanamanioa, Chaugoggagoggmanchagaugagoochaubungungamogg.
>>> 
>>> 2) I think the upper limit for English morphemes is three metrical feet.  When I make up a four-foot word it sounds odd to me, e.g. ?Okaloaseppacoochee.  The famous lake Chaugoggagoggmanchaugagoggchaubungungamogg is not an exception; it pronounced as three separate phonological words:  Chaugoggagogg, manchaugagaug, ch[schwa]bunagungamogg.
>>> 
>>> 3) If you adopt the phonotactic model of Hayes and Wilson (LI 2008), then if you include a contraint of the type *Struc, and train up the grammar, you get the right predictions: *Struc gets a modest weight, which predicts a descending-exponential probability function for words of ever-increasing length.  In this theory, the extreme unlikelihood of extremely long words is simply an extrapolation from the moderate unlikelihood of somewhat-long words.
>>> 
>>> Best regards,
>>> Bruce
>>> 
>>> Bruce Hayes
>>> Professor and Chair
>>> Department of Linguistics, UCLA
>>> Los Angeles CA  90095-1543
>>> bhayes at humnet.ucla.edu
>>> www.linguistics.ucla/people/hayes
>> 
>>>>> 
>>>>>> Please excuse the double-posting. I haven't worked
>>> on this stuff for a while, so I will undoubtedly show my ignorance of
>>> some large body of research, but I was wondering (due to a question from
>>> a colleague) whether there is any work that tries to derive a maximum
>>> morpheme length (I wouldn't think this would be the way to address the
>>> issue, frankly, but I could be quite wrong). As the question was put to
>>> me: "It seems to me that almost all morphemes are quite short?probably
>>> not easy to find one with e.g. 12 phoneme segments. The question is is
>>> there anything in known phonological theories which predict this?or is
>>> it just assumed that morphemes can be of any length and that the reason
>>> there are none of length e.g. 624, 578 is simply that they would be
>>> unlearnable? The latter would be my ideal view, just as the reason that
>>> no one uses a sentence of length 624,578 words has to do with practical
>>> performance limitations." I know that there is work on "resizing theory"
>>> (Pycha 2008) and various other approaches linking morphology and
>>> metrical structure. But those approaches so far as I know offer no
>>> principled upper bound to morpheme length. Any help would be
>>> appreciated. Happy holidays to all, Dan Everett
>>> 
>>> 
>>> 
>>> Links:
>>> ------
>>> [1]
>>> mailto:DEVERETT at bentley.edu