[Lingtyp] Explicating morpheme boundaries
Martin Haspelmath
martin_haspelmath at eva.mpg.de
Mon Oct 21 05:13:10 UTC 2024
The general issue that Mae Carroll brought up is indeed important to
keep in mind: There is often no unique solution to grammatical analysis,
something that has been known to linguists since Yuen-Ren Chao's classic
paper on phonemic non-uniqueness from 1934 (see
https://dlc.hypotheses.org/3381).
Somehow linguists keep forgetting this old insight, acting as if there
were a unique solution after all and we simply have not searched hard
enough. Cognitively oriented linguistics shifted the focus to mental
grammars, hoping that the phychological perspective would help select
the true grammar out of the many possibilities (e.g. by an evaluation
metric, or by a parameter-setting model). But of course different
speakers could have different mental grammars, and perhaps all they
share is a set of social conventions – and these social conventions can
be described in multiple ways.
For the goal of rigorous and complete description, this is not a problem
– one may even say that variety is good because of different esthetic
preferences (and because students have multiple ways of guessing a right
answer :-)).
But for the goal of general understanding (of Human Language, i.e.
g-linguistics), we need cross-linguistic generalizations, which must be
based on comparative concepts (see my 2021 paper:
https://ling.auf.net/lingbuzz/005158).
So I think that for comparative work on grammar, such as Phillip
Rogers's and Tim Zingler's work on morph length, we do need a way of
identifying morphs in a uniform way, using the same criteria in all
situations and in all languages. (The issue is actually the same as was
brought up on this list a few weeks ago, by people like Volker Gast and
Mark Donohue:
https://listserv.linguistlist.org/pipermail/lingtyp/2024-September/011692.html)
Best,
Martin
On 21.10.24 01:37, Mae Carroll via Lingtyp wrote:
> Hi Joe
>
> Not really in direct answer to your question but you raise a really
> interesting point that we all rub against when both teaching and
> describing languages with complex morphology,
>
> Sacha Beniamine (Surrey) and I have a paper under review discussing
> this and related problems in the context of morphological theory and
> typology. It's what people have called 'The Segmentation Problem'
> (Manning 1998; Beniamine & Naranjo 2021); basically while linguists
> can be trained to do segmentation, there is no single algorithm that a
> linguist or a computer can apply to the segmentation of words into
> component morphs to produce a consistent output that also aligns with
> how linguists traditionally think about these things. Andy Spencer
> (2012, p.92) has talked about this, he says: “when we look at the
> practice of grammarians, it should be clear from a cursory glance
> through descriptions of familiar morphologically complex languages
> that there is no consensus on segmentation even for extremely
> well-studied languages.”
>
> The Nida example provided by Christian is really still the best
> version out there but it can't be formalised in an explicit way but I
> have also found it very useful for teaching. I think it's a really
> fascinating problem.
>
> I also think this makes sense. Descriptive linguists, myself included,
> like to use the concept of morphemes when writing grammars because
> they are intuitive, useful and we all learn them but they are a
> theoretical notion, i.e. a minimal pairing of form and meaning. But so
> much of morphological typology has shown that below the word level
> (and above in many cases) the alignment of meaning and form is rarely
> so neat or isomorphic. Further, many, if not most, morphological
> theories either eschew the notion or use a very specific definition
> that doesn't really correspond to the traditional definition most of
> us were trained to think about.
>
> Anyway, just some thoughts on the topic.
>
> Best
> Mae
>
> Beniamine, S. and M. Guzmán Naranjo. 2021. Multiple alignments of
> inflectional paradigms. Proceedings of the Society for Computation in
> Linguistics 4(21) .
>
> Manning, C.D. 1998. The segmentation problem in morphology learning.
> In Proceedings of the Joint Conferences on New Methods in Language
> Processing and Computational Natural Language Learning, NeMLaP3/CoNLL
> ’98, Stroudsburg, PA, USA, pp. 299–305. Association for Computational
> Linguistics.
>
> Spencer, A. 2012. Identifying stems. Word Structure 5(1): 88–108
>
>
> Dr Mae Carroll (she/her)
> Lecturer in Linguistics
> School of Languages and Linguistics
> University of Melbourne
> www.maecarroll.com <http://www.maecarroll.com/>
>
> /I acknowledge the Traditional Owners of the land where I live and
> work, the Wurundjeri people of the Kulin Nations, and pay my respects
> to Elders past and present. I recognise that sovereignty was never ceded./
>
>
>
>
> On Sun, Oct 20, 2024 at 10:56 PM Zingler, Tim via Lingtyp
> <lingtyp at listserv.linguistlist.org> wrote:
>
> Dear Joe, dear all,
>
>
> Phillip Rogers and I have a paper coming out with LT any day now
> where we look at that issue for a worldwide sample of 25 languages
> and about 1,500 affixes. The bottom line is as follows: both types
> are usually monosyllabic, but when they are not, prefixes tend to
> be sub-syllabic (i.e., vowelless) more often than suffixes, whose
> second-favorite option is disyllabic size. Those results are
> statistically significant. So in a way, both your guiding
> principles are correct.
>
>
> Previous research on that issue, with compatible results, can be
> found in Bybee et al. (1990), in this volume:
>
> https://benjamins.com/catalog/tsl.20?srsltid=AfmBOopGoDDksgrf9l02wZ41FuMVBL8d5wdfgVZ6EW_uOndXCFIWFl63
>
> Best,
>
> Tim
> ------------------------------------------------------------------------
> *Von:* Lingtyp <lingtyp-bounces at listserv.linguistlist.org> im
> Auftrag von Joe Blythe via Lingtyp <lingtyp at listserv.linguistlist.org>
> *Gesendet:* Sonntag, 20. Oktober 2024 07:22
> *An:* lingtyp at listserv.linguistlist.org
> *Betreff:* [Lingtyp] Explicating morpheme boundaries
> Dear typologists,
>
> I’d like some help please coming up with a better rule of thumb
> for explicating morpheme boundaries to students. I teach a third
> year undergraduate morphosyntax unit. When beginning with orthodox
> agglutinative morphology, I usually give my students a kind of
> rule of thumb, that when presented with a particular problem set,
> they should generally allocate as much phonological material as
> possible to the root (or stem), and the remainder can be allocated
> to an affix. I don’t recall where I learned this, but it always
> seemed to me the sensible thing to do. Whether it is justified I’m
> not so sure. There are however occasions when it seems sensible to
> ensure that an affix contains at least a vowel. Yet there are
> certainly some affixes that lack a nucleus (the examples I’m
> thinking of are all suffixes). So rightly or wrongly, I have two
> guiding principles here that are pushing in opposite directions.
> And there are always a few students who get all the glosses
> correct but have carved the joints differently. These solutions
> look odd to me but I struggle to explain why I find them
> unsatisfactory. Now for every language it is usually possible
> determine where to carve the joints, given enough data. But when
> presented with a small dataset for some random language (perhaps
> without even a name), how should we explain why prefixes generally
> need vowels? Actually, is this last point even true? I feel these
> issues has probably been discussed somewhere at length. If so, can
> you point me to some relevant literature, as I’d like to provide
> some advice that’s typologically grounded, not on what feels like
> the right thing to do.
>
> All the best
>
> Joe
>
>
> *Joe Blythe*
> Associate Professor | Department of Linguistics | Macquarie University
> Room 514, 12 Second Way, North Ryde, NSW 2109, Australia
> Vice President (Conferences), Australian Linguistic Society
>
> joe.blythe at mq.edu.au | www
> <https://researchers.mq.edu.au/en/persons/joe-blythe>
> Ph: +61-2-9850-8089 <tel:+61-2-9850-8089> | Mob:
> +61-409-88-1153 <tel:+61-409-88-1153>
> /Macquarie Linguistics Conversation Analysis Lab
> <https://www.mq.edu.au/about/about-the-university/our-faculties/medicine-and-health-sciences/departments-and-centres/department-of-linguistics/our-research/conversation-analysis>/
> /Conversational Interaction in Aboriginal and Remote Australia
> <https://www.ciaraproject.com/>/
> /OzSpace:/ /Landscape, language and culture in Indigenous
> Australia/ <https://ozspace.org/>
>
>
>
>
>
>
>
>
>
>
>
>
> _______________________________________________
> Lingtyp mailing list
> Lingtyp at listserv.linguistlist.org
> https://listserv.linguistlist.org/cgi-bin/mailman/listinfo/lingtyp
>
>
> _______________________________________________
> Lingtyp mailing list
> Lingtyp at listserv.linguistlist.org
> https://listserv.linguistlist.org/cgi-bin/mailman/listinfo/lingtyp
--
Martin Haspelmath
Max Planck Institute for Evolutionary Anthropology
Deutscher Platz 6
D-04103 Leipzig
https://www.eva.mpg.de/linguistic-and-cultural-evolution/staff/martin-haspelmath/
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.linguistlist.org/pipermail/lingtyp/attachments/20241021/39f4369c/attachment.htm>
More information about the Lingtyp
mailing list