[Lingtyp] wordhood: responses to Haspelmath

Dryer, Matthew dryer at buffalo.edu
Sun Nov 12 22:22:03 UTC 2017

I am in Papua New Guinea with inconsistent access to the internet so there are delays in my responding to issues in this discussion.

But my reply to Bill Croft’s question to me:

The existence of problematic cases may be a problem for a crosslinguistic category (for those who believe in them), but it is not a problem for comparative concepts for two reasons. First, there is often a way to define one’s comparative concept in a “narrower” way so that at least some problematic cases can be classified. Second, there are, in my experience, almost always borderline cases in classifying languages according to a particular comparative concept.

Bear in mind that the issue of applying a comparative concept of “word” means defining a comparative concept in some way for the purposes of classifying languages for a particular typological study. Thus the only relevance of problematic cases like these Bickel et al studies is whether they present a problem for classifying these languages for the purposes of a particular typological study. It is easy to imagine cases where they do present a problem, but then these languages can simply be classified as unclear for the purposes of that study.


From: William Croft <wcroft at unm.edu<mailto:wcroft at unm.edu>>
Date: Sunday, November 12, 2017 at 9:52 AM
To: "Matthew S. Dryer" <dryer at buffalo.edu<mailto:dryer at buffalo.edu>>
Cc: "lingtyp at listserv.linguistlist.org<mailto:lingtyp at listserv.linguistlist.org>" <lingtyp at listserv.linguistlist.org<mailto:lingtyp at listserv.linguistlist.org>>
Subject: Re: [Lingtyp] wordhood: responses to Haspelmath


  I am having a hard time reconciling your statement

I am completely in agreement with Martin about the problems with the notion of morphosyntactic word - in fact I would go even further than him

with your statement

Thus it is not clear that there is any problem with a comparative concept of word.

unless you mean that a morphosyntactic definition of a word is even more problematic than Martin made it out to be, but that a phonological definition of a word (whatever criteria those are based on) is unproblematic. For the latter, these two papers based on studies of phonological criteria in Tibeto-Burman languages indicate that phonological criteria are also problematic, that is, do not converge on the same morpheme strings as words:

Bickel, Balthasar, Kristine A. Hildebrandt and René Schiering.  2009. The distribution of phonological word domains: a probabilistic typology. Phonological domains: universals and deviations, ed. Bariş Kabak and Janet Grijzenhout, 47-75. Berlin: Mouton de Gruyter.

Schiering, René, Balthasar Bickel and Kristine A. Hildebrandt. 2010. The prosodic word is not universal. Journal of Linguistics 46.657-709.

    Incidentally, your 2015 ALT talk is a counterexample to my statement in an earlier email that nobody has applied a consistent crosslinguistic criterion to wordhood. You took a very strict but clearly crosslinguistically applicable notion of wordhood (or more precisely, affixhood) -- i.e. one that everyone would agree with -- and demonstrated that the suffixing preference applies to these affixes. I apologize for the oversight.


On Nov 11, 2017, at 4:35 PM, Dryer, Matthew <dryer at buffalo.edu<mailto:dryer at buffalo.edu>> wrote:

What is missing from this discussion is that Martin’s paper discusses problems with morphosyntactic criteria for words as proposed in recent literature in morphological theory, not problems with phonological criteria. But typologists base their claims on language descriptions and language descriptions primarily use phonological criteria in deciding what to represent as a word, supplemented some by morphosyntactic criteria. I am completely in agreement with Martin about the problems with the notion of morphosyntactic word - in fact I would go even further than him. But this has little bearing on typological claims that make reference to words.

This is not to say that there may not be problems with phonological criteria. I would say that for every language I have worked on, there are morpheme boundaries were neither phonological criteria nor morphosyntactic criteria provide a basis for deciding whether to treat something as a word. But such morpheme boundaries represent a relatively small percentages of morpheme boundaries in the language. Thus it is not clear that there is any problem with a comparative concept of word.

In the 2015 ALT talk of mine that Martin referred to, I argued that if one restricts attention to grammatical morphemes that are nonsyllabic or that exhibit morphophonemic alternations, we find clear evidence of a suffixing preference. This illustrates how the problems that Martin discusses do not present a problem for showing that there is a preference for suffixes over prefixes.


From: Lingtyp <lingtyp-bounces at listserv.linguistlist.org<mailto:lingtyp-bounces at listserv.linguistlist.org>> on behalf of "Anstey, Matthew" <MAnstey at csu.edu.au<mailto:MAnstey at csu.edu.au>>
Date: Sunday, November 12, 2017 at 9:06 AM
To: "lingtyp at listserv.linguistlist.org<mailto:lingtyp at listserv.linguistlist.org>" <lingtyp at listserv.linguistlist.org<mailto:lingtyp at listserv.linguistlist.org>>
Subject: Re: [Lingtyp] wordhood: responses to Haspelmath

Have there been any studies into the cognitive salience of the notion 'word', say, along the lines of segmentivity?

Such that, that all people and hence languages chunk and split language symbols one way or another, driven by a number of constraints (eg limiting semantic complexity, facilitating processing speed, allowing combinatorial possibilities like juncture/nexus joints, etc).

If then this a universal cognitive requirement, could the culture-language specific version of the most 'privileged' chunk, no matter much it differs around the world, be what we identify as this elusive 'word'?

Not sure how cognitive typology works though....

With regards

On 12 Nov 2017, 8:23 AM +1030, Daniel Ross <djross3 at illinois.edu<mailto:djross3 at illinois.edu>>, wrote:
Just a quick clarification: my metaphor was just a comparison to suggest there might be different sizes/levels of words (depending on what we mean by that), not a theoretical point about how to analyze words in particular or any claims about morphology.

But your clarification is appreciated!


On Sat, Nov 11, 2017 at 1:04 PM, Peter Arkadiev <peterarkadiev at yandex.ru<mailto:peterarkadiev at yandex.ru><mailto:peterarkadiev at yandex.ru>> wrote:
Dear all.

just to add to the cunundrum, many contemporary morphologists do not believe that "words are boxes for morphemes", as Danny put it, and do not use the concept of "morpheme" at all, operating with features and their exponents and paradigmatic relations between words instead (see work by Stephen Anderson, Gregory Stump, Jim Blevins and many, many others). For me, as a morphologist, this makes much sense, because I know that, first, there are languages where much if not most morphological information is expressed by internal modification rather than by affixes (cf. the Western Nilotic language Dinka as decsribed by Torben Andersen as a possibly extreme case), and, second, even in those cases where it is arguably possible to segment words into discrete formatives, the relations between those and the meanings expressed in the word are notoriously complex (cf. Nen and its relatives as described by Nick Evans and his associates as a possibly extreme case). Whether this bears on the universal applicability of the notion of "word" is unclear to me; however, what is clear to me is that if "word" is not well-defined, then "morpheme" is still worse.

Best regards,


Peter Arkadiev, PhD
Institute of Slavic Studies
Russian Academy of Sciences
Leninsky prospekt 32<https://maps.google.com/?q=Leninsky+prospekt+32&entry=gmail&source=g>-A<https://maps.google.com/?q=Leninsky+prospekt+32&entry=gmail&source=g%3E-A> 119991 Moscow
peterarkadiev at yandex.ru<mailto:peterarkadiev at yandex.ru><mailto:peterarkadiev at yandex.ru>

11.11.2017, 23:49, "Daniel Ross" <djross3 at illinois.edu<mailto:djross3 at illinois.edu><mailto:djross3 at illinois.edu>>:
I did not mean anyone in particular was being too extreme, sorry if I gave that impression. I just wanted to point out that words can still exist language-internally.

I agree with the last two replies to he thread.

To me, the question is whether words are the same across languages, in the same way that nouns and verbs may not be. But we still do talk about nouns and verbs (and, yes, families, however unclear the definition may be).

One other question I realized I wanted to add is that words are like boxes for morphemes, packaging them in groups of some kind. I wonder if the idea of words therefore must be unique, even within a language. Could it be that words exist on a continuum, just like larger and smaller boxes you might use to pack up all of the items in your house when moving? Of course I'm not sure what if anything would be left as the "essence" of words then. If that is not the case I'd like to know why. (Maybe just because it would make typology more complex so we'd rather avoid it?)


On Saturday, November 11, 2017, William Croft <wcroft at unm.edu<mailto:wcroft at unm.edu><mailto:wcroft at unm.edu>> wrote:
I am not arguing for an extreme position like writing grammars without word boundaries either. I am just trying to bring to people’s attention that wordhood is problematic, and to persuade someone to look at wordhood without presupposing an essentialist concept of ‘word’, that would get us past appealing to intuitions which are actually rather unclear on closer inspection. There might be a common core, i.e. a set of crosslinguistically valid criteria which form universal patterns like a typological prototype (as the latter is defined in my “Typology and Universals” textbook). But I don’t know what the criteria are or what their typological relationships are. I would really like to know.

Actually, I *don’t* know what a family is, in a cross-cultural sense, and even in my own culture, given the notions of immediate, nuclear and extended family, foster children, adoption, divorce etc. I don’t even know if ‘family’ makes sense cross-culturally, given the variety of kin systems and the organization of society they reflect.


On Nov 11, 2017, at 12:16 PM, Östen Dahl <oesten at ling.su.se<mailto:oesten at ling.su.se>> wrote:

I am sorry if I gave the impression that I'm arguing for an extreme position (such as writing grammars without word boundaries). I'm rather trying to see what the ultimate consequences are of Martin's proposals. But what I am wondering about is whether there isn't a common core to the language-specific concepts of "word", although it need not involve precise criteria. I think "word" may be a concept rather much like "family". Consider Wikipedia's definition of "family", which hardly provides any criteria that can be used to identify families cross-culturally:

"In the context of human society, a family (from Latin: familia) is a group of people affiliated either by consanguinity (by recognized birth), affinity (by marriage or other relationship), or co-residence (as implied by the etymology of the English word "family"[1]) or some combination of these. Members of the immediate family may include spouses, parents, brothers, sisters, sons, and daughters. Members of the extended family may include grandparents, aunts, uncles, cousins, nephews, nieces, and siblings-in-law. Sometimes these are also considered members of the immediate family, depending on an individual's specific relationship with them."

Still, we think we know what a family is.


-----Ursprungligt meddelande-----
Från: Lingtyp [mailto:lingtyp-bounces at listserv.linguistlist.org] För William Croft
Skickat: den 11 november 2017 20:06
Till: lingtyp at listserv.linguistlist.org<mailto:lingtyp at listserv.linguistlist.org>
Ämne: Re: [Lingtyp] wordhood: responses to Haspelmath

The problem that we need to guard against is using language-specific definitions for a supposedly crosslinguistic (comparative) concept of ‘word’. One has to use a crosslinguistically valid criterion for wordhood, and apply the same criterion across languages. I have yet to see anyone do this.

As usual, the problem is the belief in which linguistic units have essences like ’noun, ‘verb’, ‘word’ etc., and all we linguists need to do is “discover” this essence through some accidental linguistic fact of a particular language (using ‘essence’ and ‘accident’ in the philosophical sense); and it doesn’t matter if the facts are different from one language to the next, or are defined in a way that works only for that language. Until, of course, someone else comes along and decides that the essence is different from what the first person thought, even by looking at the same accidental facts; or maybe that they don’t even believe in the essence.

The solution, in my opinion, is to look at the “accidental" facts, that is, the different criteria for wordhood (defined in a crosslinguistically valid fashion), and find out what the typological universals are that govern those facts. I would expect that (a) the criteria won’t match, within or across languages, as with parts of speech etc.; but (b) the criteria would pattern typologically in such a way that most of the morpheme strings that we would intuitively call “words” would have a fairly high degree of syntagmatic unity most of the time. (Yes, “morpheme” raises some of the same issues -- but if we don’t address these issues, we can’t really trust our results.)


On Nov 11, 2017, at 11:47 AM, Edith A Moravcsik <edith at uwm.edu<mailto:edith at uwm.edu>> wrote:

I agree with Fritz (if I interpret his message correctly).  As far as I can see, we can work with any definition of "word" in crosslinguistic research and then see if that definition is useful or not - i.e., whether it does or does not yield typological correlates. If we try this approach,  I cannot see that we could go wrong; or is there any possible problem that we need to guard against?

Edith Moravcsik

-----Original Message-----
From: Lingtyp [mailto:lingtyp-bounces at listserv.linguistlist.org] On
Behalf Of Frederick J Newmeyer
Sent: Saturday, November 11, 2017 11:04 AM
To: Martin Haspelmath <haspelmath at shh.mpg.de<mailto:haspelmath at shh.mpg.de>>
Cc: lingtyp at listserv.linguistlist.org<mailto:lingtyp at listserv.linguistlist.org>
Subject: Re: [Lingtyp] wordhood: responses to Haspelmath

Let's say that there are no rigid consistent criteria that distinguish words, prefixes, and suffixes. I don't see why that would necessarily prevent us from making valid generalizations about prefixes and suffixes. Consider an analogy. We can make valid generalizations about men and women (their preferences for whatever, their likelihood to do whatever, etc.) even though gender is to a certain extent fluid. There are adults who consider themselves neither male or female and others who consider themselves both. Different criteria lead to different assignments for being a man or for being a woman. It seems like an analogous issue would come up for virtually any 'natural' category. What is the essential problem here?


Frederick J. Newmeyer
Professor Emeritus, University of Washington Adjunct Professor, U of
British Columbia and Simon Fraser U

On Sat, 11 Nov 2017, Martin Haspelmath wrote:

As far as I'm aware, only one typologist has taken up the challenge
of my 2011 paper: Matthew Dryer in his 2015 ALT talk at Albuquerque (I have copied his abstract below, as it seems to be no longer available from the UNM website).

Otherwise, the reaction has generally been that this is old news (for
those with no stake in the syntax-morphology distinction), or that
the distinction is fuzzy, like almost all distinctions in language.
But the latter reaction misses the point that it's not clear whether
there are any cross-linguistic regularities to begin with (apart from
orthographic conventions) that point to the cross-linguistic
relevance of something like a "word" notion. (The results of the
recent work by Jim Blevins and colleagues do seem to point in this
direction, but it is only based on four European languages.)

An interesting case is OUP's recent handbook on polysynthesis: While
all definitions of polysynthesis make reference to the "word" notion, almost none of the authors and editors try to justify it, instead simply presupposing that there is such a thing as polysynthesis.

(The one paper that addresses the issue, by Bickel & Zúñiga, agrees
with my skepticism in that it finds that "polysynthetic "words" are often not unified entities defined by a single domain on which all criteria would converge". OUP's handbook is hard to access, but a manuscript version of Bickel & Zúñiga can be found here:



Evidence for the suffixing preference

Matthew S. Dryer

University at Buffalo

Haspelmath (2011) argues that there are no good criteria for
distinguishing affixes from separate words, so that claims that make
reference to a distinction between words and affixes are suspect. He
claims that there is therefore no good evidence for the suffixing
preference (Greenberg 1957). since that assumes that one can distinguish affixes from separate words. He implies that decisions that linguists describing languages make in terms of what they represent as words may at best be based on inconsistent criteria and he has suggested that we have no way of knowing whether the apparent suffixing preference reflects anything more than the fact that the orthography of European languages far more often represents grammatical morphemes as suffixes than as prefixes.

In this paper, I provide evidence that the suffixing preference is unlikely to be an artifact of orthographic conventions, at least as it applies to tense-aspect affixes.
I examined the phonological properties of tense-aspect affixes in a sample of over 500 languages, distinguishing two types on the basis of their phonological properties.
Type 1 affixes are either ones that are nonsyllabic, consisting only
of consonants, or ones that exhibit allomorphy that is conditioned
phonologically by verb stems. Type
2 affixes are those that exhibit neither of these two properties. The
reason that this distinction is relevant is that grammatical
morphemes of the first sort are almost always represented as affixes
rather than as separate words in grammatical descriptions, so that we
can safely assume that in the vast majority of cases, grammatical morphemes of this sort that are represented as affixes really are such. Haspelmath’s suggestion that the suffixing preference might be an artifact of orthographic conventions thus predicts that we should not find a significant difference in the relative frequency of Type 1 prefixes and suffixes, but only with Type 2 prefixes and suffixes.

The results of my study show that this prediction is not confirmed.
They show that for both types of affixes, suffixes outnumber prefixes
by a little over 2.5 to 1. The number of languages in my sample with
Type 1 suffixes outnumber the number of languages with Type 1 prefixes by 181 to 67, or around 2.7 to 1, while the number of languages with only Type 2 suffixes outnumber the number of languages with only Type 2 prefixes by 223 to 85, approximately 2.6 to 1. Thus the prediction that the suffixing preference should be found primarily with Type 2 affixes, is not borne out. To the contrary, we find the same suffixing preference among both types of affixes.

This provides evidence that, at least for tense-aspect affixes, the suffixing preference is real and not an artifact of orthographic conventions.


Haspelmath, Martin. 2011. The indeterminacy of word segmentation and
the nature of morphology and syntax. Folia Linguistica 45: 31-80.>

On 10.11.17 06:11, Adam J Tallman wrote:
     I am writing a paper about wordhood - has anyone responded to Haspelmath's 2011 Folia Linguistica paper on the topic?

I have only found two sources that mention the paper and seem to put forward an argument against its conclusions, but its mostly in en passant fashion.

On is Blevins (2016) Word and Paradigm Morphology and another is Geertzen, Jeroen, James P. Blevins & Petar Milin. ‘Informativeness of unit boundaries’
[pdf]. Italian Journal of Linguistics 28(2), 1–24.

Any correspondence in this regard would be greatly appreciated,


Adam J.R. Tallman Investigador del Museo de Etnografía y Folklore, la
Paz PhD candidate, University of Texas at Austin

Martin Haspelmath (haspelmath at shh.mpg.de<mailto:haspelmath at shh.mpg.de>) Max Planck Institute for
the Science of Human History
Kahlaische Strasse 10<https://maps.google.com/?q=Kahlaische+Strasse+10&entry=gmail&source=g>
D-07745 Jena
Leipzig University
IPF 141199
Nikolaistrasse 6-10<https://maps.google.com/?q=Nikolaistrasse+6-10+%3E%3E%3E+D-04109+Leipzig&entry=gmail&source=g>
D-04109 Leipzig<https://maps.google.com/?q=Nikolaistrasse+6-10+%3E%3E%3E+D-04109+Leipzig&entry=gmail&source=g>

Lingtyp mailing list
Lingtyp at listserv.linguistlist.org<mailto:Lingtyp at listserv.linguistlist.org>

Lingtyp mailing list
Lingtyp at listserv.linguistlist.org<mailto:Lingtyp at listserv.linguistlist.org>

Lingtyp mailing list
Lingtyp at listserv.linguistlist.org<mailto:Lingtyp at listserv.linguistlist.org>

Lingtyp mailing list
Lingtyp at listserv.linguistlist.org<mailto:Lingtyp at listserv.linguistlist.org><mailto:Lingtyp at listserv.linguistlist.org>

Lingtyp mailing list
Lingtyp at listserv.linguistlist.org<mailto:Lingtyp at listserv.linguistlist.org>

Lingtyp mailing list
Lingtyp at listserv.linguistlist.org<mailto:Lingtyp at listserv.linguistlist.org>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.linguistlist.org/pipermail/lingtyp/attachments/20171112/a9e5b1a3/attachment.htm>

More information about the Lingtyp mailing list