<html>
  <head>
    <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
  </head>
  <body text="#000000" bgcolor="#FFFFFF">
    <p><br>
    </p>
    <p>Hi Johanna, I wouldn't call your option (2) a 'disjunctive'
      definition but an '(annotation) decision tree', and I would
      actually consider this good practice if the criteria are
      systematically related to each other (I think Martin made that
      point, too). They may, for instance, form hierarchies of
      specificity, e.g. insofar as Criterion A implies Criterion B etc.
      If the most specific criterion is not available you move on to the
      next, just like you wrote. In the project that I mentioned we used
      a tree for one of the semantic variables (not hierachically
      ordered though) and it worked very well, at least in terms of
      inter-annotator agreement. It also makes coding much faster (but
      has to be treated with care, obviously).<br>
    </p>
    <p>Volker<br>
    </p>
    <br>
    <div class="moz-cite-prefix">Am 18.11.2017 um 13:40 schrieb Johanna
      NICHOLS:<br>
    </div>
    <blockquote type="cite"
cite="mid:CAHDpjwoTorbtc4_B-mBqKWbC-tVwZXrr8ugM0+bGZRXixBwsBA@mail.gmail.com">
      <div dir="ltr"><br>
        I want to come back to the question of consistency in
        disjunctive definitions, i.e. the danger that linguists pick and
        choose which diagnostics to apply for which languages.  Without
        a more careful review of the literature I can't tell whether
        what's been done is (a) cherry-picking or (b) use of as many of
        the criteria as are applicable to the language in question.  If
        it's the latter, or any other strict procedure for choosing,
        things are more rigorous and consistent than is assumed.  <br>
        <br>
        In the cherry-picking view a set of criteria are treated like
        this:<br>
        <br>
        (1)    Criterion A, and/or<br>
                Criterion B, and/or<br>
                Criterion D, and/or<br>
                Criterion D, and/or<br>
                Criterion E.<br>
        <br>
        What would be the impact of using them like this:<br>
        <br>
        (2)    Criterion A if applicable; if not<br>
                Criterion B if applicable; if not<br>
                Criterion C if applicable; if not<br>
                Criterion D if applicable; if not<br>
                Criterion E.<br>
        <br>
        The first way gives you a free choice to choose what works.  The
        second way requires you to go through the criteria in a set
        order, starting with a specific one as first candidate, and
        proceeding to the second only if the first is inapplicable, i.e.
        if the language lacks the means to use it.  If the language does
        have the means for criterion A but fails to deploy them in such
        a way that we could use it to identify wordhood, then we have to
        stop, and the outcome is No, the object in question is not a
        word.  Only if criterion A is completely inapplicable to the
        language do we get to try B.  And so on.  The procedure stops at
        the first success: if Criterion A is applicable and
        distinguishes some unit, that unit is a word.  If none are
        applicable the language has nothing that passes muster as a
        word.<br>
        <br>
        If the criteria are carefully chosen and defined and ordered,
        this procedure is rigorous and consistent.  (Both theoretical
        and empirical work would have to go into choosing and ordering
        them.)  (2) is probably too strict and would give many false
        negatives (languages with no such thing as a word). 
        Alternatives  might be:<br>
        <br>
        (3)    All of Criteria A-E that are applicable.<br>
        (4)    A majority of Criteria A-E that are applicable.<br>
        (5)    (Some more complex combination of ordering and
        applicability.)<br>
        (6)    (Some constraint other than applicability, plus rules for
        choosing.)<br>
        <br>
        Of these, (3) probably overdoes the rigor.  (4) allows for
        inconsistency since you can pick and choose to make up your
        majority.  (2), (3), and (4) are theoretically unprincipled
        since applicability is due to random typological or historical
        contingencies.  Something like (5) or (6) might strike a good
        balance.<br>
        <br>
        So, to what extent have procedures like (2)-(6) been implicit in
        what has actually been done in the literature?  Or what such
        approach to disjunctive criteria might work?  Maybe fuzzy
        concepts can have some bones in them after all.<br>
        <br>
      </div>
      <div class="gmail_extra"><br>
        <div class="gmail_quote">On Sat, Nov 18, 2017 at 3:27 PM,
          Johanna NICHOLS <span dir="ltr"><<a
              href="mailto:johanna@berkeley.edu" target="_blank"
              moz-do-not-send="true">johanna@berkeley.edu</a>></span>
          wrote:<br>
          <blockquote class="gmail_quote" style="margin:0 0 0
            .8ex;border-left:1px #ccc solid;padding-left:1ex">
            <div dir="ltr">
              <div>
                <div>Volker,  <br>
                  <br>
                </div>
                If there's a way to do this diplomatically, could you
                cite an example or two of  "important publications by
                major authorities of the field where these criteria are
                simply not applied"?   In linguistics we don't have as
                much technical comment on publications as some other
                fields do, and maybe we should.  In journals where I see
                technical comments sections those comments are refereed,
                edited, brief, and focused on factual and methodological
                matters, i.e. about empirical fundamentals and not
                debate on theoretical frameworks.</div>
              <div><br>
              </div>
              <div>If there's no way to do it diplomatically, never
                mind.<br>
              </div>
              <div><br>
              </div>
              Johanna<br>
            </div>
            <div class="gmail_extra"><br>
              <div class="gmail_quote">
                <div>
                  <div class="h5">On Sat, Nov 18, 2017 at 12:37 PM,
                    Volker Gast <span dir="ltr"><<a
                        href="mailto:volker.gast@uni-jena.de"
                        target="_blank" moz-do-not-send="true">volker.gast@uni-jena.de</a>></span>
                    wrote:<br>
                  </div>
                </div>
                <blockquote class="gmail_quote" style="margin:0 0 0
                  .8ex;border-left:1px #ccc solid;padding-left:1ex">
                  <div text="#000000" bgcolor="#FFFFFF">
                    <div>
                      <div class="h5">
                        <p>Matthew -- are you saying that "one cannot
                          rule out disjunctively defined comparative
                          concept" because this is what you did?</p>
                        <p>I am not convinced by "disjunctive
                          comparative concepts". Now, that's nothing for
                          you to worry about -- I'm just one reader
                          (actually, audience of your ALT/2015-talk) who
                          doesn't buy your conclusions because he
                          doesn't accept your operationalizations.</p>
                        <p>But if we want "to talk TO each other (not
                          only PAST each other)", as Martin writes, it
                          would be good to have what other fields call
                          "standards of empirical research". We have
                          copied a lot of statistical methods from
                          fields such as the social sciences and
                          biology. I think it would also be beneficial
                          to take a look at their standards at the
                          "lower" level -- for instances, wrt how data
                          is gathered, processed and classified, how
                          hypotheses are operationalized, etc., to make
                          sure that the results obtained by somebody are
                          also accepted by others (just think of the
                          5%-threshold for statistical significance,
                          which is just a matter of convention).<br>
                        </p>
                        <p>I'm aware that this type of remark is
                          annoying for some of you. I teach both corpus
                          linguistics and typology. In corpus
                          linguistics our students deal with very basic
                          questions of empirical research -- like the
                          traditional 'quality criteria' -- e.g.
                          (external, internal) validity, objectivity,
                          reliability -- and then, in typology, we read
                          important publications by major authorities of
                          the field where these criteria are simply not
                          applied, sometimes the statistics are faulty,
                          and students do enquire about this. What can I
                          say? There are no research standards in
                          typology? There is an ongoing discussion about
                          "arbitrary/subjective/random/d<wbr>isjunctive
                          comparative concepts" on the Lingtype-list?
                          I'm afraid it wouldn't convince them. What I
                          say is that typology still has some way to go
                          to in terms of research methods. There are
                          many non-trivial problems, as we have seen in
                          various discussions on this list, and we
                          should be aware that linguistic data is sui
                          generis (for instance, I think we can't adopt
                          just any method/software package from
                          genetics). But we shouldn't use "authority" as
                          a criterion in our methodological choices, and
                          the choices shouldn't be made in such a way to
                          legitimize our own research 'ex post'.<br>
                        </p>
                        <p>Volker<br>
                        </p>
                        <br>
                        <div
                          class="m_-3530313933156395715m_-3772461755557999611moz-cite-prefix">Am
                          18.11.2017 um 07:36 schrieb Dryer, Matthew:<br>
                        </div>
                      </div>
                    </div>
                    <blockquote type="cite">
                      <div>
                        <div class="h5">
                          <div>
                            <p class="MsoNormal">With respect to
                              Martin’s comment</p>
                            <p class="MsoNormal"> </p>
                            <p class="MsoNormal">“<span
                                style="font-size:14.0pt;font-family:Calibri">It
                                is my impression that such ortho-affixes
                                (= forms written as affixes) are perhaps
                                even more common than “phonologically
                                weak” ortho-affixes, but this is an
                                empirical question (in his 2015 ALT
                                abstract, Matthew mentions 248 languages
                                with weak affixes, but 308 languages
                                with only affixes of the Tauya type,
                                apparently confirming my impression).</span>”</p>
                            <p class="MsoNormal"> </p>
                            <p class="MsoNormal">I realize that this is
                              a reasonable inference from my abstract,
                              but one often has to simplify things for
                              the purposes of an abstract. My definition
                              of a weak affix is very narrow and many if
                              not most affixes that are not weak affixes
                              by my narrow criteria can still be shown
                              to be attached phonologically by broader
                              criteria. Furthermore, I also treat a
                              morpheme as an affix for the purposes of
                              this study if it triggers phonologically
                              conditioned allomorphy in stems it
                              attaches to and it is clear from
                              Macdonald’s description of Tauya that some
                              of the ortho-affixes Martin mentions do
                              trigger phonologically conditioned
                              allomorphy in stems they attach to (pp 54,
                              72, 74, 79). </p>
                            <p class="MsoNormal"> </p>
                            <p class="MsoNormal">I counted an affix as
                              weak for the purposes of the study in my
                              2015 ALT talk only if the description of
                              it in a grammar makes clear that it is
                              nonsyllabic (or has nonsyllabic
                              allomorphs) or that it exhibits
                              phonologically allomorphy or triggers
                              phonologically conditioned allomorphy in
                              adjacent stems. But in many grammars, it
                              is only in the discussion of phonology
                              that it becomes clear that a given affix
                              exhibits phonologically conditioned
                              allomorphy or that it triggers
                              phonologically conditioned allomorphy in
                              adjacent stems. But because I wanted to
                              include a large sample of languages and
                              because it is often unclear from
                              discussions of phonology whether
                              particular rules apply to particular
                              affixes or stems such affixes combine
                              with, I adopted the procedure of not
                              consulting the discussions of phonology in
                              classifying ortho-affixes as weak. This
                              made sense for my 2015 ALT talk since I
                              was examining whether there is a suffixing
                              preference and restricting attention to
                              weak affixes so defined applies equally to
                              prefixes and suffixes. For a different
                              type of typological study, this would have
                              been inappropriate. This illustrates how
                              comparative concepts are specific to
                              particular typological studies.</p>
                            <p class="MsoNormal"> </p>
                            <p class="MsoNormal">Furthermore, there are
                              other factors that I did not examine that
                              are relevant to whether a given
                              ortho-affix is attached phonologically.
                              There may be clear evidence from
                              allophonic rules, but it is often very
                              unclear from grammatical descriptions
                              whether particular allophonic rules apply
                              to particular ortho-affixes or stems to
                              which ortho-affixes are attached. And even
                              if the information is there in the
                              grammatical description, it may take a lot
                              of work to see whether they apply to a
                              particular affix. For example, careful
                              examination of Macdonald’s description of
                              Tauya implies that the benefactive
                              ortho-affix <i>-pe</i> that Martin
                              mentions is attached phonologically, since
                              she gives examples of phonetic
                              representations of forms containing this
                              morpheme where it takes the form [-be]
                              after /m/ ([tembe] on page 54).</p>
                            <p class="MsoNormal"> </p>
                            <p class="MsoNormal">There might also be
                              evidence from stress, but still be unclear
                              how stress is assigned to forms including
                              ortho-affixes. For example, Tauya has
                              word-final stress, but it is not clear
                              from Macdonald’s description whether this
                              means that nouns bearing the ortho-affixes
                              that Martin mentions take stress on the
                              ortho-affix.</p>
                            <p class="MsoNormal"> </p>
                            <p class="MsoNormal">Some of you may have
                              noticed that what I say here contradicts
                              what I said in my earlier email about
                              comparative concepts needing to be
                              exhaustive. The comparative concept I used
                              in my 2015 ALT talk was not exhaustive and
                              was in fact disjunctive. Since that seemed
                              appropriate for that study, this suggests
                              that one cannot rule out disjunctively
                              defined comparative concepts. I sympathize
                              with Martin’s objecting to disjunctive
                              comparative concepts as a way to continue
                              to use confusing and ambiguous terms and I
                              agree that there is something odd about
                              arbitrary disjunctive comparative
                              concepts, but it is a mistake to simply
                              rule out disjunctive comparative concepts.</p>
                            <p class="MsoNormal"> </p>
                            <p class="MsoNormal">I should note finally
                              that while it is clear that the
                              ortho-affixes that Martin mentions are
                              attached phonologically, they are actually
                              not affixes by either his criteria or mine
                              since they are clitics that attach to
                              postnominal modifiers. [Martin has written
                              about problems with the use of the term
                              “clitic”. I am in complete agreement with
                              him about this. But I use the term here
                              and elsewhere in my research (including my
                              upcoming ALT talk on the encliticization
                              preference) as a label for a comparative
                              concept for grammatical morphemes that are
                              phonologically attached but attach to
                              stems of more than one stem class.]</p>
                            <p class="MsoNormal"> </p>
                            <p class="MsoNormal">Matthew</p>
                          </div>
                          <div><br>
                          </div>
                        </div>
                      </div>
                      <span
                        id="m_-3530313933156395715m_-3772461755557999611OLK_SRC_BODY_SECTION">
                        <div>
                          <div class="h5">
                            <div
style="font-family:Calibri;font-size:11pt;text-align:left;color:black;BORDER-BOTTOM:medium
                              none;BORDER-LEFT:medium
none;PADDING-BOTTOM:0in;PADDING-LEFT:0in;PADDING-RIGHT:0in;BORDER-TOP:#b5c4df
                              1pt solid;BORDER-RIGHT:medium
                              none;PADDING-TOP:3pt"> <span
                                style="font-weight:bold">From: </span>Lingtyp
                              <<a
                                href="mailto:lingtyp-bounces@listserv.linguistlist.org"
                                target="_blank" moz-do-not-send="true">lingtyp-bounces@listserv.ling<wbr>uistlist.org</a>>
                              on behalf of Martin Haspelmath <<a
                                href="mailto:haspelmath@shh.mpg.de"
                                target="_blank" moz-do-not-send="true">haspelmath@shh.mpg.de</a>><br>
                              <span style="font-weight:bold">Date: </span>Thursday,
                              November 16, 2017 at 7:14 PM<br>
                              <span style="font-weight:bold">To: </span>"<a
href="mailto:lingtyp@listserv.linguistlist.org" target="_blank"
                                moz-do-not-send="true">lingtyp@listserv.linguistlist<wbr>.org</a>"
                              <<a
                                href="mailto:lingtyp@listserv.linguistlist.org"
                                target="_blank" moz-do-not-send="true">lingtyp@listserv.linguistlist<wbr>.org</a>><br>
                              <span style="font-weight:bold">Subject: </span>Re:
                              [Lingtyp] wordhood: bonded vs. bound<br>
                            </div>
                            <div><br>
                            </div>
                          </div>
                        </div>
                        <div>
                          <div bgcolor="#FFFFFF" text="#000000">
                            <div>
                              <div class="h5"> Matthew Dryer thinks that
                                wordhood is generally understood by
                                grammar authors in terms of <b>bondedness</b>
                                (= phonological weakness, as shown by
                                nonsyllabicity and phono-conditioned
                                allomorphy), not in terms of <b>boundness</b>
                                (= inability to occur in isolation).
                                <p class="MsoNormal"> </p>
                                I don’t know if this is true, but
                                Matthew actually recognizes that
                                grammars often describe grammatical
                                markers as “affixes” even when they do
                                not show the two “phonological weakness”
                                (or bondedness) features.
                                <p class="MsoNormal"> </p>
                                For example, Tauya (a language of New
                                Guinea) is said to have (syllabic) case
                                suffixes, but these never show any
                                allomorphy, e.g.
                                <p class="MsoNormal"> </p>
                                fena’a-ni [woman-ERG]<br>
                                na-pe [you-BEN]<br>
                                wate-’usa [house-INESS]<br>
                                Aresa-nani [Aresa-ALL]<br>
                                Tauya-sami [Tauya-ABL] (MacDonald 1990:
                                119-126)
                                <p class="MsoNormal"> </p>
                                It is my impression that such
                                ortho-affixes (= forms written as
                                affixes) are perhaps even more common
                                than “phonologically weak”
                                ortho-affixes, but this is an empirical
                                question (in his 2015 ALT abstract,
                                Matthew mentions 248 languages with weak
                                affixes, but 308 languages with only
                                affixes of the Tauya type, apparently
                                confirming my impression).
                                <p class="MsoNormal"> </p>
                                For this reason, I have suggested that
                                the stereotypical “affix” notion should
                                perhaps be captured in terms of
                                boundness together with
                                single-root-class adjacency. Since the
                                Tauya case-markers attach only to nouns,
                                they count as affixes; by contrast, if a
                                bound role marker attaches to both nouns
                                (English “for children”) and adjectives
                                (“for older children”) as well as to
                                other elements (“for many children”), we
                                do not regard it as an affix (but as a
                                preposition), even if it is bound (=
                                does not occur in isolation; English
                                "for" does not).
                                <p class="MsoNormal"> </p>
                                Matthew quite rightly points out that
                                this notion of boundness (which goes
                                back at least to Bloomfield 1933: §10.1)
                                implies that most function words in
                                English are bound, and in fact most
                                function words in most languages are
                                bound – but this is exactly what we
                                want, I feel, because the best way to
                                define a “function word” is as a bound
                                element that is not an affix. Linguists
                                often think of function words (or
                                “functional categories”) as defined
                                semantically, but it is actually very
                                hard to say what is the
                                semantic(-pragmatic) difference between
                                a plural marker and a word like
                                “several”, between a dual marker and the
                                word “two”, between a past-tense marker
                                and the expression “in the past”, or
                                between a comitative marker and the word
                                “accompany”. It seems to me that these
                                distinctions are best characterized in
                                terms of boundness, i.e. inability to
                                occur in isolation.
                                <p class="MsoNormal"> </p>
                                It may be true that occurrence in
                                isolation is a feature of an element
                                that is not easy to elicit from
                                speakers, but in actual language use,
                                there are a very large number of very
                                short utterances, so at least positive
                                evidence for free status (=non-bound
                                status) is not difficult to obtain.
                                <p class="MsoNormal"> </p>
                                In any event, it seems clear to me that
                                some key concepts of grammatical
                                typology such as “flag” (= bound role
                                marker on a nominal) and “person index”
                                (= bound person marker, generally on a
                                verb) require the Bloomfieldian
                                boundness notion, and that these
                                concepts are much easier to work with in
                                typology than the traditional
                                stereotypical notions of “case”,
                                “adposition”, “agreement marker”, and
                                “pronominal clitic”. (For bound person
                                forms, this was a major lesson of Anna
                                Siewierska’s 2004 book “Person”.)
                                <p class="MsoNormal"> </p>
                                Best,<br>
                                Martin<br>
                                <br>
                                <div
                                  class="m_-3530313933156395715m_-3772461755557999611moz-cite-prefix">On
                                  14.11.17 07:02, Dryer, Matthew wrote:<br>
                                </div>
                              </div>
                            </div>
                            <blockquote type="cite">
                              <div><span class="">
                                  <p class="MsoNormal"
                                    style="font-size:14px">I have a
                                    number of problems with Martin’s
                                    proposal:</p>
                                  <p class="MsoNormal"
                                    style="font-size:14px"> </p>
                                  <p class="MsoNormal"
                                    style="font-size:14px"><span
                                      style="font-size:14pt">"<b>Here’s
                                        a proposal for defining a notion
                                        of “affix”, in such a way that
                                        the results do not go too much
                                        against our intuitions or
                                        stereotypes:</b></span></p>
                                  <p class="MsoNormal"
                                    style="font-size:14px"><span
                                      style="font-size:14pt"> </span></p>
                                  <p class="MsoNormal"
                                    style="font-size:14px"><b><span
                                        style="font-size:14pt">An affix
                                        is a bound form that always
                                        occurs together with a root of
                                        the same root-class and is never
                                        separated from the root by a
                                        free form or a non-affixal bound
                                        form."</span></b><span
                                      style="font-size:14pt"></span></p>
                                  <p class="MsoNormal"
                                    style="font-size:14px"> </p>
                                  <p class="MsoNormal"
                                    style="font-size:14px">If one
                                    examines the notion of “bound” from
                                    his 2013 paper, I believe it implies
                                    a comparative concept of affix that
                                    differs greatly from what most
                                    linguists (at last most
                                    non-generative linguists) understand
                                    by the term. That’s not a problem
                                    for it as a comparative concept, but
                                    it is a comparative concept that
                                    differs considerably from the
                                    stereotype.</p>
                                  <p class="MsoNormal"
                                    style="font-size:14px"> </p>
                                  <p class="MsoNormal"
                                    style="font-size:14px">Martin’s
                                    definition of “free and “bound” from
                                    his 2013 paper is as follows:</p>
                                  <p class="MsoNormal"
                                    style="font-size:14px"> </p>
                                  <p class="MsoNormal"><b><span
                                        style="font-family:Times;font-size:18px">"But
                                        distinguishing in a general way
                                        between bound elements and free
                                        elements is quite
                                        straightforward, because there
                                        is a single criterion: Free
                                        forms are forms that can occur
                                        on their own, i.e. in a complete
                                        (possibly elliptical) utterance
                                        (Bloomfield 1933: 160). This
                                        criterion correlates very highly
                                        with the criterion of
                                        contrastive use: Only free forms
                                        can be used contrastively."</span></b></p>
                                  <p class="MsoNormal"
                                    style="font-size:14px"><span
                                      style="font-size:13.0pt;font-family:Times"> </span></p>
                                  <p class="MsoNormal"
                                    style="font-size:14px"><span
                                      style="font-family:Times">First, I
                                      find the notion of complete
                                      utterance ambiguous. Does it mean
                                      utterances in normal speech or
                                      does it include metalinguistic
                                      uses (like “What is the last word
                                      in the sentence “Who are you going
                                      with”? Answer “with”). I would
                                      assume that it does not include
                                      such metalinguistic uses. But then
                                      many if not most so-called
                                      function words in English would
                                      count as bound since they cannot
                                      be used as complete utterances.
                                      Perhaps other speakers of English
                                      would have different intuitions,
                                      but if so that only indicates the
                                      lack of clarity in the notion.
                                      Furthermore, for many function
                                      words in English, I am not sure
                                      how to judge whether they can
                                      occur alone as utterances. Many
                                      such so-called function words
                                      would appear to count as bound by
                                      Martin’s definition, though they
                                      would not count as affixes since
                                      they lack other properties in his
                                      definition of “affix”.</span></p>
                                  <p class="MsoNormal"
                                    style="font-size:14px"><span
                                      style="font-family:Times"> </span></p>
                                  <p class="MsoNormal"
                                    style="font-size:14px"><span
                                      style="font-family:Times">Second,
                                      many languages have grammatical
                                      morphemes that must occur adjacent
                                      to an open class word but which
                                      behave as separate words
                                      phonologically. These would all
                                      apparently count as affixes by
                                      Martin’s definition. Again, I have
                                      no problem with this as a
                                      comparative concept, only that it
                                      means his notion of affix deviates
                                      considerably from the stereotype.</span></p>
                                  <p class="MsoNormal"
                                    style="font-size:14px"><span
                                      style="font-family:Times"> </span></p>
                                  <p class="MsoNormal"
                                    style="font-size:14px"><span
                                      style="font-family:Times">Third,
                                      Martin says that his criterion “</span><span
                                      style="font-family:Times">correlates
                                      very highly with the criterion of
                                      contrastive use</span><span
                                      style="font-family:Times">”. But
                                      by my intuitions, the ability to
                                      occur as complete utterances does
                                      not correlate closely with the
                                      criterion of contrastive use,
                                      since most so-called function
                                      words CAN occur with contrastive
                                      use (such as can in this
                                      sentence!), as can some morphemes
                                      that are conventionally treated as
                                      affixes, like <i>un-</i> in “I’m
                                      not happy, I’m UNhappy”. Of
                                      course, Martin might argue that <i>
                                        un-</i> is more like so-called
                                      function words and less like
                                      morphemes conventionally treated
                                      as affixes. But the fact remains
                                      that <i>un-</i> is easily the
                                      locus of contrast but cannot be
                                      used as a complete utterance. I
                                      thus see no evidence of a close
                                      correlation between the ability to
                                      occur as a complete utterance and
                                      the ability to be the locus of
                                      contrast.</span></p>
                                  <p class="MsoNormal"
                                    style="font-size:14px"><span
                                      style="font-family:Times"> </span></p>
                                  <p class="MsoNormal"
                                    style="font-size:14px"><span
                                      style="font-family:Times">Finally,
                                      it is my experience that languages
                                      differ in their conventions
                                      regarding what can be a complete
                                      utterance. Imagine two closely
                                      related languages that differ in
                                      their grammatical rules governing
                                      what is a complete utterance. By
                                      Martin’s definition, there might
                                      be a large number of morphemes
                                      that count as separate words in
                                      one language but as affixes in the
                                      other language. This strikes me as
                                      odd. It seems odd to have a
                                      criterion for what is a word and
                                      what is an affix so dependent on
                                      the grammatical rules in the
                                      language for what constitutes a
                                      complete utterance.</span></p>
                                  <p class="MsoNormal"
                                    style="font-size:14px"><span
                                      style="font-family:Times"> </span></p>
                                </span>
                                <p class="MsoNormal"
                                  style="font-size:14px"><span
                                    style="font-family:Times">Matthew</span></p>
                              </div>
                              <div style="font-size:14px"><br>
                              </div>
                              <span
                                id="m_-3530313933156395715m_-3772461755557999611OLK_SRC_BODY_SECTION"
                                style="font-size:14px">
                                <div
style="font-family:Calibri;font-size:11pt;text-align:left;color:black;BORDER-BOTTOM:medium
                                  none;BORDER-LEFT:medium
none;PADDING-BOTTOM:0in;PADDING-LEFT:0in;PADDING-RIGHT:0in;BORDER-TOP:#b5c4df
                                  1pt solid;BORDER-RIGHT:medium
                                  none;PADDING-TOP:3pt"><span class="">
                                    <span style="font-weight:bold">From:
                                    </span>Lingtyp <<a
                                      href="mailto:lingtyp-bounces@listserv.linguistlist.org"
                                      target="_blank"
                                      moz-do-not-send="true">lingtyp-bounces@listserv.ling<wbr>uistlist.org</a>>
                                    on behalf of Martin Haspelmath <<a
href="mailto:haspelmath@shh.mpg.de" target="_blank"
                                      moz-do-not-send="true">haspelmath@shh.mpg.de</a>><br>
                                  </span><span style="font-weight:bold">Date:
                                  </span>Sunday, November 12, 2017 at
                                  10:47 PM<span class=""><br>
                                    <span style="font-weight:bold">To: </span>"<a
href="mailto:lingtyp@listserv.linguistlist.org" target="_blank"
                                      moz-do-not-send="true">lingtyp@listserv.linguistlist<wbr>.org</a>"
                                    <<a
                                      href="mailto:lingtyp@listserv.linguistlist.org"
                                      target="_blank"
                                      moz-do-not-send="true">lingtyp@listserv.linguistlist<wbr>.org</a>><br>
                                    <span style="font-weight:bold">Subject:
                                    </span>Re: [Lingtyp] wordhood<br>
                                  </span></div>
                                <div><br>
                                </div>
                                <div>
                                  <div bgcolor="#FFFFFF" text="#000000"><span
                                      class="">
                                      <p class="MsoNormal">Mattis List
                                        and Balthasar Bickel rightly
                                        emphasize that “word” is not a
                                        Platonic entity (a natural kind)
                                        that exists in advance of
                                        language learning or linguistic
                                        analysis – few linguists would
                                        disagree here, not even
                                        generativists (who otherwise
                                        liberally assume natural-kind
                                        catgeories).</p>
                                      <p class="MsoNormal"> </p>
                                      <p class="MsoNormal">But I think
                                        many linguists still ACT AS IF
                                        there were such a natural kind,
                                        because the “word” notion is a
                                        crucial ingredient to a number
                                        of other notions that linguists
                                        use routinely – e.g. “gender”,
                                        which is typically defined in
                                        terms of “agreement” (which is
                                        defined in terms of inflectional
                                        marking on targets; and
                                        inflection is defined in terms
                                        of “word”).</p>
                                      <p class="MsoNormal"> </p>
                                      <p class="MsoNormal">So is it
                                        possible to define a comparative
                                        concept ‘word’ that applies to
                                        all languages equally, and that
                                        accords reasonably with our
                                        stereotypes? Note that I didn’t
                                        deny this in my 2011 paper, I
                                        just said that nobody had come
                                        up with a satisfactory
                                        definition (that could be used,
                                        for instance, in defining
                                        “gender” or “polysynthesis”). So
                                        I’ll be happy to contribute to a
                                        discussion on how to make
                                        progress on defining “word”.</p>
                                      <p class="MsoNormal"> </p>
                                      <p class="MsoNormal">Larry Hyman
                                        notes that other notions like
                                        “syllable” and “sentence” are
                                        also problematic in that they
                                        also “leak”. However, I think it
                                        is important to distinguish two
                                        situations of “slipperiness”:</p>
                                      <p class="MsoNormal"> </p>
                                      <p class="MsoNormal">(1) “Leakage”
                                        of definitions due to vague
                                        defining notions</p>
                                      <p class="MsoNormal"> </p>
                                      <p class="MsoNormal">(2)
                                        Incoherence of definitions due
                                        to the use of different criteria
                                        in different languages</p>
                                      <p class="MsoNormal"> </p>
                                      <p class="MsoNormal">The first can
                                        be addressed by tightening the
                                        defining notions, but the second
                                        is fatal.</p>
                                      <p class="MsoNormal"> </p>
                                      <p class="MsoNormal">To take up
                                        Östen Dahl’s example of the
                                        “family” notion: In one culture,
                                        a family might be said to be a
                                        set of minimally three living
                                        people consisting of two adults
                                        (regardless of gender) living in
                                        a romantic relationship plus all
                                        their descendants. In another
                                        culture, a family might be
                                        defined as a married couple
                                        consisting of a man and a woman
                                        plus all their living direct
                                        ancestors, all their (great)
                                        uncles and (great) aunts, and
                                        all the descendants of all of
                                        these.</p>
                                      <p class="MsoNormal"> </p>
                                      <p class="MsoNormal">With two
                                        family concepts as different as
                                        these, it is obviously not very
                                        interesting to ask general
                                        cross-cultural questions about
                                        “families” (e.g. “How often do
                                        all family members have meals
                                        together?”). So the use of
                                        different criteria for different
                                        cultures is fatal here.</p>
                                      <p class="MsoNormal"> </p>
                                      <p class="MsoNormal">What I find
                                        worrying is that linguists often
                                        seem to accept incoherent
                                        definitions of comparative
                                        concepts (this was emphasized
                                        especially in my 2015 paper on
                                        defining vs. diagnosing
                                        categories). Different
                                        diagnostics in different
                                        languages would not be fatal if
                                        “word” were a Platonic
                                        (natural-kind) concept, but if
                                        we are not born with a “word”
                                        category, typologists need to
                                        use the SAME criteria for all
                                        languages.</p>
                                      <p class="MsoNormal"> </p>
                                      <p class="MsoNormal">So here’s a
                                        proposal for defining a notion
                                        of “simple morphosyntactic
                                        word”:</p>
                                      <p class="MsoNormal"> </p>
                                      <p class="MsoNormal"><b>A simple
                                          morphosyntactic word is a form
                                          that consists of (minimally) a
                                          root, plus any affixes.</b></p>
                                      <p class="MsoNormal"> </p>
                                      <p class="MsoNormal">Here’s a
                                        proposal for defining a notion
                                        of “affix”, in such a way that
                                        the results do not go too much
                                        against our intuitions or
                                        stereotypes:</p>
                                      <p class="MsoNormal"> </p>
                                      <p class="MsoNormal"><b>An affix
                                          is a bound form that always
                                          occurs together with a root of
                                          the same root-class and is
                                          never separated from the root
                                          by a free form or a
                                          non-affixal bound form.</b></p>
                                      <p class="MsoNormal"> </p>
                                      <p class="MsoNormal">These
                                        definitions make use of the
                                        notions of “root” and
                                        “root-class” (defined in
                                        Haspelmath 2012) and<span>  </span>“bound
                                        (form)” vs. “free (form)”
                                        (defined in Haspelmath 2013).
                                        All these show leakage as in (1)
                                        above, but they are equally
                                        applicable to all languages, so
                                        they are not incoherent. (I
                                        thank Harald Hammarström for a
                                        helpful discussion that helped
                                        me to come up with the above
                                        definitions, which I had not
                                        envisaged in 2011.)</p>
                                      <p class="MsoNormal"> </p>
                                      <p class="MsoNormal">(What I don’t
                                        know at the moment is how to
                                        relate “simple morphosyntactic
                                        word” to “morphosyntactic word”
                                        in general, because I cannot
                                        distinguish compounds from
                                        phrases comparatively; and I
                                        don’t know what to do with
                                        “phonological word”.)</p>
                                      <p class="MsoNormal"> </p>
                                      <p class="MsoNormal">Crucially,
                                        the definitions above make use
                                        of a number of basic concepts
                                        that apply to ALL languages in
                                        the SAME way. David Gil’s
                                        proposal, to measure “bond
                                        strength” by means of a range of
                                        language-particular phenomena,
                                        falls short of this requirement
                                        (as already hinted by Eitan
                                        Grossman). Note that the problem
                                        I have with David’s proposal is
                                        not that it provides no
                                        categorical contrasts (recall my
                                        acceptance of vagueness in (1)
                                        above), but that there is no way
                                        of telling which phenomena
                                        should count as measuring bond
                                        strength.</p>
                                      <p class="MsoNormal"> </p>
                                      <p class="MsoNormal">David’s
                                        approach resembles Keenan’s
                                        (1976) attempt at defining
                                        “subject” (perhaps not by
                                        accident, because Ed Keenan was
                                        David’s PhD supervisor), but I
                                        have a similar objection to
                                        Keenan: If different criteria
                                        are used for different
                                        languages, how do we know that
                                        we are measuring the same
                                        phenomenon across languages?
                                        Measuring X by means of Y makes
                                        sense only if we know
                                        independently that X and Y are
                                        very highly correlated. But do
                                        we know this, for subjects, or
                                        for bond strength?</p>
                                      <p class="MsoNormal"> </p>
                                      <p class="MsoNormal">Best,</p>
                                      <p class="MsoNormal">Martin</p>
                                      <p class="MsoNormal"> </p>
                                      <br>
                                    </span><span class="">
                                      <pre class="m_-3530313933156395715m_-3772461755557999611moz-signature" cols="72">-- 
Martin Haspelmath (<a class="m_-3530313933156395715m_-3772461755557999611moz-txt-link-abbreviated" href="mailto:haspelmath@shh.mpg.de" target="_blank" moz-do-not-send="true">haspelmath@shh.mpg.de</a>)
Max Planck Institute for the Science of Human History
Kahlaische Strasse 10   
D-07745 Jena  
&
Leipzig University 
IPF 141199
Nikolaistrasse 6-10
D-04109 Leipzig    





</pre>
                                    </span></div>
                                </div>
                              </span></blockquote>
                            <span class=""> <br>
                              <pre class="m_-3530313933156395715m_-3772461755557999611moz-signature" cols="72">-- 
Martin Haspelmath (<a class="m_-3530313933156395715m_-3772461755557999611moz-txt-link-abbreviated" href="mailto:haspelmath@shh.mpg.de" target="_blank" moz-do-not-send="true">haspelmath@shh.mpg.de</a>)
Max Planck Institute for the Science of Human History
Kahlaische Strasse 10   
D-07745 Jena  
&
Leipzig University 
IPF 141199
Nikolaistrasse 6-10
D-04109 Leipzig    





</pre>
                            </span></div>
                        </div>
                      </span><span class=""> <br>
                        <fieldset
                          class="m_-3530313933156395715m_-3772461755557999611mimeAttachmentHeader"></fieldset>
                        <br>
                        <pre>______________________________<wbr>_________________
Lingtyp mailing list
<a class="m_-3530313933156395715m_-3772461755557999611moz-txt-link-abbreviated" href="mailto:Lingtyp@listserv.linguistlist.org" target="_blank" moz-do-not-send="true">Lingtyp@listserv.linguistlist.<wbr>org</a>
<a class="m_-3530313933156395715m_-3772461755557999611moz-txt-link-freetext" href="http://listserv.linguistlist.org/mailman/listinfo/lingtyp" target="_blank" moz-do-not-send="true">http://listserv.linguistlist.o<wbr>rg/mailman/listinfo/lingtyp</a><span class="m_-3530313933156395715HOEnZb"><font color="#888888">
</font></span></pre>
                        <span class="m_-3530313933156395715HOEnZb"><font
                            color="#888888"> </font></span></span></blockquote>
                    <span class="m_-3530313933156395715HOEnZb"><font
                        color="#888888"> <br>
                        <span class="HOEnZb"><font color="#888888">
                            <pre class="m_-3530313933156395715m_-3772461755557999611moz-signature" cols="72">-- 
Prof. Volker Gast
English and American Studies
Ernst-Abbe-PLatz 8
D-07743 Jena

Fon: ++49 3641 9-44546
Fax: ++49 3641 9-44542</pre>
                          </font></span></font></span></div>
                  <span class="">
                    <br>
                    ______________________________<wbr>_________________<br>
                    Lingtyp mailing list<br>
                    <a href="mailto:Lingtyp@listserv.linguistlist.org"
                      target="_blank" moz-do-not-send="true">Lingtyp@listserv.linguistlist.<wbr>org</a><br>
                    <a
                      href="http://listserv.linguistlist.org/mailman/listinfo/lingtyp"
                      rel="noreferrer" target="_blank"
                      moz-do-not-send="true">http://listserv.linguistlist.o<wbr>rg/mailman/listinfo/lingtyp</a><br>
                    <br>
                  </span></blockquote>
              </div>
              <br>
            </div>
          </blockquote>
        </div>
        <br>
      </div>
      <br>
      <fieldset class="mimeAttachmentHeader"></fieldset>
      <br>
      <pre wrap="">_______________________________________________
Lingtyp mailing list
<a class="moz-txt-link-abbreviated" href="mailto:Lingtyp@listserv.linguistlist.org">Lingtyp@listserv.linguistlist.org</a>
<a class="moz-txt-link-freetext" href="http://listserv.linguistlist.org/mailman/listinfo/lingtyp">http://listserv.linguistlist.org/mailman/listinfo/lingtyp</a>
</pre>
    </blockquote>
    <br>
    <pre class="moz-signature" cols="72">-- 
Prof. Volker Gast
English and American Studies
Ernst-Abbe-PLatz 8
D-07743 Jena

Fon: ++49 3641 9-44546
Fax: ++49 3641 9-44542</pre>
  </body>
</html>