<html>
  <head>
    <meta content="text/html; charset=utf-8" http-equiv="Content-Type">
  </head>
  <body bgcolor="#FFFFFF" text="#000000">
    Thanks - would anyone else like to comment?  I'm currently not sure
    whether simply enriching ICONS (and possibly HCONS) with a ...
    notation would be enough to give us what's needed in terms of
    behaviour but I suspect that needs to be part of the solution.<br>
    <br>
    Ann<br>
    <br>
    <div class="moz-cite-prefix">On 18/02/16 00:26, Emily M. Bender
      wrote:<br>
    </div>
    <blockquote
cite="mid:CAMype6cKCD6B42nhaXxmauZGAK1mTg=zZgOA0yx5b1Eo3mvwhA@mail.gmail.com"
      type="cite">
      <div dir="ltr">Just a quick and belatedly reply to say that from
        where I sit your analysis
        <div>of the situation makes a lot of sense.</div>
        <div><br>
        </div>
        <div>Emily</div>
      </div>
      <div class="gmail_extra"><br>
        <div class="gmail_quote">On Sun, Feb 7, 2016 at 2:12 AM, Ann
          Copestake <span dir="ltr">&lt;<a moz-do-not-send="true"
              href="mailto:aac10@cam.ac.uk" target="_blank">aac10@cam.ac.uk</a>&gt;</span>
          wrote:<br>
          <blockquote class="gmail_quote" style="margin:0 0 0
            .8ex;border-left:1px #ccc solid;padding-left:1ex">
            <div bgcolor="#FFFFFF" text="#000000"> Thanks! and thanks
              all!  I've come to a view on this which I think is
              consistent with what everyone has been saying.<br>
              <br>
              First of all, note that in the MRS syntax, we do not
              distinguish between terminated and non-terminated
              lists/bags.  If we think about it from the perspective of
              typed feature structures, it is clear that there is a
              distinction - for instance a type `list' is the most
              general type of list, and the type `e-list' (empty list)
              is usually a maximally specific type.    Coming back to
              the notation I used in an earlier message, there is a
              distinction between { ... } (analogous to list in a TFS)
              and {} (cf e-list).  <br>
              <br>
              Now, there are two possible interpretations of ICONS as it
              arises from a DELPH-IN grammar (i.e., as it is output
              after parsing):<br>
              1. information structure<br>
              2. information structure as it arises from morphosyntax<br>
              <br>
              In the `normal' sentences of the `Kim chased the dog'
              type, no information structure elements arise from
              morphosyntax.  We can, however, expect that various
              contexts (e.g., discourse) give rise to information
              structure in association with such a sentence.  Hence,
              with respect to interpretation 1, ICONS is not strictly
              empty but underspecified (and similarly a one element
              ICONS may be underspecified with respect to a two-element
              ICONS and so on).  I think this is consistent with what
              Sanghoun and Emily are saying.  Under this interpretation,
              an MRS with no specification of ICONS should indeed
              generate all the variants sentences we've been
              discussing.  And so on.<br>
              <br>
              However, with respect to interpretation 2, the ICONS
              emerging from the parse of such a sentence is terminated. 
              Once we've finished parsing, we're guaranteeing no more
              ICONS elements will arise from morphosyntax, whatever
              someone does in discourse.  Under this interpretation, if
              I say I want to generate a sentence with an empty ICONS, I
              mean I want to generate a sentence with no ICONS
              contribution from morphosyntax.  This is also a legitimate
              use of the realiser, considered as a stand-alone module.<br>
              <br>
              Since ICONS is something which I have always thought of as
              on the boundary of morphosyntax and discourse, I want to
              be able to enrich ICONS emerging from parsing with
              discourse processing, so interpretation 1 makes complete
              sense.  However, I believe it is also perfectly legitimate
              to be able to divide the world into what the grammar can
              be expected to do and what it can't, and that is
              consistent with interpretation 2.<br>
              <br>
              As a hypothetical move, consider an additional
              classification of ICONS elements according to whether or
              not they arise from morphosyntax.  Then we can see that a
              single ICONS value could encompass both interpretations. 
              i.e., what would arise from a parse would be a terminated
              list of morphosyntactic-ICONS elements but the ICONS as a
              whole could be non-terminated.<br>
              <br>
              I think there may be reasons to be able to distinguish
              ICONS elements according to whether they are intended as
              grammar-derived or not, though I do see this might look
              messy.  But anyway, I want to first check that everyone
              agrees with this analysis of the situation before trying
              to work out what we might do about it in terms of
              notation. <br>
              <br>
              Incidentally - re Dan's message - my overly brief comment
              about Sanghoun's use of DMRS earlier was intended to point
              out that if DMRS had the necessary links for demonstrating
              ICONS, then in principle this was something we know how to
              extract.  But right now, I'm not clear whether or not we
              do need all the underspecified elements, and that's
              something I would like Dan to comment on before we go
              further.<br>
              <br>
              All best,<br>
              <br>
              Ann
              <div>
                <div class="h5"><br>
                  <br>
                  <br>
                  <div>On 06/02/2016 17:59, Sanghoun Song wrote:<br>
                  </div>
                  <blockquote type="cite">
                    <div dir="ltr">My apologies for my really late
                      reply!<br>
                      <br>
                      I am not sure whether I fully understand your
                      discussion, but I would like to leave several my
                      ideas on using ICONS for generation. <br>
                      <br>
                      First, in my analysis (final version), only
                      expressions that contribute to information
                      structure introduce an ICONS element into the
                      list. For example, the following unmarked sentence
                      (a below) has no ICONS element (i.e. empty ICONS).
                      <br>
                      <br>
                      a. plain: Kim chases the dog.<br>
                      b. passivization: The dog is chased by Kim.<br>
                      c. fronting: The dog Kim chases.<br>
                      d. clefting: It is the dog that Kim chases.<br>
                      <br>
                      Using  the type hierarchy for information
                      structure in my thesis, I can say the followings<br>
                       <br>
                      (i) The subject Kim and the object the dog in a
                      plain active sentence (a) are in situ. They may or
                      may not be focused depending on which constituent
                      bears a specific accent, but in the sentence-based
                      processing their information structure values had
                      better remain underspecified for flexible
                      representation.<br>
                      <br>
                      (ii) The promoted argument the dog in the passive
                      sentence (b) is evaluated as conveying
                      focus-or-topic, while the demoted argument Kim is
                      associated with non-topic. <br>
                      <br>
                      (iii) In (c), the fronted object the dog is
                      assumed to be assigned focus-or-topic in that the
                      sentence conveys a meaning of either "As for the
                      dog, Kim chases it". or (d), while the subject in
                      situ is evaluated as containing neither topic nor
                      focus (i.e. background). (Background may not be
                      implemented in the ERG, I think.)<br>
                      <br>
                      (iv) The focused NP in (d) carries focus, and the
                      subject in the cleft clause Kim is also associated
                      with bg. <br>
                      <br>
                      Thus, we can create a focus specification
                      hierarchy amongst (a-d) as [clefting &gt; fronting
                      &gt; passivization &gt; plain].<br>
                      <br>
                      What I want to say is that a set of sentences
                      which share some properties may have subtle shades
                      of meaning depending on how focus is assigned to
                      the sentences. Paraphrasing is made only in the
                      direction from the right to the left of [clefting
                      &gt; fronting &gt; passivization &gt; plain],
                      because paraphrasing in the opposite direction
                      necessarily causes loss of information. For
                      example, a plain sentence such as (a) can be
                      paraphrased into a cleft construction such as (d),
                      but not vice versa.<br>
                      <br>
                      In a nutshell, a more specific sentence might not
                      better to be paraphrased into a less specific
                      sentence in terms of information structure. <br>
                      <br>
                      Second, I provided many dependency graphs in my
                      thesis. The main reason was that nobody outside of
                      the DELPH-IN can fully understands the complex
                      co-indexation in ICONS/MRS. At that time, I didn't
                      work on DMRS with respect to ICONS. If there is a
                      way to represent ICONS in DMRS (direct from TFS or
                      via MRS), I am interested in the formalism. <br>
                      <br>
                      <br>
                      Sanghoun<br>
                      <br>
                    </div>
                    <div class="gmail_extra"><br>
                      <div class="gmail_quote">On Sat, Feb 6, 2016 at
                        1:26 AM, Ann Copestake <span dir="ltr">&lt;<a
                            moz-do-not-send="true"
                            href="mailto:aac10@cam.ac.uk"
                            target="_blank"><a class="moz-txt-link-abbreviated" href="mailto:aac10@cam.ac.uk">aac10@cam.ac.uk</a></a>&gt;</span>
                        wrote:<br>
                        <blockquote class="gmail_quote" style="margin:0
                          0 0 .8ex;border-left:1px #ccc
                          solid;padding-left:1ex">
                          <div bgcolor="#FFFFFF" text="#000000"> Briefly
                            (more this evening maybe) - I don't see a
                            particular problem with filling in the ICONS
                            since what you describe are relationships
                            that are overt in the *MRS anyway, aren't
                            they?  I thought, in fact, that these are
                            pretty clear from the DMRS graph - which is
                            why Sanghoun uses it to describe what's
                            going on.  <br>
                            <br>
                            I believe we can build the DMRS graph direct
                            from the TFS, incidentally - don't need to
                            go via MRS ...<br>
                            <br>
                            Cheers,<br>
                            <br>
                            Ann
                            <div>
                              <div><br>
                                <br>
                                <div>On 05/02/2016 23:40, Dan Flickinger
                                  wrote:<br>
                                </div>
                                <blockquote type="cite">
                                  <div
style="font-size:12pt;color:#000000;background-color:#ffffff;font-family:Calibri,Arial,Helvetica,sans-serif">
                                    <p>As I understand the Soon and
                                      Bender account, an MRS for a
                                      sentence should include in the
                                      ICONS list at least one element
                                      for each individual (eventuality
                                      or instance) that is introduced. 
                                      In the ERG this would mean that
                                      the value of each ARG0 should
                                      appear in at least one ICONS
                                      entry, where most of these would
                                      be of the maximally underspecified
                                      type `info-str', but possibly
                                      specialized because of syntactic
                                      structure or stress/accent or
                                      maybe even discourse structure.</p>
                                    <p><br>
                                    </p>
                                    <p>I see the virtue of having these
                                      overt ICONS elements even when of
                                      type `info-str', to enable the
                                      fine-grained control that Stephan
                                      notes that we want for generation,
                                      and also to minimize the
                                      differences between the ERG and
                                      grammars being built from the
                                      Matrix which embody Sanghoun's
                                      careful work.</p>
                                    <p><br>
                                    </p>
                                    <p>If the grammarian is to get away
                                      with not explicitly introducing
                                      each of these ICONS elements in
                                      the lexical entries, as Sanghoun
                                      does in the Matrix, then it would
                                      have to be possible to predict and
                                      perhaps mechanically add the
                                      missing ones after composition was
                                      completed.  I used to hope that
                                      this would be possible, but now
                                      I'm doubtful, leading me to think
                                      that there is no good alternative
                                      to the complication (maybe I
                                      should more kindly use the term
                                      `enrichment') of the grammar with
                                      the overt introduction of these
                                      guys everywhere.  Here's my
                                      reasoning:</p>
                                    <p><br>
                                    </p>
                                    <p>I assume that what we'll want in
                                      an MRS for an ordinary sentence is
                                      an ICONS list that has exactly one
                                      entry for each pair of an
                                      individual `i' and the eventuality
                                      which is the ARG0 of each
                                      predication in which `i' appears
                                      as an argument.  Thus for `the cat
                                      persuaded the dog to bark' the
                                      ICONS list should have four
                                      elements: one for cat/persuade,
                                      one for dog/persuade, one for
                                      bark/persuade, and one for
                                      dog/bark.  Now if I wanted to have
                                      the grammar continue to only
                                      insert ICONS elements during
                                      composition for the non-vanilla
                                      info-str phenomena, and fill in
                                      the rest afterward, I would have
                                      to know not only the arity of each
                                      eventuality-predication, but which
                                      of its arguments was realized in
                                      the sentence, and even worse,
                                      which of the realized syntactic
                                      arguments corresponded to semantic
                                      arguments (so for example not the
                                      direct object of `believe'). 
                                      Maybe I give up too soon here, but
                                      this does not seem doable just
                                      operating on the MRS resulting
                                      from composition, even with access
                                      to the SEM-I.</p>
                                    <p><br>
                                    </p>
                                    <p>So if the necessary ICONS
                                      elements have to be introduced
                                      overtly by the lexicon/grammar
                                      during composition, then I would
                                      still like to explore a middle
                                      ground that does not result in the
                                      full set of ICONS elements Soon
                                      and Bender propose for a
                                      sentence.  That is, I wondered
                                      whether we could make do with
                                      adding to the ERG the necessary
                                      introduction of just those ICONS
                                      elements that would enable us to
                                      draw the distinctions between
                                      `unmarked', 'topic', and 'focus'
                                      that we were used to exploiting in
                                      the days of messages.   But since
                                      pretty much any preposition's or
                                      adjective's or verb's complement
                                      can be extracted, and any verb's
                                      subject can be extracted, and most
                                      verbs' direct and indirect objects
                                      can be passivized, I think we'll
                                      still end up with an ICONS entry
                                      for each eventuality/argument pair
                                      for every predication-introducing
                                      verb, adjective, and preposition
                                      in a sentence, and maybe also for
                                      some nouns as in "who is that
                                      picture of?".  This still lets us
                                      exclude ICONS elements involving
                                      adverbs and maybe also the
                                      arguments of conjunctions,
                                      subordinators, modals.  If we went
                                      this route, I think it would be
                                      possible to make modest additions
                                      to certain of the constructions,
                                      and not have to meddle with
                                      lexical types, to get these ICONS
                                      elements into the MRS during
                                      composition.</p>
                                    <p><br>
                                    </p>
                                    <p>Such a partial approach does not
                                      have the purity of Soon and
                                      Bender's account, but might be
                                      more practical, at least as a
                                      first step, for the ERG.  It would
                                      at least enable what I think is a
                                      more consistent interpretation of
                                      the ICONS elements for generation,
                                      and should give us the
                                      fine-grained control I agree that
                                      we want.  Thus to get the
                                      generator to produce all variants
                                      from an MRS produced by parsing a
                                      simple declarative, one would have
                                      to remove the info-str ICONS
                                      element whose presence excludes
                                      the specialization to focus or
                                      topic because of our friend
                                      Skolem.</p>
                                    <p><br>
                                    </p>
                                    <p>Counsel?</p>
                                    <p><br>
                                    </p>
                                    <p> Dan<br>
                                    </p>
                                    <br>
                                    <div style="color:rgb(0,0,0)">
                                      <hr
                                        style="display:inline-block;width:98%">
                                      <div dir="ltr"><font
                                          style="font-size:11pt"
                                          face="Calibri, sans-serif"
                                          color="#000000"><b>From:</b> <a
                                            moz-do-not-send="true"
                                            href="mailto:developers-bounces@emmtee.net"
                                            target="_blank"><a class="moz-txt-link-abbreviated" href="mailto:developers-bounces@emmtee.net">developers-bounces@emmtee.net</a></a>
                                          <a moz-do-not-send="true"
                                            href="mailto:developers-bounces@emmtee.net"
                                            target="_blank">&lt;developers-bounces@emmtee.net&gt;</a>
                                          on behalf of Ann Copestake <a
                                            moz-do-not-send="true"
                                            href="mailto:aac10@cam.ac.uk"
                                            target="_blank"><a class="moz-txt-link-rfc2396E" href="mailto:aac10@cam.ac.uk">&lt;aac10@cam.ac.uk&gt;</a></a><br>
                                          <b>Sent:</b> Friday, February
                                          5, 2016 1:43 PM<br>
                                          <b>To:</b> Emily M. Bender;
                                          Stephan Oepen<br>
                                          <b>Cc:</b> developers; Ann
                                          Copestake<br>
                                          <b>Subject:</b> Re:
                                          [developers] ICONS and
                                          generation</font>
                                        <div> </div>
                                      </div>
                                      <div>Thanks!<br>
                                        <br>
                                        <div>On 05/02/2016 21:30, Emily
                                          M. Bender wrote:<br>
                                        </div>
                                        <blockquote type="cite">
                                          <div dir="ltr">Not sure if
                                            this answers the question,
                                            but a couple of comments:
                                            <div><br>
                                            </div>
                                            <div>(a) I do think that
                                              written English is largely
                                              underspecified for
                                              information structure.</div>
                                            <div>It's part of what makes
                                              good writing good that the
                                              information structure is
                                              made apparent</div>
                                            <div>somehow.</div>
                                            <div><br>
                                            </div>
                                          </div>
                                        </blockquote>
                                        <br>
                                        OK.  should I understand you as
                                        saying that composition (as in,
                                        what we do in the grammars)
                                        leaves it mostly underspecified,
                                        but that discourse level factors
                                        make it apparent?  or that it
                                        really is underspecified?<br>
                                        <br>
                                        <blockquote type="cite">
                                          <div dir="ltr">
                                            <div>(b) I think the "I want
                                              only the unmarked form
                                              back" case might be
                                              handled by either</div>
                                            <div>a setting which says
                                              "no ICONS beyond what as
                                              in the input" (i.e. your
                                              ICONS { }) or</div>
                                            <div>a
                                              pre-processing/generation
                                              fix-up rule that takes
                                              ICONS { ... } and outputs
                                              something</div>
                                            <div>that would be
                                              incompatible with anything
                                              but the unmarked form.  Or
                                              maybe the</div>
                                            <div>subsumption check goes
                                              the wrong way for this
                                              one?</div>
                                            <div><br>
                                            </div>
                                          </div>
                                        </blockquote>
                                        Yes, I think the ICONS {} might
                                        be a possible way of thinking
                                        about it.  I should make it
                                        clear - I don't think there's a
                                        problem with constructing an
                                        implementation that produces the
                                        `right' behaviour but I would
                                        much prefer that the behaviour
                                        is specifiable cleanly in the
                                        formalism rather than as another
                                        parameter to the generator or
                                        whatever.<br>
                                        <br>
                                        <blockquote type="cite">
                                          <div dir="ltr">
                                            <div>I hope Sanghoun has
                                              something to add here!</div>
                                            <div><br>
                                            </div>
                                            <div>Emily</div>
                                          </div>
                                          <div class="gmail_extra"><br>
                                            <div class="gmail_quote">On
                                              Fri, Feb 5, 2016 at 1:01
                                              PM, Stephan Oepen <span
                                                dir="ltr"> &lt;<a
                                                  moz-do-not-send="true"
href="mailto:oe@ifi.uio.no" target="_blank"><a class="moz-txt-link-abbreviated" href="mailto:oe@ifi.uio.no">oe@ifi.uio.no</a></a>&gt;</span>
                                              wrote:<br>
                                              <blockquote
                                                class="gmail_quote"
                                                style="margin:0 0 0
                                                .8ex;border-left:1px
                                                #ccc
                                                solid;padding-left:1ex">
                                                colleagues,
                                                <div><br>
                                                </div>
                                                <div>my ideal would be a
                                                  set-up where the
                                                  provider of generator
                                                  inputs has three
                                                  options: (a) request
                                                  topicalization (or
                                                  similar), (b) disallow
                                                  it, or (c)
                                                  underspecify and get
                                                  both variants.
                                                  <div><br>
                                                  </div>
                                                  <div>we used to have
                                                    that level of
                                                    control (and
                                                    flexibility) in the
                                                    LOGON days where
                                                    there were still
                                                    messages: in the
                                                    message EPs, there
                                                    were two
                                                    optional ‘pseudo’
                                                    roles (TPC and PSV) <span></span>to


                                                    control
                                                    topicalization or
                                                    passivization of a
                                                    specific instance
                                                    variable.
                                                     effectively, when
                                                    present, these
                                                    established a binary
                                                    relation between the
                                                    clause and one of
                                                    its
                                                    nominal constituents.
                                                     if i recall
                                                    correctly, blocking
                                                    topicalization was
                                                    accomplished by
                                                    putting an otherwise
                                                    unbound ‘anti’-variable

                                                    into the TPC or PSV
                                                    roles.</div>
                                                  <div><br>
                                                  </div>
                                                  <div>could one imagine
                                                    something similar in
                                                    the ICONS realm, and
                                                    if so, which form
                                                    would it have to
                                                    take?</div>
                                                  <div><br>
                                                  </div>
                                                  <div>best wishes, oe</div>
                                                  <div>
                                                    <div>
                                                      <div><br>
                                                        <br>
                                                        On Friday,
                                                        February 5,
                                                        2016, Woodley
                                                        Packard &lt;<a
                                                          moz-do-not-send="true"
href="mailto:sweaglesw@sweaglesw.org" target="_blank"><a class="moz-txt-link-abbreviated" href="mailto:sweaglesw@sweaglesw.org">sweaglesw@sweaglesw.org</a></a>&gt;


                                                        wrote:<br>
                                                        <blockquote
                                                          class="gmail_quote"
                                                          style="margin:0
                                                          0 0
                                                          .8ex;border-left:1px
                                                          #ccc
                                                          solid;padding-left:1ex">
                                                          I can confirm
                                                          that under
                                                          ACE, behavior
                                                          is what you
                                                          indicate, i.e.
                                                          generating
                                                          from parsing
                                                          the
                                                          topicalized
                                                          feline-canine-playtime
                                                          I get just the
                                                          topicalized
                                                          variant out,
                                                          but when
                                                          generating
                                                          from parsing
                                                          the ordinary
                                                          word order I
                                                          get all 5
                                                          variants out.<br>
                                                          <br>
                                                          I believe this
                                                          was designed
                                                          to imitate the
                                                          long-standing
                                                          condition that
                                                          the MRS of
                                                          generation
                                                          results must
                                                          be subsumed by
                                                          the input
                                                          MRS.  The
                                                          observed
                                                          behavior seems
                                                          to me to be
                                                          the correct
                                                          interpretation
                                                          of the
                                                          subsumption
                                                          relation with
                                                          ICONS
                                                          involved. 
                                                          Note that an
                                                          MRS with an
                                                          extra
                                                          intersective
                                                          modifier would
                                                          also be
                                                          subsumed, for
                                                          example, but
                                                          such MRS are
                                                          never actually
                                                          generated
                                                          since those
                                                          modifier
                                                          lexical
                                                          entries never
                                                          make it into
                                                          the chart.<br>
                                                          <br>
                                                          It’s certainly
                                                          reasonable to
                                                          ask whether
                                                          (this notion
                                                          of)
                                                          subsumption is
                                                          really the
                                                          right test. 
                                                          I’ve met lots
                                                          of folks who
                                                          prefer to turn
                                                          that
                                                          subsumption
                                                          test off
                                                          entirely.  I
                                                          guess it’s
                                                          also possible
                                                          that the
                                                          subsumption
                                                          test is right
                                                          for the RELS
                                                          portion of the
                                                          MRS but not
                                                          for the ICONS,
                                                          though that
                                                          seems a bit
                                                          odd to
                                                          consider. 
                                                          However, given
                                                          that we don’t
                                                          have many
                                                          ideas about
                                                          truth-conditional
                                                          implications
                                                          of ICONS,
                                                          maybe not so
                                                          odd.<br>
                                                          <br>
                                                          I don’t really
                                                          have much to
                                                          offer in terms
                                                          of opinions
                                                          about what the
                                                          right behavior
                                                          should be.  I
                                                          (believe I)
                                                          just
                                                          implemented
                                                          what others
                                                          asked for a
                                                          couple years
                                                          ago :-)<br>
                                                          <br>
                                                          -Woodley<br>
                                                          <br>
                                                          &gt; On Feb 5,
                                                          2016, at 8:03
                                                          AM, Ann
                                                          Copestake &lt;<a
moz-do-not-send="true" href="mailto:aac10@cl.cam.ac.uk" target="_blank"><a class="moz-txt-link-abbreviated" href="mailto:aac10@cl.cam.ac.uk">aac10@cl.cam.ac.uk</a></a>&gt;


                                                          wrote:<br>
                                                          &gt;<br>
                                                          &gt; I'm part
                                                          way through
                                                          getting ICONS
                                                          support
                                                          working in
                                                          Lisp, testing
                                                          on the version
                                                          of the ERG
                                                          available as
                                                          trunk. I have
                                                          a question
                                                          about
                                                          generation. 
                                                          If I
                                                          implemented
                                                          the behaviour
                                                          described in <a
moz-do-not-send="true" href="http://moin.delph-in.net/IconsSpecs"
                                                          target="_blank"><a class="moz-txt-link-freetext" href="http://moin.delph-in.net/IconsSpecs">http://moin.delph-in.net/IconsSpecs</a></a>
                                                          there doesn't
                                                          seem to be a
                                                          way of
                                                          specifying
                                                          that I want a
                                                          `normal'
                                                          ordering for
                                                          English.<br>
                                                          &gt;<br>
                                                          &gt; e.g., if
                                                          I take the MRS
                                                          resulting from<br>
                                                          &gt;<br>
                                                          &gt; that dog,
                                                          the cat
                                                          chased.<br>
                                                          &gt;<br>
                                                          &gt; without
                                                          ICONS check,
                                                          there are 5
                                                          realizations,
                                                          including the
                                                          `null ICONS'
                                                          case `The cat
                                                          chased that
                                                          dog.'  With an
                                                          exact ICONS
                                                          check, I can
                                                          select
                                                          realizations
                                                          with the same
                                                          ICONS (modulo
                                                          order of ICONS
                                                          elements, of
                                                          course, in the
                                                          case where
                                                          there's more
                                                          than one
                                                          element).  But
                                                          with the <a
                                                          moz-do-not-send="true"
href="http://moin.delph-in.net/IconsSpecs" target="_blank"> </a><a
                                                          moz-do-not-send="true"
href="http://moin.delph-in.net/IconsSpecs" target="_blank"><a class="moz-txt-link-freetext" href="http://moin.delph-in.net/IconsSpecs">http://moin.delph-in.net/IconsSpecs</a></a>
                                                          behaviour,
                                                          there's no way
                                                          of specifying
                                                          I want a
                                                          `normal' order
                                                          - if I don't
                                                          give an ICONS,
                                                          I will always
                                                          get the 5
                                                          realisations.
                                                          In fact, as I
                                                          understand it,
                                                          I can always
                                                          end up with
                                                          more icons in
                                                          the
                                                          realisation
                                                          than in the
                                                          input, as long
                                                          as I can match
                                                          the ones in
                                                          the input.<br>
                                                          &gt;<br>
                                                          &gt; So:<br>
                                                          &gt; - is the
                                                          IConsSpec
                                                          behaviour what
                                                          is desired for
                                                          the ERG (e.g.,
                                                          because one
                                                          can rely on
                                                          the
                                                          realisation
                                                          ranking to
                                                          prefer the
                                                          most `normal'
                                                          order)?<br>
                                                          &gt; - or does
                                                          the ERG behave
                                                          differently
                                                          from Emily and
                                                          Sanghoun's
                                                          grammars, such
                                                          that different
                                                          generator
                                                          behaviour is
                                                          desirable? and
                                                          if so, could
                                                          we change
                                                          things so we
                                                          don't need
                                                          different
                                                          behaviours<br>
                                                          &gt;<br>
                                                          &gt; Ann<br>
                                                          &gt;<br>
                                                          &gt;<br>
                                                          &gt;<br>
                                                          <br>
                                                          <br>
                                                        </blockquote>
                                                      </div>
                                                    </div>
                                                  </div>
                                                </div>
                                              </blockquote>
                                            </div>
                                            <br>
                                            <br clear="all">
                                            <div><br>
                                            </div>
                                            -- <br>
                                            <div>
                                              <div dir="ltr">Emily M.
                                                Bender<br>
                                                Professor, Department of
                                                Linguistics<br>
                                                Check out CLMS on
                                                facebook! <a
                                                  moz-do-not-send="true"
href="http://www.facebook.com/uwclma" target="_blank"> </a><a
                                                  moz-do-not-send="true"
href="http://www.facebook.com/uwclma" target="_blank"><a class="moz-txt-link-freetext" href="http://www.facebook.com/uwclma">http://www.facebook.com/uwclma</a></a><br>
                                              </div>
                                            </div>
                                          </div>
                                        </blockquote>
                                        <br>
                                      </div>
                                    </div>
                                  </div>
                                </blockquote>
                                <br>
                              </div>
                            </div>
                          </div>
                        </blockquote>
                      </div>
                      <br>
                      <br clear="all">
                      <br>
                      -- <br>
                      <div>
                        <div>=================================</div>
                        <div>Sanghoun Song</div>
                        <div>Assistant Professor</div>
                        <div>Dept. of English Language and Literature</div>
                        <div>Incheon National University</div>
                        <div><a moz-do-not-send="true"
                            href="http://corpus.mireene.com"
                            target="_blank">http://corpus.mireene.com</a></div>
                        <div>phone: +82-32-835-8129 (office)</div>
                        <div>=================================</div>
                      </div>
                    </div>
                  </blockquote>
                  <br>
                </div>
              </div>
            </div>
          </blockquote>
        </div>
        <br>
        <br clear="all">
        <div><br>
        </div>
        -- <br>
        <div class="gmail_signature">
          <div dir="ltr">Emily M. Bender<br>
            Professor, Department of Linguistics<br>
            Check out CLMS on facebook! <a moz-do-not-send="true"
              href="http://www.facebook.com/uwclma" target="_blank">http://www.facebook.com/uwclma</a><br>
          </div>
        </div>
      </div>
    </blockquote>
    <br>
  </body>
</html>