<?tei:SOF p2cose.odd ?>
<!-- Date:         Fri, 26 Jun 92 11:34x:07 MST  -->
<!-- From:         Terry Langendoen <LANGENDT@ARIZVM1.ccit.arizona.edu -->
<!-- P2.23d : S (arbitrary segmentation)                      -->
<!-- Revisions:                                               -->
<!-- 92-12-09 : MSM : spellcheck for publication              -->
<!-- 92-12-09 : MSM : cosmetic copy editing                   -->
<!-- 92-12-09 : MSM : add space after some end-tags           -->
<!-- 92-12-07 : MSM : review for publication                  -->
<!-- 92-11-09 : MSM : suppress leading blanks in some elements -->
<!-- 27 oct LB : change as per DTL suggestions                -->
<!-- 6 Oct 92 : LB : change IDs; re-introduce some material cut -->
<!--                    by DTL; expand and simplify his e.g.  -->
<!-- 25 Jun 92 : DTL    : revise, add examples                -->
                          delete part on ambiguous punctuation-->
<!-- 25 May 92 : CMSMcQ : revise                              -->
<!-- 21 Apr 92 : CMSMcQ : retag for TEI P2, add tag docs.     -->
<!-- 16 Oct 90 : CMSMcQ : TEI P1 version 1.1                  -->
<!-- 17 Jul 90 : CMSMcQ : TEI P1 version 1.0                  -->
 
<div2 id=COSE><head>Arbitrary Segments</head>
<!-- TEI P1, sec. 6.8                                           -->
 
 
<p>For many purposes, one would like to be able to segment text
systematically into pieces smaller than the paragraph.  The element
<tag>s</tag> (segment) is provided for this purpose.  Segments of any
size or type may be marked using this tag.
 
<tagList>
<tagDesc tagDoc=S atts='type'>
</tagList>
 
<p>The global <att>id</att> and <att>n</att> attributes may be used to
associate standard reference identifiers with the segment for purposes
of citation or retrieval.
 
<p>The <att>type</att> attribute may be used to categorize the segment,
for example as an <term>orthographic sentence</term>, or as a
<term>phrase</term> of some specific type.  For some purposes, it may be
desired to characterize segments as <term>orthographic words</term>, or
even explicitly to mark sequences of <term>punctuation</term>.
 
<p>To illustrate this flexibility we first consider the simplest common
requirement:  the ability to mark orthographic sentences.  These we term
<term>S-units</term> to prevent confusion with the linguistic notion of
<q>sentence</q>, which is commonly taken to depend implicitly or
explicitly on a formal grammatical analysis.  S-units are signalled by
initial capitalization and particular punctuation marks (e.g. full stop,
exclamation mark, question mark), for example:
 
<!-- Leonard Bloomfield, "Literate and Illiterate Speech",    -->
<!-- American Speech 2, p. 432, 1927.                         -->
 
<eg><![ CDATA [
<s>Literate and illiterate speech in a language like English
   are plainly different.</s>
<s>This in fact is the layman's chief interest in linguistics.</s>
]]>
</eg>
 
<p>These S-units are both full sentences, but consideration of any real
life text will rapidly demonstrate the need to mark as S-units phrases
or even single words.  For example:
 
<eg><![CDATA[
<s>I pause to record that I feel in extraordinary form.</s>
<s>Delirium perhaps.</s>
]]>
</eg>
<!-- Beckett, Malone dies (1959) p258 -->
 
<p>S-units are commonly used in this way to demarcate manageable units
of text, for example as input to an automatic word-class tagger.  The
global <att>id</att> attribute may be used to associate an identifier
with each such segment.  This identifier may then be used to link the
segment with syntactic or other kinds of analysis.
 
<eg><![ CDATA [
<s id=BL0034>Literate and illiterate speech in a
   language like English are plainly different.</s>
<s id=BL0035>This in fact is the layman's chief
   interest in linguistics.</s>
]]>
</eg>
 
The <gi>s</gi> element may also be used to mark constituent elements
of any kind, such as are typically output by an automatic parsing
system:
 
<eg><![ CDATA [
<s id=BL0034>
    <s type=noun-phrase> Literate and illiterate speech </s>
    <s type=prep-phrase> in a language like English </s>
    <s type=complement> are plainly different.     </s>
</s>
]]>
</eg>
 
<p>As indicated above, the <gi>s</gi> element may nest within itself,
and thus provides a simple but powerful mechanism for representing
segmentation down to an arbitrary level of complexity, provided that the
structure can be represented by a single tree.  Mechanisms for
representing segmentations of a text which do not resolve to a single
tree, or in which constituent features are discontinuous, are discussed
in section <xref target=SA>.
 
<p>As a final example, we present the same two sentences segmented down
to the level of words and punctuation.  Each word or punctuation sequence
is given a unique identifier.
 
<note>White space and self-explanatory code names have been liberally
introduced in this example, in order to aid legibility:  a real life
example would, of course, be considerably more terse.</note>
 
<eg><![CDATA[
<p>
<s type=orth-sent id=S1>
   <s type=phrase rendition='small caps' id=PHR1>
      <s type=orth-word id=S1.1>    Literate     </s>
      <s type=orth-word id=S1.2>    and          </s>
      <s type=orth-word id=S1.3>    illiterate   </s>
      <s type=orth-word id=S1.4>    speech       </s>
   </s>
   <s type=orth-word id=S1.5>       in           </s>
   <s type=orth-word id=S1.6>       a            </s>
   <s type=orth-word id=S1.7>       language     </s>
   <s type=orth-word id=S1.8>       like         </s>
   <s type=orth-word id=S1.9>       English      </s>
   <s type=orth-word id=S1.10>      are          </s>
   <s type=orth-word id=S1.11>      plainly      </s>
   <s type=orth-word id=S1.12>      different    </s>
   <s type=punct     id=PUNC1>      .            </s>
</s>
 
<s type=orth-sent id=S2>
   <s type=orth-word id=S2.1>      This         </s>
   <s type=punct     id=PUNC16>    ,            </s>
   <s type=orth-word id=S2.2>      in           </s>
   <s type=orth-word id=S2.3>      fact         </s>
   <s type=punct     id=PUNC17>    ,            </s>
   <s type=orth-word id=S2.4>      is           </s>
   <s type=orth-word id=S2.5>      the          </s>
   <s type=orth-word id=S2.6>      layman
       <s type=punct id=PUNC18>    '        </s>
                                   s            </s>
   <s type=orth-word id=S2.7>      chief        </s>
   <s type=orth-word id=S2.8>      interest     </s>
   <s type=orth-word id=S2.9>      in           </s>
   <s type=orth-word id=S2.10>     linguistics  </s>
   <s type=punct     id=PUNC19>    .            </s>
   </s>
]]>
</eg>
 
<p>In the first sentence, the phrase <q>literate and illiterate
speech</q>, identified by its distinctive rendition, has been marked as
a segment of a different type from the remainder.  In the second
sentence, the <q>orthographic word</q> with identifier s2.6
(<q>layman's</q>) contains a nested <gi>s</gi> element of type
<q>punct</q>.  Each s-unit has an unique identifier which may be used to
associate it with a more detailed linguistic analysis, as further
discussed in sections <xref target=SA> and <xref target=FS>.
 
<p>The formal declaration for the <gi>s</gi> element is as follows:
 
<dtdFrag n='Arbitrary Segments' id=DCOSE>
<tagDecl tagDoc=S>
</dtdFrag>
 
<?tei:EOF p2cose.odd ?>
