Using the TEI Scheme in Compiling a Korean Dictionary

  1. 1. Beom-mo Kang

    Department of Linguistics - Korea University

Work text
This plain text was ingested for the purpose of full-text search, not to preserve original formatting or readability. For the most complete copy, refer to the original conference program.

1. A Dictionary Project
At Korea University in Seoul, we are currently
compiling a Korean monolingual dictionary. We
are trying to use computers as much as possible to
make the compilation process very efficient and
ultimately to make a good dictionary. In this process, we have used and intend to use the TEI
scheme in the following two ways.
2. Text Headers
First, we follow the lead of compilers of major
dictionaries such as COBUILD (Sinclair 1987) in
building and using a corpus as a resource of authentic examples and other valuable information
such as sense frequency. The problem is how to
encode texts on the computer so that we can extract relevant information efficiently. In the process of building a Korean language corpus called
“KOREA-1 Corpus” (Kim and Kang 1996), now
of size of 10,000,000 words, we have used tags
provided by TEI P3 (Sperberg-McQueen and Burnard 1994), mainly for text header information
(<teiHeader>). In the body of a text several tags
such as <q> and <l> have been used in some cases
but only the tag <p> has been consistently inserted.
Besides many <teiHeader> tags that we have
adopted as they are, we have used a modified tag
<catRef> to classify Korean texts according to our
needs of classification. In short, we use four digits
to represent 1) written/spoken distinction, 2) media–newspaper, magazine, book, unpublished material, and others (including originally prepared in
electronic form–, 3) fields(topics)–general, literary, humanities, social sciences, natural sciences,
etc–, and 4) more detailed field (content) classification within a major field.
For example, the following encoding means that
the source of the present text is a book whose topic
is in the field of history.
<catRef scheme=’kcrc’ target=’k1355’>
Notice that while the original <catRef> is defined
as an empty element in TEI P3, in our revised
scheme we allow it to contain a short explanatory
Since it is possible to prefix a KWIC line with this
kind of classification code and to sort lines according to it, it is potentially a useful means for
lexicographers writing definitions and usage notes
of a lexical item. An example of a KWIC concordance follows:
3. Dictionary Entries
Second, since we want to use computers in writing
and publishing the dictionary as well as in building
a corpus, the problem that we are now faced with
is how to represent a dictionary entry on the computer in its electronic form. Since TEI P3 offers
ways to encode dictionary items, we intend to
adopt the TEI encoding scheme and use it in some
stages of dictionary compilation.
Although TEI suggestions for dictionary encoding
are very comprehensive to cover various kinds of
dictionaries, its current commitment is to consider
only dictionaries of western languages (Ide and
Veronis 1995: 168). We are struggling with problems encountered in encoding Korean dictionary
entries in conformance with TEI. We try to extend
and modify the TEI encoding scheme in the way
suggested by TEI. In addition, we restrict content
models to a certain degree so that the encoded
dictionary might be viewed more as a database
than as a simple computerized (originally printed)
Among other things, we revise the <entry> model
so that it can have a number of proverbs <prov>
and idioms <idiom>, which consistently appear
on the entry level in Korean dictionaries.
<!ELEMENT %n.entry; - O ( (%n.hom; | %n.sense; |
(prov | idiom)* )
+(anchor) >
<prov> and <idiom>, in turn, can be defined so
that they can contain any dictionary parts such as
<form>, <def>, and <eg>:
<!ELEMENT prov - - (%paraContent |
%m.dictionaryParts)* >
<!ATTLIST prov;
%a.dictionaries; >
<!ELEMENT idiom - - (%paraContent |
%m.dictionaryParts)* >
<!ATTLIST idiom;
%a.dictionaries; >
One major revision which affects the hierarchical
structure of dictionary entries would be allowing
recursion for <hom>, as <div> is allowed to be
self-embedded. In Korean dictionaries, some entries have two levels of homography; namely 1)
different parts of speech, and 2) different subcategorizations. For example, some form (one entry)
is both a verb and an adjective (and a suffix with
a related meaning, too). Sometimes, a verb form
can be an intransitive verb, a transitive verb, or an
auxiliary verb. Of course, some theoretical considerations might allow us to disregard this kind of
complex homography levels and have different
entries for different parts of speech, so that we can
stick to the TEI scheme. However, respecting the
tradition of Korean lexicography, we want to
maintain at least the two levels of homography
mentioned above.
<!ELEMENT %n.hom; - O (%n.sense; | %n.hom |
-(entry) >
<!ATTLIST %n.hom;;
type (homPos | homSubc) homSubc
TEIform CDATA ’hom’ >
Here is an example, where ‘....’ represents some
Korean characters. Grammar codes in <pos> and
<subc>, such as ‘verb’, ‘adj’, ‘trans’, ‘intrans’, are
transliterations from the Korean counterparts. (For
the elements <lenHyph>, <irreg> and <irrForm>,
see below.)
<hom type=homPos n=’I’>
<hom type=homSubc n=’1’>
<hom type=homSubc n=’2’>
<hom type=homPos n=’II’>
Also, we add a dictionary top level element
<sciName> for scientific names, which appear
prominently in Korean dictionaries, by defining an
x-dot parameter entity in the TEI.extensions.ent
<!ENTITY % x.dictionaryTopLevel ’sciName |’>
For dictionary entry forms, which conventionally
show the major morphological immediate constituent break (by a hyphen) and long syllables (by a
colon) at the same time, we add <lenHyph> as a
member of the class “formInfo” in the same way.
In addition, to indicate the irregular inflectional
classes and to show typical inflected forms, which
usually appear along with grammatical category
information, we add <irreg> and <irrForm> as
members of the class “gramInfo”. These are slight
revisions to the TEI suggestions.
<!ENTITY % x.formInfo ’lenHyph |’>
<!ENTITY % x.gramInfo ’irreg | irrForm |’>
Academic domains (special fields), other domains
(such as ‘old Korean’), and dialect areas, which
are also prominent in Korean dictionaries, are
encoded with new tags defined within <usg>.
They are <domAca>, <domEtc>, and <dialArea>.
Also, since the content and format of etymology
(<etym>) and cross reference (<xr>) in a Korean
dictionary is constrained in certain ways, some
modifications of the DTD definitions of these
elements are needed. For <etym>, we add a new
attribute ‘hdType’ whose value should be one of
the following: ‘hj’ (hanja, i.e. of Chinese origin:
content given in Chinese characters), ‘foreign’ (of
any other foreign origin), and ‘kor’ (of Korean
origin proper). Incidentally, more than half of the
entries in a large Korean dictionary are of Chinese
origin and can be written in Chinese characters as
well as in Hangul, the Korean alphabet.
For <xr>, we define various “empty” elements
which mark the kinds of cross reference to be used
in the dictionary. Among them are ‘synonym’,
‘antonym’, ‘long form’, ‘short form’, ‘honorific
form’, etc. One of these elements should be used
in the first part of <xr>. The relevant part of the
DTD extension is given below:
<!ELEMENT %n.xr; - - ( (xrsee | xrstd | xrxstd |
xrant | xrsame | xrsyn | xrshort | xrlong |
xrstr2 | xrstr | xrsoft | xrlarge | xrsmall |
xrhon | xrint | xrchg | xrcfwd | xrvar | xrof),
(%paraContent | %n.usg | %n.lbl)* ) >
<!ATTLIST %n.xr;;
TEIform CDATA ’xr’ >
<!-- "See" the word for definition -->
<!ELEMENT xrsee - O EMPTY >
<!ATTLIST xrsee;
%a.dictionaries; >
<!-- synonym -->
<!ELEMENT xrsyn - O EMPTY >
<!ATTLIST xrsyn;
%a.dictionaries; >
......, etc.
Here is an example with an etymology <etym>
and a cross reference of type synonym <xrsyn>.
(Again, ‘......’ are parts in Korean.)
<form><orth>.....</orth><lenHyph>...... </lenHyph></form>
<etym hdType=hj> ...... </etym>
<sense n=’1’><def> ................ </def>
<eg><q> ...... <oRef>.</q></eg></sense>
<sense n=’2’><xr><xrsyn><ref> ...... </ref></xr>
<eg><q> ...... <oRef>.</q></eg></sense>
We might have constrained the “type” of <xr>,
e.g. <xr type=‘syn’>, in the DTD instead of introducing empty elements such as <xrsyn>.
4. Character Representation Problem
Finally, the character representation problem for
the 11,172 modern Hangul (Korean Alphabet)
characters and tens of thousands of Chinese characters used in Korean texts and dictionaries should
be addressed. Unlike Roman alphabets which require only one byte to encode a character, at least
two bytes are required for Hangul and Chinese
characters. Once UNICODE/UCS (ISO 10646-1)
has been adopted by program developers, the character problem would no longer be a serious one,
but for the time being, we should be satisfied with
the current Korean standards. The standard we are
adopting now uses the control code area above
ASCII 127 (C2 area).
Therefore, if a default SGML declaration for an
SGML parser like nsgmls (Clark 1995) prohibits
the use of both control code areas of C1 and C2,
we should revise the SGML declaration. The relevant part of the sgml declaration used for parsing
by NSGMLS follows:2
<!SGML "ISO 8879:1986"
International Reference Version
(IRV)//ESC 2/5 4/0"
10 1 10
13 1 13
14 12 UNUSED
26 1 UNUSED -- eof --
32 95 32
127 1 UNUSED
128 127 128 -- used by Hangul code:
255 1 UNUSED
5. SGML Parsing and Processing
Sample encodings of a Korean dictionary, together with dtd extensions and modifications,
TEI.extensions.ent and TEI.extensions.dtd, have
been validated by nsgmls, an SGML parser (Clark
1995). In addition, SoftQuad PANORAMA Pro
has been able to process the sample encodings
successfully. The beginning part of a sample Korean dictionary which is to be parsed by nsgmls is
as follows:
<!DOCTYPE tei.2 SYSTEM "c:\sgml\dtd\tei2.dtd" [
<!ENTITY % TEI.dictionaries "INCLUDE">
<!ENTITY % TEI.corpus "INCLUDE">
<!ENTITY % TEI.extensions.ent SYSTEM "hdic.ent">
<!ENTITY % TEI.extensions.dtd SYSTEM "hdic.dtd">
Clark, J. (1995) NSGMLS – a Validating SGML
Parser, software available at ftp://ftp.jclark.
Ide, N. and J. Veronis (1995) “Encoding Dictionaries”, in Computers and the Humanities 29-
2, 167–179.
Kim, H. and B. Kang (1996) “KOREA-1 Corpus:
Design and Composition”, in Korean Linguistics 3, 233-258 [written in Korean].
Sinclair, J. (ed.) (1987) Looking Up, London: Collins.
Sperberg-McQueen, C.M. and L. Burnard (eds.)
(1994) Guidelines for Electronic Text Encoding and Interchange (TEI P3), Chicago and
Oxford: TEI.
1 Ide and Veronis (1995) and Chapter 12 (Print
Dictionaries) of TEI P3 (Sperberg-McQueen
and Burnard, eds., 1994) discuss three views
of dictionaries: (a) the typographic view; (b)
the editorial view; (c) the lexical view. The
first view is concerned with the two-dimensional printed page while the last view is concerned with underlying information represented
in a dictionary, without concern for its exact
form. (The editorial view is in between.) Since
we are not encoding an existing dictionary on
the computer but preparing a lexical database
which is to be used in printing later, the last
view should be adopted in our project. 2 Michael Sperberg-McQueen helped me to revise the SGML declaration while I was participating in the 1995 Summer Seminar
organized by Center for Electronic Texts in the
Humanities, Princeton University

If this content appears in violation of your intellectual property rights, or you see errors or omissions, please reach out to Scott B. Weingart to discuss removing or amending the materials.

Conference Info

In review


Hosted at University of Bergen

Bergen, Norway

June 25, 1996 - June 29, 1996

147 works by 190 authors indexed

Scott Weingart has print abstract book that needs to be scanned; certain abstracts also available on dh-abstracts github page. (

Conference website:

Series: ACH/ICCH (16), ALLC/EADH (23), ACH/ALLC (8)

Organizers: ACH, ALLC