Multimedia, Multilingual Hyperdictionaries: A Japanese <-> English Example

Extended Abstract
1 Introduction
Japanese is often characterized as the most difficult foreign language for a Westerner to learn.
While differences in vocabulary, grammar, and
culture contribute to this difficulty, the main problem is the complex writing system which, utilizing
many thousands of characters, requires such an
expenditure of time that for a very long period, the
student of Japanese simply cannot read. This
learner illiteracy in the target language represents
a nearly insurmountable barrier to acquisition (c.f.
Krashen, 1987, 1989, 1994). In addition to the
usual bilingual dictionaries, the student must use
dictionaries not only for the ordinary uses of characters, but also for unique and idiosyncratic uses
of characters in personal and place names. Modern
storage media such as CD-ROMs and magnetooptical disks permit large amounts of data to be
stored so that the information contained in a set of
bilingual and character dictionaries can be compactly represented. The information thus represented can be treated as a linguistic database which
can be deductively accessed, combining many
linear dictionaries into a single multi-linear dictionary. Furthermore, traditional methods of character lookup can be generalised easily. The applicability of the notion of a deductive dictionary for
human users is, of course, not confined to any
single language or language pair.
2 Japanese Orthography
The Japanese writing system, deftly characterised
by James D. McCawley as ‘without doubt the
screwiest writing system in the world’ (Constantine, 92), uses thousands of kanji (characters of
Chinese origin), hiragana and katakana – two sets
of kana syllabaries of 46 characters each (highly
simplified Chinese characters which are used
phonetically) and sometimes even the Roman alphabet. Although more characters are used in writing Chinese, the Japanese writing system is more
complicated than the Chinese writing system because there are usually several pronunciations associated with each kanji. For example: the same
kanji – individually or in combination with others
– can be given different pronunciations, each
meaning the same thing, but used in different
social contexts in decreasing order of formality:
myounichi – tomorrow
asu – tomorrow
ashita – tomorrow
A student attempting to read Japanese thus requires not only the bilingual dictionaries familiar
to any language student, but also a character dictionary which gives the various pronunciations of
the characters and their definitions in the student’s
language, as well as pronunciations and definitions of compounds, words which are written with
two or more characters.
One major initial problem faced by a student of
Japanese is how to search for a kanji in a dictionary. Some dictionaries are arranged phonetically
or have phonetically-organized indexes, but this is
useful only if one knows how to pronounce the
kanji in the first place. Otherwise, a given kanji is
simply a graphic that must be reduced to primitives in order to facilitate a search.
In some newer dictionaries, kanji are arranged by
shape, but this is an arbitrary classification scheme
that is not apparent to all users. The most common
search method is to break the kanji into two parts:
one called a radical that carries (at least in theory)
semantic value, and another that indicates pronunciation – not necessarily in Japanese, however, but
rather in the original Chinese. There are 214 radicals, organised in order of the number of brush
strokes required to write them; these provide the
initial search index. Kanji with a given radical are
3
then organized further according to stroke count.
Thus a student who wishes to use a character
dictionary must first memorise the 214 radicals (or
be condemned to a linear search through a chart).
S/he must also learn some basic rules of writing
kanji, since the number of strokes sometimes differs from the number a naive user might calculate
by eye. Even then, however, difficulties remain,
since most complex kanji contain more than one
standard radical, and the student must determine
which of these radicals is the true index of the
kanji. Moreover, since most “words” in Japanese
combine two or more kanji, the user must then
search through another list, this time of combinations.
Current Japanese↔English dictionaries, whether
in book or electronic form, do not serve Englishlanguage users very well. Because so much learning “overhead” is required, they cannot be used
until the learner is at a fairly advanced stage. As
indicated above, they require too much advance
knowledge for the beginning user, who often finds
that all kanji of more than four strokes look alike.
The more sophisticated user still finds look-up a
slow process which may require several separate
dictionaries to determine the meaning of a single
kanji combination; the result is that she or he is
often stuck at an awkward stage, able to decode a
text but not really to read it. Since the better
English↔Japanese dictionaries are produced with
the Japanese user in mind, he is in a slightly better
position, but still receives inadequate help on matters of usage. Thus there is a need for a dictionary
that gives all levels of users rapid access to complete and accurate information. This can be done
only through a system that utilizes multiple-access
paths and interlocking databases such as we propose below.
3 What is a hyperdictionary?
The problems discussed above can be ameliorated
by introducing the concept of a “hyperdictionary”
as defined below:
Hyperdictionary
A relational and deductive database containing the words of a language or languages,
together with an open-ended set of access
and display methods so as to present at least
their orthography, pronunciation, signification, part of speech, and use, their history,
synonyms, homonyms, antonyms, derivation, relationships to one another, and any
other aspect of the words which may be
necessary for reference, teaching or study
purposes. Additional information about the
language or languages, including grammar,
morphology, semantics, pragmatics, machine tractable representations, etc., as well
as relevant information concerning geography, names, literature, society, culture, history and so on, is not excluded from the
database.
This hyperdictionary should be an encyclopedic
database of the words of a language or languages
coupled with an open-ended set of deductive access methods and multi-media display methods.
Current hardware and software technology, with
the exception of very high quality portable screens
and reliable miniature optical character readers,
are adequate for the purpose, so the real problems
of implementing hyperdictionaries lie in the design of the database and the access and display
methods. The design process must be carried out
from the fundamental point of view that the resulting hyperdictionary will be implemented not as a
book, that is, as a list of words, but with media
which permit efficient random, non-linear associative access and a variety of powerful input and
display techniques.
It is absolutely essential to design the hyperdictionary in such a way that will allow the user to
access information in any number of ways, according to the user’s cognitive and learning styles
and the demands of the task at hand. In doing this,
we need to consider the findings of previous studies focusing on the orientation of users within a
hypermedia environment, useful features to include in a hyperdictionary, and various methods
of dictionary organization (cf. Tripp and Roby,
1992, 1994; Tinkham, 1993), as well as the failure
of traditional dictionaries in preventing communication breakdown due to lexical deficit (cf. Christianson, 1995).
Finally, numerous studies have been conducted
into dictionary use and misuse by language learners. For instance, Christianson (1995) found that
over 40% of words looked up in various dictionaries by Japanese EFL learners were used incorrectly in their writing. A hyperdictionary has the
potential to drastically reduce such numbers, as it
would allow for fast, extensive access to cross-references and grammatical information.
4 Implementation Issues
The need to find kanji in a variety of ways requires
providing multiple input methods. The traditional
lookup methods involving pronunciation, stroke
count, and radical must be provided as well as
lookup by any of the JIS coding systems. These
search methods, though, were invented as a means
of indexing into linear lists of characters. It is
preferable to free the user from any such kind of
procedure, for example by facilitating direct input
of characters on a screen. This requires either a
video camera or optical scanner for input of characters which are already written or printed on
4
some other medium (e.g., in a book or on a sign)
or a penlike device with which characters can be
hand written on the screen. Alternatively, a menu
might provide graphical possibilities for synthesising the character. Fundamental component
shapes can be hyperlinked to a graphical display
of more common primitives. Together with spatial
information, and with mechanisms for couplings
with other similarly generated components, partial
queries to a shape database may automatically be
generated. Prolog-style matching and search may
be used to access a small set of candidate kanji to
be presented for user selection, with direct links to
the dictionary database. In this way character
lookup can be accomplished not only by being
able to recognize the radical, but rather, given a
suitable character analysis recorded in the database, by any reasonably identifiable constituent of
the character. For example, the character
should be discoverable by searching either on
or on , or indeed even on or on any
of the horizontal or vertical lines. (See (Abramson,
95) and (Abramson et al, 95) for further discussion
of this.) In order to find kanji in this manner, it is
necessary to have an analysis of all characters into
their identifiable components (eg, )
(see (Dürst, 93)) and also of all drawn characters
into components which are parsable in the sense
of two dimensional grammars for the specification
of graphic languages. The capability to search for
kanji containing similar components, which
hardly exists with present approaches, can be an
advantage in vocabulary building and language
study.
Various access paths, multiple search methods to
be supported, and the contents of the dictionary
database all depend on user access needs for locating a reference in the hyperdictionary.
In order to support multiple accessing requirements, a database needs to be created to provide
multithreaded access to stored data resources.
Based on analyses of individual access needs, an
optimum search plan is implemented. In addition,
the essential database components that may be
necessary are identified. On completion of the
analysis of individual access plans, the final content of the database includes:
• words, meanings and encoded pronunciations;
• access information such as JIS codes;
• shape encoding information to facilitate
search based on similarity of kanji shapes;
• radical based encoding to create a family of
kanji having a common radical;
• special encodings as may be necessary to
support the identified access paths;
• multiple indexes to support search paths.
The hyperdictionary must also support complex
searching options that are not provided by conventional dictionaries. For example,
• a complex search stating ‘identified radicals’
and ‘a range of number of strokes’ for a
character.
• all compounds containing a given kanji.
• thesaurus-like search to provide a collection
of words with similar senses and meanings.
For supporting the complex searching options,
query language support needs to be provided that
can allow use of ’AND’, ’OR’, and ’NOT’ options
over the given access methods. More complex
language manipulation capability may also be
needed to simplify the user interface for unskilled
users. For example, the notion of dictionary
lookup can be extended by providing a general
morphological analyser as a front end to the deductive database.
We illustrate these ideas by showing the core of
an exemplary database, using Prolog as a specification and prototyping language. The database
consists of definitions of radicals, kanji, and compounds. (In the actual database, there may be more
fields than those shown below.)
Starting from this core, further components of the
database may be derived:
• English, Japanese and Chinese wordlists
with pointers to the radicals, kanji, and compounds in which they occur.
• From each compound, definitions of which
characters are in the compound. For example, from the following are derived:
• From the Chinese and Japanese pronunciations of radicals, kanji, and compounds, katakana and hiragana equivalents are generated.
(The hyphen in the roman is present only as
5
an aid to students who are learning pronunciation.)
chinese ( ‘betsu’ )
japanese ( ‘bekkou-iro’ )
The specification of in_compound facilitates
searches for compounds on the basis of any character in the compound. This, combined with the
kanji lookup by any identifiable constituent (see
[Abramson et al, 1996] in this conference), makes
it possible to look up compounds provided that
some part of each (or any) character in the compound is recognized. By “some part” we mean
either radical or other constituent structure, stroke
count, pronunciation (indicated in romaji or kana)
or meaning. Furthermore, the hyperdictionary
may be extended in may ways. Audio for the
pronunciation of Japanese and/or English might
be added directly to the database, or alternatively,
speech generators could generate sound from the
kana or English spelling already in the database.
Space does not permit a listing of all the possible
probes into or extensions to the database, but we
hope this brief description gives an indication of
what is possible.
5 Wider implications
The difficulties faced by non-native readers of
Japanese have created barriers between the Japanese and other peoples that are hard to surmount.
These barriers contribute to a feeling of isolation
on the part of the Japanese and to a myth of
uniqueness that deceives both Japanese and foreigners alike into believing that Japanese culture
is impenetrable. The psychological distance between Japan and most other nations can be attributed in part to this communication problem, and
has continuing implications for issues of trade,
diplomacy, and scholarly exchange. A multimedia
dictionary can help to bridge this communication gap.
References
Abramson, H. (1995). The Web of Kanji, Deductive Dictionaries and Logic Programming,
Proceedings Natural Language Understanding
and Logic Programming Conference, Lisbon,
May 26–29 l995, pp. 1–22.
Abramson, H., Bhalla, S., Christianson, K., Goodwin, J., Goodwin, J., Sarraille, J., Schmitt, L.
(l995) The Logic of Kanji Lookup in a Japanese↔English Hyperdictionary. Proceedings
Joint International Conference ALLC-ACH
’96, Bergen, Norway, June 25–29, 1996.
Christianson, K. (1995). When FL writing errors
occur despite dictionary use. Paper presented
at the 21st Annual Conference of the Japanese
Association of Language Teachers (JALT),
Nagoya, Japan, November 4, l995.
Constantine, P. (1992). Japanese Street Slang.
New York: Tengu Books. Foreword by James
D. McCawley.
Dürst, M. (1993). Coordinate-independent Font
Description using Kanji as an Example. Electronic Publishing, Vol. 6(3), pp. 133–143.
Krashan, S. D. (1987). Principles and practice in
second language acquisition. New York:
Prentice-Hall.
. (1989). Language teaching technology: A lowtech view. In: Altatis, J. E. (Ed.). Georgetown
University Round Table on Languages and
Linguistics 1989, pp. 393–407. Washington,
D.C.: Georgetown University Press.
. (1994). Beyond the input hypothesis. Plenary
session presented at the 28th Annual TESOL
Conference, Baltimore, MD, March 11, 1994.
Tinkham, T. (1993). The effect of semantic clustering on the learning of second language vocabulary. System, 21(3), pp. 371–80.
Tripp, S. D. & W. Roby. (1992). The effects of
congruent orienting strategies on learning
from hypertext bilingual lexicon. Journal of
Hypermedia and Multimedia Studies, 2(2), pp.
6–12.
. (1994). The effects of various information
resources on learning from a hypertext bilingual lexicon. Journal of Research on Computing in Education, 27(1), pp. 92–103