Japanese Word Processor

1. OVERVIEW

Japanese is a very unique language in its variety of letters. It has several kinds of letters: Hira-Kana (Japanese alphabet;&?,~,, 3 ,---), Kata-Kana (Japanese alphabet used for foreign words; 7 , .-( , ·) , ---), Kanji (Chinese character;lll ,JII, J.l~,---), and alphanumerics (a,b,c,---,O,l,2,---).

Kanas consist of about fifty letters. Popular Kanjis amount to about two thousand. These letters appear in ordinary Japanese documents such as newspapers, business letters, and so on. The abundant letters invoke difficulties in typing Japanese documents. The conventional Japanese typewriter has to be equipped with all letters, and typists must look for letters one by one. Thus "letter-searching" is an intrinsic problem in the Japanese typewriting. In order to overcome this problem, Kana-to­ Kanji translation thechnology is applied in the Japanese Word Processor (JWP). A Kana sentence is just a sequence of phonetic sym­ bols. So the Kana sentences are easy to type because of the small number of keys, but they

are difficult to understand.

The new techniques to translate Kana sentences into ordinary Japanese sentences are discussed henceforth.

2. KANA-TO-KANJI TRANSLATION

Grammatical analysis is applied to get correct translation by JWP. Kana sentences are trans­ lated into Kanji sentences phrase by phrase.

The definition of the phrase is given as follows:

Phrasec(prefix)substantive word(suffix) (functional word*)

Substantive word=noun/pronoun/verb/adjective/ adverb/conjunction/interjection

Functional word=particle/auxiliary verb

Here parentheses indicate optionality, the asterisk indicates one or more occurences, and

the slant indicates alternatives.

Generally speaking, nouns are accompanied with a particle which indicates cases. Verbs are accompanied with conjugating auxiliary verbs, which indicate tenses, moods, and voices, with­ out any separation between them (Fig.l).

?tid: tt tl.~ f.'£ 13 fd: /)\-, ft

minakerebanaranakatta Thad to ~;) -- -

Each underlined part denotes a verb, auxiliary verbs, and a particle.

Fig.l An example of Japanese verb phrase

The grammatical analysis means analyzing the phrase grammatically. There are two ·main difficulties here, conjugation and concate­ nation of verbs and auxiliary verbs, and appearance of homonyms.

2 . 1 GRAMMATICAL ANALYSIS

Japanese sentences consist of a series of the phrases. The role of the Japanese phrases in sentences is approximately similar to English one (Fig.2), but structure of the phrases is different as shown by the above definition.

Here pl is a noun phrase, p2 is a pre­ positional phrase, and p3 is a verb phrase.

Fig,2 An example of the role of phrases

The analysis of verb phrase will be cited below since it is the most intricate. Japanese verbs do not express tenses, moods,and voices by their conjugation. If it is necessary to express ·them, auxiliary verbs which denote them respectively are concatenated to the verb . The verbs are classified into five conjugational groups and there are 25 auxilia­ ry verbs. They have six or less conjugational forms. These numerous conjugational forms of all verbs and auxiliary verbs make the verb phrases very complex.

The conjugational forms indicate the condition of the concatenations of verb and auxiliary verbs in the phrase. That is, they indicate what kind of word is concatenated to the verb or auxiliary verb. We made a connection table for all verbs and auxiliary verbs to check the concatenation. Figure 3 showes examples of the structure of the verb phrases.

iL6 (hasiru) run

.i:l2 ~ (hasit~)

iL IJ /.:~' (hasiritai)

ran

want to run

Fig.3 Examples of the structure of the verb phrase

In the figure 3, each underlined part indi­ cates a verb and an auxiliary verb.

2.1.1 Dictionary consultation

The first step of the grammatical analysis is to consult a dictionary for finding the sub­ stantive words. All possible candidates are extracted from the dictionary. We take the phrase "l.ll,'?~>~ '" (hasiranai;do not run) as an example to explain the grammatical analysis. Candidates for "liLS~·P" areshownby figure 4. Only stems are stored for verbs to reduce the number of items in the dictionary. The dictionary has the following contents for a word; Kanas, Kanjis, grammatical informations such as part of speech, conjugational type,etc., and frequency of use.

2.1.2 Analysis of conjugation and concatenation

The conjugational part of verb "j£ "(hasi) must be S (ra) ,I') (ri), 0 (ru), t\ (re), or ?>(ro) according to the grammatical information of the dictionary. As the third Kana of Ul,O:,~~' is (ra), the verb "~" satisfies the conjugational condition. Then it must be checked that the conjugating form of " it~:> "(hasira) can be accompanied with the string "f.< ~'" (nai). It is known by the connection table that the auxilia­ ry verb "f.··~'" can be attached to "il:.S ". "ltS'd:~'"(hasiranai) satisfies the condition of the concatenation. Therefore ~~~~::;ts~'" is one of possible translations. Nouns are accompanied with specific particles. These particles are stored in a table. The attached Kana-string to "U "( lil.,S;hasira) is "16~'" (nai). But this string is not found in the table~ accordingly "H ~-·~' "does not satisfy the condition of the concatenation as the phrase. By the same pro­ cedure, all the nouns of the candidates are also rejected. Thus only one translation" it~)~ ~' " is got.

2.2 Homonyms

2.2.1 Reduction of homonyms by grammatical analysis

There are a lot of homonyms in Japanese. It results from the fact that plural Chinese characters have the identical pronunciation expressed by Kanas. Chinese character itself is a letter and not a word in Japanese. The words are composed of one or more Chinese characters. The words have homonyms, too. That is, plural words have the identical Kanas. JWP reduces the homonyms by the grammatical analysis. For example, the phrase "~.~; ?I?J: <.t" (kyouryokuna; powerful) is translated uniquely into " ~tJu " by analyzing it grammatically. If the grammatical analysis is omitted and only consultation of the dictionary is applied two homonyms occurred: "5!llhf.J:" and "lntJfd: ". Here " IC&tJfd:" is meaningless · since " lntJ " is a noun meaning "cooperation" and "fd: "(na) is a conjugating part of the adjective with which noun can not be accompanied. The meaningless translations like this often happen without the grammatical analysis.

In case of the simplest translation technique, a table of Kanjis and their corresponding Kanas is used (Fig. 5).

There are a lot of homonyms to a phonetic unit which is expressed by Kanas. For example, a phonetic unit " <~ J: -) "(kyou) corresponds to 137 Chinese characters; "'""· "~","~",and so on. "t)J:<." has six; "JJ", "H", "HJ", and so on. Moreover " ·~ J; ? I') J; <."has other possi­ bilities of combination of phonetic units except "kyou-ryoku";that is, "kyo-u-ryo-ku", "kyo-uryo-ku", and so forth. If including these, homonyms increase enormously and many meaningless translations occur.

2.2.2 Reduction of semantical homonyms by learning

JWP has two kinds of memories to avoid seman­ tical homonyms: a short term memory and a long term memory. The former is valid during typing one document and is reset when other documents are typed. When homonyms appear on a CRT dis­ play of JWP and one of them is selected by the typist, the word is memorized by JWP. When the same homonyms occur , the memorized word is indicated on the CRT display as the most possi­ ble candidate word in the context of the text. The latter is implemented as an administrator of the frequency of word use: JWP administrates its dictionary regarding to the frequency. When one of homonyms is selected as mentioned above, the frequency of the word increases by one. Accordingly the dictionary is becoming suitable to a certain field such as science, economics, literature, etc., since JWP refers the frequency during translation and th~ most frequent word is given priviledge to appear on the CRT display first. This method is effec­ tive for the semantical homonyms depending fields.

3. CONCLUSION

The automatic Kana-to-Kanji translation made the touch method possible in Japanese typing and released the typists from the burden of searching letters among .the huge number of

letters on the keyboard.

JWP occupies a remarkable status against the conventional Japanese typewriter and other related machines for its ability of automatic translation. Several new techniques applied here are quite effective to realize high performance translation. Figure 6 showes an example of translation which is a Japanese sentence of the abstract of this paper.

~u~~?-~7ut7~~~~i~~b& ~~~75~~~U~?k<.~~~~~k6 G <. /JI~\I,L:>~ tt/;:IJ"Jil ·) ~J; ~)I;:: IJ~<::' tl)..S~~ G.t ~ J; -) H,::J.;-.)/JI.;l~ ~~-(/51' 9~H'<!J.-)�J;<t.

--- Typed Kana sentence

EJ ;j\~9- ~/r:l 1~-:tltl<t.:h.t (· tl)nJ)(~ ~/~~9~U~<.Wt~~�<.~R~hk n~~J;~R*~tl)~~ksa~~~~.;l~ 9~I7~9H({)t�1lJ".

--- Translated sentence

Fig.6 An example of Kana-to-Kanji

translation

4. ACKNOWLEDGEMENT

We would like to thank Dr. Hiroshi Genchi and Dr. Ken-ichi Mori, our supervisors at Toshiba R&D Center for their encouragement and support in the course of the work reported here. We also wish to acknowledge Kimihito Takeda for his cooperation to us in developping JWP.