Among the five vowels in Modern Japanese, the vowel u is a highcentral unrounded sound [w]. Accordingly, the vowel triangle of Modern Japanese is somewhat anomalous, lacking a rounded high-back vowel [u]. How did this sort of vowel system form historically?
A reconstruction of eight vowels in Old Japanese, the central dialect of the Nara period (710-84), is commonly accepted. Although interpreted otherwise by some, the so-called ko-otsu distinctions must have been due to the vocalic oppositions, i≠ï, e≠ë, and o≠ö. The vowel diagrams of Old Japanese drawn by scholars adopting the eight-vowel theory are, however, in conflict with what some consider to be phonetic universals. In particular, there are only three vowels (i, i, and u) at the high level but four (e, ë, ö, and o) at the mid.
There is no problem with assuming that the sound values of (C) a, (C) i, (C) u, (C) e, and # o were identical with those of the corresponding vowels in Modern Japanese. The problem reduces to determining the sound values of Cï, Cë, Co, and Cö. Judging from the positions of the Chinese Characters, used as man' yogana, in the Ancient Chinese rhyme tables, one may infer that -ö(following a consonant) was phonetically identical with # o (not following a consonant), which was [o], and that the sound value of -o was [u], a high-back vowel. Thus, o with no preceding consonant should be described as ö. Likewise, ï was very likely high-central and ë mid-central. Drawing a diagram based on the inferred sound values, i, ï, u, and o form a high vowel series, front to back; e, ë, and ö form a mid vowel series, front to back; and a is a low-central vowel, just as in the vowel triangles of Altaic languages.
Internal reconstruction on the basis of the verbal stems of Old Japanese indicates that all eight vowels could be coupled with all consonants, including zero, in Archaic (pre-Nara) Japanese. The eighth century was the close of a transitional era, when the ko-otsu distinction remained only in a few environments. At the outset of the Heian period (794-1192), the phonetic changes ï>i, ë>e, and o>ö took place in all environments. The remaining five vowels have continued into the modern period, forming the “anomalous” vowel system of Modern Japanese.

抄録全体を表示

Jurchen is the language of the Jurchen tribes, which founded the Chin dynasty in the early twelfth century, and is the oldest of the Tunguz languages for which there are extant materials in the native script.
It is, however, almost impossible to reconstruct Jurchen phonology and establish the phonological system in that period due to the scarcity of linguistic materials. Although Chin-kuo yü-chieh金国語解has survived, the dictionary produced by the Jurchen themselves, the Nü-chen-tzu tzu-mu女真字字母, is not extant.
Nevertheless, inasmuch as the Jurchen section of the Hua-i i-yü華夷訳語 (Chinese-Barbarian Glossaries) produced by the Ming Translators Bureau, namely Nü-chen-kuan i-yü女真館訳語, exists, one may reconstruct the Jurchen phonemes in the Ming period by means of the Chinese characters in the text used for the transcription of Jurchen. Therefore, the present state of knowledge of Ming Chinese phonology furnishes an important key to the solution of Jurchen sound equivalences transcribed by Chinese characters. It goes without saying also that linguistically the Manchu language plays a supplementary role in the solution of Jurchen phonology.
Thus, the pronunciations of individual Jurchen characters, of which some are phonograms and some are ideograme, can be reconstructed by comparison of all occurrences of a given character with its transliterations into Chinese, and then by setting up the phonemes that leave the fewest contradictions. After establishing Ming Jurchen phonology in this fashion, it is then possibleto reconstruct Chin Jurchen phonology via historico-linguistic methods.
As a result, one may discover such facts as: the Tunguz initial *p- was still preserved in Chin Jurchen, which then changed to *f-in Ming Jurchen; the back vowel series *a, *o, and *u contrasted with the front vowels *e, *ö, and *ü in addition to the neutral *i in the Chin period, whereas in the Ming period are found only five vowels, *a and *o of the back series, only *e in the front, and two neutral vowels *i and *u; etc. In addition, although it is known that there are no native Manchu words ending in a consonant but -n, some words ending in the consonant-*γ are to be found in Jurchen; there is no labial attraction in Jurchen, unlike Manchu; there is no phonological distinction between final-*n and-*η even in Chinese loans, as there is in Manchu; etc.

抄録全体を表示

In spoken Manchu, i. e. the Sibo dialect in the Ili valley, west of Dzungaria, the consonant s is always palatalized as [š] before the highvowel i, wherever it may occur. Also, the consonant that was transcribedby the Manchu letter normally representing s was Romanized by some Western scholars in the XVIIth through XIXth centuries as š, s, sh, etc., which presumably indicated [š], when followed immediately by i. Althoughsome other scholars in that period used the Latin alphabet s (or Cyrillic alphabet c) for the same consonant in the same environment, their Romanization was surely not a phonetic transcription but rather atransliteration of the Manchu script into the Latin (or Cyrillic) alphabet.
Since the transcription of phonemes of a language in its native scriptusually reflects the speakers' concept of the phonemic level, and not ofthe phonetic level, the syllable transcribed with the Manchu letters sand i denotes /si/. Thus, it is certain that /s/ in /si/ in literary Manchurepresented [š], both initially and intervocalically as an environmentalvariant of /s/. In fact, judging from an investigation of the Chinesetranscription of Jurchen words appearing in the Sino-Jurchen glossaryproduced in the XVth century, the syllable /si/ in Manchu words correspondsto *ši in any position in the cognate words in Jurchen, a pre-Manchulanguage. The Jurchen syllable *si is found only in Chinese loansderived from Chinese si.
On the other hand, Manchu had a consonant that was transcribed bythe Manchu letter normally representing š. This consonant, i.e./š/, however, always preceded the vowels other than i since [ši] was regarded as/si/ then, as a result of a sound change, *ši>si, which had taken placebefore the appearance of the Jurchen language. The syllable /ši/, so transcribedin Manchu script, appears only in Chinese loans of which theoriginal sound was si with a retroflex in its syllable initial.
To sum up, in literary Manchu the sound value of /s/ preceding vowelsother than i was [š], but when preceding i it was [š]. On the otherhand, the sound value of /š/ was [š] before vowels other than i, but [š], a loan pronunciation, when preceding i.
Thus, it can be said that conclusions on Altaic or Tunguz comparativelinguistics such as in Nikolaus Poppe's and Johannes Benzing's, which were reached by regarding the Manchu /si/ as [si], and not as [ši], have to be revised.

抄録全体を表示

日本語とウイグル語は共に膠着語であり, 構文的に類似した点が多い. したがって. 日本語からウイグル語への機械翻訳においては, 形態素解析によって得られた各単語を逐語翻訳することにより, ある程度の翻訳が可能となる. しかし, 従来の日本語文法は動詞が活用することを前提としていたため, ウイグル語への翻訳の前に, 動詞の活用処理が必要であった. 本論文では, 日本語, ウイグル語を共に派生文法で記述することにより, 日本語の活用処理を不要とすると同時に, 両言語間の形態論的類似点を明確にし, 単純でかつ体系的な機械翻訳が可能になることを示す, しかし, 日本語とウイグル語との問の文法的差異から, 単純な逐語翻訳では不自然な翻訳となる場合がある. 本論文では, 単語間の接続関係を考慮した訳語置換表を用いることによりこの問題を解決し, より自然な翻訳を実現した. さらに, この手法に基づく日本語-ウイグル語機械翻訳システムを作成した. このシステムでは, 日本語形態素解析システムとウイグル語整形システムを, それぞれ独立のモジュールとして構成している. この設計は, 他の膠着語間における翻訳にも応用可能であると考えられる. また, 実験によりその翻訳精度を評価した. 本論文では, 特に両言語において文の中心的役割を果たす動詞句の翻訳について述べる.

抄録全体を表示

日本語とウイグル語は共に膠着語であり, 語順がほぼ同じであるなどの構文的類似性が見られる. そのため, 日本語-ウイグル語機械翻訳においては, 日本語文を形態素解析した後, 逐語訳を行うだけでもある程度の翻訳が可能となる. これは, 名詞や動詞などの自立語の文中での役割が助詞, 助動詞といった付属語によって示されており, そうした付属語においても, 日本語とウイグル語との間で対応関係があるからである. 特に名詞に接続する格助詞は, 文中での他の語との関係を決めるという, 言語構造上重要な機能を持っている. そのため, 格助詞を正しく翻訳できなければ, 違和感のある翻訳文になるだけでなく, ときには致命的に誤った意味となる翻訳文を生成することがある. そこで, 本論文では, 日本語-ウイグル語機械翻訳における格助詞の取り扱いについて論じる. まず, 計算機用日本語基本動詞辞書IPALを用いて動詞と格助詞の使われ方を調べるとともに, それぞれの格助詞の機能に対応するウイグル語格助詞を決定する. さらに, この調査結果から作成した動詞の格パターンを利用して複数の格助詞の訳語候補の中から, 適切な訳語を選択する手法を提案する. また, 本提案手法に対する評価実験では, 環境問題関連の新聞社説3編の日本語138文を対象にし, 我々が本論文で提案するアプローチに基づいて実験を行った. その結果, 99.3%の正解率を得ることができた.

抄録全体を表示

To be an entire and intact word, it is necessary for any lexical item to get properly terminated. In this paper the author tries to make a brief sketch of the terminative structures in the verbal system of the Oga subdialect of Akita. The Verb of the Akita Dialect consists of (i)the Stem and (ii)the Extention, which form the Base, to which the Terminator must be connected to complete a verbal Word. The lexical items are divided into four types: a) □-(left side closed, right side open): verbs/adjectives[bases] and nouns b) -□(left side open, right side closed): suffixes c) -□-(both sides open): auxiliary verbs[postbases] d) □(both sides closed): adverbs by nature, interjections, particles The Terminator applies for the right arm of a lexical item, which must have been closed on the left side beforehand, and makes the word complete and intact.

抄録全体を表示

本稿では, 動詞の語尾変化について体系的な扱いが可能な派生文法に基づいて, 日本語形態素解析システムのための形態素文法を記述した. 但し, 派生文法における音韻単位での扱いを日本語の文字単位の扱いに変更する方法を示し, より形態素解析処理に適した形で記述した. さらに, これを実働する形態素解析システムに適用し, EDRコーパスと比較することによって精度を測定した.

抄録全体を表示

機械翻訳に対する要求の高まりに伴い, 日本語や英語, 韓国語といった言語の翻訳に関する研究が進み, 実用的なシステムが構築されつつある. その一方で, そうした研究があまり進んでいない言語が存在する. こうした言語においては, 翻訳の要である対訳辞書の整備も遅れている場合が多い. 一般に対訳辞書の構築には高いコストが必要であり, 機械翻訳システムを実現する上での障害となっている. しかし, 人間が翻訳作業をする場合, 対訳辞書に記載がない単語を別の表現に言い換えて辞書を引くことにより, この問題に対処する場合がある. 本研究ではこの手法を模倣し, 未登録語を登録語に言い換えることにより対訳辞書を拡充することを提案する. 本論文では, 対訳辞書の拡充に必要な単語の言い換え処理を収集段階と選抜段階の二つに分割し, 前者において語義文に基づく手法を, 後者において類似度に基づく手法をそれぞれ適用した. また, 類似度に基づく手法では, シソーラスにおける概念問の距離に加え, 単語を構成する漢字の語義を利用した. これによって, 語法や概念が近く意味的にも等価な言い換えを獲得できた. さらに, 獲得した言い換えを翻訳システムで翻訳して日本語一ウイグル語対訳辞書への追加を試みたところ, 未登録語300語のうち, その68.3%に対して利用可能な対訳が得られた.

抄録全体を表示