@article{oai:repository.ninjal.ac.jp:00000572,
 author = {前川, 喜久雄 and MAEKAWA, Kikuo},
 issue = {5},
 journal = {国語研プロジェクトレビュー, NINJAL Project Review},
 month = {Jun},
 note = {国立国語研究所, National Institute for Japanese Language and Linguistics, 日本語ザ行子音における調音様式の変異([z]～[dz])の要因を定量的かつ組織的に検討するために,『日本語話し言葉コーパス』の「Core」に施された音声ラベリングデータを分析した。分析結果は,語頭では破擦音,語中では摩擦音という従来の定説では自発音声中の異音の生起パタンが説明できないことを明らかに示していた。現実の生起パタンは,子音/z/の調音に利用可能な時間長が重要であることを示唆している。この特徴を捉えるためにTACA(Time Allotted for Consonant Articulation)という指標を考案した。再分析の結果,TACAは現実の生起パタンと高い相関を示すことが明らかになった。例えば決定木分析ではTACAのみでデータ全体(14,603サンプル)の74%が正しく分類され,これに個人差の要因と語中位置情報を追加することによってほぼ80%の分類精度が得られた。論文全体の最後に自発音声コーパスを利用した音声研究の得失を論じた。, Phonetically annotated part of the Corpus of Spontaneous Japanese was analyzed in order to examine the variation in the manner of articulation of Japanese consonant /z/. Results of quantitative analyses revealed clearly that the traditional account based upon the distinction of word-initial versus word-internal consonant location could not explain the occurrence patterns observed in the 'real' corpus data. Observation of the real occurrence patterns suggests the importance of the time that a speaker can use for the articulation of the consonant. To capture this property of speech, a new phonetic measure called TACA (Time Allotted for Consonant Articulation) was proposed. Reanalysis using TACA revealed high correlation between the TACA and the variation of /z/. For example, results of a decision tree analysis revealed that 74% of the /z/ (out of the total of 14,603 samples) could be correctly classified using only the data of TACA. The correct classification rate went up to about 80% if the information of individual difference of speakers and location of /z/ in a word were also provided. The whole paper ends with a short discussion about the pros and cons of the corpus-based studies in phonetics., application/pdf},
 pages = {21--45},
 title = {/z/の調音様式の変異 : コーパスによる分析},
 year = {2011},
 yomi = {マエカワ, キクオ}
}