@techreport{oai:ipsj.ixsq.nii.ac.jp:00057056,
 author = {カルロストロンコーソ and 河原, 達也 and 山本, 博史 and 菊井, 玄一郎 and Carlos, Troncoso and Tatsuya, Kawahara and Hirofumi, Yamamoto and Genichiro, Kikui},
 issue = {131(2004-SLP-054)},
 month = {Dec},
 note = {大語彙連続音声認識において，n-gramモデルより長距離の単語共起をモデル化するトリガー言語モデルについて検討する．一般に言語モデルの構築においては，タスクにマッチした学習コーパスのサイズは小さいため，統計量の学習が十分に行えず，逆に，大規模なコーパスでは一般的過ぎて，タスク依存性がなくなるという問題がある．本研究では，タスクにマッチしたコーパスからトリガーペアを抽出し，大規模なテキストコーパスからトリガーペアの生起確率を推定するアプローチを提案する．ATRの旅行会話コーパス(BTEC)，及び日本語話し言葉コーパス(CSJ)の模擬講演において評価を行った結果を報告する．, We study the trigger-based language model (LM) for large vocabulary continuous speech recognition (LVCSR), which can model dependencies between words longer than those modeled by the n-gram LM. In general, in language modeling for LVCSR, when the training corpus matches the target task, its size is typically small, and therefore insufficient for providing us with reliable probability estimates. On the other hand, large corpora are often too general to capture task dependency. The proposed approach tries to overcome this generality-sparseness trade-off problem by constructing a trigger-based LM in which task-dependent trigger pairs are first extracted from the corpus that matches the task, and then the occurrence probabilities of the pairs are estimated from a huge text corpus. We report evaluation results in ATR's Basic Travel Expression Corpus (BTEC) as well as in the Corpus of Spontaneous Japanese(CSJ).},
 title = {異種コーパスの組合せによるトリガー言語モデルの構築},
 year = {2004}
}