本論文觀察台灣年輕人說閩南語的國語口音，建立語料庫作為分析基礎，並根據優選理論中的浮動制約模型（Floating Constraint, Reynolds 1994）分析語料。 本文語料主要分為兩大組：A組及B組。除AB兩組均保留雙唇濁塞聲母 /b/之外，A組的國語口音較明顯，亦即軟顎濁塞聲母 /g/ 較常丟失、齒齦清塞音韻尾 /t/ 較常丟失、雙唇及軟顎清塞音韻尾 /p, k/ 常丟失或發為 /t/。B組的國語口音較不明顯，亦即較常保留上述輔音。 本文針對AB兩組的聲母及韻尾語料數據，各提出一組浮動制約。相較於A組，B組在聲母的浮動制約多了一個信實制約MAX-PLACE、在韻尾部分也多了另一個信實制約MAX-CONSONANT(CODA)。因此B組的信實制約有較多機會浮動至高層，也導致較多信實的輸出值。由此可見，兩組差異之關鍵在於浮動而非固定的制約，因此認為浮動制約模型適合用於處理此種語言變異現象。 / This thesis establishes a corpus of Taiwan youngsters’ Mandarin-accented pronunciation of Taiwanese voiced stop onsets /b/ and /g/ and voiceless stop codas /p, t, k/. The Floating Constraint model (Reynolds 1994) within the Optimality Theory framework is adopted to analyze the corpus data. The corpus includes two groups: Group A and Group B. Both groups retain the voiced bilabial stop onset /b/. Group A has higher percentages of omitting /g/ onset and /t/ coda, and codas /p/ and /k/ are often altered to /t/ or omitted. On the other hand, Group B tends to preserve those sounds. For each group, a set of floating constraints is proposed to account for their onset and coda data. The differences between Group A and Group B lie essentially in the number of floating faithfulness constraints. The Group B grammar allows one more floating constraint MAX-PLACE for onsets, and one more floating constraint MAX-CONSONANT(CODA) for codas. I have shown that the Mandarin accent is better accounted for in terms of the “floating” constraints, but not the “fixed” constraints.