本研究以紀堯姆・阿波里奈爾（Guillaume Apollinaire）的詩作〈米哈波橋〉為個案，探討人工智慧是否能取代人類譯者進行詩歌翻譯。本文援引 Katharina Reiss（2000）的文本類型理論，將詩歌歸類為「表意型」（expressive）的文本類型，並比較三種機器翻譯工具（Microsoft Bing、Google Translate、DeepL Translate）與三位人類譯者（沈寶基、戴望舒、吳敏華）的翻譯成果。研究結果顯示，人類譯本往往能夠自覺地回應中西傳統詩學，並更有效地呈現法文原詩的詩體特徵，例如：詩句長度、詩節形式、押韻形態與詩語風格等。人類譯本在語義的精確度與深度方面，亦普遍優於機器譯文；後者難免出現或明顯或荒謬的錯誤，且難以再現詩歌傳統所涵詠的詩學特質。由此翻譯個案分析可見：在詩歌翻譯的審美層面、語義層面，乃至語法層面，機器譯文雖擁有迅雷般的翻譯速度，卻難免譯出扞格之謬誤；反觀，人類譯者因能融貫詩學傳統，進而匠心獨運，每每機杼獨出，呈現機器譯文與人類譯文之間難以跨越的楚漢鴻溝。
This study examines whether AI can replace human translators in poetry rendition using Guillaume Apollinaire’s poem “Le Pont Mirabeau” as a case study. The paper draws on Katharina Reiss’s (2000) theory of text typology to categorize poetry as “expressive” text-type. The researcher conducts a comparative translation study between machine-generated translations, via neural machine translation systems such as Bing, Google Translate, and DeepL, and human translations by three translators, including Shen Baochi, Dai Wangshu and Min-Hua Wu. The results show that human translations tend to be more conscious of traditional poetics and are better at representing the original poetic features of the French poem, such as line length, rhyming pattern, and poetic diction. Human translations also tend to have greater semantic accuracy and profundity compared to machine translations, which may commit gross errors and fail to represent the poem’s poetic features. Thus, human translators outperform machine translators in both aesthetic and semantic dimensions of poetry rendition.