中文新聞語境中的擬人化分析：以人工智慧論述為例

Abstract

近幾年人工智慧（AI）發展迅速，為了讓社會大眾更容易理解這項前瞻性的技術，媒體經常採用擬人化的方式來報導AI。然而，過度將人類特質投射到AI身上，可能導致公眾對AI的實際能力產生誤解，進而影響民眾對相關科技風險與倫理議題的判斷。因此，本研究選擇台灣主要數位新聞平台ETtoday作為研究案例，分析其AI相關報導中的擬人化現象。 研究採用史丹佛大學自然語言處理團隊開發的AnthroScore框架，並針對中文AI內容特性進行了三項重要改進：(1) 採用兩步驟預測法，讓模型在評估前充分理解完整語境；(2) 調整為中文適用的代詞列表，如「他」、「她」、「它」；(3) 引入基於閾值的過濾機制，提高結果可靠性。 本論文分析2016年至2024年間的AI新聞，發現三個重要趨勢：第一，AI新聞的擬人化程度呈現顯著上升趨勢；第二，娛樂、社會及體育新聞相較其他類別展現較高的擬人化分數,反映這些領域更傾向將AI描述為具有人性特質;第三，ChatGPT相關報導比一般AI實體有更高的擬人化程度，顯示對話式AI更容易使用擬人化描述。 本研究成功將AnthroScore框架應用於中文語境，為未來跨語言的擬人化研究提供了新的研究途徑。本論文結果揭示了台灣媒體對AI的擬人化趨勢，提醒新聞從業人員在報導AI議題時應更加謹慎，以避免誤導讀者。本研究結果不僅有助於了解AI在華語媒體中的描繪方式，更對推動負責任的科技新聞報導產生深遠的影響。

With the rapid advancement of artificial intelligence (AI), news media often employ anthropomorphic representations to help the public better understand this emerging technology. However, excessive attribution of human characteristics to AI can lead to misconceptions about its actual capabilities and affect public perception of technological risks and ethical issues. This study examines AI-related news from ETtoday, a major Taiwanese digital news platform, to explore anthropomorphism in AI reporting. Applying the AnthroScore framework, developed by Stanford’s NLP Group, this study introduces three key modifications for Chinese-language analysis: (1) a two-step prediction method to enhance contextual understanding, (2) an adapted pronoun list including "他" (he), "她" (she), and "它" (it), and (3) a threshold-based filtering mechanism for improved reliability. Analyzing AI-related news articles published between 2016 and 2024, this study finds: (1) a significant increase in AI anthropomorphism over time, (2) higher anthropomorphism in entertainment, society, and sports news, and (3) ChatGPT-related articles showing greater anthropomorphism than general AI reports, indicating conversational AI elicits stronger human-like descriptions. This study contributes by adapting AnthroScore for Chinese, offering insights for cross-linguistic research. It also highlights Taiwan’s media tendency to anthropomorphize AI, underscoring the need for responsible journalism to prevent public misinterpretation.