在WOZ研究中使用自動仿真工具模擬仿真器的行為

Simulators' mimetic behaviours with automated simulation tools in wizard of oz study

「綠野仙蹤」（WOZ）使用「人」作為模擬器來模擬智能系統，這些系統超出了當前的技術水平或實現所需的時間和成本。在WOZ仿真中，越來越多的採用自動仿真工具。傳統的研究通常集中在終端用戶身上，對模擬器的行為所知甚少。為了彌合這一差距，我們進行了一項比較研究，以研究作為使用自動模擬工具模擬智能語音對話系統的模擬器性能的可信度、嚴謹性、一致性和效率。結果表現為兩種模仿行為：故意反應和本能反應。具體來說，不管經驗水平如何，模擬器都能很好的模擬故意的反應，在這種反應中，它們的反應是可信和有效的。然而，模擬器暴露了不一致和非嚴格的模仿行為的本能反應。此外，經驗豐富的和業餘的模擬器在響應速度、解釋最終用戶輸入和做出相應決策方面的本能模擬表現出顯著差異。本能模仿行為對模擬的最終用戶感知沒有顯著影響，這也反應出，討論WOZ作為一種高效設計工具的意義。

Wizard of oz (WoZ) employs human simulators to mimic intelligent systems that are beyond current technology levels or time- and cost-consuming to implement. Automated simulation tools are increasingly adopted in WoZ simulation. Conventional studies are usually focused on end users and little is known about the simulators' behaviours. To bridge this gap, we conducted a comparative study to investigate credibility, rigorousness, consistency, and efficiency in the performances of simulators as simulating smart spoken dialogue systems with automated simulation tools. The results reported two types of mimetic behaviours: the intentional and instinct response. Specifically, the simulators, regardless of experience levels, performed well as mimicking intentional responses, in which their responses were credible and efficient. However, the simulators exposed inconsistent and non-rigorous mimetic behaviours in instinctive responses. Furthermore, the experienced and amateur simulators showed significant differences in instinctive simulations in response speed, interpreting the end user inputs and making corresponding decisions. Nevertheless, the instinctive mimetic behavior had no significant effects on the perception of end user for the simulations. The implications for WoZ considering as an efficient design tool are discussed.