Defining a Human-Machine Teaming Model for AI-Powered Human-Centered Machine Translation Agent by Learning from Human-Human Group Discussion: Dialog Categories and Dialog Moves

Abstract

The vision of human-machine symbiosis is that a human will work closely and harmoniously with the machine. We study interactions among human translators to help define potential Human-Machine Interface (HCI) and Human-Machine Teaming (HMT) models for human-centered machine translation systems. The role of the machine, in this context, is to be an AI-based agent serving as a real-time partner. The questions we ask are that if we have such an agent, what are the main functions of the agent, how does the agent interact with a human translator in a way where they can work symbiotically as partners, what human deficiencies can be augmented by an AI-based agent and how, and what kind of human behaviors should the AI-based agent mimic. We used a data set collected from an online translation study group composed of certified and highly experienced translators (both English-to-Chinese and Chinese-to-English), and altogether we analyzed several hundred dialogs between these translators related to the translation results. Each dialog always started with an initial comment and would focus on one of many possible categories. The first question we asked was which categories were discussed more frequently than the others, and why. At both the word level and above the word level, three dominating categories were found: source misunderstanding, target expression problem, and confirmation on good translation. In addition, we found that the more than half of the dialogs focused only at the word-level. The second question we asked was whether a discussion (represented by a dialog) was effective or not. What we found was that the most common pattern was the one associated with “simple but effective” dialog, while constructive dialogs were conversely very infrequent. Based on these findings, we derive the HCI/HMT design implications for an AI-based translation agent: provide better capability beyond the word/phrase level to complement human deficiencies; focus on building algorithms to support better source understanding and target expression delivery; provide quick information search and retrieval to support real-time interaction; provide confirmation to a human partner’s good translation with reasons and explanations; provide help regarding source understanding and target language delivery based on the native language of the human partner; act in the role of a “lead translator” who has better domain and linguistic knowledge, superior cognitive capability, and unique analytic perspectives to complement human deficiency; and perform in-depth constructive dialogs with human partners by stimulating thought.