自然語言一直是個很熱門的研究領堿。一個能夠自動產生對話的代理人一直是所有人工智慧研究者的夢想。對話的產生包括兩部份，一是代理人要決定對話的內容。二是從對話的內容產生對話的實體。對話的內容包括對話的目的，對話相關的物件，對話所產生的影響。而對話的實體包括對話的文法，對話的選詞，對話的語氣，等等。本篇論文的重點在第一部份，也就是如何決定對話的實體內容。
為了使對話有意義，我們必需使對話能連結到要說話的人，也就是代理人的真實狀況。所以對話中談到的物體，需是代理人真的能理解的物體，如果代理人處在一個虛擬的環境中，我們就要透過那個虛擬的環境中存在的物品來設計對話，除此之外，對個對話因為必需是有意義的，所以每個對話也應該牽扯到到一個對代理人有意義的計畫。如此一來，就會延申出更多的問題。

The natural language generating is an exciting research topic. An agent that may speak is a dream for all artificial intelligent researchers. There are two domains for NLG. One is the grammar of the message and the other is the content of the message. In this paper, we will discuss some problem when the content of the message is involved in an executable plan. The difference of knowledge and believe between two agents may lead the conversation to a failure situation. We’ll discuss how the failure occurs and how to avoid it.

[1] David Traum, Jeff Rickel, Jonathan Gratch and Stacy Marsella. Negotiation over Tasks in Hybrid Human-Agent Teams for Simulation-Based Training. Proceedings of the 2nd Joint International Conference on Autonomous Agents and Multi-Agent Systems (AAMAS 03), Melbourne, Australia.
[2] Munindar P. Singh. Agent Commmunication Languages: Rethinking the Principles. IEEE Computer, 1998.
[3] David Traum, Jeff Rickel, Jonathan Gratch and Stacy Marsella. Negotiation over Tasks in Hybrid Human-Agent Teams for Simulation-Based Training. Proceedings of the 2nd Joint International Conference on Autonomous Agents and Multi-Agent Systems (AAMAS 03), Melbourne, Australia.
[4] Sarit Kraus,Katia Sycara, Amir Evenchik. Reaching agreements through argumentation: a logical model and implementation. Artificial intelligence 104(1998)