What is a chatbot? What is an AIML? Before these questions being answered, firstly we need to know something about Natural Language Process (NLP). According to the definition of NLP from Natural Language Processing Research Group (NLPRG), the largest NLP research group in the UK. NLP (Natural Language Process) focuses on investigating the properties of human written language and developing innovative practical applications involving the intelligent processing of written human language by computers.
Then back to the main topic – chatbot, the word made up with chat and robot, the chatbot (actually you can call it conversational agent, chatterbot or virtual human, etc.) is a kind of artificial living entity designed to have NLP conversations with real human beings. That normally is a text conversation via computer screen, but sometime can be a spoken conversation via some particular equipment (for example a robot with microphone) or even a non-verbal conversation. The original chatbot is named Eliza and accepted as the most famous chatbot, Eliza is a program which can start a conversation between the user and itself in a psychiatrist talk, developed at MIT in 1966.
AIML the word mentioned in the beginning of the article is the short form of Artificial Intelligence Makeup Language. AIML is a product from a 7-year cooperating development of a worldwide free software community started in 1995, based on a highly evolving Eliza which is called ALICE (Artificial Linguistic Internet Computer Entity). Most chatbots on the internet are generated by using AIML. AIML provides two aspects to implement Natural Language Process for the chatbots. First, parsing the text from human beings and extracting meaning of it. Second, generating dialogs from the current database, these two processes are called Pattern Matching Process (PMP). The technology is to write a huge database of patterns in one massive XML file. Each pattern contains specific words and some words may be mentioned together in high possibility, so different variations of the same text are able to recognize and label. Then the chatbot can make a choice to response in a corresponding pattern. Now people can use free AIML knowledge base (most AIML interpreters are offered under a free or open source license) for their own chatbots.
In the field of Artificial Intelligence of computer games, some similar technology have been already used in computer game context, for example, you can have a contents limited talk with NPCs (Non Player Characters) in some Role Playing Games (RPG) like Fallout3, the NPCs will response differently depends on your kind behaviors and evil behaviors in games. At the meanwhile scientists are still trying to create smarter NPCs in computer games.
Things seems to be perfect, however, there are still problems. In my opinion, if we try to create some much smarter chatbots or computer game characters (you can hardly tell whether it is a visual agent or not) in the future, we will never succeed by keep using the style of generating chatbots or NPCs from a prepared knowledge base like AIML. From nowadays chatbots’ performance, we can find some distinct defects of AIML. These defects limit the capability of AIML.
The first defect is that AIML need time to match the patterns. Since AIML responses on a Pattern Matching Process, and PMP needs a knowledge base as huge as possible to keep millions of different sentence pattern or even different language types. If we put all these in a computer game and design the NPCs by using such system, how long could we wait for the NPCs’ answers? Just imaging you would have to wait for 10 or 20 seconds to get a response from the NPC in some adventure games, such time cost can not be accepted in real applications. This limits the applicability of the technology to specific parts of games that don’t feature enough interaction. If a visual agent can not response in time when you order or ask, it will be quite disappointing. So, I think a limited “chatbot” style NPCs in computer games is still the best choice for computer game design in a long time, excepting our scientists create some revolutionary computers with much higher calculation ability (Such as a quanta computer).
The second defect is that AIML can only response to the user’s last statement, it makes the whole conversation between computer and human not very logical. Because the patterns in the knowledge base doesn’t have logical relationship or just have some simple logical rules. Even assuming we add as many as possible logical rules (I think it will still make the conversation not very logical, because the human language is fuzzy and unpredictable), the question will go back to the first question: the time cost. In any computer games or chatbot the AIML never really understand any natural language words, they just find the possible response to you.
Even assuming we create the big enough knowledge base and collect enough language types, conversation patterns and perhaps the knowledge of whole human history from Stone Age to Twenty-First century. And we also set very strict logical rules of natural language and generate a very outstanding language generating system. Why we should use them in computer games? The visual characters with limited functionality can still performance very well in today’s computer games. Of course, the limitations that forbid a human-level chatbot’s appearance are still there, we need non-Turing machine computer (Personal idea).
Bibliography
[WAC09] What is A Chatbot? http://www.chatbots.org/directory/static/what_is_a_chatbot/, [2009].
[TDNL] The Definition of Natural Language Process, http://nlp.shef.ac.uk/index.html, [2009].
[Ozveren09] C. Süheyl Özveren, AIML Artificial Intelligence Mark up Language and related technologies in the context of Games. [2009].
[WAAI] What is AIML Artificial Intelligence Markup Language, http://www.alicebot.org/aiml.html
[2009].
扫一扫
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
