by Tony Veale

单口相声经常哀叹这样一个事实:“每个人都他妈的是喜剧演员!”这是真的:每个人(在某种程度上)都欣赏幽默，而且大多数人都能自发地产生某种形式的幽默。但正是这种普遍性使得幽默更难而不是更容易被正式理解，因为幽默有许多伪装，并在无数的语境中以微妙的差异运作。历史上，许多哲学家和其他思想家都提出过关于幽默的理论，但没有一个是完全成功的，因为他们都倾向于强调幽默的某一方面。我最喜欢的观点是亚里士多德提出的，他认为幽默是一种“有教养的傲慢”。当然，你必须忽略亚里士多德学派学派式分析的枯燥细节，以及他喜欢把每个主题分成这七种类型和那五种类型。在考虑幽默理论时，不要去看细节(它们通常都是非常粗略的)，而要试着提炼出其本质，看看它是否能公正地反映现象。

说“幽默是受过教育的傲慢”是什么意思?亚里士多德用两个关键词说明了一切。首先，“受过教育”并不是说幽默只能从正规的教育中产生，而是说幽默是一个基于知识的过程。要取笑这个世界或它的居住者，我们需要挖掘知识，而知识越丰富，幽默就越微妙和复杂。对于一个计算主义者来说，这意味着知识表示是幽默理论的一个基本元素。人工智能领域的理论家们研究这个问题已经有50年了，他们为这些方案产生了大量的表示方案和推理机制。一些雄心勃勃的项目，如奥斯汀的Cyc项目，旨在使用这些方案和机制来代表典型人类所拥有的常识知识的总和。虽然我对Cyc作为计算式幽默的表现形式持怀疑态度(我在这个项目上工作了一年)，但我相信，正如Cyc的指导学者Doug Lenat所描述的那样，“知识瓶颈”是幽默研究中的一个关键问题:没有知识，就没有幽默，也就没有幽默产生。

第二个词“傲慢”同样重要，因为它描述了一个人应该如何运用所有这些知识来理解和制造幽默。在这里，人们可能会将傲慢视为一种侮辱性行为，因此将幽默视为喜剧演员优越感的一种表现。这个普遍的理论，被称为幽默的优越性理论，认为幽默具有攻击性，所以每个笑话都有一个靶子或一个屁股。然而，傲慢是一个更普遍的概念，指的是一种颠覆性的态度，不只是对其他人(因为一些幽默是明显的侵略性)，而是对情况，信仰和知识本身。就我个人而言，我倾向于在幽默中看到我们在科学思维实验中看到的那种颠覆性行为。在这样的实验中，一位科学家提出了一个简单的实验，不需要任何仪器，但可以在大脑的实验室中进行。伽利略和爱因斯坦都是格丹肯实验的大师，他们利用格丹肯实验来说服人们相信传统科学智慧是愚蠢的。一旦这种智慧被实验证明会导致荒谬的局面，就为新理论的出现扫清了舞台。因此，我看到许多笑话具有压缩的思想实验的特点。Try this one on for size:

"If God wanted us to be vegetarians, he wouldn't have made animals out of meat!".

哪些观念在这里被颠覆了，素食，动物还是肉类?答案似乎是三个都有，因为我们似乎看到了三个非常特殊的物体，它们同时颠覆了三个不同的类别。首先，我们被引导去想象一个例外的动物类别的成员，动物作为肉类机器，所有非功利的方面都与它分离;如果这样的动物没有知觉，素食主义就没有道德基础。其次，我们被要求想象一种特殊的肉，一种拥有传统肉的所有生物学特性，但可能不是来源于动物的肉。第三，我们被引导去想象一种特殊的素食者，他们会吃非动物来源的肉。这三种颠覆共同导致了对素食主义这一范畴的颠覆，因为如果素食者可以自由地吃肉却仍然是素食主义者，这种生活方式还能保留什么样的道德力量呢?上面这个笑话实际上是一个高度压缩的思想实验，因为它试图破坏传统观念，即素食主义是一种道德上优越的生活方式，同时为肉食者的道德自由放任辩护。颠覆的一些最有效的用法旨在产生更本能的影响，比如“饮食被高估了”。记住，食物只是等待发生的大便。”

When most people hear that computer scientists are trying to model humor processes on a computer, their reactions typically range from the "why bother" (or "don't expect tenure") to "it's clearly impossible". As a topic of computational research, humor seems both wasteful and futile; even it succeeds, do we really need a computer with a sense of humor? Those people that already believe that computers are too smart for them would surely not be pleased to think that their computers are also laughing at them. It doesn't help, of course, that the archetype of intelligent computers in pop-culture is HAL from 2001, who murdered his crew. When I worked at Cyc, my boss looked forward to the day that Cyc would exhibit language capabilities like HAL, but presumably (and I can't be sure) he wasn't looking forward to Cyc murdering me and my co-workers.

More seriously, few computer scientists work on humor as their main topic of research, and for most, like myself, it is an interesting (but relevant) side-line. For one, is not an entirely wise career choice. I've met graduate students at humor conferences (yes, they exist, but they can be very dry indeed) who are there against the advice of their supervisors, who suggest it is better to study these topics after tenure has been secured (this places humor research in the same scientific category as paranormal studies!). Second, and more realistically, there are so many problems to do with general human intelligence and language competence that must be solved first before we can even begin to think about genuinely humorous computers. The state of the art in computer-generated humor is still in the school-yard, intellectually speaking. Computers can do a very good job of generating puns, and even humorous acronyms (such as CIA = Central Incompetence Agency, to pick one at random). To understand and generate truly conceptual humor, where ideas rather than just words are manipulated, requires that we first understand other aspects of creative language use. To my mind, the most important aspect of human language is metaphor. This is the primary focus of my research (though its only slightly more respectable than humor in computer science circles), since metaphor underpins our ability to stretch the conventions of language and describe people and ideas in strikingly novel ways. You don't need to read poetry to encounter the need to process metaphors. Almost all natural language texts are permeated with metaphors, from the Bible to the Wall Street Journal, so there is a real financial imperative to make substantial engineering progress on this topic. Once our computers can understand and produce metaphors, they will possess the educational requirement of Aristotle's theory. Then it will be a matter of using this education for insolent purposes. Choice insults like "Baldrick, your family tree has Dutch Elm disease" are just around the corner.

For those readers interested in knowing more about computational approaches to metaphor, and indirectly, humor, please do visit my group's research web-site at:afflatus.ucd.ie. (btw: "Afflatus" is not a bowel-complaint, but a pretentious term for the "creative urge" foisted upon our server by a departing graduate student). Alternatively, you can contact me directly attony.veale@UCD.ie.

托尼·维尔(Tony Veale)是都柏林大学学院的计算机科学家。他也是我们今天幽默节目的嘉宾。