托福阅读材料推荐:埃及象形文字

　　8:24

　　Given such patterns, here was our idea. The idea was to use a computer to learn these patterns, and so we gave the computer the existing texts. And the computer learned a statistical model of which symbols tend to occur together and which symbols tend to follow each other. Given the computer model, we can test the model by essentially quizzing it. So we could deliberately erase some symbols, and we can ask it to predict the missing symbols. Here are some examples. You may regard this as perhaps the most ancient game of Wheel of Fortune.

　　9:04

　　What we found was that the computer was successful in 75 percent of the cases in predicting the correct symbol. In the rest of the cases, typically the second best guess or third best guess was the right answer. There's also practical use for this particular procedure. There's a lot of these texts that are damaged. Here's an example of one such text. And we can use the computer model now to try to complete this text and make a best guess prediction. Here's an example of a symbol that was predicted. And this could be really useful as we try to decipher the script by generating more data that we can analyze.

　　9:36

　　Now here's one other thing you can do with the computer model. So imagine a monkey sitting at a keyboard. I think you might get a random jumble of letters that looks like this. Such a random jumble of letters is said to have a very high entropy. This is a physics and information theory term. But just imagine it's a really random jumble of letters. How many of you have ever spilled coffee on a keyboard? You might have encountered the stuck-key problem -- so basically the same symbol being repeated over and over again. This kind of a sequence is said to have a very low entropy because there's no variation at all. Language, on the other hand, has an intermediate level of entropy; it's neither too rigid, nor is it too random. What about the Indus script? Here's a graph that plots the entropies of a whole bunch of sequences. At the very top you find the uniformly random sequence, which is a random jumble of letters -- and interestingly, we also find the DNA sequence from the human genome and instrumental music. And both of these are very, very flexible, which is why you find them in the very high range. At the lower end of the scale, you find a rigid sequence, a sequence of all A's, and you also find a computer program, in this case in the language Fortran, which obeys really strict rules. Linguistic scripts occupy the middle range.

　　10:49

　　Now what about the Indus script? We found that the Indus script actually falls within the range of the linguistic scripts. When this result was first published, it was highly controversial. There were people who raised a hue and cry, and these people were the ones who believed that the Indus script does not represent language. I even started to get some hate mail. My students said that I should really seriously consider getting some protection. Who'd have thought that deciphering could be a dangerous profession? What does this result really show? It shows that the Indus script shares an important property of language. So, as the old saying goes, if it looks like a linguistic script and it acts like a linguistic script, then perhaps we may have a linguistic script on our hands. What other evidence is there that the script could actually encode language?