科学家发现人类在学习语言的时候，大脑中会进行大量而复杂的运算。
讲话以及听人说话在生活中似乎是件很平常的事，大概很少人会想到为什么自己能够在一句长而流利毫不间断的话语之中，能够听出每个字，并且了解每个字的涵意。因此我们听人家说话才会有「大珠小珠落玉盘」般，字字分明的感觉。在Rochester University中，由Elissa Newport以及Richard Aslin所领导的团队正在研究这样的问题。
首先他们创造了一种语言，这个语言当中利用一些音节组成没有意义的字，然后以随机的组合让受试者听20分钟。在听的时候，受试者会接受音节的信息，像是音节出现的频率以及音节之间的关系等等，听完了20分钟之后再做测试。
实验结果发现，在成人的受试者当中，高达85％的时候他们可以分辨出哪些是刚刚听到的字，哪些是其它用英文中前缀字尾音节所组合的字，甚至连五岁小孩也可以分辨得出来。因此他们认为人类的大脑有能力对特定音节出现的频率以及音节与音节之间的关系去进行复杂的运算。
接着他们设计了另外一个实验，在实验当中包含三种语言，每种语言的字都是三个音节。
第一种语言是非相邻音节（non-adjacent syllables）的规则性，也就是第一跟第三音节不变，只改变第二个音节。
第二种语言是使用固定的子音，但是改变元音，像是：ring、rang、rung。
第三种语言则是固定元音，改变子音，在土耳其语当中就存在这种型式的规则性。
成人的受试者对于第一种语言的辨识率极低，不过对于第二种及第三种语言的辨识率就提高很多，科学家推测这可能是因为人们可以分辨某些子音关系的规则性，并且利用它们作为断句的依据。实验结果也显示，人类对于已经存在于目前人类语言当中的非相邻式规则性的辨识率，会比那些不存在于目前人类语言中的非相邻式规则性的辨识率来得好。因此他们推测，也许在语言成形时，人类在先天上就有某些偏好的撷取音节方式，而且这种偏好在语言形成上可能扮演了某种重要的角色。
另外他们也使用棉冠狨猴（Cotton- top tamarin）进行同样的实验，结果发现狨猴对于第一种跟第三种语言的表现较佳，这显示人类与其它灵长类在感知语言及计算音节的能力上也许是有所不同的。
由此看来，能听得懂别人在说些什么，还真是一件不简单的事啊！
原始论文：
Elissa L. Newport, Richard N. Aslin. Learning at a distance I. Statistical learning of non-adjacent dependencies, Cognitive Psychology , Volume 48, Issue 2, March 2004, Pages 127-162.
Elissa L. Newport , Marc D. Hauser , Geertrui Spaepen and Richard N. Aslin. Learning at a distance II. Statistical learning of non-adjacent dependencies in a non-human primate, Cognitive Psychology, Available online 3 March 2004.
Human brain works heavy statistics learning language
A team at the University of Rochester has found that the human brain makes much more extensive use of highly complex statistics when learning a language than scientists ever realized. The research, appearing in a recent issue of Cognitive Psychology, shows that the human brain is wired to quickly grasp certain relationships between spoken sounds even though those relationships may be so complicated they're beyond our ability to consciously comprehend.
"We're starting to learn just how intuitively our minds are able to analyze amazingly complex information without our even being aware of it," says Elissa Newport, professor of brain and cognitive sciences at the University and lead author of the study. "There is a powerful correlation between what our brains are able to do and what language demands of us."
Newport and Richard Aslin, professor of brain and cognitive sciences, began by looking at how people are able to recognize the division between spoken words when spoken language is really a stream of unbroken syllables. They wanted to know how it is that we perceive breaks between spoken words, when in fact there are no pauses. This is why it often seems as if speakers of foreign languages are talking very quickly; we don't perceive pauses.
So how is a baby supposed to make out where one word begins and another ends? Newport and Aslin devised a test where babies and adults listened to snippets of a synthetic language: a few syllables arranged into nonsense words and played in random order for 20 minutes. During that time, the listeners were taking in information about the syllables, such as how often each occurred, and how often they occurred in relation to other syllables. For instance, in the real words "pretty baby," the syllable "pre" is followed by "ty," which happens more frequently in English than the syllable "ty" being followed by "ba"--thus the brain notes that "ty" is more likely to be associated with "pre" than with "ba," and so we hear a pause between those two syllables.
After listening to the synthesized string of syllables for the full 20 minutes, adults were played some of the invented words along with some words made up of syllables from the beginning and ending of words--like "ty-ba." More than 85 percent of the time, adults were able to recognize words from non-words. Five-year-olds also reacted definitively to words and non-words, showing that the human mind is wired to statistically track how often certain sounds arise in relationship to other sounds.
"If you were given paper and a calculator, you'd be hard-pressed to figure out the statistics involved," says Newport. "Yet after listening for a while, certain syllables just pop out at you and you start imagining pauses between the 'words.' It's a reflection of the fact that somewhere in your brain you're actually absorbing and processing a staggering amount of information."
Newport and Aslin take the research a step farther in the Cognitive Psychology piece. Language does not only consist of relationships between adjacent syllables or other language elements. For instance, in the sentence, "He is going," the element "is" is linked to the element "ing," even though they are not adjacent to each other. Newport and Aslin devised a new, more complex, synthetic language where three-syllable words had constant first and last syllables, but the middle syllable was interchangeable. Despite being somewhat similar to the original test, "people were terrible at this," notes Newport. One test subject never identified a single pattern, despite taking the test numerous times.
Though the new test was significantly more complicated than the first, Newport and Aslin were surprised that people performed so poorly. The team looked carefully at the non-adjacent aspects of languages, like Hebrew, which is replete with non-adjacent elements, and discovered that while whole syllables were rarely related in this way, vowels and consonants often were. They restructured the test so that the invented words had consistent consonants and variable vowels--like "ring", "rang," and "rung." Immediately, test scores skyrocketed. People were able to distinguish the regularity of certain consonant relationships and use them to properly divide the stream of sounds into words even though the statistics involved were at least as complicated as the earlier test that was universally failed.
Even switching the roles of consonants and vowels so that the vowels remained steady as the consonants varied, resulted in the test subjects picking out the words with great accuracy. Turkish, as an example, uses this "vowel harmony" quite regularly.
"These results suggest that human learning ability is not just limited to a few elementary computations, but encompasses a variety of mechanisms," says Newport. "A question to explore now is: How complex and extensive are these learning mechanisms, and what kinds of computational abilities do people bring to the process of learning languages?"
