语法和词汇是两套系统。
拖了好久才看平克这本words and rules，他的讲述还是比较平易近人的。
1 The Infinite Library
首先正如标题，words和rules是本书着重强调的两个概念：
the word, is based on a memorized arbitrary pairing between a sound and a mean
The entry for a word is simply its address in one’s memory
First, the rules are productive … that allows us to convey unprecedented combinations of idea
Second, the symbols contained by the rules are symbolic and hence abstract.
Third, the rules are combinatorial
words and rules are two modes of operation of a single faculty
Bishop John Wilkins在启蒙运动时期为哲学语言作出贡献。
wug-test 在平克其他书也有提到。
然后引出了本文主要的讨论内容——irregular verb
Irregular forms are just words
Irregular and regular forms therefore would be the inevitable outcome of two mental subsystems, words and rules, trying to do the same thing,
2 Dissection by Linguistics
explain morphology as well as other English grammar to native speaker, esp. to those who never learn a foreign language. 讲语法的过程好像回到了小学上XX英语。
下面这句话纯粹是有趣
Science is not always kind to folklore from the natural world. Elephants do forget, lemmings don’t commit mass suicide, two snowflakes can be alike, we use more than 5 percent of our brains, and Eskimos don’t have a hundred words for snow.
We need to communicate, and language is the fulfillment of that need. For every idea there is a word and vice-versa, and we utter the words in an order that reflects the connections among ideas. If this common-sense view is true, there would be little need to speak of language being a complex system. The complexity would reside in the meanings, and language would reflect that complexity directly.
Language does express meaning as sound, of course, but not in a single step
One is a storehouse of memorized words, the mental lexicon. Another is a team of rules that combine words and parts of words into bigger words, a component called morphology. A third is a team of rules that combine words into phrases and sentences, a component called syntax. The three components pass messages about meaning back and forth with the rest of the mind
Morphology may be divided into derivation—rules that form a new word out of old w
and inflection—rules that modify a word to fit its role in a sentence
Syncretism—one form, several roles—is one kind of violation of the simplest conceivable system in which every sound has one meaning and vice-versa
why do we need separate boxes for semantics (the thoughts expressed in language) and the lexicon?
First, the English irregular verbs could not have arisen simply from a communal effort to optimize clarity.
Also, irregular forms do not correlate with any kind of meanin
Compared to ordinary verbs they are less filling; a light verb doesn’t have a meaning that stays with it, but takes on dozens of meanings, especially in combination with particles such as in, out, up, off, over, and around
irregular verb in English, the so-called strong verbs. They belong to alliances with similar sounds, and despite this solidarity, they have been dwindling for millennia.
The rules themselves, however, did not survive.
3 Broken Telephone
irregular nouns
(1) Many nouns ordinarily don’t take any plurals: mass nouns such as mud, celery, furniture, and evidence are treated as seamless stuff rather than countable things.
exactly seven change their vowel instead of adding -s:(2)Several names for gregarious animals that are hunted, gathered, or farmed are identical in the singular and plural
(3)Another three irregular plurals take the old Anglo-Saxon suffix -en rather than -s:
(4) A fourth class of nouns takes the regular -s ending but changes its final consonant, usually f but sometimes th or s, from unvoiced to voiced
(5) Finally, there are nouns that take Latin or Greek plurals
irregular verbs
Be, together with go, they are the only verbs whose past tense is a completely unrelated word, a relation that linguists call suppletion. Suppletion arises from a merger of two verbs
no-change verbs arose in Middle English and Early Modern English
why is plural form of foot is feet ? [umlaut 元音变音] umlaut: original fot -> foti (plural) o and i harmonize -> feti (in pronunciation)
The perpetrator is a process of language change that is the opposite of the various slurrings and swallowings and cutting of corners that we have seen so far. All of those changes make it easier for the speaker to speak but do nothing for the listener, who would rather have the speaker enunciate clearly
4 In single Combat
What model has been applied to the development of past tense of English language - on why irregular verbs deny changing (to be regular)
if the irregulars were truly acquired one by one they could just as easily have been the rule.
The irregular patterns refuse to die
霍布斯
associationism
empiricismIt’s been three hundred and fifty years since Leviathan, and scholars are still debating rationalism and empiricism. Both sides appeal to theoretical coherence, intuitive plausibility, political ramifications, and harmony with the morals of modern science, but these are pretty squishy criteria, and the debate rages on.
Two model introduced:
(1)Chomsky [again and again], and (2) Rumelhart & Mcclelland
Chomsky
generative phonology developed by Noam Chomsky and Morris Halle
What are the facts to be explained? Irregular verbs defy the suggestion that they are memorized by rote because they show three kinds of patterning.
First, irregular past-tense forms are similar in sound to their base forms.
Second, a few kinds of change from a stem to its past are seen over and over among the 164 irregular verbs.
Third, the verbs undergoing a given irregular change are far more similar than they have to be.
Noam Chomsky and Morris Halle’s 1968 magnum opus The Sound Pattern of English
Another rein on rules keeps each one in a stratum (a component or subcomponent) so that it does not run wild and apply where it shouldn’t.
power of ‘economy’ and the great vowel shift
This economy and power also accrues to rules that fiddle with vowels if the vowels are dissolved into features
the Great Vowel Shift in the fifteenth century
long vowel tensing & Towards orthographic
Chomsky and Halle therefore proposed that English has a rule of Long Vowel Tensing, which tenses all long vowels, and a rule of Diphthongization, which adds the little ys and ws that give us the two-part vowels in lake (leh-eek), glide (gla-eed), need (nee-y’d), loud (la-ood), and road (ro-ood).
English spelling is not only exonerated of the charge that it is an illogical, sadistic mess, but “comes remarkably close to being an optimal orthographic system.”
The defunct-rule explanation has an advantage over the Chomsky-Halle-Mohanan theory.
First, why would the child bother if the rules are there only to generate the surface form, and the child already has the surface form?
cons of Chomsky’s theory
Chomsky-Halle-Mohanan theory cannot explain the third kind of similarity running through the irregulars: the similarities among stems, as in sting, string, sling, stink, sink, swing, and spring.
Irregular clusters are family resemblance categories with fuzzy boundaries.
irregular clusters are family resemblance categories
They don’t have strict, all-or-none definitions that specify which verbs are in and which verbs are out. Instead they have fuzzy boundaries and members that are in or out to various degrees depending on how many properties they share with one another
长得很像神经网络的那个什么东西。key idea is not words themselves but their features
morphology box—sitting between meaning and sound
The mainspring of the model, then, is forming associations between features and features, and that duplicates the human habit that embarrassed the words-and-rules theory: generalizing irregular patterns to similar words.
Rumelhart and McClelland
First, Rumelhart and McClelland’s pattern associator memory is a device that only produces past-tense forms
Second, the model computes every detail of the pronunciation of the past-tense form. Yet we saw that many of these details, such as the choice among -t, -d, and id, are found in fifteen different parts of the language system
Third, by forgoing the use of lexical entries and relying entirely on a word’s sound to compute its past-tense form, the model cannot tell the difference between two words that have the same sound.
A fourth problem is that Rumelhart and McClelland had to use some jiggery-pokery to get the model to duplicate children’s stages of language development.
Comparison
One phenomenon, two models, both explaining too much to be completely wrong, both too flawed to be completely right. Prince and I have proposed a hybrid in which Chomsky and Halle are basically right about regular inflection and Rumelhart and McClelland are basically right about irregular inflection.
defunct-rule explanation
Wickelphone problem
5. Word Nerds
这一章节讲了之前的words and rules是怎么为我们的记忆和思维所用的。
This chapter looks at how words and rules pop into mind as we use language in real time.
According to the theory, regular forms are generated by rules, and irregular forms are retrieved from memory; the memory, however, is not a list of slots but is partly associative, linking patterns with patterns as well as words with words.
通过活用/变形产生新词是语言的生产力的体现。
The great challenge for any theory of language is productivity, the ability to generate and understand an unlimited number of new forms.
The morphology module allows it to store just the stems of words and to compute the inflected and derived forms by rule.
英语里的不规则动词都是最常用的动词和名词。这也是可以理解的，如果不规则动词都不常用的话，那些变形过个几百年就没人记得
A simple property of memory is that the more often you hear something the better you remember it.
Irregular verbs are the most common verbs and vice-versa, in English and in most other languages. The explanation is simple. Irregular forms have to be memorized repeatedly, generation after generation, to survive in a language, and the commonly heard forms are the easiest to memorize. If an irregular verb slips in popularity, a generation of children will fail to hear its past tense often enough to remember it
下面举了forgo和bear两个例子，说明了一些不规则变形会让然觉得怪怪的，是因为这些动词本身不常以过去式出现。
Take the verb forgo. Though uncommon, it retains a certain liveliness, particularly in the sarcastic phrase to forgo the pleasure of, as in You’ll excuse me if I forgo the pleasure of watching the video of your wife giving birth. Now try to put it in the past tense. The word is certainly not forgoed, but the alternative is not quite right either: Last night I forwent the pleasure of watching Hank’s vacation slides.6 Similarly, it is perfectly natural to say I don’t know how she can bear the guy, but something is odd about I don’t know how she bore the guy.
一个有趣的现象叫hapax legomena说的是在古文献中只出现一次的词，无法通过其他例子确定这些词的意思。越多这种词汇说明语言的生产力越强。
Equally informative are the clues found among the verbs with the lowest frequencies of all, those that appear exactly once. There is a lovely technical term for a word that appears once in a body of text: a hapax legomenon, plural hapax legomena, Greek for “once said.” The term comes from philology, the study of old texts. Hapax legomena can be a nuisance to scholars of ancient languages because with only one instance of a word one can never be sure what it means
The body of water is the English language. A fish is a suffixed word. A million casts of the fishing rod is a million-word corpus. A fish that has been caught ten times is a word with a frequency of ten per million. A fish that has been caught only once is a word with a frequency of one: a hapax legomenon. If the creel is filled with hapax legomena, the words must be breeding quickly. That is, the suffix must be productive and the set of words accepting it open-ended. The vague notion that a rule of language is “productive” or “open-ended” can therefore be translated into a number.
当我们使用不规则动词的时候，我们并不是直接回忆起变形，而是要先“抑制”+ed的规则变形。这可以是说明一个单向性：我们更容易看到give说gave，看到gave却不容易第一时间想起give。
When you produce an irregular form, you not only have to dredge it out of memory but also must repress the “Add -ed” rule so you don’t say breaked or broked. Linguists call this principle blocking—the irregular form blocks the rule—
They found that when people heard asked, it was easier to read ask, but when they heard gave, it was no easier to read give
A word can prime a word with a similar meaning only if the second one appears immediately after the first; the priming by shared meaning dies down quickly. A word can prime itself, however, over many seconds, minutes, sometimes even hours or days.
另外一种说法是因为人脑记忆空间不足，所以不规则动词在减少。这个理论存疑，因为我们的记忆能力事实上还是很富裕的。
another prediction of the theory does not square with the facts. The theory’s main precept is that any trace of redundant information is pried out of memory and computed by rule; the theory assumes that in the human brain, memory is expensive but computation is cheap. The ultimate redundancy is in the regular verbs, where storage of the past-tense forms would be unconscionably wasteful. If predictable forms are never stored, then surely regular verbs are not stored.
Human memory is not a scarce resource reserved for the incompressible nuggets that cannot be generated by rules. If a word form is common enough, we can look it up directly, rather than breaking it into parts and looking up the parts.
总结是，记忆和计算分别对应words and rules，而我们在使用语言的时候是不同功能的组合。
the most important lesson of the chapter is that the mind, like any complex device, is a system of mechanisms optimized for different jobs. Any theory that has one mechanism doing all the work is proposing a kind of crippleware that the human brain is bound to outperform. If the mechanism is a set of rules, it loses the advantage of cacheing the results of frequently performed computations so that it can look them up quickly rather than recomputing them every time. If the mechanism is a set of associations, it loses the advantage of variables and the rules that combine them.
6. Of Mice and Men
正餐前的插曲，关于棒球里的flyout（应该是接杀的意思？）
You won’t find it in most dictionaries, but flied is the past tense of fly in one specialized field: baseball. You could not say of the batter who hoisted a can of corn to the center fielder that he “flew out”; you must say he “flied out.”
下面介绍本章内容
In the next two chapters we will see that no magical combination of settings is likely to work for all inflections in all languages. But in this chapter, I set aside questions about numbers. Rules will show their worth, and pattern associators their limitations, because of the kind of information captured in words, not how many words of various types there are.
语义并不影响词形变化（后面会解释，语音上的影响更多）
Semantic stretching in itself has no effect on a word’s past tense or plural. There are hundreds, perhaps thousands, of examples in which the meaning of an irregular word is stretched, sometimes to the breaking point, and people do not abandon its irregular forms; Using a noun as a metaphor also does nothing to its irregularity.
词汇是根，规则是头
This chapter will show how the word structure theory explains dozens of puzzles about English words that fill the language columns and cartoon pages. Since it appeals to the essence of words (the root) and to the essence of rules (the head), the success of the theory will stand as a confirmation of the words-and-rules theory more generally. We begin with words that cannot find their roots; then we will turn to words that have lost their heads.
An irregular plural or past-tense form is a root linked to another root: sank to sink, feet to foot:
没有根的动词不会被大脑记住，所以直接应用规则，例子有：(1) 拟声词，(2) 引语，(3) 专有名词，(4) 外来语，(5) 非自然含义，(6)词性变换
When a word is rootless and thereby disconnected from inflected forms stored in memory, however, it is not left without a past tense or plural; the rule rushes in and turns it into a regular form by adding a suffix.
The first example of this effect is onomatopoeia,
The engine pinged [not pang or pung].
My grant got dinged [not dang or dung].second kind of sound that lacks a canonical root is a quotation. A quotation does not have to use canonical words; it reflects a stretch of sound that someone else has said; quotations are not perceived as roots and fail to link to roots and their associations in memory.
A third way a word can be rootless is to be based on a name.
A fourth class of rootless words are foreign borrowings such as latke and cappuccino,
A fifth class consists of words that are recognized as rootless because they were concocted by artificial means. To synch is a truncation of to synchronize, and its past tense is generally synched
The sixth and final example consists of words that lack their own roots because they are converted from a root of a different part-of-speech category.
类似地，我们可以考察“没有头”的词语，规则变形对于他们不适用
How does a word lose its head? One way is to be a compound that doesn’t refer to the kind of thing indicated by its rightmost word. Instead it refers to something else,
cutthroat is not a kind of throat, nor is a lazybones a kind of bones. Linguists call these bahuvrihi compounds
Headlessness explains a fourth curiosity, this one a quirk of spelling. English spelling is a rule system that connects the sounds of words with their written forms. As with grammar, spelling is rife with irregularities and complications, especially in frequent words such as eye, of, have, and would.
Dollies (the eponymous cloned sheep
对上述的总结
the exceptions to the regularization effect are exceptions that prove the rule: When people don’t perceive a word as headless, they don’t plug the pipeline that sends irregular forms up from memory either. That secures a major kind of evidence for the word structure theory and for its parent, the words-and-rules theory. Irregular forms are word roots stored in memory; regular forms are computed when memory fails to cough up a form, for any reason. That in turn shows that a rule is a mental operation that manipulates variables, such as “verb” or “noun,” rather than an association to concrete memories of particular words and their sound patterns. It also shows that people erect an abstract mental scaffolding around words. The memory blockage in the examples in this chapter come from the nature of the mental scaffolding: People sense whether a word is stored as a root, and whether a word has a structure that allows information about the root to percolate up from memory
A regular rule is a powerful instrument, creating inflected forms for a motley collection of rare, strange, and eccentric words. Is there any place it cannot work? Indeed there is, and it is my final demonstration of the difference in kind between regular and irregular inflection.
Regular plurals don’t like appearing inside compounds.
qualitative difference between irregular and regular forms: Irregulars are roots and can be the input to the process of word formation; regulars are the products of rules and have to be the output of the process of word formation
当然任何理论都不是完美的，都会有反例。
There are counterexamples, as there always are in linguistics.
first law of language: Strings are nothing, trees are everything
The examples also show that people consider more than sound when they form new words: An input sound like fly can emerge in the output as either flew or flied, depending on the person’s analysis of the whole word. But people consider more than meaning, too; a word’s meaning may be stretched to a wispy filament, as in threw up or cut a deal, and people will inflect it as if nothing had happened.
最后，我们转入下一章的话题：儿童是如何学习词汇和规则的呢？
Can we catch children in the act of learning a rule as they master their mother tongue? Do rules work in all the world’s languages the way they do in English? And can we distinguish words and rules in the human brain?
7. Kids Say the Darnedest Things
Nothing is more important to the theory of words and rules than an explanation of how children acquire rules and apply them—indeed overapply them—to words. The simplicity of these errors is deceptive. As we shall see, it is not easy to explain why children start making them, and it’s even harder to explain why they stop.
本章的重点是考察儿童如何掌握语言和规则。研究发现儿童的掌握规律呈现U形，意思是说一开始儿童掌握程度很高，基本不会有错，后来随着学习会出现错误，再后来错误逐渐减少，熟练运用语言和规则。这是唯一一个呈现U形的学习趋势，儿童在其他领域的学习通常呈现直线上升——时间越久掌握程度越高。
if the percentage of past-tense forms of irregular verbs that are correct is plotted over time, the shape of the graph looks a bit like a U. “U-shaped development” fascinates child psychologists because with almost anything else you measure, children get better as they get older
产生这种现象的重要原因是过度普适化。儿童一开始会学习一定词汇，这部分是一个一个学的，不怎么会出错；之后学习规则，他们会把规则普适到所有词汇里，所以导致他们在不规则变形上出错；后来他们才会掌握所有规则和不规则变形。
Overgeneralization errors are a symptom of the open-ended productivity of language,
even the English plural shows statistics unlike those of the past tense. The handful of irregular nouns known to children (men, children, feet, teeth) never could dominate their noun vocabularies the way that irregular verbs, at least in theory, could dominate their verb vocabularies. Yet children show the same U-shaped development with plurals as with past-tense forms. When they begin to speak, all of their plurals are correct, and then they begin to overgeneralize at a low rate for several years
另外就是在这里家长的纠正可能并不是很重要
parental feedback cannot be crucial
8. The Horrors of the German Language
在上面几章我们都是以英语为中心进行讨论，在这个章节我们要从英语的近亲德语说起，然后也会讨论和英语相对比较遥远的阿拉伯语和汉语。
For many reasons, those mother tongues are a motherlode for the understanding of language and mind.
First, no one is biologically disposed to speak a particular language.Also, to understand language we have to test hypotheses about cause and effect, but linguists don’t have the luxury of synthesizing a language in a test tube and seeing how it is spoken, learned, and changed. The differences among languages already out there make up the only laboratory apparatus that allows a linguist to vary one factor and see how it affects another.
Finally, no one supposes that language evolved six thousand times.
德语以变形复杂著称。
Of the thousand commonest verbs in English, a majority, 86 percent, are regular, but of the thousand commonest verbs in German, a minority, only 45 percent, are regular.
the first seven chapters we explored the tracks and traces of a rule in action, but only for two suffixes in a single language. In this chapter we have spotted them in eight other languages ranging from the closest siblings of English to the most distant strangers
通过考察不同语言，我们发现并不是所有语言都像英语一样朝着规则的方向前进，规则变形和不规则变形是相互转化的。
As predicted, the subjects preferred irregular plurals for the roots, and the preference shrank when the roots didn’t rhyme with existing irregular nouns
A common misconception is that because Old English had more irregular verbs than Modern English, languages always evolve from irregular to regular. Languages don’t consistently evolve in either direction, because different psycholinguistic processes constantly create and destroy the two kinds of words or convert one into the other. These processes have made scattered appearances throughout the book;
9. The Black Box
这一个章节讨论了与语言和语法功能对应的神经科学。在过于，大脑对于语言学家来说就是一个黑箱，我们只能看到输入和输出而不知道其中工作的原理，通过现代的仪器分析我们可以看到语言功能的区域，甚至还有对应的基因位点。
If words and rules are the ingredients of language, we should be able to tell them apart in the brain. Parts of the brain that handle memory for words should be implicated in the use of irregular forms, and parts that handle rules should be implicated in the use of regular forms.
语言是复杂的，需要许多功能区域相互配合。
Anyone who has recently installed a software package on a personal computer knows that even a simple program needs hundreds of files scattered all over the machine to coordinate the program with memory, input, output, and other programs. And language has even more features than the latest bloat from the software industry
SPCH1, the first genetic region specifically linked to a speech and language disorder
electroencephalogram, or EEG
10. A Digital Mind in an Analog world
在本书的结尾，我们要讨论的是不规则变形的词汇的共性。如果从传统分类学角度，单独看他们似乎无迹可寻，但是如果从范畴学的家族相似的角度去看，可以找到他们的共同特点
a remarkable parallel between regular and irregular inflection and something completely different. The parallel cannot be a coincidence, and it hints that the distinction between regular and irregular forms may expose even deeper principles about the nature of the mind and how it reflects the world.1
family resemblance categories rather than classical categories
First, with most categories it is almost impossible to find a set of membership conditions.
Second, the members of a category are not created equal, which is what one would expect if they were admitted into the category by meeting the definition.
Third, the categories of the mind have fuzzy borders.
Fourth, most of our everyday categories, and not just games, show Wittgenstein’s family resemblance and crisscrossing features.
Fifth, categories have stereotyped features: traits that everyone associates with the category, even if they have nothing to do with the criteria for membership.
All five of their distinguishing traits can be found in the irregulars
First, despite the contortions of centuries of language scholars, no one has been able to craft a set of rules that properly pick out the different kinds of irregular verbs.
Second, in every irregular family some members are more equal than others.
Third, in the halo around the poor relations in an irregular family there are verbs so poor that no one knows whether they belong in the family at all
Fourth, the members of irregular families resemble each other in crisscrossing ways, rather than by sharing any trait
Fifth, irregular families have stereotyped features that run in the family but play no role in defining the past-tense form
也提到了Lakoff的著作《女人、火焰和危险的事物》
Women, Fire, and Dangerous Things (a family resemblance category in an Australian aboriginal language)
结语：
Many things we find around us could not be deduced by any body of laws, because they are shaped by myriad events of history no longer visible to us.
The human mind can think in idealizations, reducing an object to an austere description of the variables manipulated by the rule system, such as generation and gender in the case of kinship or the outcome of the electoral process in the case of law.