The Language Instinct: How the Mind Creates Language

by Steven Pinker

  “Would it be of any use, now,” thought Alice, “to speak to this mouse? Everything is so out-of-the-way down here, that I should think very likely it can talk: at any rate, there’s no harm in trying.” So she began. “O Mouse, do you know the way out of this pool? I am very tired of swimming about here, O Mouse!” (Alice thought this must be the right way of speaking to a mouse: she had never done such a thing before, but she remembered having seen, in her brother’s Latin Grammar, “A Mouse—of a mouse—to a mouse—a mouse—O mouse!”)

  English speakers tag a noun phrase with a case by seeing what the noun is adjacent to, generally a verb or preposition (but for Alice’s mouse, the archaic “vocative” case marker O). They use these case tags to match up each noun phrase with its verb-decreed role.

  The requirement that noun phrases must get case tags explains why certain sentences are impossible even though the super-rules admit them. For example, a direct object role-player has to come right after the verb, before any other role-player: one says Tell Mary that John is coming, not Tell that John is coming Mary. The reason is that the NP Mary cannot just float around tagless but must be casemarked, by sitting adjacent to the verb. Curiously, while verbs and prepositions can mark case on their adjacent NP’s, nouns and adjectives cannot: governor California and afraid the wolf, though interpretable, are ungrammatical. English demands that the meaningless preposition of precede the noun, as in governor of California and afraid of the wolf, for no reason other than to give it a case tag. The sentences we utter are kept under tight rein by verbs and prepositions—phrases cannot just show up anywhere they feel like in the VP but must have a job description and be wearing an identity badge at all times. Thus we cannot say things like Last night I slept bad dreams a hangover snoring no pajamas sheets were wrinkled, even though a listener could guess what that would mean. This marks a major difference between human languages and, for example, pidgins and the signing of chimpanzees, where any word can pretty much go anywhere.

  Now, what about the most important phrase of all, the sentence? If a noun phrase is a phrase built around a noun, and a verb phrase is a phrase built around a verb, what is a sentence built around?

  The critic Mary McCarthy once said of her rival Lillian Hellman, “Every word she writes is a lie, including ‘and’ and ‘the.’” The insult relies on the fact that a sentence is the smallest thing that can be either true or false; a single word cannot be either (so McCarthy is alleging that Hellman’s lying extends deeper than one would have thought possible). A sentence, then, must express some kind of meaning that does not clearly reside in its nouns and verbs but that embraces the entire combination and turns it into a proposition that can be true or false. Take, for example, the optimistic sentence The Red Sox will win the World Series. The word will does not apply to the Red Sox alone, nor to the World Series alone, nor to winning alone; it applies to an entire concept, the-Red-Sox-winning-the-World-Series. That concept is timeless and therefore truthless. It can refer equally well to some past glory, a hypothetical future one, even to the mere logical possibility, bereft of any hope that it will ever happen. But the word will pins the concept down to temporal coordinates, namely the stretch of time subsequent to the moment the sentence is uttered. If I declare “The Red Sox will win the World Series,” I can be right or wrong (probably wrong, alas).

  The word will is an example of an auxiliary, a word that expresses layers of meaning having to do with the truth of a proposition as the speaker conceives it. These layers also include negation (as in won’t and doesn’t), necessity (must), and possibility (might and can). Auxiliaries typically occur at the periphery of sentence trees, mirroring the fact that they assert something about the rest of the sentence taken as a whole. The auxiliary is the head of the sentence in exactly the same way that a noun is the head of the noun phrase. Since the auxiliary is also called INFL (for “inflection”), we can call the sentence an IP (an INFL phrase or auxiliary phrase). Its subject position is reserved for the subject of the entire sentence, reflecting the fact that a sentence is an assertion that some predicate (the VP) is true of its subject. Here, more or less, is what a sentence looks like in the current version of Chomsky’s theory:

  An auxiliary is an example of a “function word,” a different kind of word from nouns, verbs, and adjectives, the “content” words. Function words include articles (the, a, some), pronouns (he, she), the possessive marker’s, meaningless prepositions like of, words that introduce complements like that and to, and conjunctions like and and or. Function words are bits of crystallized grammar; they delineate larger phrases into which NP’s and VP’s and AP’s fit, thereby providing a scaffolding for the sentence. Accordingly, the mind treats function words differently from content words. People add new content words to the language all the time (like the noun fax, and the verb to snarf, meaning to retrieve a computer file), but the function words form a closed club that resists new members. That is why all the attempts to introduce gender-neutral pronouns like hesh and than have failed. Recall, too, that patients with damage to the language areas of the brain have more trouble with function words like or and be than with content words like oar and bee. When words are expensive, as in telegrams and headlines, writers tend to leave the function words out, hoping that the reader can reconstruct them from the order of the content words. But because function words are the most reliable clues to the phrase structure of the sentence, telegraphic language is always a gamble. A reporter once sent Cary Grant the telegram, “How old Cary Grant?” He replied, “Old Cary Grant fine.” Here are some headlines from a collection called Squad Helps Dog Bite Victim, put together by the staff of the Columbia Journalism Review:

  New Housing for Elderly Not Yet Dead

  New Missouri U. Chancellor Expects Little Sex

  12 on Their Way to Cruise Among Dead in Plane Crash

  N.J. Judge to Rule on Nude Beach

  Chou Remains Cremated

  Chinese Apeman Dated

  Hershey Bars Protest

  Reagan Wins on Budget, But More Lies Ahead

  Deer Kill 130,000

  Complaints About NBA Referees Growing Ugly

  Function words also capture much of what makes one language grammatically different from another. Though all languages have function words, the properties of the words differ in ways that can have large effects on the structure of the sentences in the language. We have already seen one example: overt case and agreement markers in Latin allow noun phrases to be scrambled; silent ones in English force them to remain in place. Function words capture the grammatical look and feel of a language, as in these passages that use a language’s function words but none of its content words:

  DER JAMMERWOCH

  Es brillig war. Die schlichte Toven

  Wirrten und wimmelten in Waben.

  LE JASEROQUE

  Il brilgue: les tôves lubricilleux

  Se gyrent en vrillant dans la guave.

  The effect can also be seen in passages that take the function words from one language but the content words from another, like the following pseudo-German notice that used to be posted in many university computing centers in the English-speaking world:

  ACHTUNG! ALLES LOOKENSPEEPERS!

  Das computermachine ist nicht fuer gefingerpoken und mittengrabben. 1st easy schnappen der springenwerk, blowenfusen und poppencorken mit spitzensparken. 1st nicht fuer gewerken bei das dumpkopfen. Das rubbernecken sightseeren keepen das cottenpickenen hans in das pockets muss; relaxen und watchen das blinkenlichten.

  Turnabout being fair play, computer operators in Germany have posted a translation into pseudo-English:

  ATTENTION

  This room is fulfilled mit special electronische equippment. Fingergrabbing and pressing the cnoeppkes from the computers is allowed for die experts only! So all the “lefthanders” stay away and do not disturben the brainstorming von here working intelligencies. Otherwise you will be out thrown and kicked andeswhere! Also: please keep still and only wat
chen astaunished the blinkenlights.

  Anyone who goes to cocktail parties knows that one of Chomsky’s main contributions to intellectual life is the concept of “deep structure,” together with the “transformations” that map it onto “surface structure.” When Chomsky introduced the terms in the behaviorist climate of the early 1960s, the reaction was sensational. Deep structure came to refer to everything that was hidden, profound, universal, or meaningful, and before long there was talk of the deep structure of visual perception, stories, myths, poems, paintings, musical compositions, and so on. Anticlimactically, I must now divulge that “deep structure” is a prosaic technical gadget in grammatical theory. It is not the meaning of a sentence, nor is it what is universal across all human languages. Though universal grammar and abstract phrase structures seem to be permanent features of grammatical theory, many linguists—including, in his most recent writings, Chomsky himself—think one can do without deep structure per se. To discourage all the hype incited by the word “deep,” linguists now usually refer to it as “d-structure.” The concept is actually quite simple.

  Recall that for a sentence to be well formed, the verb must get what it wants: all the roles listed in the verb’s dictionary entry must appear in their designated positions. But in many sentences, the verb does not seem to be getting what it wants. Remember that put requires a subject, an object, and a prepositional phrase; He put the car and He put in the garage sound incomplete. How, then, do we account for the following perfectly good sentences?

  The car was put in the garage.

  What did he put in the garage?

  Where did he put the car?

  In the first sentence, put seems to be doing fine without an object, which is out of character. Indeed, now it rejects one: The car was put the Toyota in the garage is awful. In the second sentence, put also appears in public objectless. In the third, its obligatory prepositional phrase is missing. Does this mean we need to add new dictionary entries for put, allowing it to appear in some places without its object or its prepositional phrase? Obviously not, or He put the car and He put in the garage would slip back in.

  In some sense, of course, the required phrases really are there—they’re just not where we expect them. In the first sentence, a passive construction, the NP the car, playing the role of “thing put” which ordinarily would be the object, shows up in the subject position instead. In the second sentence, a wh-question (that is, a question formed with who, what, where, when, or why), the “thing put” role is expressed by the word what and shows up at the beginning. In the third sentence, the “place” role also shows up at the beginning instead of after the object, where it ordinarily belongs.

  A simple way to account for the entire pattern is to say that every sentence has two phrase structures. The phrase structure we have been talking about so far, the one defined by the super-rules, is the deep structure. Deep structure is the interface between the mental dictionary and phrase structure. In the deep structure, all the role-players for put appear in their expected places. Then a transformational operation can “move” a phrase to a previously unfilled slot elsewhere in the tree. That is where we find the phrase in the actual sentence. This new tree is the surface structure (now called “s-structure,” because as a mere “surface” representation it never used to get proper respect). Here are the deep structure and surface structure of a passive sentence:

  In the deep structure on the left, the car is where the verb wanted it; in the surface structure on the right, it is where we actually hear it. In the surface structure, the position from which the phrase was moved contains an inaudible symbol that was left behind by the movement transformation, called a “trace.” The trace serves as a reminder of the role that the moved phrase is playing. It tells us that to find out what the car is doing in the putting event, we should look up the “object” slot in the entry for the verb put; that slot says “thing put.” Thanks to the trace, the surface structure contains the information needed to recover the meaning of the sentence; the original deep structure, which was used only to plug in the right sets of words from the lexicon, plays no role.

  Why do languages bother with separate deep structures and surface structures? Because it takes more than just keeping the verb happy—what deep structure does—to have a usable sentence. A given concept often has to play one kind of role, defined by the verb in the verb phrase, and simultaneously a separate role, independent of the verb, defined by some other layer of the tree. Consider the difference between Beavers build dams and its passive, Dams are built by beavers. Down in the verb phrase—the level of who did what to whom—the nouns are playing the same roles in both sentences. Beavers do the building, dams get built. But up at the sentence (IP) level—the level of subject-predicate relations, of what is being asserted to be true of what—they are playing different roles. The active sentence is saying something about beavers in general, and happens to be true; the passive sentence is saying something about dams in general, and happens to be false (since some dams, like the Grand Coulee Dam, are not built by beavers). The surface structure, which puts dams in the sentence’s subject position but links it to a trace of its original verb phrase position, allows the cake to be both eaten and had.

  The ability to move phrases around while still retaining their roles also gives the speaker of a rigid-word-order language like English a bit of wiggle room. For example, phrases that are ordinarily buried deep in the tree can be moved to early in the sentence, where they can hook up with material fresh in the listener’s mind. For example, if a play-by-play announcer has been describing Nevin Markwart’s progression down the ice, he could say Markwart spears Gretzky!!! But if it was Wayne Gretzky the announcer had been describing, he would say Gretzky is speared by Markwart!!!! Moreover, because a passive participle has the option of leaving the doer role, ordinarily the subject, unfilled in deep structure, it is useful when one wants to avoid mentioning that role altogether, as in Ronald Reagan’s evasive concession Mistakes were made.

  Hooking up players with different roles in different scenarios is something that grammar excels at. In a wh-question like

  What did he put [trace] in the garage?

  the noun phrase what gets to live a double life. Down in the who-did-what-to-whom realm of the verb phrase, the position of the trace indicates that the entity has the role of the thing being put; up in the what-is-being-asserted-of-what realm of the sentence, the word what indicates that the point of the sentence is to ask the listener to provide the identity of something. If a logician were to express the meaning behind the sentence, it would be something like “For which x, John put xin the garage.” When these movement operations are combined with other components of syntax, as in She was told by Bob to be examined by a doctor or Who did he say that Barry tried to convince to leave? or Tex is fun for anyone to tease, the components interact to determine the meaning of the sentence in chains of deduction as intricate and precise as the workings of a fine Swiss watch.

  Now that I have dissected syntax in front of you, I hope your reaction is more favorable than Eliza Doolittle’s or Jack Cade’s. At the very least I hope you are impressed at how syntax is a Darwinian “organ of extreme perfection and complication.” Syntax is complex, but the complexity is there for a reason. For our thoughts are surely even more complex, and we are limited by a mouth that can pronounce a single word at a time. Science has begun to crack the beautifully designed code that our brains use to convey complex thoughts as words and their orderings.

  The workings of syntax are important for another reason. Grammar offers a clear refutation of the empiricist doctrine that there is nothing in the mind that was not first in the senses. Traces, cases, X-bars, and the other paraphernalia of syntax are colorless, odorless, and tasteless, but they, or something like them, must be a part of our unconscious mental life. This should not be surprising to a thoughtful computer scientist. There is no way one can write a halfway intelligent program without defining variables and d
ata structures that do not directly correspond to anything in the input or output. For example, a graphics program that had to store an image of a triangle inside a circle would not store the actual keystrokes that the user typed to draw the shapes, because the same shapes could have been drawn in a different order or with a different device like a mouse or a light pen. Nor would it store the list of dots that have to be lit up to display the shapes on a video screen, because the user might later want to move the circle around and leave the triangle in place, or make the circle bigger or smaller, and one long list of dots would not allow the program to know which dots belong to the circle and which to the triangle. Instead, the shapes would be stored in some more abstract format (like the coordinates of a few defining points for each shape), a format that mirrors neither the inputs nor the outputs to the program but that can be translated to and from them when the need arises.

  Grammar, a form of mental software, must have evolved under similar design specifications. Though psychologists under the influence of empiricism often suggest that grammar mirrors commands to the speech muscles, melodies in speech sounds, or mental scripts for the ways that people and things tend to interact, I think all these suggestions miss the mark. Grammar is a protocol that has to interconnect the ear, the mouth, and the mind, three very different kinds of machine. It cannot be tailored to any of them but must have an abstract logic of its own.