1    Foundational issues

Universal Grammar

According to behaviorism, the dominant paradigm in psychology in the 1940s and 50s, learning consists of associating a stimulus with a response. The classic example of such learning is that exhibited by Pavlov's dogs.

In the late 1950s and 60s, behaviorist learning theory was confronted with some serious and ultimately fatal challenges. For instance, the psychologist John Garcia and others performed a number of experiments in the classic behaviorist mold whose results nevertheless contradicted fundamental behaviorist assumptions. Even more relevantly to our concerns in this course, the linguist Noam Chomsky

called attention to two fundamental facts about language. First, virtually every sentence that a person utters or understands is a brand-new combination of words, appearing for the first time in the history of the universe. Therefore a language cannot be a repertoire of responses; the brain must contain a recipe or program that can build an unlimited set of sentences out of a finite list of words. That program may be called a mental grammar (not to be confused with pedagogical or stylistic "grammars," which are just guides to the etiquette of written prose). The second fundamental fact is that children develop these complex grammars rapidly and without formal instruction and grow up to give consistent interpretations to novel sentence constructions that they have never before encountered. Therefore, he argued, children must innately be equipped with a plan common to the grammars of all languages, a Universal Grammar, that tells them how to distill the syntactic patterns out of the speech of their parents. (Pinker 1994:22)

The proper way to think of Universal Grammar is as an innate mental organ specific to the species Homo sapiens. Just as our eyes and ears allow us to perceive and interpret electromagnetic radiation and vibration within certain bandwidths, so does the human language faculty (= Universal Grammar) allow us to perceive and interpret information governed by certain formal constraints. Exactly what is meant by 'formal constraints' will be explicated in the course of the term.

First evidence for syntactic structure

It is worth emphasizing from the outset that the information that the human language faculty perceives and processes is quite abstract. In particular, before being trained in syntax, people tend to think of the words in a phrase or sentence as being arranged in a linear order in time much as beads on a string can be arranged in a linear order in space. It is true that this one-dimensional arrangement of linguistic elements is important for certain subdisciplines of linguistics. For instance, according to a well-known generalization concerning discourse, 'old' information precedes 'new' information. But time is not the only dimension that is of importance in human language. Instead, the study of syntax clearly reveals a second dimension: hierarchical structure. The central role that hierarchical structure plays in human language is sometimes referred to as structure dependence.

A constraint on the interpretation of pronouns

Structure dependence can be illustrated with a simple fact from English. Consider (1).

(1) a.   Zelda's brother helped her.
b. Zelda helped her.

In both of the sentences in (1), the pronoun her is referentially dependent on some antecedent in the discourse.

Note: The term 'antecedent' is potentially misleading. Etymologically, it derives from Latin antecedens 'one who walks before'. An antecedent might therefore reasonably be expected to precede a referentially dependent expression. However, linear precedence is neither a necessary nor a sufficient condition for antecedenthood (see (4) below). A less misleading term might be 'referential anchor.' We will nevertheless continue to use 'antecedent' because it is the standard term in the literature.

But the sentences in (1) also differ in a way that you may never have thought of, but that is perfectly obvious once it is pointed out. In (1a), the antecedent of her can be either Zelda or some other woman. In (1b), on the other hand, the antecedent of her cannot be Zelda, but must be some other woman. If we specify the referent of a noun phrase, whether a pronoun or a full noun phrase (= nonpronominal noun phrase), by means of an index, we can conveniently represent this state of affairs as in (2). As the contrast in (2b) shows, a sentence's grammaticality is always determined relative to an intended interpretation.

Note: It is standard practice to use the letters of the alphabet as indices, beginning with i. However, later on, we will use i, j, k, … for other purposes, and we therefore press the natural numbers into service here.

(2) a. i. [ Zelda's ]1 brother helped [ her ]1.
ii. [ Zelda's ]1 brother helped [ her ]2.
b. i. * [ Zelda ]1 helped [ her ]1.
ii. [ Zelda ]1 helped [ her ]2.

The facts in (2) are usually represented even more concisely as in (3). The order of the subscripts in (3b) is intended to represent the scope of the asterisk unambiguously.

(3) a.   [ Zelda's ]1 brother helped [ her ]1,2.
b. [ Zelda ]1 helped [ her ]2,*1.

Linear order?

What is it that gives rise to the difference in available interpretations in (1)? That is, what allows Zelda to serve as the antecedent for her in (1a), but not in (1b)?

It is clear that the semantic contrast in (1) cannot be due to linear order, because the (potential) antecedent Zelda precedes the referentially dependent pronoun her in both cases. Thus, linear precedence is not a sufficient condition for antecedenthood. Neither is linear precedence a necessary condition for antecedenthood. This is evident from the sentences in (4), which differ minimally from those in (1) in that the pronoun and the full noun phrase are switched.

(4) a.   [ Her ]1 brother helped [ Zelda ]1,2.
b. [ She ]1 helped [ Zelda ]2,*1.

It is worth noting that for some speakers, so-called backwards pronominalization (or cataphora, to use the traditional term), where the referentially dependent expression precedes the semantic antecedent can be quite awkward in examples like (4a). But even for such speakers, there remains a contrast in (4), and they, like all other speakers of English, cheerfully accept coreference in examples like (5). Linear precedence is not, therefore, a necessary condition for semantic antecedenthood.

(5) a.   After [ she ]1 had eaten dinner, [ Zelda ]1 went to the movies.
b. After [ Zelda ]1 had eaten dinner, [ she ]1 went to the movies.

The conclusions from this discussion are important enough to reiterate. The availability of Zelda as an antecedent for her in (3a) and (4a), but not in (3b) or (4b), shows that linear order is irrelevant in explaining the semantic contrasts among these sentences.


Could it be the distance between the referentially dependent expression and its antecedent that is relevant? After all, there are two words intervening between the two expressions in the (a) examples in (3) and (4), but only one in the (b) examples. Perhaps replacing the single word in the (b) examples by two words would allow the pronoun to refer to Zelda? But the facts in (6) force us to reject this hypothesis.

(6) a.   [ Zelda ]1 never helped [ her ]2,*1.
b. [ She ]1 never helped [ Zelda ]2,*1.

(7) and (8) provide further evidence that distance is not a relevant factor in explaining when an expression can be an antecedent.

(7) a.   [ Zelda ]1 certainly never helped [ her ]2,*1.
b. [ She ]1 certainly never helped [ Zelda ]2,*1.
(8) a.   [ Zelda ]1 almost certainly never helped [ her ]2,*1.
b. [ She ]1 almost certainly never helped [ Zelda ]2,*1.

The c-command relation

In order to solve the puzzle presented by (3) and (4), it is necessary to appeal to hierarchical structure. In particular, we must recognize that not only are the phrases Zelda's/her brother longer than Zelda/she, but that they also have a different structure. This is shown in (9).

(9) a.       b.  

In (9a), there are two levels of abstract structure above the word level: NP, a phrasal node (= element of syntactic structure) for the entire noun phrase, and N, a node for the noun that heads the phrase (= forms its grammatical core). At the same hierarchical level as the head is a further node, labelled PossNP (= possessive noun phrase). This gives the topmost NP node in (9a) a branching structure. Let us turn now to (9b). Since Zelda/she has much the same distribution as Zelda's/her brother (that is, it can occupy much the same slots in a sentence), it stands to reason that even single words can be noun phrases. If we make the further reasonable assumption that all noun phrases, even ones consisting of single words, have heads, then (9b) has two levels of abstract structure above the word level, just like (9a). However, since the entire noun phrase consists of a single word, the NP node in (9b) is nonbranching, unlike the branching NP node in (9a).

Notice that in so-called tree structures like (9), the two logical dimensions of linear order and hierarchical structure are represented in space along the horizontal and vertical axes, respectively.

A structural description of the constraint

We will now show that it is the difference in structure just described that lies at the root of the semantic contrast between the (a) and (b) sentences in (3) and (4). We begin with (3b), which we represent as in (10).


Let us focus on the relationship between the NP node that dominates (= contains) Zelda (the subject NP) and the one that dominates her (the object NP).

Note: Be sure to focus not on the actual words, but on the NP nodes that dominate the words. The reason is that it is entire noun phrases that refer, not simply their head nouns. For instance, when we answer questions like What did you buy? with a sentence fragment, the sentence fragment must be a full noun phrase (say, a new laptop), not simply a head noun (say, laptop).
Observe that the node that dominates the subject NP (namely, 'Sentence') also dominates the object NP. We can exploit this fact to define a complex structural relation between the subject and the object called c-command. The definition of this relation is given in (11).

(11)     A c-commands B iff (= if and only if) the first branching node that dominates A also dominates B.

Note: The notion of c-command is defined in terms of dominance. The definition makes no mention of linear precedence. Beginners in syntax routinely, but mistakenly, assume that c-command logically implies linear precedence, or vice versa. Assignment 1, Exercise 4 addresses this misconception.

Since Zelda and her cannot be interpreted as referring to the same person in (3b), let us hazard the hypothesis that pronouns like her are allergic to being c-commanded by a noun phrase bearing the same index. Using the terminology defined in (12), we can restate this hypothesis more concisely as in (13). (Further on in the course, we will see that (13) needs to be revised, but for present purposes, the formulation is adequate.)

(12) a.   A and B are coindexed iff they bear the same index.
b. A binds B iff
i. A c-commands B, and
ii. A and B are coindexed.
c. A is free iff it is not bound.
(13) Pronouns must be free.

Note: In accordance with the next to last note, remember to interpret 'pronoun' as 'NP node dominating a pronoun.'

Having formulated the hypothesis in (13) on the basis of (3b), we must now test it against (3a), which has the structure in (14).


In order for (14) to be consistent with (13), the object NP must be free. That is, no noun phrase in the sentence can both c-command the object and be coindexed with it. We know from the interpretation of the sentence that the possessive NP (the PossNP node dominating Zelda's) and the object NP may be coindexed, so the crucial question is whether the possessive NP c-commands the object NP. Since the first branching node dominating the possessive NP (namely, the subject NP) does not dominate the object NP, the answer is 'no.' In the absence of a c-command relation between the possessive NP and the object NP, the condition in (13) is satisfied regardless of whether the two NPs are coindexed. This is what allows the referential dependence of her on Zelda's in (3a).

On the basis of the discussion so far, you are now ready to take a stab at explaining the facts in (4) and related facts (see Assignment 1, Exercises 1 and 2).