Noun phrases and semantic representation

The principal respect in which I find existing versions of symbolic logic insufficient for the representation of meaning has to do with noun phrases. Consider the sentence

(15)The man killed the woman.

 

If one accepts the position (expounded and defended in e.g. McCawley 1968 a) that each noun phrase occurrence in a syntactic representation must have attached to it an ‘index’, which corresponds to the ‘intended reference’ of that noun phrase occurrence, then the structure which underlies (15) will have to have some index X1 attached to the man and some index x2 attached to the woman. The meaning of (15) will then involve the assertion that x1 participated as agent and x2 as patient in a certain event y of killing, the assertion that y took place prior to the speech act, and the assertions that x± is a man and that x2 is a woman. One might propose that the semantic representation of (15) is obtained simply by conjoining all these assertions, with perhaps some additional terms to cover the meaning of the, which I have ignored:

(16) kill_y (x₁, x₂) Λ Past (y) Λ Man(x₁) A Woman(x₂).¹

 

However, (16) does not correctly represent the meaning of (15). Note that if one says

(17) I deny that the man killed the woman.

he is not denying (16). To deny a conjunction is to assert that at least one of the conjuncts is false. However, in (17) the speaker is not merely asserting that one of the four terms of (16) is false but is asserting specifically that the first term is false and assuming the other three terms to be true: it would not be correct to say (17) when one means that x₁ did in fact kill x₂ but that x₁ is not a man. Similarly, when one asks

(18) Did the man kill the woman?,

he is not asking whether the conjunction (16) is true (i.e. whether ail four terms are true) but is assuming the truth of the last three terms and asking whether the first term is true. It thus appears that in some sense the meanings of the expressions the man and the woman play a subordinate role in the meaning of (15).

 

To represent meaning correctly, symbolic logic will have to be supplemented by a way of representing this type of ‘ subordination ’. The fact that no such device has been used in symbolic logic so far is a result of the fact that symbolic logic has largely been used as a device for representing the content of propositions of mathematics. This ‘subordination’ relates to an important way in which the sentences of natural languages differ from mathematical propositions. In mathematics one enumerates certain objects which he will talk about, defines other objects in terms of these objects, and confines himself to a discussion of objects which have been either explicitly postulated or explicitly defined and which thus have been assigned explicit names; these names are in effect proper names. However, one does not begin a conversation by giving a list of postulates and definitions. One simply starts talking about whatever he intends to talk about, and the bulk of the things which he talks about will be things for which either there is no proper name (e.g. there is no proper noun Glarf meaning ‘ the nail on the third toe of Lyndon Johnson’s left foot ’) or the speaker does not know any proper name (e.g. an expression such as ‘ the little redhead that Max was talking to in the coffee shop yesterday ’ used to refer to someone whose name one does not know). Moreover, people often talk about things which either do not exist or they have identified incorrectly. Indices exist in the mind of the speaker rather than in the real world: they are conceptual entities which the individual creates in interpreting his experience.

 

Communication between different persons is possible because (i) different individuals often correctly identify things or make similar misidentifications, so that what one speaker says about an item in his mental picture of the universe will jibe with something in his hearer’s mental picture of the universe, and (2) the noun phrases which speakers use fulfill a function comparable to that of postulates and definitions in mathematics: they state properties which the speaker assumes to be possessed by the conceptual entities involved in what he is saying and are used chiefly to give the listener sufficient information to identify the things that the speaker is talking about.² I conclude that it is necessary for semantic representation to separate a clause into a ‘ proposition ’ and a set of noun phrases, which provide the material used in identifying the indices of the ‘ proposition ’, e.g.

 

That representations such as the above play a role in grammar is shown by an interesting class of ambiguities which appears to have escaped the notice of linguists until recently, although it has been discussed by philosophers since the middle ages as the distinction between de dicto and de re interpretation. The sentence

(19) Willy said that he has seen the woman who lives at 219 Main St.

is appropriate either to report Willy’s having said something such as ‘ I saw the woman who lives at 219 Main St ’ (the de dicto interpretation) or to report his having said something such as ‘ I saw Harriet Rabinowitz ’, where the speaker identifies Harriet Rabinowitz as ‘the woman who lives at 219 Main St’ (the de re interpretation). This ambiguity is brought out by the fact that the sentence can be continued in two ways, each of which allows only one of the two interpretations:

(20a) . . .but the woman he had in mind really lives in Pine St. [de dicto]

(20b) . . .but he doesn’t know that she lives there, [de re]

Similarly, while

(21) Boris says that he didn’t kiss the girl who he kissed.

might conceivably be a de dicto report of Boris’s having uttered a contradictory sentence such as ‘ I didn’t kiss the girl who I kissed ’, it is more likely to involve a sentence such as ‘I didn’t kiss Nancy’ reported de re by a person who is convinced that Boris really did kiss Nancy. Similarly with

(22) Harry admits that he kissed the girl who he kissed.

(23) Joe doesn’t know that your sister is your sister.

 

These facts indicate that in certain kinds of embedded sentences the lexical material relating to noun phrases in the embedded sentence may be semantically either a part of the embedded sentence or part of a higher sentence. The proposal of the last paragraph makes it possible for representations to show just such a distinction. For example, the two meanings of (19) can be represented as

 

In the first tree the woman who lives at 2ig Main St is part of what Willy allegedly said; in the second tree it is not. Distinctions relating to what sentence a noun phrase is a constituent of are also involved in ambiguous sentences such as

(24) Nancy wants to marry a Norwegian.

which may mean either that there is a Norwegian who Nancy wants to marry or that Nancy wants her future husband to be Norwegian, although she may not yet have found a Norwegian that she would want to marry. In the first case a Norwegian is a constituent of the main sentence, and in the second case it is a constituent of the sentence which is the underlying object of want. There is a similar ambiguity in

(25) John wants to find the man who killed Harry.

 

In cases of multiple embeddings, it is possible to get multiple ambiguities. For example, Bach (1968) points out that

(26) John says that Nancy wants to marry a Norwegian.

is ambiguous between the three senses (i) there is a person who John says Nancy wants to marry and who the speaker identifies as a Norwegian, (ii) John says that Nancy wants to marry a certain person who John identifies as a Norwegian, and (iii) John says that Nancy wants her future husband (whoever he might be) to be Norwegian. It is difficult to see how these three senses could be assigned different ‘ deep structures ’ unless those structures allowed noun phrases to occur separate from the propositions that they are involved in and to be constituents of sentences in which those propositions are embedded.³

Similarly, the ambiguity of ⁴

(27) John thinks he is smarter than he is.

between a sense in which John subscribes to the contradiction ‘I am smarter than I am ’ and the more normal sense which asserts that John is not as smart as he thinks he is, is an ambiguity between an underlying structure in which something such as the extent to which John is smart is part of the complement of think and one in which it is not. In the one case the extent to which John is smart will be John’s smartness as identified by John, in the other as identified by the speaker. The sentence

(28) I wonder why more men don’t beat their wives than do.

 

is ambiguous between (i) I wonder why the number of men who don’t beat their wives exceeds the number who do and (ii) I wonder why the number of men who beat their wives is as small as it is. In meaning (ii), the complement of wonder is the question ‘Why isn’t the number of men who beat their wives greater than n? ’ and the lexical material corresponding to n (The number of men who heat their wives or something such) is an adjunct to the whole sentence rather than part of the complement of wonder.

 

The hypothesis that English sentences are derived from semantic representations of the form proposed above entails the conclusion that English has a transformation which attaches each noun phrase to an occurrence of the corresponding index. In the proposed underlying structure of (19), this transformation would attach the woman who lives at 219 Main St to the single occurrence of x₂ and Willy to the first of the two occurrences of x₁. In this case only the first occurrence of x₁ is a possible place to attach Willy: note that in

(29) He said that Willy had seen the woman who lives at 219 Main St.

the he may not refer to Willy. However, in some cases it is possible to attach a noun phrase to any of several occurrences of the index in question, so that there are alternate surface forms such as

(30) After John left his apartment, he went to the pool hall.

(31) After he left his apartment, John went to the pool hall.

 

Since the occurrences of an index to which no full NP is attached are realized as pronouns, no pronominalization transformation as such is needed.⁵ The important constraint on pronominalization noted by Ross (1967 a) and Langacker (in press) must thus be reformulated as a constraint on the NP-attachment transformation.⁶ Ross and Langacker, working from the assumption that pronouns are derived by a transformation which replaces one of two identical noun phrases by a pronoun, concluded that a noun phrase may trigger the pronominalization of another noun phrase either if it precedes it or if it follows it and is in a sentence which the other noun phrase is in a clause subordinate to. Thus, he may refer to John in

(32) John went to the pool room after he left his apartment.

but not in

(33) He went to the pool room after John left his apartment.

 

However, in (31), where he is in a clause subordinate to that containing John, it may refer to John. The effect of this constraint can be imposed on the NP-attachment transformation by saying that a noun phrase may be attached to any occurrence of the corresponding index which either precedes or is in a ‘ higher ’ sentence than all other occurrences of that index.

 

The treatment of pronominalization which I have just proposed is supported by sentences of a type first investigated by Emmon Bach and P. S. Peters and also discovered independently by William Woods and Susumu Kuno:

(34)A boy who saw her kissed a girl who knew him.

 

Here her is to be interpreted as referring to the girl mentioned in the sentence and him referring to the boy. Under the conception of pronominalization which derives a pronoun from a copy of its antecedent, her would have to come from a copy of a girl who knew a boy who saw her, which would in turn have to come from a copy of a girl who knew a boy who saw a girl who knew him, etc., and each of the two noun phrases would have to be derived from an infinitely deep pile of relative clauses. However, under the conception of pronominalization which I propose, this anomaly would vanish. (34) would be derivable from a structure roughly representable as The attachment of noun phrases to index occurrences takes place sequentially.

 

The process may begin with either X₁ or X₂. What results under the Proposition node will be respectively

(35) A boy who saw x₂ kissed x₂.

(36) X₁ kissed a girl who knew x₁.

 

In (35), both occurrences of x₂ are possible places for the attachment of the remaining noun phrase; attaching it to the first occurrence of x₂ yields

(37) A boy who saw a girl who knew x₁ ( = him) kissed x₂ ( = her).

and attaching it to the second occurrence of x₂ yields (34). In (36), only the first occurrence of x₁ meets the constraint formulated above, and attaching the noun phrase there yields (34). I call the reader’s attention to the fact that there are thus two derivations which convert the tree given above into the surface structure of (34), which may or may not be a defect of this account of pronominalization.

 

¹ If p and q are propositions, then p Λ q is the proposition that p and q are both true. One major respect in which (16) fails to be an adequate semantic representation is that it fails to represent the semantic structure underlying kill. Details of the correct representation are given in McCawley (1968 b).

² See Donnellan (1966) for some highly insightful observations on this use of NPs.

³ A valiant attempt at a description of these sentences is given in Bach (1968).

⁴ I am grateful to Charles J. Fillmore for calling this sentence to my attention.

⁵ This is an oversimplification. While ordinary personal pronouns need not be derived by a pronominalization transformation such as that of Ross (1967 a) or Langacker (1969), sentence pronominalizations in examples such as

Marvin said that there was a unicorn in the garden, and so there was.

Margaret is rumored to have been arrested, but I don’t believe it.

must in fact be derived through a transformation that replaces one of two identical structures by a pronoun.

⁶ One promising alternative is suggested in Lakoff (1968): that the ‘antecedent of’ relation is marked in surface structures (in addition to the information usually considered to be present in surface structures) and that the Ross-Langacker constraint is an output constraint (in the sense of Ross 1967 b) on the antecedent relation.

The treatment of personal pronouns that I proposed above eliminates one important anomaly that pronominalization appeared to have in earlier treatments: it was the only exception to the principle (Ross 1967b 1340) that ‘feature-changing’ rules are ‘upper bounded’, i.e. that an element may trigger feature changes only in elements of the same clause or of a ‘lower’ clause. However, this anomaly is eliminated only at the expense of introducing a new anomaly, since NP-attachment would violate Ross’s ‘coordinate structure constraint’, the principle that material may not be moved into or out of a coordinate structure; NP-attachment has to move material into a coordinate structure in cases such as

The girl you like and her brother’s roommate have just eloped.

مواضيع ذات صلة

Some lexical structures and their empirical validity Conclusion

Metaphor and the form of lexical entry

Systematic and accidental lexical gaps explained by lexical hierarchies

Hierarchies are non-convergent

COMPARATIVES

FEATURE ASSIMILATION

SELECTIONAL RESTRICTIONS

The implications of lexical hierarchies for some ‘syntactic’ phenomena

Hierarchies among lexical items

A hierarchy of semantic features

Binary features

Some lexical structures and their empirical validity