Where do noun phrases come from?¹

Background

The contents of this paper is clearly transformational grammar but not so clearly generative grammar. In discussing English, I will be treating the English language not as a class of sentences but as a code which relates messages (semantic representations) to their encoded forms (surface structures). The possibility or impossibility of a given (surface form of a) sentence depends on two quite separate factors, namely the details of the code and the restrictions on possible messages: a surface structure is possible only if there is a message which the code pairs with it. This paper is transformational grammar in that I maintain that the code has roughly the form of the ‘transformational component’ of a grammar as discussed in Chomsky (1965), i.e. that the ‘encoding’ of a message can be regarded as involving a series of intermediate stages, each obtained by applying a transformation to the preceding stage. It is not obviously generative grammar, in that to a large extent I leave open the question of what a ‘ possible message ’ is.

There are several kinds of constraints on ‘ possible message ’. One kind of constraint relates to what might be called the ‘logical well-formedness’ of the message. For example, ‘ or ’ is a predicate which may be combined with two or more propositions. However, ‘ or ’ by itself is not a ‘ possible message ’, nor is ‘ or ’ plus a single proposition, nor is ‘ or ’ plus things which are not all propositions; the following loosely represented structures are thus not ‘ possible messages ’:

or

or (Max drinks vodka martinis)

or (China is industrializing rapidly; the Pope).

These constraints on ‘possible message’ correspond roughly to what Chomsky (1965) calls ‘strict subcategorization’.

In Chomsky (1965), another kind of constraint is discussed, namely ‘selectional restrictions ’, which are supposed to exclude sentences such as

(1) *That idea is green with orange stripes.

in which each predicate is combined with the right number of things but the sentence is odd because of ‘ incompatible ’ choices of lexical material. Chomsky treats selectional restrictions as idiosyncrasies of lexical items: each lexical item is assigned ‘ selectional features’, which express restrictions on what material it may be combined with in ‘ deep structures ’, e.g. the verb surprise might have the restriction that it ‘ requires an animate object ’, i.e. that it may only appear in a deep structure in which it is followed by a noun phrase² having the property ‘ animate’. It is not clear that such restrictions, to the extent that they are valid restrictions, have anything to do with deep structures and lexical items rather than with semantic representations and semantic items that appear in them. If it in fact turns out that the ‘selectional restrictions’ of all lexical items are predictable from their meanings, then they are not restrictions on how lexical items may be combined but rather restrictions on how semantic material may be combined, i.e. restrictions on ‘possible message’.³ It is worthwhile at this point to consider some examples which are occasionally cited as cases where a selectional restriction is not predictable from the meaning of the item in question, i.e. cases in which there are two or more words that allegedly have the same meaning but different selectional restrictions. I maintain that in each case the words actually do not have the same meaning. Consider, for example, the Japanese verbs kaburu, hameru, haku, etc., which one might gloss as ‘put on, said of headwear’, ‘put on, said of gloves’, ‘put on, said of footwear’, etc., thus suggesting that these verbs have the same meaning but different selectional restrictions. Such a description is incorrect, since the verbs in fact refer to the quite different actions involved in putting on the articles of clothes in question, as is demonstrated by the fact that if one puts on an article of clothing in an unnatural manner (e.g. puts a pair of socks on his hands, uses a necktie to hold up his trousers, etc.), the choice of verb is dictated not by the article of clothing but by the manner in which it is put on, e.g. covering one’s head with a pair of gloves would be described by kaburu rather than hameru. Similarly, one might propose defining the English verbs kick, slap, and punch as ‘ strike with the foot ’, ‘ strike with the open hand ’, and ‘ strike with the fist ’, suggesting that they have the meaning ‘strike’ with different selectional restrictions. However, J. R. Ross (personal communication) has observed that in the bizarre situation in which a person had been subjected to surgery in which his hands and feet were cut off and grafted onto his ankles and wrists respectively, it would be perfectly normal to speak of that person as kicking someone with his fist or slapping someone with his foot. This implies that the verbs refer to the specific motion⁴ which the organ in question performs and are thus not simply contextual variants of strike.⁵ For an excellent compendium of selectional restrictions, some of which are not so obviously predictable from meanings as are those just discussed, I refer the reader to Leisi (1966).

Many so-called selectional restrictions are actually not real restrictions, since ‘ violations ’ of them are quite normal in reports of dreams, reports of other people’s beliefs,⁶ and science-fiction stories:

(2) I dreamed that my toothbrush was pregnant.

(3) I dreamed that I poured my mother into an inkwell.

(4) I dreamed that I was a proton and fell in love with a shapely green-and-orange-striped electron.

(5) Max thinks that electrons are green with orange stripes.

(6) Harry thinks that his toothbrush is trying to kill him.

(7) Boris believes that ideas are physical objects and claims to have seen several that were green with orange stripes.

While some linguists might suggest that a person who says things like

(8) My toothbrush is alive and is trying to kill me.

observes different selectional restrictions than normal people do, it is pointless to do so, since the difference in ‘selectional restriction’ will correspond exactly to a difference in beliefs about one’s relationship with inanimate objects. A person who utters (8) should be referred to a psychiatric clinic, not to a remedial English course.

Note, however, that dreams, etc. are not completely free as to how semantic material may combine. For example, the constraint that the complement of the progressive be must be headed by an ‘activity verb’ may not be violated even in sentences such as

(9) *1 dreamed that Arthur was knowing the answer.

(10) *Max believes that Arthur is knowing the answer.

Similarly with the restriction that only a quantity of time can elapse:

(11) *1 dreamed that my toothbrush elapsed.

These constraints appear to be real constraints on ‘ possible message ’; it is evident that to enumerate these constraints one will have to tackle some of the classical problems of philosophy, namely the question of categories and the question of the distinction between ‘ essential ’ and ‘ accidental ’ properties of things.

¹ This paper is an extensively revised version of a paper of the same title published in Roderick Jacobs and Peter S. Rosenbaum (eds.), Readings in English Transformational Grammar (Blaisdell, 1970); part of its contents first appeared in McCawley (1967). Conclusions very similar to those presented here were arrived at by Postal independently of me and slightly earlier than I arrived at them (Postal, unpublished).

² Actually, Chomsky treats the condition as imposed not on the entire noun phrase but only on its head noun. The intenability of that proposal is demonstrated in McCawley (1968a), where I also observe that selectional restrictions cannot be regarded as requiring a noun phrase to possess a property such as ‘animate’ but only as excluding those noun phrases having semantic representations incompatible with that property (see also McCawley 1968 c).

³ The position that ‘category mistakes’, which may be the only anomalies that can be correctly described as ‘selectional violations’, do not correspond to ‘possible thoughts’ is ably defended in Drange (1966).

⁴ Dwain Parrack has called to my attention the fact that the meaning of these verbs includes not merely the motion of the organ but the type of surface contact which results. Parrack points out that slap requires contact with a more or less flat surface and that one could thus speak of the soles of someone’s feet slapping the surface of the water. Similarly, a certain motion of the arm would be a punch if the hand were clenched but a poke it it were open and only the fingertips were involved in the contact.

⁵ It is occasionally suggested (e.g. Bierwisch 1967:8) that there is a linguistically significant distinction between interpretations which are possible only by imagining some bizarre situation and interpretations which require no such effort of the imagination. However, it is not clear that this criterion really defines a classification of sentences. How easy one finds it to imagine a situation in which a given sentence would be appropriate depends on such extra-linguistic factors as his factual knowledge, the strength of his imagination, and the possible presence of LSD in his bloodstream. I suspect that the sentences which one can interpret without thinking up some story to embed them in are simply those which it is so easy to imagine someone’s using that it would require no effort to think up such a story.

⁶ Jakobson (1941, §§26-7) points out that many persons, especially children, associate colors with sounds and, for example, will not hesitate to say that the vowel [a] is red.