We can think of such situations as involving double instantiation. A particular sentential argument is instantiated in an ordinary syntactic position with a semantically empty constituent (in all cases the word "it") and it is also instantiated later in the sentence, in the form of an expression which gives its semantics.
The term extraposition is due to the Danish Anglicist J. Otto Jespersen, but the construction we will be examining is only one of the types of phenomena Jespersen had in mind when he coined the term, and moreover our view of that sub-type differs somewhat from his. Jespersen's definition (1933/1964, p. 95) is:
To Jespersen, this category of expressions included anything in which it was possible to isolate a complete sentence, to find either before or after this complete sentence an expression which was not syntactically a part of it, and to find in the sentence to which this expression was adjoined, a pronoun corresponding to that expression or part of it. '
The examples accompanying Jespersen's definition included those in . (We have added bracketing around the extraposed elements, and we represent the coreferential constituents in italics.)
In his discussion of the function of this pronoun he says (1933/1964, pp. 154f):
The collection of examples that Jespersen offers at this point includes those given in . (In these examples, we have bracketed the extraposed elements.)
The construction which is the topic of this chapter belongs to the latter group, but with a few qualifications. First, we cannot insist that what remains, when the extraposed element is taken out of consideration, has to be a potential complete sentence. In reality, neither did Jespersen, since the "It occurred to me" and "It seems to me" of examples 4-5 are not complete sentences. Secondly, we will restrict our attention to cases where the extraposed element is a verbal or clausal constituent, giving only slight attention to examples like 7, where the extraposed constituent is "the number of mistakes he always makes." Thirdly, we will not require ourselves to follow Jespersen's intention of situating the extraposed constituents outside the boundaries of a sentence, but will see it instead as a constituent in the phrasal VP. And lastly, we will not identify Jespersen's description of the "it" as "representing" the extraposed element with the way a pronoun is linked to its antecedent. It will suit our purposes equally well to assume that the word "it" merely serves the role of occupying the grammatical slot in the sentence which the valence element it stands for might have occupied. In our terms, this means that we will not require the description of the lexeme "it" in these contexts to have sem features which unify with the sem features of the extraposed constituent.
In offering our description of IT-Extraposition, we will need to sort out what kinds of constituents occur in extraposition, what grammatical functions are involved, and what constructional or lexical constraints there are on its optional or obligatory occurrence. We will later consider what other constructions it interacts with, and we will discuss the question of whether our treatment produces the correct constituent structure for these sentences.
But first, our discussion will seek to put the IT-Extraposition construction in perspective, situating it within two other families of constructions in English: (i) those involving non-anaphoric or "empty" uses of the pronoun "it", and (ii) those in which a valence element is instantiated as more than one constituent in its sentence.
Sentences about "The Current Local Situation".
These are the cases in which the pronoun "it" is the subject of sentences whose content is about the weather, the atmosphere, or the time. Examples:
We will not at this time participate in the discussion of whether the subjects of these sentences are merely empty fillers of the subject slot, serving only to satisfy the requirement in the grammar of English that every sentence has to have a subject, or whether "it" in these sentences is best understood as referring to something like 'the current situation'.
There are sentences in which a given argument is, so to speak, lifted out for special attention and followed by what looks like (but isn't) a relative clause whose missing element ("gap") is construed as the specially fronted element. These "cleft sentences" have various functions, as can be noticed from the examples in  but which will not be discussed here now.
Idiomatic Uses of IT.
These are sentences in which the "it" is idiomatically associated with some verbal expression in which it does not obviously refer to anything at all.
To the idiomatic uses involving specific lexical collocations we might add uses of "it" in connection with some authoritative text (the Bible, a dictionary, the textbook for a school course, etc.) in a setting in which the appropriateness of appealing to this text is taken for granted. These expressions always have the function of relating some proposition or imperative to the authority in question.
Constructions with Multiple-Constituent Realizations of Single Valence Elements
IT-Extraposition belongs to a class of constructions in which single valence elements are realized as two different constituents of the actual sentence. There are many other such constructions, some of which have already been adduced.
Left Dislocation and Right Dislocation.
The modern names for the Jespersen examples given in , repeated here, are Left Dislocation and Right Dislocation.
In the first of these we have an example of Left Dislocation: the name "Charles Dickens" is introduced first, and then a complete sentence, containing a pronoun referring to Charles Dickens, follows. In the second the phrase interpreting the pronoun "he" is presented at the end of the sentence, in Right Dislocation. In the typical cases of Right Dislocation, the "dislocated" constituent is pronounced with low pitch and its force is often described as that of an "afterthought".
In Left Dislocation we find that the first constituent often has a kind of "topic-introducing" function. Contrasting with Topicalization structures, Left-Dislocation structures are often used when "true" topicalization is impossible or awkward. In Topicalization, one of the constituents of the sentence is placed in left- most position in the sentence rather than where it might otherwise be expected. (Topicalization is one of the "Left Isolation" constructions discussed earlier.) But in Left Dislocation, a corresponding pronoun appears in the appropriate place inside the sentence. Since a subject always comes in initial position in a simple assertive sentence, topicalization of subjects would not affect the sentence's word order: sentence 1 avoids that unclarity. And there can be no topicalized equivalent to sentence , since English offers no way of topicalizing a possessor in a possessive construction.
Relative Clause Extraposition.
In a number of contexts it is possible to find a relative clause, separated from its head noun, at the end of a sentence. Examples:
Noun Complement Extraposition.
In many cases, the complement of a noun is found at the end of the sentence. Examples:
In 1 the multiply-realized element is "The suggestion ... that the director should be fired"; in 2 it is "the decision ... to eliminate overtime pay"; and in 3 it is "A brilliant review ... of Chomsky's latest book."
Adjective Complement Extraposition.
Complements of adjectives can also be extraposed in certain contexts. [footnote]
In 1, the multiply realized element is "aware ... of the problems you've caused me"; in 2 it is "willing...to help with family chores".
In all of the examples just surveyed, a valence element is instantiated in the form of two separate sentence constituents. Instances of such discontinuities are not limited to the instantiation of valence elements. There are also cases of adjective phrases that straddle the nouns they modify, and adverb phrases that straddle the adjectives they modify. The former situation is illustrated in example 1, the latter in 2. (Obviously we're not going to have time to treat this problem this semester.)
IT- Extraposition is also obligatory in cases for which the expected position of the element is that of a prepositional object, as is shown in 1,2.
Since some IT-Extraposition structures resemble other constructions using "it", it will be useful to find criteria for determining when we're looking at a true instance of IT-Extraposition. The main question to ask is whether the putative extraposed constituent has a semantic role in the sentence that could be associated with the syntactic position occupied by the word "it". Except for contexts in which extraposition is obligatory, one way of deciding whether something is or is not an instance of IT-Extraposition is by asking whether the "it" could be replaced by the extraposed constituent, preserving meaning. Compare the sentences in  and .
The substitutability test shows that the "that"-clause in  is the sole argument of "wonderful" (by virtue of being the subject of "was wonderful"), but no such argument can be made for sentence (1) in [xx]. Since the verb "put" has an obligatory locative complement, it is clear that "over there" has to be taken as an argument of the "put" clause; hence 1 has to be seen as an instance of the Cleft Sentence construction.
There is a variety of IT-Extraposition that we decided (in discussing Jespersen's example 7 not to include in the scope of our examination just now, but which is a part of the full story of this construction, namely that in which the extraposed element is not, strictly speaking, a clause, and for which the distinction between IT- Extraposition and Right Dislocation becomes blurred. Compare the sentences in .
We would see 2 as clearly an instance of Right Dislocation, because the plural pronoun ("they") agrees in number with the extraposed nominal expression ("the things children say"). Sentence 1, however, does indeed look like a case of IT- Extraposition. We might note, first of all, that many ordinary-looking NPs in extraposition are paraphrasable by expressions containing interrogative pronouns. Compare the following:
Now if we try this sort of paraphrasability test on the sentences in , we get the following results. Since "what things children say" is not a plural NP, 1 cannot stand as a well-formed Left Dislocation sentence.
Note now that 3 does not really mean 1, but rather 2.
This semantic property of sentences of this type is perhaps easiest to observe in its absence:
Example 2 is impossible because a fact cannot be blue and the main predicate in this construction has a fact as its argument, not a dress. In these cases the extraposed noun phrase does not directly express the argument whose normal syntactic position is occupied by "it". We conclude that this variety of extraposition, although it supplies "it" in an argument position and places later a constituent which leads to the construal of the meaning of the argument replaced by "it", is a little different from the type of IT-extraposition under study here.
"IT" as Subject
"IT" as Object
"IT" as Prepositional Object
The examples are sorted according to two criteria: (i) the syntactic position occupied by the direct argument (subj, obj, pobj) or the word"it" in the corresponding extraposition case and (ii) the syntactic form of the extraposed (or non-extraposed) constituent. It appears from this survey that, as a rough generalization, IT- extraposition of a propositional (clausal or infinitival) complement is possible whenever the predicator permits that form of syntactic expression of the argument (as subject, object or prepositional object). Moreover, extraposition of such complements is sometimes obligatory, as we have already noted and will be discuss further below.
The verb "hope" has the characteristic that its complement can have a future meaning with present-tense inflection. Thus:
The passive of "hope", with IT-Extraposition, however, does not have this feature. Thus:
There are examples with gerundial clauses or VPs that look like extraposition but are probably better classed as right dislocation. Illustrations of such sentences were omitted from the display but are given now.
In a right dislocation structure, the "it" and the final clause are truly coreferential and the final clause represents a repeat of information already present in the discourse context. (The "it" is, thus ‘cataphoric’, i.e., forward referring.) Note in examples  the comma, indicating pause, and the fact that the clausal part is most naturally pronounced with the low level tone that is indicative of repeated material. Note further that the final clauses contain anaphoric elements ("like that", "that speech"), which further suggests that information already present in the discourse context is being expressed by these clauses. These are all clues suggesting a right dislocation analysis in preference to an extraposition analysis.
In the case of "that"-clauses, there could also be much to say about the conditions under which the marker "that" is obligatory versus optional, as can be seen in the following examples.
It appears from the data display that the positions that can readily host the "it" of IT-Extraposition are subject, direct object, and prepositional object. These are discussed in turn.
Subject. In all contexts permitting subjects to be occupied by any of the permitted clause types, IT-Extraposition is permitted as well. It appears, moreover, that all ordinary intransitive verbs that take that-clause arguments require IT-Extraposition. In a great many cases, these same verbs allow infinitival complements with Subject Coinstantiation (Raising) (see the examples in  and ). This is true, for example, of "seem", "appear", "happen", etc. Compare:
There is a small number of intransitive verbs taking "that"- clause complements which do not require extraposition. (These also happen not to be Raising verbs.) Examples are "stink" and "suck"; these may well be the only examples.
There is also a class of adjectives that allows all three possibilities: clausal arguments, IT-Extraposition, and Raising. This class includes "likely", "certain", and many others.
We see, then, that a number of adjectives and intransitive verbs have at least two, and in come cases three, possibilities for expressing in their fully specified valences a single propositional argument. Later on we will look at the interaction of the verb "seem" and the adjective "likely", since interactions of these two types of predicators will require us to make a distinction between cases of IT- Extraposition which are built into the valence structure of particular words and a general-purpose linking rule which can unify with a wide range of lexical items.
Objects of Null Prepositions.) The occurrence of "it" as prepositional object appears to be restricted to a listable number of fixed verb-preposition collocations. In each case we are dealing with what we will describe as a null (semantically empty) preposition. Some examples follow:
The fact that extraposition in these cases is obligatory follows from the fact that prepositions do not take as objects the kinds of syntactic objects (clause or V[to]) which figure in the extraposition alternation. For example, the most common syntactic type of prepositional object is a noun phrase, and IT-extraposition does not apply straightforwardly to noun phrases.
Direct Objects. In the case of IT-Extraposition involving subjects and prepositional objects, there are some simple generalizations to make. For example, IT-Extraposition is always possible in subject position, and it is always required in prepositional object position.
When it comes to direct objects the relevant generalizations are bit more complicated. First, we need to distinguish simple transitive verbs and verbs with more complex valence requirements. Second, we need to distinguish cases of IT-Extraposition involving direct objects in which Secondary Predication figures, and cases in which it does not. It will also be important to notice what categories of constituent the verb allows as direct object: if nominal objects are not allowed IT-extraposition is impossible and if clausal objects are not allowed IT-extraposition is required.
We notice that certain semantically similar simple transitive verbs can differ with respect to their toleration of IT-Extraposition. Compare "think" with "believe" in the following examples.
What we see is that the verb "think" requires a clausal complement, but the verb "believe" does not, allowing extraposition as an option. One part of the explanation has to be that "believe", unlike "think", allows NP objects in general. That is, we find sentences like those in  but not like those in [xx].
Transitive verbs can differ from each other with respect to their ability to take NP vs. clausal objects. It should be pointed out that the restriction in the case of verbs like "think" involves not the content argument as such, but only the [gf obj] realization of that argument. As with other transitive verbs, the verb "think" is available for the passive valence. When the content argument is the subject of a passive sentence, IT-Extraposition is not only possible but obligatory. Notice example .
In the case of the complex-valence transitives, we need to distinguish those in which the third valence element is lexically specified from those in which it is merely functionally specified. Consider first the examples in  and .
The point of this comparison is that "take NP for granted" is a lexically "filled" VP idiom, whereas an expression of the form"consider NP Pred" is a reflection of the fact that the verb "consider" requires a patient (or 'topic') argument expressed as a NP and a secondary predicate whose subject is controlled by the topic NP. The difference is that, in the latter case, the sequence "NP Pred" will also characterize simple copular sentences of the form "NP is/are Pred". Compare the following:
(In the case of "take X for granted", a ‘heavy’ X can be moved to the end of the VP, as in 1. We might (erroneously) then regard 2 as an instance of an ordinary "that"-clause object which shows up at the end of the VP by ordinary ‘heavy shift’ ordering principles. In 3 we have a NP object in post-verbal position. From 4-5, however, we have to conclude that the idiom "take for granted" prohibits a clausal object in immediate post-verbal position.
That is, in the case of "take for granted", clausal objects are either "heavy shifted", as in 2, or extraposed, as in 5, but are not allowed to appear right after the verb, as in 4.)
Among the intransitive verbs having such dual valence possibilities are "seem" and "happen"; among adjectives sharing such dual possibilities are "likely" and "certain". The two types are illustrated in examples  and  [fn].
In describing the double use of such predicators we will distinguish the "Finite Clause Complement" or FC version of the word's valence (as shown in the a-b examples of [xx] and [xx]), and the Infinitival Complement" or IC version (as in the c examples of [xx] and [xx]). Of interest in the chain of reasoning we are about to engage in is the fact that the grammaticality of IT-Extraposition varies with different structural configurations of FC and IC predicators.
(A) We observe (in ) that simple sentences with verbs like "seem" in their FC version cannot have the clause as their subject. Can we then be satisfied with the simple generalization that the verb "seem" does not take a clausal subject? No, because of sentences like .
(B) We observe (in 1-2) that simple sentences with adjectives like "likely" in their FC version allow either "that"-clauses as subjects or the "it" of an Extraposition structure. But even though both "seem" and "likely" welcome "that"-clause complements, sentence 1 is ungrammatical.
What we need to notice in these examples is that in structures in which predicators taking FC arguments are embedded as complements to IC predicators, it is the valence possibilities of the complement ("downstairs") predicator which determines the acceptability or unacceptability of a clausal subject for the governing ("upstairs") verb. In particular, it is the fact that "seem" does not allow in its minimal valence for a simple "that"-clause expression of its single propositional argument (as shown by 1) which accounts for the unacceptability of 1 as well. Whatever solution we eventually come up with must allow us to account for the observations contained in (A) and (B) above.
Figure 1 gives the configuration which agrees with a literal interpretation of Jespersen's original notion, where the extraposed constituent sits to the right of a complete sentence. Figure 2 has the extraposed constituent sitting outside of the VP.
The configuration suggested by Figure 1 would require us to have one sentence-level constituent ("it seems to be likely") inside another sentence-level constituent ("it seems to be likely that ..."), and the configuration suggested by Figure 2 would require us to have one VP ("seems to be likely") inside another ("seems to be likely that ..."). Either of these analyses would require us to add to the grammar
Further complexities abound. For example, each of these analyses would require the grammar to treat free or "weight"- motivated variation in constituent ordering not in terms of alternative orderings of sisters of a single mother, but in terms of alternative structurings. An example of the kind of problem that would arise is seen in the sentences in , where the puzzle we face has to do with the location of the "by"-phrase. If the extraposed constituent is outside of the sentence, or outside of the VP, where is the "by"-phrase located?
These difficulties do not by themselves constitute a knock- down argument against the correctness of the constituent-structure proposals embodied in Figures 1 or 2; but they present complications that should be avoided if possible. The intuition behind the proposal in Figure 2 is that the relationship between "seems to be likely" and "that she agrees with me" is one of predication. "That she agrees with me" is the subject, "seems to be likely" is the predicate, and the two are adjacent members of a two-part construction, here just as in the familiar Subject-Predicate Construction. The argumentative force of that intuition, however, is weakened on two grounds: the first is the set of complexities, just touched on, that such an analysis would have to deal with; the second is the fact that the separation of the [gf subj] argument from the rest of the clause, the "predicate", is something that characterizes the particular left-to-right pattern we find in the basic structuring of English simple finite sentences, but is by no means necessary in other contexts. For example, in , the a sentence has the "predicate" intact and adjacent, as a whole, to the "subject"; but the b sentence has the subject serving happily as a sister to the head verb and its other complement.
The structural configurations suggested by Figures 3-5 require no new apparatus for our grammar, since the extraposed constituent can simply be realized as a complement of the "seems" (Figure 3), of "be" (Figure 4), or of the adjective "likely" (Figure 5), and can be positioned as a sister of the lexical head of the relevant AP or VP – in the way that any non-subject argument is regularly instantiated. But how do we decide among these three possibilities?
The difference between the solutions of Figures 3-5 has to do with the question of what argument in the valence of a complement predicator is coinstantiated by what argument in the valence of the governing predicator. Shall we say, for example, that IT-Extraposition is to be stated as a possibility for "likely", and that in Figure 5 "that she agrees with me" is a complement of "likely" and "it" is coinstantiated by the subject of "be" and secondarily by the subject of "seems"? Or shall we say that the "that"-clause is coinstantiated by ("raised to") an argument of "be", and that the valence that allows both "it" and an extraposed constituent is available for the verb "be" (Figure 4), and the "it" is coinstantiated by the subject of "seems"? Or shall we say that the "that"-clause is coinstantiated by the subject of "be" and secondarily by the subject of "seems" and that it is at the level of the phrase headed by "seems" that the extraposed clause appears (Figure 5)?
It is difficult to find decisive arguments to resolve this issue, but our choice is the structure found in Figure 3. The result is a compromise between two intuitions. The first is that the "it" and the extraposed clause should not be on vastly different levels of the tree, as they are in Figure 13, because, although admittedly not coreferrential, they do in some sense ‘cash out’ a single semantic requirement. This intuition is supported by the fact that the usual intonation is not suggestive of the extraposed clause being buried at maximum depth in the VP. The second intuition is that IT-extraposition must be formulated as a valence (linking) construction in order to capture the generalization that regardless of grammatical function (subj, obj, pobj) one finds null "it" in the relevant structural position (daughter of S, daughter of VP, daughter of PP) and the semantics-bearing clause at the end of the sentence. Our compromise will be in effect to let the clausal argument undergo coinstantiation up to the highest verb in a chain of raising predicates and then let "it" and the semantics-bearing clause both be valence elements of this predicator and get realized accordingly. (A separate mechanism will be required when the "seem"-type predicators occur downstairs, because these verbs don’t allow clausal arguments in their minimal valences to start out with.)
It is simplest to begin our explanation with the special, "seem"-type verbs. The verb "seem" in Figures 3-5 is used in its IC (hence, Raising) version. In the case of the FC version of the "seem" verbs, however, it is necessary to provide the IT-Extraposition valence from the beginning, i.e., as a part of its minimal valence, because "that"clause subjects to "seem" are not possible. That valence will include (i) a [role [gf subj, q null]] argument to which the lexeme "it" is assigned, and (ii) a "that"-clause which bears a new grammatical function that we will call "xp" (a mnemonic for "extraposition"). The constituent with [role [gf xp, q content]], being one of the non-subject arguments of the verb, will be instantiated as a sister to the verb "seem". But the "it", being the subject, is of course available for coinstantiation by higher embedding predicates, and can therefore participate in a chain of raising.
Thus the sentence "It is likely to seem that ..." is possible because the "it" that was assigned [gf subj] for "seem" can be coinstantiated by the subject of "is likely". The non-sentence "*That he is asleep is likely to seem" is not possible because extraposition is obligatory for the lexical verb "seem": there is no possibility that minimal "seem" could have the "that"-clause as its subject, and hence a "that"-clause complement of "seem" could not be coinstantiated as ("raised to") the subject of "is likely" and ‘thence’ directly instantiated as the subject of the sentence.
For (most) intransitive verbs taking propositional arguments (e.g., "seem", "happen", etc., but not "stink") we must provide, as argued above, minimal valence sets including an empty "it" subject and an extraposed "that"-clause. Thus, for verbs of the "seem" class, we would have the following [fn]:
It is important to realize that this is a sample lexical valence, not a linking construction. (The extraposition construction which covers all the other cases, however, is a linking construction. It will look the same as Figure 6, except for the features 'loc +' on the "it" element and the absence of any theta role specification on the xp element – and of course the psuedo-specification 'lxm X'.)
In a sentence like "It is likely to seem that he is sleeping", the most deeply embedded predicator, "seem", has the two part minimal valence shown in Figure 6. The clause, "that he is sleeping" is realized as a sister to "seem", with the xp gf. Since this element is assigned gf xp, it cannot be subject to coinstantiation (raised), because raising requires that the governed element to be ‘gf subj’. The other valence element, "it", is successively raised to subject of "likely" and subject of "be", whence it is directly instantiated via S-P.[fn]
With regard to the more general case, the linking construction for IT-extraposition looks like that shown in Figure 7. As mentioned, this is just like the minimal valence template for "seem"- type verbs shown in Figure 6, except for the addition of the ‘loc +’ feature on the "it" element and suppression of the '[q content]' notation on the xp element.
The 'loc +' feature, which will play an important role in the next chapter, represents the fact that the "it" must be locally instantiated as a direct argument of the verb whose valence set has this feature. When we speak of 'local instantiation' of a valence element, we mean that this element is directly instantiated in a constituent headed by its governor: either within the VP (if the element bears the obj, pobj or obl(ique) function) or within an S-P constituent (if the element bears the subj function). In particular, a locally instantiated valence element is neither raised by coinstantiation nor 'extracted' by distant instantiation. Figure 8 shows the necessary amendment of the Coinstantiation construction.
In examining Figure 7, we notice that the rule applies only to verbs. This means that extraposed constituents are never daughters of adjective phrases. The requirement that the "it" must be locally instantiated guarantees that multiply-embedded structures (i.e., structures with chained coinstantiation links) will be structurally unambiguous: the extraposition construction will only apply to the 'top' verb in a chain of raising predicators, since the "it" introduced by extraposition cannot itself undergo raising, on account of being marked 'loc +'. So a sentence like
receives the following analysis. The passive verb "discovered" allows for a "that"-clause subject (Compare: "That he was sleeping was discovered.") The syn and sem values of the element in the valence of "discovered" representing this clause is constantiated by "be", then "likely", then "is". "It- extraposition" unifies with the valence of "is" to licence "it" as the subject of the sentence and the xp clause "that he is sleeping" as a daughter of the VP headed by "is". (We leave as an exercise to the reader to figure out why in (63) the IT-extraposition construction cannot unify with either the "be" or "likely" valences, which might produce ambiguity.)
The second respect in which the IT-Extraposition construction of Figure 7 differs from the valence template for "seem"- type verbs of Figure 6 is that the former lacks the indication '[q content]'. Recall that coinstantiation only unifies inherent (non-role) features. Compare sentence 1, in which there is no raising of the element to be extraposed, and 2, in which the extraposed element is at the top of a raising chain.
In 1 the xp constituent "that you didn't call" will receive the theta value of content from the minimal valence of "worry". In a raising case like 2, however, the theta role of content will be, as it were, left behind, in the relevant valence element of "worry". "Appear" has only only one theta role to give out, that of content, and the phrase receiving that role is the VP headed by "worry". Similarly "likely" assigns its content role to the VP headed by "appear" and "is" assigns its content role to the AP headed by "likely". In a sentence like [xx]x we don't want to say that the extraposed clause has any theta role – although we do in [xx]x – and omitting mention of theta role in the formulation of the IT-extraposition construction (Figure 7) produces this result.
The second minor generalization we still need to take official cognizance of is that a certain class of VP idioms, such as those illustrated in  and , require extraposition from object position.