Abstract

Existential constructions (ECs) offer two great puzzles to linguists.

I.  Why are only certain types of associates possible, as in (1)?

(1)	a.		There is a wolf at the door.
	b.	*	There is the wolf at the door.
	c.		There were several people cycling along the creek.
	d.	*	There were John and Mary cycling along the creek.
	e.		There was an article mentioned.
	f.	*	There was Frank's article mentioned.
(Milsark)

II. Why does the verb agree with the associate, as in (2)?

(2)	a.	There is a strange man outside.
	b.	There are some strange men outside.

A priori, it seems likely that the answers are related.  It is otherwise too much of a coincidence that one construction should have two such idiosyncratic properties.

Minimalist answers to (II) generally involve movement of the associate at LF.  Chomsky (1986) suggested that there has no semantic content at all, and is replaced by its associate at LF.  Chomsky (1993) said instead that the associate adjoins to there.  Either way, these accounts use LF movement to account for agreement properties clearly visible at PF.  The proposed explanation involves a definition of "strong" features, which need to be satisfied before spell-out.  These explanations do nothing to account for (I).

My original intent was to put forth the following alternative theory:

1. The associate is, indeed, the subject of the sentence.
2. there is a clause-level existential quantifier.
3. The agreement between the verb and the associate results from subject-verb inversion, similar to interrogative constructions.

However, I was unable to support this theory.  Rather, the data seem to indicate that there bears many of the characteristics of the subject, with the associate bearing the other characteristics.  I discuss the basis of a possible explanation using principles from generative phonology.

Semantic Contribution

Milsark observed that some noun phrases can be used as an associate[1] of a there existential, while others cannot, as in (1).  He claimed that associates are limited to only those NP's which he terms expressions of cardinality, in contrast to expressions of quantification.  This observation led him to conclude that there is itself a type of quantifier.

I disagree with Milsark's choice of terminology.  Traditional quantifiers such as some or every occur in the determiner position; there occurs by itself in overt syntax.  However, I agree with Milsark that there carries semantic weight, specifically as an assertion of existence.[2]

Intuitively, an EC can only take an associate that is new information.  Particularly, the new information is something that the speaker is asserting to exist, assuming that the listener was unaware of its existence or presence.

This intuition helps explain why ECs only take indefinite associates – or unquantified associates, if we accept Milsark's analysis.  A weak determiner phrase such as some students or an indefinite phrase such as a student does not assume that the referent is known.  Hence it can be new information; hence it can be used with an EC.  A strong determiner phrase such as every student or a definite phrase such as the students assumes that the referent is already known.  Hence it cannot be new information; hence it cannot be used with an EC.  This contrast is shown in (3).

(3)	a.		There are some students in the classroom.
	b.		There is a student in the classroom.
	c.	*	There is every student in the classroom.
	d.	*	There are the students in the classroom.

This intuition also explains apparent exceptions to the definiteness principle of ECs, as in (4).

(4)	a.	A: No one in our family can cook.  B: Well, there is Aunt Sue.
	b.	There is every reason to hope that we will come out of this alive.

In (a), Aunt Sue is a known entity, not new information.  However, her role in this situation is previously unrecognized by speaker A.  Speaker B's assertion of her existence in this capacity licenses use of the EC.

In (b), every appears to have an unusual usage.  Normally, the quantifier selects universally from a known set, which prohibits it from use in an associate DP, as in (5).

(5)	a.	*	There was everyone in the room.
	b.	*	There is every student in the classroom.

In (4b), on the other hand, the expression asserts the presence of members of a set that was presumed to be empty, i.e. reasons to hope.  Since it is asserting new information, it licenses the EC.

This intuition also straightforwardly explains why there cannot be two there existentials in the same sentence, as in (6).

(6)

*

There seems there to be a man in the garden.

Since each there highlights new information, having two in the sentence leads to competition.  One way to think of this involves existential closure.  If the first there existentially closes the proposition, then the second there has nothing to close.

Syntactic explanation

It is possible that these semantic phenomena can be explained syntactically.  One theory might involve some sort of CP-level morpheme, perhaps existence or focus, which provides existential closure to the sentence.  Another explanation might suggest that existence is somehow encoded in partitive Case (Lasnik 1995a).  I will not pursue either of these theories in detail here.

Subjecthood

To determine whether there is the subject of an EC, we compare it to other characteristics of known subjects.[3]

Verb agreement

In English, it is typical for the verb to show agreement for number and person with the subject, as in (7).

(7)	a.	A strange man is at the door.
	b.	Some strange men are at the door.
	c.	Some strange men are at the door.

To the best of my knowledge, in no cases other than ECs does the verb agree with an object or complement.  Even when expletive it[4]

takes a plural complement, the verb continues to agree with the subject, as in (8).

(8)	a.		It's a strange man at the door.
	b.	?	It's some strange men at the door.

By this criterion, there is not the subject of the sentence.  Rather, its associate is the subject.

Subject/auxiliary inversion

Interrogatives or focus movement can trigger the reverse of the normal word order between a subject and auxiliary verb, triggering do-support if no auxiliary exists.  Examples are in (9).

(9)	a.	(The/Some) strange men are in the garden.
	b.	Are the strange men in the garden?
	c.	Where are the strange men?
	d.	In the garden are some strange men.

At first glance, ECs may appear to be another trigger for subject/auxiliary inversion.  By this analysis, the subject follows the verb, and there fills a higher position in the sentence, perhaps C or [Spec, CP].

However, this account breaks down when we observe that most triggers for inversion can occur with an EC, as in (10).

(10)	a.		There are strange men in the garden.
	b.		Are there strange men in the garden?
	c.	?	Where are there strange men?
	d.	*	In the garden are there strange men.

This suggests that a basic EC is in its standard, uninverted word order.  By this test, there would appear to be the subject.

Note that (10d) is bad.  Why is this?  Remember the observation that an EC introduces new information.  This is similar to the function of focus.  Perhaps this example is bad because the fronted PP and the EC are both trying to introduce/focus new information.  This is similar to the problem we saw in (6), with two there existentials in the same sentence.

Ellipsis in a conjoined clause

The analysis of sentences such as (11) is controversial. 

(11)	Peter brushed his teeth and went to bed.

In one analysis, (11) is a conjoined sentence, where the subject of the second conjunct has been deleted.  Thus, (11) is equivalent to (12).

(12)	Peteri brushed his teeth and hei went to bed.

If this is, indeed, the correct analysis, it suggests another test for subjecthood: When the subject of the second clause in a conjoined sentence bears some relationship to the subject of the first clause, it can be deleted.  The exact nature of the licensing relationship is yet to be determined.

Note that this ellipsis process is highly productive.  For instance, it can occur when one of the verbs is active and the other is passive, as in (13).

(13)	a.	Tony was picked and (he) walked to the outfield.
	b.	Tony was picked and (he) walked to the outfield.

It can even occur with the expletive subject it with only slight degradation, as in (14).

(14)	a.		It is cold and it is snowing.
	b.		It is cold and snowing.
	c.	?	It is cold and is snowing.

However, subject ellipsis cannot occur with there, as shown in (15).

(15)	a.		There are happy people and there are sad people.
	b.		There are happy people and sad people.
	c.	*	There are happy people and are sad people.

Unless some non-syntactic criterion can justify this behavior, this test would suggest that there is not a subject.

Raising

The subject of a subordinate clause can raise to either subject or object position of the matrix clause in English, depending on the matrix predicate.  Examples are in (16).

(16)	a.	Harry likes ice-cream.
	b.	Harryi seems [ti to like ice-cream].
	c.	Martha believes Harryi [ti to like ice-cream].

Crucially, only the subordinate subject can raise in these constructions.  Bad attempts to raise the object are shown in (17).

(17)	a.	*	Ice-creami seems [Harry to like ti].
	b.	*	Martha believes ice-creami [Harry to like ti].

In cases where the subordinate object appears to have raised, it is plausible to suppose that it first raised to subject position of the subordinate clause, as in (18).[5]

(18)	Ice-creami seems [ti to be liked ti].

There is capable of raising under the same circumstances, as in (19).

(19)	a.	There is a full moon.
	b.	Therei seems [ti to be a full moon].
	c.	Martha believes therei [ti to be a full moon].

On the other hand, its associate is not capable of such movement, as shown in (20).

(20)	a.	*	A full mooni seems [there to be ti].
	b.	*	Martha believes a full mooni [there to be ti].

According to this test, there appears to be a subject.

Tag questions

English allows productive use of a tag question following up a declarative sentence, as in (21).[6]

(21)	a.	John is here, isn't he?
	b.	You've lost weight, haven't you?
	c.	Your plant died, didn't it?

In each case, the tag question is formed using an auxiliary verb and a pronoun co-referential with the subject of the main sentence.  The n't
clitic is used only if the main sentence is positive, as shown in (22).

(22)	a.		John isn't here, is he?
	b.	*	John isn't here, isn't he?
	c.		You haven't lost weight, have you?
	d.	*	You haven't lost weight, haven't you?
	e.		Your plant didn't die, did it?
	f.	*	Your plant didn't die, didn't it?

The tag question is possible even with an expletive subject, as in (23).

(23)	a.	It isn't raining, is it?
	b.	It's cold, isn't it?

A tag question is also possible with there, as in (24).

(24)	a.	There isn't someone outside, is there?
	b.	There is a full moon, isn't there?

This test suggests that there is a subject.

Summary

Out of five tests, three suggest that there is a subject, and two suggest that it is not.  An explanation will require investigation into the nature of a subject.

Agreement checking

In the minimalist tradition, the subject is defined as the element in [Spec, AGR_S].  The tests which indicate that there is a subject suggest that it does, indeed, reside in this position, unless it raises to [Spec, AGR_O].  The challenge, then, is to explain why the verb agrees with the associate, an element that does not reside in [Spec, AGR_S].[7]

By observation, a finite verb must have agreement features in order to be pronounced.  This suggests that the verb must check agreement features at (or before) PF.

Underspecified agreement feature

In generative phonology, a phoneme can be underspecified for a specific feature.  Typically, the value spreads from a neighbor, according to language-specific guidelines – for instance, it may only spread from the next phoneme to the right.  If no neighbor is available, the phoneme is pronounced with a default setting for the feature.[8]

Whatever there may be, it is not a typical noun phrase.  Perhaps it has the DP feature that satisfies the EPP, but is underspecified for number and/or person.[9]

Under this hypothesis, when there moves to [Spec, AGR_S][10], it is unable to check the number agreement feature, because it is underspecified for this feature.  As in phonology, this triggers feature spread from an appropriate neighbor.  It happens that the only acceptable neighbor is the associate.  Following Chomsky (1986) and Hornstein & Witkos, it might seem that the feature can only spread from another element in the same A-chain.

This treatment bears some resemblance to the notion of feature movement suggested by Chomsky (1995) and Lasnik (1995b).  The only feature that moves (i.e. spreads) is the one that is underspecified for the element residing in [Spec, AGR_S].

Incidentally, this process could also explain why the associate can only be a DP, since only a DP has agreement features.  This blocks, for instance, the sentences in (25).

(25)	a.	*	There are hungry.
	b.	*	There is raining outside.

Default value

If this analogy with phonology is valid, one might expect to find cases in which a default feature value is pronounced because no neighbor is available.  Consider the sentences in (26), listed in order of acceptability.[11]

(26)	a.		There are some strange men in the garden.
	b.		There seem to be some strange men in the garden.
	c.	?	There seems to be some strange men in the garden.
	d.	?	There's some strange men in the garden.
	e.	?	There seems to me to be some strange men in the garden.
	f.	??	There seems to me to be some strange men in the garden.
	g.	*	There is some strange men in the garden.

Note that in (c)-(e), the verb fails to agree in number with the associate, and yet the sentences are only slightly degraded.  More serious degradation is seen in (f), in which the verb does, in fact, agree with the associate.

The examples suggest that there is a limit to how far the cognitive faculty can search to find the correct agreement.  As (b) shows, it is perfectly acceptable to agree with an associate in a subordinate clause.  Yet (c) suffers hardly noticeable degradation.  Presumably the distance between the verb and the associate makes it difficult to determine the agreement.  This licenses the use of the singular, which I take to be the default.  Furthermore, (e) and (f) show that when it becomes sufficiently difficult to determine agreement, the singular form is more natural.

Example (d) is in some ways the most interesting.  Surely, there is no difficulty in determining the correct agreement, and yet the sentence is rated perfect by most of my informants.[12]  In this, the use of the clitic is distinctly different from the use of the full verb in (g).  I will leave a detailed analysis to others.

Non-standard dialects

This account is based on the observation that a finite verb demands agreement features.  Were this not the case, we would expect to see ECs use the uninflected (i.e. featureless) form of the verb.  Thus, we get the pirate dialect of the movies, as in (27).

(27)	There be sharks in them waters.

I suspect this phenomenon may also occur in Ebonics, although I do not have evidence to confirm this.

Conclusions

I sought to address two perplexing questions related to ECs.

1. Why are only certain types of associates possible?
2. Why does the verb agree with the associate?

Following Milsark, I suggest a semantic answer to (I).  ECs assert existence.  Therefore, the only DPs that can be used as associates are those for which existence is not presupposed.  These phrases are often called indefinites.

To address (II), I suggest that there is underspecified for number agreement, triggering feature spread from its associate.  When this spread is blocked for some reason, we see evidence of a default feature value.  It would be interesting to see whether other syntactic phenomena might be explained by a similar mechanism.

A complete theory of ECs will address both of these questions.  I suspect the answers will be related.  Perhaps there is like a quantifier in some ways, and like a DP in others.  I leave the resolution of the problem for another day.