More on English Preposition Stranding

Joseph Sabbagh
MIT

and

Lotus Goldberg
McGill University

This squib is dedicated to Jorge Hankamer on the occasion of his sixtieth birthday. Jorge has been an especially great influence on us both, and on the course which our linguistic paths (and lives) have taken. It has been a blessing and privilege to have been able to work with him during our time at UCSC, and we are very grateful to be able to honor him in this way.

Goldberg adds: The experience of being taught by Jorge as it develops over the span of a quarter or a year probably instantiates the deepest respect I have ever felt from another person in a human interaction. When I remember time spent in Jorge's classroom or office, I am so thankful to know and have been able to interact with someone who understands so profoundly how to stimulate a love of thinking and a sense of excitement at the intricacy of structure of all kinds. I will always owe an immense debt to Jorge for generously sharing all of this with me, directly as well as through his nurturing of the UCSC linguistics program which has in turn nurtured so many of us.

Sabbagh adds: During the Winter of '98, I took Syntax II from Jorge. I had no idea during that time what the point of deriving sentences with more than 15 S nodes was. It was not until a year or so later, when I began meeting students from other departments both in and out of UCSC, and when I began conducting my own research, that I realized the intellectual impact that Jorge's teaching had on me. Jorge has served as much more than a teacher over the years. At the LSA Summer Institute, he was a friend, and a mentor (not to mention a bowling competitor). I only hope that, in the future, I will be able to inspire students just as Jorge has done for me and many others.

Please accept this squib as nothing more than a five page birthday card with a lot of questionably judged English sentences in it.

1 Introduction

As mentioned in the introductory notes to Postal's (1997) book, Langendoen and Pullum (henceforth L&P) published in a 1977 CLS book of squibs a short presentation of various puzzles relating to instances of stranded (as opposed to pied piped) Ps in what are mostly double object constructions of some sort. We will aim here to recall some of the data from that 1977 squib, and to suggest that the prohibition on P stranding noted by L&P might be viewed as part of a more general prohibition on extraction from specifier positions. That it is possible now to consider this approach—which we sketch below in only a very nascent form—is a result of the consequences for movement posed by recent developments in the theory of internal VP structure, and by other current phrase structural conceptions. Thus, although the generalization suggested here is not conclusive by any means, it does suggest a line of inquiry that might not have been possible twenty years ago.

2 The L&P Data

We wish to focus mainly upon those sentences presented in L&P's examples (2–4). In these examples, an NP extracted from a PP without pied piping of the P results in ungrammaticality; this appears to occur specifically when the PP at issue immediately follows the V to which it serves as complement, with a second complement to the V appearing to the immediate right of the PP. The observed ungrammaticality appears to occur in all standard A-bar extraction types, including wh-questions, relative clauses, and topicalization structures. We include the relevant wh-question examples here; all sentences not designated with an L&P example number are own additions:

(1)*Who does the chair feel to __ very hard? (L&P's (2a))
(2)*Which room is there in __ a strange beast? (L&P's (3a))

NP extraction stranding the same P is fully licit when the PP position with respect to the other complement of V is reversed:

(3) Who does the chair feel very hard to __ ?
(4) Which room is there a strange beast in __ ?

In fact, the marked PP position displayed in (1–2) appears to be quite degraded even in examples without co-occurring extraction and P stranding:

(5)??This chair feels to me very hard.
(6)??There is in this room a strange beast.

However, the cause of the degradation in (5–6) cannot be the sole cause of the ungrammaticality in (1–2): not only are judgements for the examples with P stranding more strongly ungrammatical than for those without it, but the former improve significantly when the complement of V which appears at the right edge of the sentence is made heavier, while the latter do not:

(7) This chair feels to me so hard that it could be made of concrete.
(8) There is in this room a very strange beast with enormous antlers and five arms.
 cf.:
(9)*Who does the chair feel to __ so hard that it could be made of concrete?
(10)*Which room is there in __ a very strange beast with enormous antlers and five arms?

Furthermore, the sort of ungrammaticality present in the cases of extraction in (1–2) appears to occur even when the relevant PP occupies a position which is fully licit without NP extraction which strands the P. Thus, the ungrammaticality of (11–13) seems on a par with that of (1–2):

(11)*Who did you receive from __ several long letters?(from L&P's (3b))
(12)*Who did you mention to __ that we'd already eaten the cookies?(from L&P's (3c))
(13)*Who were you shown by __ that we'd eaten all the cookies?(from L&P's (3d))

As was true for (1–2), (11–13) become gramamtical when the PP in which P stranding occurs is the linearly second complement to follow the V, as shown in the (a) examples of (14–16). However, contrasting with the ungrammaticality of (5–6), the PP position from which extraction is ungrammatical in (11–13) is licit when P stranding does not take place. This latter point is shown in the (b) examples of (14–16).

(14)a.Who did you receive several long letters from __ ?
 b.We received from Jorge several long letters.
(15)a.Who did you mention that we'd already eaten the cookies to __ ?
 b.I mentioned to Jorge that we'd already eaten the cookies.
(16)a.Who were you shown that we'd eaten all the cookies by __ ?
 b.They were shown by Jorge that we'd eaten all the cookies.

In contrast to these patterns, PP complements of talk appear grammatical with P stranding, regardless of whether the stranded P occurs immediately following the V, or separated from the V by an additional argument:

(17)  Who did you talk to __ about Dot? (L&P's (4a))
(18)  Who did you talk about __ to Dot? (L&P's (4b); cf. (1–2), (11–13) above)
(19)  Who did you talk about Dot to __ ? 
(20)  Who did you talk to Dot about __ ? 

L&P also show that, where such extraposition structures are possible, the presence of expletive it immediately following the V in examples such as (15) produces contrastingly full grammaticality:

(21)  Who did you mention it to __ that we'd already eaten the cookies? (from L&P's (5c))

To add to this empirical picture, and as a segue to the next section, we note additionally that extraction out of a CP which immediately follows the V of a double object construction appears to be as unacceptable for these Vs as was extraction out of the PP:

(22)* What did you mention that we'd already eaten __ to Jorge?
(23)* What were you shown that we'd eaten __ by Jorge?
 cf.:
(24)  What did you mention to Jorge that we'd already eaten __ ?
(25)  What were you shown by Jorge that we'd eaten __ ?

We can thus note so far that, with the exception of the V talk, the patterns in the traits of the English examples seen so far have all been consistent with a generalization that extraction of an element out of the position to the immediate linear right of the main V and preceding a second verbal complement appears to result consistently in ungrammaticality. Extraction of the entire complement in this position—in contrast to extraction of just a subpart of this element—appears to be licit:1

(26) To who(m) did you mention __ that we'd eaten the cookies?
(27)From who(m) did you receive __ several long letters?

3 Additional Data and Theoretical Implications

In this section, we suggest a possible direction that an account of the facts discussed above might take. The considerations in this section are quite tentative, and are not intended to address all of the facts at this point. The first observation that we would like to make is that there could be a connection between the ungrammatical extractions in (1–2), (9–13), and (22–23), and the classic CED effects illustrated in (28):

(28)a.  Which politician did you see [pictures of __ ]?
 b.* Which politician did [pictures of __ ] upset the voters?

On this view, the contrast in (28) and the contrasts observed above might both arise from a sensitivity to phrase structure position in terms of the acceptability of extraction. More concretely, the grammaticality contrast shown in (28a–b) and the contrasts just shown in Section 2 could both be attributed to whether or not the XP from which extraction has taken place is a specifier position or a complement position.

Traditional analyses of (28a-b) have made crucial mention of the complement vs. specifier status of the NP from which extraction has taken place. Chomsky (1986), for example, distinguishes between the specifier position in (28b), from which extraction is ungrammatical, and the complement position in (28a), from which extraction is licit, by the notion of L-marking. Thus, [Spec, IP] is a non-L-marked position, whereas a verbal complement is L-marked by the V; extraction is therefore much worse in (28b) than in (28a). Extending this kind of analysis to the cases documented in the previous section would be arduous and ultimately unenlightening, because even if we could determine which relevant syntactic elements were L-markers and which were not, we would then have only a list, but no clear generalization. A more general account might involve claiming (with admitted brute force) that extraction from an XP in a specifier position is ruled out, while extraction out of a complement position is acceptable. This kind of generalization would hold irrespective of category or A- versus A-bar distinctions.

Let us consider the examples from Section 2 with the hypothesis just discussed in mind. If the V mention in (12) and (22) is a double object verb on a par with Vs like put or give, then we could follow approaches taken in some recent work (see e.g. Johnson 1991, McCloskey 2000) and assign to these sentences the structures shown in (29a–b). (We abstract away from V movement here.)

(29)a. Who did you mention VP[ PP[to__ ] V'[ V [that we'd already eaten the cookies]]]?
 b. What did you mention VP[ CP[that we'd already eaten __ ] V'[ V [to Jorge]]]?

Consider in this light a sentence like (30), another case of ungrammatical P stranding observed by L&P:

(30)* I saw the box which Amy said [out of __ ] came a toad. (L&P's (1b))

The bracketed PP plausibly occupies [Spec, IP] position (see e.g. Bresnan 1994), as follows:

(31)  I saw the box which Amy said IP[[out of __ ] I'[ I0 [came a toad]]].

In these examples, then, it seems that extraction from out of an element in specifier position results in ungrammaticality. This occurs with extraction from [Spec, VP] in (29), and from [Spec, IP] in (31) and in the CED example in (28b).

Consider now examples (32a–b). (32a) is based on L&P's (6), and (32b) is based on a classic example of Paul Postal, used as an argument against successive cyclicity:

(32)a.* Who was it to __ that you spoke?
 b.* Who do they believe to __ that the student's spoke?

In each of these sentences, the PP from which extraction has taken place is likely best analyzed as being in [Spec, CP] position:

(33)a. Who was it CP[[to __ ] C'[that [you spoke __ ]]]?
 b. Who do they believe CP[[to __ ] C'[that [the students spoke __ ]]]?

We again have extraction from a specifier position which proves to be illicit.

As a final example, let us consider (9–10) from Section 2, which involve Heavy XP Shift (HXPS). Putting aside for now many issues which arise here, HXPS is often assumed to involve a rightward movement of the heavy XP. However, others propose radically different analyses. For example, Kayne (1994) and Shlonsky and Belletti (1995) both assume that this linear order is in fact derived via leftward movement of the non-heavy XP to a specifier position which c-commands the heavy XP, with the latter not undergoing any rightward movement. If these analyses are on the right track, then they assign structures to instances of HXPS which are again consistent with the generalization being considered here. Important details aside, the structures to consider, then, would be as in (34a–b):

(34)a. Who does this chair feel FP[[to __ ] F'[ F0 VP[[so hard that...] V'[ V __ ]]]]?
 b. Which room is there FP[[in __ ] F'[ F0 VP[[a strange beast with...] V'[ V __ ]]]]?

If the structures for these ungrammatical extractions are indeed as shown in (29), (31), and (33–34), then we have come close to a generalization that accounts for the facts discussed here in a uniform fashion. Note that, at present, this generalization only concerns extraction out of an XP in a specifier position, but, as consistent with the grammaticality of extraction with pied piping in (26–27), would not apply to full extraction of a specifier XP. Although more research is clearly needed on our part to find independent motivation for the structures above, we wish at this point to point out simply that they are plausible.

4 Specifiers

If the considerations above are on the right track, then the question to ask is what is responsible for causing extraction out of an XP in a specifier position to be impossible. Recent proposals by Kayne (1994), and related proposals by Uriagereka (1999) and Nunes and Uriagereka (2000, henceforth N&U) provide a way to answer this question. Kayne suggests, based on his Linear Correspondence Axiom (LCA), that no linear ordering can be determined between the specifier to a head, the head itself, and the complements of the head. He thus proposes that the relation of specifier be reanalyzed as a case of adjunction, thereby guaranteeing satisfaction of the LCA. If all specifiers were to be reanalyzed as adjuncts, then the ungrammaticality of extracting from an XP in specifier position suggested here might plausibly follow as part of something like the Adjunct Island Condition (see Chomsky 1986 and others).

N&U (2000) also assume LCA and propose relatedly that specifiers to a head cannot be linearized with the head or its complements. Instead of entailing that specifiers be reanalyzed as adjuncts in order to satisfy the LCA, however, N&U propose as an alternative that all specifiers (and adjuncts) in a syntactic structure must be "Spelled-Out" (i.e. sent to the PF component) separately from the other elements in that structure (namely those which represent the head-complement relation). Under their approach, the separate Spelling-Out of the specifier means essentially that no syntactic operations—including movement and deletion—may access its internal parts. Their proposal is developed to account for the CED contrast in (28a–b), but has not, as far as we are aware, been considered, as suggested here, as a possible explanation for the L&P-type examples presented above.

Both the Kayne and N&U proposals provide a plausible "deeper explanation" for the generalization being considered here to account for the facts laid out above. Further study is again necessary to determine whether this is an appropriate line of research, and, if so, whether the proposal by Kayne or that by N&U is best suited to handle these facts. We end this squib, however, with two additional pieces of the puzzle.

5 Some Final Considerations

Two additional and interesting aspects of the N&U system are perhaps worth mentioning. N&U claim that their system developed to account for contrasts like (28a–b) can, with a few additional assumptions, be made to accommodate in addition the weakened ungrammaticality of sentences like (28b) when the gap left by extraction is a parasitic gap:

(35) Which politician did pictures of __(pg) upset __ ?

We note, then, that their system might predict as well that the ungrammaticality of the L&P-type examples considered here would also be weakened if the gap left by extraction were a parasitic gap. Our own judgements on such examples are a bit unreliable at this point, and so we leave it to the reader(s) to judge for themselves:

(36)a. Which society did it appear to __(pg) that greed had destroyed __ ?
 b. Which politician were you persuaded by __(pg) to believe that corporate interest had destroyed __ ?
 c. Which politician did you mention to __(pg) that the public despised __ ?
 d. Which politician did you mention that the public despised __(pg) to __ ?

Although we must leave checking of these issues and of the details of N&U's predictions for future research here, any improved acceptability for the examples in (36) would provide an intriguing addition to evidence that the ungrammaticality of CED and L&P-type facts might be connected.

A clearer prediction made by the N&U system but again not considered by these authors is that operations like sluicing should be impossible in the contexts of L&P- and CED-type examples. This is because sluicing would involve extraction from the positions where it is supposed to be prohibited, even though that position has subsequently been deleted. Our initial judgements on the relevant examples is that they are in fact acceptable—or at least much better than their non-sluiced counterparts. Here again, we will leave it to the readers to provide their own judgements, and leave a more systematic investigation to future work:

(37)a. Pictures of some politician upset the voters, but we were not sure which one [e].
  [e] = which one [pictures of __ ] upset the voters.
 b. Jack had mentioned to someone that he had already eaten the cookies, but he refuses to say who [e].
  [e] = who he had mentioned [to __ ] that he had already eaten the cookies.
 c. There is in one of these rooms a strange beast with enormous antlers and five arms, but I have yet to figure out which one [e].
  [e] = which one there is [in __ ] a strange beast with enormous antlers and five arms.

If these sentences are grammatical, then they pose a challenge to the proposal developed by N&U. On the other hand, they are perfectly expected under Kayne's analysis (along with whatever other mechanisms are needed to account for the weakening of the extraction violation under deletion).

6 Conclusion

To conclude, the question we leave the reader with is whether or not the constructions discussed in this squib can be given a uniform account in terms of the phrase structural contrast between complement positions and specifier positions. We have also explored briefly ways in which the generalization posed here—if it should turn out to be correct—might be explained in terms of recent proposals about phrase structure and related types of constraints. In addition to the needed checking of the grammaticality judgements and fleshing out of the analysis, deeper questions emerge from this inquiry as well. In particular, why should the assumptions described above need to hold, and, more generally, why should phrase structure pose such conditions on the range of permissible extractions? We hope, however, to have presented enough here to show that this approach is at least interesting and worth thinking about in more explicit terms in the time to come.

References

Bresnan, Joan (1994) "Locative Inversion and the Architecture of Universal Grammar", Language 70: 73–131.

Chomsky, Noam (1986) Barriers, MIT Press, Cambridge, MA.

Kayne, Richard (1994) The Antisymmetry of Syntax, MIT Press, Cambridge, MA.

Johnson, Kyle (1991) "Object Positions", Natural Language and Linguistic Theory 9: 577–636.

Langendoen, D. Terrence, and Geoffrey K. Pullum (1977) "Preposition Stranding in English: A problem and a mystery", in CLS Book of Squibs/Cumulative Index 1968–1977, Samuel E. Fox, Woodford A. Beach, and Shulamith Philosoph, eds., pp. 64–65, Chicago Linguistic Society, Chicago.

Larson (1988) "On the Double Object Construction", Linguistic Inquiry 19.3: 335–391.

McCloskey, James (2000) "Quantifier Float and Wh-Movement in an Irish English", Linguistic Inquiry 31.1:57–84.

Nunes, Jairo and Juan Uriagereka (2000) "Cyclicity and Extraction Domains", Syntax 3.1: 20–43.

Postal, Paul M. (1998) Three Investigations of Extraction, MIT Press, Cambridge.

Shlonsky, Ur and Adriana Belletti (1995) "The Order of Verbal Complements: A comparative study", Natural Language and Linguistic Theory 13: 489–526.

Uriagereka, Juan (1999) "Multiple Spell-out", in Working Minimalism, Samuel Epstein and Norbert Hornstein, eds., pp. 251–282, MIT Press, Cambridge, MA.

Endnote

1 We make the plausible (we think) assumption here that the moved PP originates as the first complement linearly following the V, although it could in principle have originated as the linearly second complement to follow the V. [Back]