2.1 The phenomenon of extraction from DPs

The Complex DP-constraint (CDPC), of which (4) is one anachronistic formulation, was coined by Ross (Reference Ross1967: 70)Footnote ³ to capture the fact that displacement of an XP is illicit once the XP's trace is contained in a complex DP.

1. (4) The Complex DP-constraint

   Elements dominated by a sequence which is dominated by a determiner phrase cannot be questioned or relativized.

Example (5) illustrates the constraint by way of wh-extracting a PP from a complex DP. I will refer to this option as PP-extraction from DPs. Example (6) shows the constraint at work with Ā-extraction from a relativized DP. Rule (4) successfully captures both.

1. (5) *About what did John {destroy/stack} [a book e]?

2. (6) *The man about who I read [a statement which was about e] is sick.

It was recognized very early that certain contexts annul the constraint, such as the type of embedding verb. Thus, for example, verbs of creation like write contribute to forming the bridge needed for Ā-extraction. Likewise, verbs which render the semantics of the embedded noun abstract (like read or edit), as opposed to concrete (like destroy or stack in (5)), have a similar ameliorating effect. This is illustrated in (7).

1. (7) About what did John {read/edit/buy} [a book e]?

The CDPC appears to be too strong, given that the right type of embedding verb appears to lift the constraint. Some ingenuity has flown into solving this problem, as I will elaborate later. The CDPC is a first descriptive approximation to the many factors that enter into the complex phenomenon “extraction from DPs”. The semantic relationship between the embedding verbs and the complement nouns is in any event one of the aspects that awaits a satisfying solution. The factor is operative in NCC-extraction in German, too.

The Subject Condition (Chomsky Reference Chomsky, Anderson and Kiparsky1973: 250) is another familiar constraint on displacement: while extraction from object-DPs is fine, subject-DPs disallow extraction out of them.Footnote ⁴

1. (8)

   1. a. Who is the class reading a book about?

   2. b. *Who is [a book about e] being read by the class?

      (Davies and Dubinsky Reference Davies and Dubinsky2003: 36)

2. (9)

   1. a. [Of which car] did they find [the driver e]?

   2. b. *[Of which car] did [the driver e] cause a scandal?

      (Chomsky Reference Chomsky, Freidin, Otero and Zubizarreta2008: 36)

Again, as will become clear later, NCC-extraction in German is also subject to this condition.

Finally, an important restriction is that no extraction may take place from definite or specific DPs, as (10) shows (Fiengo and Higginbotham Reference Fiengo and Higginbotham1981, Davies and Dubinsky Reference Davies and Dubinsky2003, Simonenko Reference Simonenko2015).

1. (10) *Who did Ashley read {the/that/Bill's} book about?

All these mentioned restrictions are subject to many complications. For example, (9a) bluntly violates the Specificity Condition illustrated by (10), as do all examples of NCC-extraction from DPs in German – the topic of this article. So while it is an important research topic to understand the grammatical factors that underly contrasts like the one between (7) and (10), it should be kept in mind that there are counterexamples like (9a). A close look is needed at the types of nouns involved, and arguably, a differentiation between specificity and definiteness might pave a way to a better understanding of these issues. It is the modest goal of this article to make a contribution at the observational and, maybe, the descriptive level.

2.2 A demarcation

To properly delineate the empirical scope of this article, this passage briefly indicates that for German, at least three constructions involving discontinuous DPs must be distinguished. Only PP-extraction from DPs (a German specimen is given in (11)) will serve as the backdrop for description of NCC-extraction from DPs; the two other construction mentioned here have no obvious bearing on the diagnostics and the analysis of NCC-extraction from DPs.

1. (11) Worüber  hast  Du ein  Buch  gelesen?

   where-over  have  you a  book  read

   ‘What did you read a book about?’

By and large, German exhibits the same restrictions as English when it comes to PP-extraction from DPs.

The second phenomenon involving a discontinuous DP is the was-für-split (see, e.g., Pafel Reference Pafel1996; Leu Reference Leu2008, Reference Leu2015, and references therein), which features an invariant wh-pronoun was ‘what’, which can optionally split off the associated DP (12a) or pied-pipe it (12b).

1. (12)

   1. a. Was  hast  Du  [e für ein  Buch]  gelesen?

      what  have  you  for a  book  read

   2. b. Was  für  ein Buch  hast  Du  gelesen?

      what  for  a book  have  you  read

      ‘What kind of book did you read?’

The was-für-split should be kept distinct from PP-extraction from DP, since the two exhibit a number of important structural differences: In the was-für-split, the launching site of Ā-extraction is the left edge of the DP, as indicated by the e in (12a), not the postnominal region. This indicates that the was-für-split involves extraction of a specifier (Pafel Reference Pafel1996) or extraction from a specifier (Leu Reference Leu2008, Reference Leu2015), not a complement of the noun, which is arguably the case with PP-extraction from DPs (as in (11)). Moreover, there are no restrictions with respect to the embedding verb. Thus, (13) are all possible, while the counterparts with PP-extraction from DPs are sharply out (14).

1. (13) Was  hast  Du  für  ein  Buch  {verloren/verbrannt/vergessen}?

   what  have  you  for  a  book  lost/burnt/forgotten

   ‘What kind of book did you {loose/burn/forget}?’

2. (14) *Worüber  hast  Du  ein  Buch  {verloren/verbrannt/vergessen}?

   where-over  have  you  a  book  lost/burnt/forgotten

A commonality between the two is the fact that PP-extraction from DPs is licit in indefinite contexts and impossible in definite ones, as (15) shows.Footnote ⁵

1. (15) *Worüber  hast  Du  {Marias/das/dieses}  Buch  gelesen?

   where-over  have  you  Mary's/the/that  book  read

The was-für-construction is inherently indefinite, that is, there is no definite counterpart, with or without split.

Let me finally address a construction featuring what looks like a discontinuous DP, so-called split topics (see Ott Reference Ott2011 and references therein).

1. (16) Studierende  sind  heute  einige  erschienen.

   students  were  today  some  appear

   ‘As for students, many showed up today.’

Like the was-für-construction and unlike PP-extraction from DPs, split topics exhibit no sensitivity to embedding verbs. Moreover, it seems as if they defy the Subject Condition, as the acceptability of (16) suggests, in contrast to both other constructions (see Jurka et al. Reference Jurka, Nakao, Omaki and Washburn2011 on was-für). Finally, notice that split topics are perfectly acceptable with definite DPs, again, in contradistinction to PP-extraction from DPs and was-für-split.

1. (17) Studierende  sind  heute  bloss  diese  erschienen.

   students  were  today  merely  these  appear

   ‘As for students, only these showed up today.’

There are other commonalities between these discontinuous DP-constructions, such as the fact that they all involve Ā-movement, as evidenced by connectivity effects, sensitivity to so-called island constraints and the like.Footnote ⁶ That said, the following summarizes important differences, which suggest that these should not be analyzed on a par.

1. (18)

   1. a. The was-für-split and split topics do not exhibit sensitivity to embedding verbs, that is, they are freely available independent of the verb that embeds them. This is not the case with PP-extraction from DPs. This means that crucially, the specific semantic relationship between the selecting verb and the noun in PP-extraction must be part of any analysis, and correspondingly has figured in all analyses up to date.

   2. b. Contrary to appearances, split topics involve obligatory movement, as Ott (Reference Ott2011) argues at length, while was-für-split or PP-extraction from DPs are clearly optional. Evidence for this claim comes from numerous cases in which split topics cannot surface continuously and force movement of the topic-DP.

   3. c. Relatedly, the was-für-split or PP-extraction from DPs plausibly involve a single DP and instances of subextraction from that DP, while split topics are likely to involve two nominal structures, as Ott persuasively shows.

   4. d. The was-für-split or PP-extraction from DPs involves prima facie violations of island constraints (CDPC and Left Branch Extraction respectively), while if Ott (Reference Ott2011) is right, split topics do not.

I ask the reader to bear in mind these important differences. NCC-extraction from DPs in German has many aspects in common only with PP-extraction from DPs, as will be shown later in the article, and should be analyzed along the same lines.

2.3 Treatments of the verb classes

Let me briefly sketch different analytical approaches to account for the sensitivity to verbs involved in PP-extraction from DPs.

Early treatments of PP-extraction from DPs take the CDPC as an inviolable constraint. In the context of verbs in which a PP originating from a DP can evade the latter, these approaches had to employ mechanisms that circumvent violations. Particularly, rebracketing rules like (19) were invoked for this class of verbs (Bach and Horn Reference Bach and Horn1976: 280, Cattell Reference Cattell1976, Chomsky Reference Chomsky1977).

1. (19) [_VP V [_DP N PP ]] → [_VP V DP PP ]

The effect of the application of the rule is that upon rebracketing, no CDPC-violation is incurred when the PP undergoes movement. This rule was said to apply for a restricted set of verbs – those that can give rise to PP-extraction from DPs. Verbs like destroy, stack and the like do not trigger the rebracketing rule; as a consequence, PP-movement incurs a CDPC-violation.

Empirical evidence for such a rule was given in Chomsky Reference Chomsky1977: 113–114. He noted that with the relevant class of verbs, ambiguities like (20) arise.

1. (20) They published their first book about Kafka.

   1. a. They have published other books before, but this one is the first that has Kafka as its subject. (first > PP)

   2. b. They didn't publish any books before and the first book they ever published happened to be about Kafka. (first ${\rm \not} \gt$ PP)

That is, the PP can scope below or above the adjective first. In terms of the rebracketing rule, one might say that the low scope reading of the PP follows from a stage in the derivation before rebracketing, and the hight scope reading of the PP arises after rebracketing. The same ambiguity does not arise with other types of verbs as (21) shows.

1. (21) They destroyed their first book about Kafka.

   They have destroyed other books before, but this one is the first that has Kafka as its subject. (first > PP)

This follows immediately, if the rebracketing rule is unavailable with the relevant class of verbs.

A quite different approach is taken in various works by Gereon Müller. He pursues the intuition that PP-extraction from DPs is contingent on the formation of a ‘natural predicate’. For example, book reading, but not book forgetting, involves the formation of a natural predicate. Technically, he cast this intuition into different types of implementations. Following Baker (Reference Baker1988), Müller (Reference Müller1995) suggests that with the relevant class of verbs, the embedding verb and the noun are coindexed at surface structure such that the noun undergoes abstract incorporation at the level of Logical Form, LF.

1. (22) [_VPV_i [_NP N_i PP ]] → LF: [_VP N_i+V_i [_NP t PP ]]

As the noun incorporates into the verb at LF, the barrier status of the NP is voided. When this happens, PP-extraction from DPs becomes possible. In Müller Reference Müller2010: 47–50 that same intuition is reformulated in terms of agree (avoiding possible problems of look-ahead afflicting the abstract incorporation approach). Müller (Reference Müller1995) makes the intriguing suggestion that the ability to PP-extract from DPs correlates with the option of forming the corresponding N-V compounds in German, cf. Vortraghalten (‘talk-giving’) vs. *Vortragvergessen (‘talk-forgetting’); see (23). The verb class problem is thus shifted to the availability of morphological forms.

1. (23) Worüber  hat  Einstein einenVortrag{gehalten/*vergessen}?

   where about  has  Einstein a takeheld/forgotten

While the intuition is not without problems,Footnote ⁷ it is a recurring theme which should not be carelessly dismissed.

As a final note, German freely allows DPs to front independently of their PP-dependant.

1. (24) Ein  Buch  hat  Fritz  über  Bienen  gelesen.

   a  book  has  Fritz  about  bees  read

   “Fritz read a book about bees.’

Given the right types of verbs, it thus seems as if these DPs can form a constituent to the exclusion of the PP. Such facts have given rise to either remnant movement analyses (see Müller Reference Müller1998) or operations that deliver independent DP- and PP-constituents (De Kuthy Reference De Kuthy2002).

3.1 Preliminaries

To the best of my knowledge, there is no previous study on NCC-extraction. A fair number of speakers utter and accept this construction, which I have exclusively witnessed in colloquial German. That is, I have not been able to obtain written instances in corpora. Among the speakers I consulted, some consider such sentences awkward or degraded. Yet all speakers agree with respect to the relative acceptability presented here, that is, all consider the starred examples significantly worse than the baseline sentences. Other speakers consider the construction outright ungrammatical, and any analysis or explanation should address this split in acceptability judgments. As a disclaimer, all judgments of NCC-extraction are to be considered highly idealized; needless to say, they represent relative judgments.

The mentioned gap in the text corpora might be due to the special information structural import of the construction, where the dislocated CP seems to mark the discourse-relevant topic, and the containing main clause is backgrounded, often representing given information or information less relevant to the discourse theme. The fronted CP is commonly a contrastive topic, or contains a contrastive topic. What is – descriptively speaking – the containing main clause gives the impression of an afterthought needed to grammatically complete the actual point of interest, the dislocated CP. In written language such arrangement is either unlikely to occur or might occur in chat exchanges at best.

I found a single mention of the impossibility of extracting NCCs in De Kuthy Reference De Kuthy2002: fn. 7. She dismisses the phenomenon outright, claiming that “[s]eparating a verbal [sic] complement from its nominal head always results in ungrammaticality”, based on and backed by examples like (27) – where (27a) is a fronted non-finite NCC and (27b) a fronted finite NCC.

1. (27)

   1. a. *Nach  Afrika  zu  reisen  hat  er  das  Angebot  abgelehnt.

      to  Africa  to  travel  had  he  the  offer  down-turned

      ‘He declined the offer to travel to Africa.’

   2. b. *Dass  er  inkompetent  sei,  hat  er  die  Behauptung zurückgewiesen.

      that  he  incompetent  be  had  he  the  claim rejected

      Intended: ‘He rebutted the claim that he was incompetent.’

The examples (27) are indeed ungrammatical. However, these examples involve negative embedding verbs, on which I will have more to say below. The contrast between (25)/(26) and (27) is noticeable, striking, and reminiscent of what was observed before regarding the sensitivity to selecting verbs in PP-extraction from DPs (recall the minimal pair (5)/(7)). In fact, the contrast seems to be identical, modulo the choices and availability of nouns and verbs that enter into forming a “semantically close” relationship.

3.2 Descriptive properties

Let me begin the descriptive survey by noting that NCC-extraction applies to various clause types so long as they can function as NCC in situ, that is, there is no principled restriction with respect to clause type. For example, while there is a restriction for verb second (V2) clauses to function as NCCs (roughly, the clause has to have intensional semantics), this restriction carries over to NCC-extraction. To survey the various clause types entering into NCC-extraction, I am presenting them side by side with their in situ or extraposed counterparts in the (b)-examples below.

The example (28a) involves a displaced V2-clausal complement, (29a) and (30a) are displaced non-finite clausal complements to nouns, and (31a) is a polar question complement to a noun, again, surfacing discontinuously from its associated noun Frage ‘question’.

1. (28)

   1. a. Wir  sollten  so  verfahren,  war  sie  der  Ansicht.

      we  should  so  proceed  was  she  the  opinion

   2. b. Sie  war  der  Ansicht,  wir  sollten  so  verfahren.

      she  was  the  opinion  we  should  so  proceed

      ‘She was of the opinion that we should proceed this way.’ V2 clause

2. (29)

   1. a. So  verfahren  zu  müssen,  war  sie  der  Ansicht.

      so  proceed  to  must  was  she  the  opinion

   2. b. Sie  war  der  Ansicht  so  verfahren  zu  müssen.

      she  was  the  opinion  so  proceed  to  must

      ‘She was of the opinion that we should proceed this way.’ non-finite clause

3. (30)

   1. a. ?Nach  Afrika  zu  reisen  hat  er  das  Angebot  angenommen.

      to  Africa  to  travel  had  he  the  offer  accepted

   2. b. Er  hat  das  Angebot  angenommen,  nach  Afrika  zu  reisen.

      he  had  the  offer  accepted  to  Africa  to  travel

      ‘He accepted the offer to travel to Africa.’ non-finite clause

4. (31)

   1. a. Ob  wir  so  verfahren  sollten,  stellte  sie  die  Frage.

      if  we  so  proceed  should  posed  she  the  question

   2. b. Sie  stellte  die  Frage,  ob  wir  so  verfahren  sollten.

      she  posed  the  question  if  we  so  proceed  should

      ‘She posed the question if we should proceed this way.’ interrogative

All these cases involve displaced sentential complements to nouns. A note of caution is in order. The terms sentential complements to nouns or NCC should, for lack of a better understanding, be understood in a functional sense. That is, the term is not necessarily supposed to mean that these CPs are complements to nouns in terms of X̄-theory or its successors. While in all the cases, the NCC functions as an argument to the noun, this should not be equated with the claim that NCCs are structural complements (i.e. sisters) to the noun. There are reasons to believe that they are not; rather, they appear to behave like adjuncts. For example, they can swap positions with relative clauses, given some tweaking of intonation (32).Footnote ⁹

1. (32)

   1. a. ?Sie  hat diejenige Aussage, mit der niemand gerechnet hatte,

      she  has the-one utterance with which no-one reckoned had

      dass die Proteste legitim sind, bekräftigt.

      that the protests legitimate are reinforced

   2. b. Sie  hat diejenige Aussage, dass die Proteste legitim sind, mit

      she  has the-one utterance that the protests legitimate are with

      der niemand gerechnet hatte, bekräftigt.

      which no-one reckoned had reinforced

      ‘She confirmed the statement by the protester that the protests are legitimate, which no one had expected.’

Moreover, postnominal genitive DP attributes – arguably actual X̄-theoretic complements – can intervene between NCCs and the associated nouns, as in (33a); placement of the NCC between the head noun and the genitive DP is sharply out (33b).

1. (33)

   1. a. Sie hat die Aussage der Demonstrantin dass die Proteste

      she has the utterance the.gen protester that the protests

      legitim sind bekräftigt.

      legitimate are reinforced

      ‘She confirmed the statement by the protester that the protests are legitimate.’

   2. b. *Sie  hat  die Aussage  dass  die  Proteste  legitim  sind  der

      she  has  the utterance  that  the  protests  legitimate  are  the.gen

      Demonstrantin bekräftigt.

      protester reinforced

This constraint holds, irrespective of whether or not the NCC is finite (as in (33)) or non-finite (as in (34)).

1. (34)

   1. a. Sie  hat  die  Bekundung  der  Demonstrantin  die Regeln

      she  has  the  avowal  the.gen  protester  the rules

      einzuhalten gutgeheißen.

      in.to.hold approved

      ‘She approved the avowal by the protester to observe the rules.’

   2. b. *Sie  hat  die  Bekundung  die  Regeln  einzuhalten  der

      she  has  the  avowal  the  rules  in.to.hold  the.gen

      Demonstrantin gutgeheißen.

      protester approved

The reason why the difference between finite and non-finite NCCs is relevant is because one could argue that (33) comes about by obligatory DP-internal extraposition of the finite CP as was observed at the clausal level (“heavy” material commonly ends up or has to end up in the Nachfeld – postfield – of the German clause). However, this reasoning is problematic in the face of (34): Extraposition of non-finite subordinate clauses is optional in the clause domain, and therefore extraposition would implausibly have to be obligatory in the nominal domain.

More generally, postnominal genitive DPs exhibit what looks like an adjacency requirement with the head noun. For example, PP-arguments of nouns like an die Kunden (‘to the costumer’) cannot swap positions with genitive DPs, as shown by the contrast between (35a) and (35b).

1. (35)

   1. a. die Lieferung der Ware an die Kunden

      the delivery the.gen goods to  the costumers

      ‘the delivery of goods to the costumers’

   2. b. *die  Lieferungan dieKunden  derWare

      the  deliverytothecostumers  the.gengoods

This adjacency requirement is likewise at work in (33b).Footnote ¹⁰ Concluding, the evidence suggests a structure like (36a) for NCCs, rather than (36b).

1. (36)

   1. a. [_N′ N’ CP]

   2. b. [_N′ N CP]

Nevertheless, I retain the term NCC for expository purposes and ask the reader to keep this in mind. In section 3.3, I will go into the issue of the impossibility of extracting other sentential adjuncts to DPs, namely relative clauses. Upon that, I will return to syntactic characteristics of NCC-extraction.

3.3 Restrictions on sentential extraction from DPs

As the name NCC suggests, the phenomenon exhibits restrictions with respect to the type of noun-associated clause that can be involved. Particularly, only clauses functioning as complements to nouns can figure in the construction, and not, say, a clause modifying the noun, like a relative clause.Footnote ¹¹

1. (37) *Die  wir  gut  nachvollziehen  konnten,  hatte  sie  eine  Auffassung

   which  we  well  understand  could  had  she  an  opinion

   vertreten.

   represented

   Intended: ‘She had defended a view which we could well understand.’

The deviance of this sentence is clear-cut. Arguably, this can be attributed to an independently operative ban on a surface order in which the relative clause precedes the associated head noun.

1. (38) Relative clauses must be linearly preceded by an overt instance of their associated head noun.

Indirect evidence for this restriction being in effect comes from fronting of remnant VPs to which a relative clause is adjoined. Full VP-fronting as in (39a) shows that right-adjunction of a relative clause is possible, provided the head-DP linearly precedes the relative clause in fulfilment of (38). Example (39b) is an instance of remnant VP-fronting, again with a relative clause-CP adjoined to the remnant VP, as the schematic analysis in (40) illustrates. Crucially, (39b) violates (38).

1. (39)

   1. a. Das Buch lesen, das ihr gerade geschenkt worden war, wollte

      the book read which her just given became was wanted

      Maria sofort.

      M. immediately

      ‘As for reading the book, which she had just been given, Mary wanted to do that right away.’

   2. b. *Lesen, das ihr gerade geschenkt worden war, wollteMaria das

      read which she just given became was wantedM. the

      Buch sofort.

      book immediately

2. (40) [_VP [_VP <das Buch> V=lesen] CP ] …das Buch

If (38) is the cause for the deviance of (39b), the condition might likewise be the reason for the ill-formedness of (37). This line of reasoning gives rise to an interesting expectation: (38) is silent on the possibility that NCCs linearly precede their associated head nouns. This captures the observed phenomenon under investigation in this article. However, it is possible to go beyond stating the observed facts and form the NCC-analogue of (39b). The expectation then is that it is at least better than their relative clause counterparts. This expectation is only partially met; remnant VP-fronting with a right-adjoined NCC is distinctively more acceptable than its relative clause kins.

1. (41)

   1. a. ?Geäussert,  dass  wir  so  verfahren  sollten,  hat  sie  die  Meinung

      uttered  that  we  so  proceed  should  has  she  the  opinion

      durchaus.

      indeed

      ‘She uttered the opinion that we should proceed this way.’

The improvement the example (41) represents could be due to a parsing error: Possibly, the matrix verb and the immediately following subordinate dass-clause are initially parsed as a predicate argument structure. Only upon hitting the DP die Meinung does this parse have to be corrected. But since the spurious clausal argument explicates the DP, the structure is intelligible, albeit arguably ungrammatical.

This line of reasoning is confirmed by the fact that once a verb is used that does not form Funktionsverbgefüge (light verb constructions) with the selected DP, acceptability decreases sharply.

1. (42)

   1. a. *Bewundert, den Mt. Blanc zu besteigen, hat Maria den Versuch dann

      admired the Mt. B;anc to climb has Maria the attempt then

      allerdings schon.

      on-the-other-hand ptcl

      Intended: ‘On the other hand, Mary admired the attempt to climb the Mt. Blanc.’

   2. b. *Gestellt,  ob wir  so  verfahren  sollten,  hat  sie  die  Frage.

      put  if we  so  proceed  should  has  she  the  question

      Intended: ‘She posed the question if we should proceed this way.’

The deviance of the examples (42) arguably has to do with two factors: Firstly, no parallel parsing and structural assignment along the lines suggested for (41) is possible, because the subordinate clauses cannot be taken to be arguments to the verbs bewundert ‘admired’ and gestellt ‘pose’ respectively. Thus, no illusion arises that the subordinate clause is such an argument. Secondly, the said verbs and the DPs den Versuch and die Frage do not relate closely, semantically, that is, they do not form a Funktionsverbgefüge. This is unlike the DP die Meinung and the verb geäussert in (44), which can essentially be paraphrased as ‘Indeed she was of the opinion that we should proceed this way.’

What cannot be concluded from this discussion is that (38) affects NCC-remnants as well: Firstly, (41) plainly evades (38). Secondly, all other acceptable cases of NCC-fronting evade (38). More needs to be said on the matter, but it should be clear that the possibility of forming Funktionsverbgefüge is one crucial factor in fronting NCCs, and likewise in applying remnant-VP-fronting with extraposed NCC. Example (42b) makes this particularly clear in that the embedded interrogative cannot function as the complement of the verb stellen ‘put’ (the cause for the deviance) while the meaning of the predicate (‘to ask’/‘pose the question’) is recoverable only at the point when the object DP die Frage gets parsed. The unacceptability of (42) can be traced to semantic or functional issues: sentential adjuncts which are explicative CPs may be extracted, while non-explicative CPs may not be extracted.

4.1 Ā-movement

This section seeks to establish that the dependency between the sentence-initial NCC and the (presumed) postnominal gap involves actual syntactic movement of the CP, and more particularly, Ā-extraction. That is, I argue that (43) derives from an independently available structure, along the lines of (44).

1. (43) Dass  wir  so  verfahren  sollten,  war  sie  [der  Ansicht  e].

   that  we  so  proceed  should  was  she  the  opinion

   ‘She was of the opinion that we should proceed this way.’

2. (44)

   

To do this, I apply a number of standard tests for Ā-movement, which will inform the analytical possibilities.

In variants of German that allow long distance movement, NCC-extraction can apply over boundaries of finite clauses.

1. (45) Dass wir so verfahren sollten, hat Peter gemeint, dass sie [der Ansicht e]

   that we so proceed should has Peter meant that she the opinion

   war.

   was

   ‘Peter said that she was of the opinion that we should proceed this way.’

This may serve as a first indication that movement is involved. NCC-extraction is, furthermore, sensitive to several kinds of islands, lending further support for the conclusion. One such island is a complex DP. Example (46) shows that NCC-extraction may not apply to a DP inside another complex (here: relativized) DP. The result is sharply unacceptable.

1. (46) *Dass wir so verfahren sollten, hat Peter [das Mädchen das der Ansicht e

   the we so proceed should has Peter the girl who the opinion

   war] getroffen.

   was met

Furthermore, to the extent that extraction out of a weak island is impossible, it is not possible to NCC-extract from an embedded polar question, as (47) shows.

1. (47) *?Dass wir so verfahren sollten, hat Peter sich gefragt, [ob Fritz der

   that we so proceed should has Peter refl asked if Fritz the

   Ansicht e war].

   opinion was

Finally, the subject condition restrains NCC-extraction, as (48) illustrates, again parallel to what was found with PP-extraction from DPs; see (49a)/(49b).Footnote ¹²

1. (48) *?Dass  wir  so verfahren  sollten,  hat [die  Auffassung e]  niemanden

   that  we  so proceed  should  has the  opinion  no-one

   überzeugt.

   convinced

2. (49)

   1. a. *Zum  ECP hat eine  Frage  den  Professor  verärgert.

      on-the  ECP has a  question  the  professor  annoyed

      Intended: ‘A question about the ECP annoyed the professor.’

      Grewendorf (Reference Grewendorf1989: 36)

   2. b. *Zur  Polenfrage  hat  eine  Versammlung  einige  Fragen

      on-the  Poland-question  has  an  assembly  some  questions

      geklärt.

      cleared

      Intended: ‘An assembly on the question regarding Poland solved some issues.’

      Fanselow (Reference Fanselow1987: 76)

In this context, it should be noted that passivization ameliorates the violation of the subject condition; see (50).

1. (50) ?Dass  wir  so  verfahren  sollten,  wurde  [die  Hoffnung e] geäussert.

   that  we  so  proceed  should  was  the  hope uttered

   ‘The hope was expressed that we would proceed this way.’

This is expected if, in passives, the grammatical subject is an underlying object. And again, such observations are familiar from corresponding effects with PP-extraction from DPs (see, e.g., Chomsky Reference Chomsky, Freidin, Otero and Zubizarreta2008: 147/153 on wh-extraction from derived subjects).

Does NCC-extraction have to target the “prefield” of the clause? Extraction of finite NCCs (51) must not target the “middle field”,Footnote ¹³ suggesting that this is so.

1. (51) *?War  sie  dass  wir  so  verfahren  sollten  [der  Ansicht e]?

   was  she  that  we  so  proceed  should  the  opinion

   Intended: ‘Was she of the opinion that we should proceed this way?’

That is, one could argue that there is no NCC-scrambling. However, notice that scrambling of a non-finite NCC leads to a distinct improvement, (52).

1. (52) ?Hat  sie  so  verfahren  zu  können  wirklich [die  Hoffnung e] geäussert?

   has  she  so  proceed  to  can  really the  hope uttered

   Intended: ‘Did she really express the hope that we could proceed this way?’

Plausibly, then, the contrast between (51) and (52) can be traced to the independently operative ban on finite subordinate CPs in the middle field. If so, and if NCC-extraction patterns like regular Ā-movement, one can reason as follows: V2-clauses are known to be distinctively worse in the middle field than, for instance, finite verb-final clausal complements. It is then expected that scrambling of NCCs is limited to the extent that clausal complements can appear at all in the middle field and that scrambling of a V2-NCC is the least acceptable among the clause types. The expectation is met as (53) shows.

1. (53) *War  sie  wir  sollten  so  verfahren  wirklich  [der  Ansicht e]?

   was  she  we  should  so  proceed  really  the  opinion

Notice that the possibility of scrambling non-finite NCCs, as shown in (52), appears to be a problem for an analysis of the NCC that involves base-generation of NCC in the left periphery in addition to a silent operator undergoing Ā-movement from within the DP. Essentially, NCCs are not confined to the clausal left periphery, just as PPs extracted from DPs are not, as (54) shows.

1. (54) dass  darüber  wohl  niemand ein  Buch  gelesen  hat

   that  there-over  ptcl  no one a  book  read  has

   ‘that no one read a book about that’

To conclude this section, based on several established diagnostics, I have argued that NCC-extraction is an instance of Ā-movement. I will now move on to the issue of the choice of verbs selecting the DP containing the gap.

4.2 Verb classes

PP-extraction from DPs exhibits a sensitivity to the selecting verb as examples (5)/(7)/(23) show. NCC-extraction from DPs features a comparable sensitivity. Acceptable cases include (55a)–(56a) with the verb-DP combinations die Frage stellen (‘pose the question’) and den Versuch unternehmen (‘make the attempt’). These contrast with instances of NCC-extraction with negative verbs like zurückweisen ‘reject’ (lit. back-show), (55b), and abbrechen ‘cancel’, (56b), which are ungrammatical.

1. (55)

   1. a. Ob  wir  so  verfahren  sollten,  stellte  sie  die  Frage.

      if  we  so  proceed  should  posed  she  the  question

      ‘She asked if we should proceed this way.’

   2. b. *Ob  wir  so  verfahren  sollten,  wies  sie  die  Frage  zurück.

      if  we  so  proceed  should  showed  she  the  question  back

      Intended: ‘She rejected the question if we should proceed this way.’

2. (56)

   1. a. Den  Mt. Blanc zu besteigen hatten sie den Versuch unternommen.

      the  Mt. Blanc to climb had they the attempt undertaken

      ‘They made the attempt to climb the Mt. Blanc.’

   2. b. *Den  Mt. Blanc zu besteigen hatten sie den Versuch abgebrochen.

      the  Mt. Blanc to climb had they the attempt cancelled

      Intended: ‘They cancelled the attempt to climb the Mt. Blanc.’

As before, this parallels instances of PP-extraction from DPs. It is no surprise that Funktionsverbgefüge enhance the acceptability of the NCC-extraction examples. Such constructions feature a semantically “light” or bleached verb or an auxiliary verb, in addition to a semantically more content-full noun. For example, den Versuch unternehmen (the attempt undertake, ‘to attempt’) conveys effectively the meaning of the verb translation. The reason for the enhanced acceptability of NCC-extraction thus appears to be that acceptable DP-V-sequences allow for the paraphrasing with a single verb.

1. (57)

   1. a. Dass wir so verfahren, dachte/meinte sie (=war sie der Ansicht).

      that we so proceed thought/meant she (=was she the opinion)

   2. b. Dass die Leute noch erscheinen würden, hoffte sie (=hatte sie die Hoffnung).

      that the people still appear would hoped she (=had she the hope)

   3. c. Ob wir so verfahren sollten, fragte sie (=stellte sie die Frage)

      if we so proceed should asked she (=posed she the question)

   4. d. Den Mt. Blanc zu besteigen, hatten sie versucht

      the Mt. Blanc to climb had they attempted

      (=unternahmen sie den Versuch)

      (=undertook they the attempt)

Essentially then, the types of semantic relationships N-V-sequences can entertain facilitate or inhibit extraction from DPs quite generally, and NCC-extraction is no exception. A thorough investigation into the types of relationships between the verbs and the nouns is an important research task, but beyond the scope of this article.

4.3 (In)definiteness/(Un)specificity

The Specificity Condition is violated in all examples hitherto considered. That is, NCC-extraction from definite DPs is permitted, and indeed, required. In this respect, NCC-extraction contrasts with PP-extraction, of which (58) is a German example.

1. (58) *Worüber  hat  Fritz  {das/dieses/Chomskys}  Buch  gelesen?

   where-over  has  Fritz  the/this/Chomsky's  book  read

The fact that all NCC-extraction cases violate the Specificity Condition as such raises questions. At the same time, it is unsurprising, given the co-occurrence pattern with different determiner types of NCCs contiguous with their associated nouns: They all require definite determiners.Footnote ¹⁴

1. (59)

   1. a. Sie  war  der  Ansicht,  dass  wir  so  verfahren  sollten.

      she  was  the  opinion  that  we  so  proceed  should

   2. b. {#/*}Sie war einer Ansicht, dass wir so verfahren sollten. / {#/*}Sie

      she was an opinion that we so proceed should she

      stellte eine Frage, ob wir so verfahren sollten.

      posed a question if we so proceed should

While general conditions of possible specificity violations remain to be understood, the violation in NCC-extraction appears to follow the rule saying NCCs generally require a definite DP. I will move on to possessed DPs. The introduction of prenominal possessors within the complex DP leads to a degradation of NCC-extraction.

1. (60)

   1. a. Maria bewunderte [Edes Versuch, [den Mt. Blanc zu besteigen]].

      Maria admired Ede's attempt the Mt. Blanc to climb

      ‘Mary admired Ede's attempt to climb the Mt. Blanc.’

   2. b. ?*[Den Mt. Blanc zu besteigen] bewunderte Maria [Edes Versuch].

      the Mt. Blanc to climb admired Maria Ede's attempt

The deviance represented by (60b) is puzzling, given the acceptability of (60a). I have no solution to offer.

Returning to the Specificity Conditions with definite articles, notice that violations, even with PP-extraction, are not unheard of, see (61a) and (61b) (from Davies and Dubinsky Reference Davies and Dubinsky2003: 18, Huang Reference Huang1982 (cited in Chomsky Reference Chomsky1986: 80)).

1. (61)

   1. a. What did they observe the production of?

   2. b. Which city did you witness the destruction of?

According to Davies and Dubinsky (Reference Davies and Dubinsky2003), the choice of noun is crucial in lifting the Specificity Condition: Extraction out of definite DPs headed by process nouns like production or destruction is fine, as opposed to such with concrete or result nouns.

Davies and Dubinsky (Reference Davies and Dubinsky2003: fn. 7) consider the possibility that definites function differently in process nominals than with result nouns in a number of other respects as well. First, the definite determiner is often optional, as in (62).

1. (62) Scientists observed revolution of a satellite around Mars.

Secondly, process nouns introduced by the definite article often do not require presupposed existence, as in (63), that is, the familiar semantic property often subsumed under existential presupposition is not adhered to with process nouns.

1. (63) To account for the observed solar wobbles, some scientists have posited (the) revolution of the sun around some unknown gravity source.

Finally, definite subject-DPs with process nouns may occur in existential constructions, which usually dislike definite DPs, as shown in (64).

1. (64) And then, there was the revolution of a satellite around Mars on the video transmission, which the scientists had not expected.

These observations indicate that the definite article plays a different role in process nouns. Possibly, process nouns might give rise to an expletive article reading, in the spirit of Longobardi (Reference Longobardi1994).

Can these tests be replicated with nouns which take NCC-complements? Caution is required before making such transfers, because (a) we are dealing with a different semantic category of noun and (b) German definite articles behave somewhat differently from English ones. Certainly, omission of the article (in singular nouns) in NCC-contexts yields degradation, as in (65).

1. (65) *Sie  hat  [Behauptung  dass  er  inkompetent  sei]  gemacht.

   she  has  claim  that  he  incompetent  be  made

Regarding existential constructions, one can observe that NCC-DPs can figure in them, while other definite DPs cannot (compare (66) with (67a) and the latter with (67b)).Footnote ¹⁵

1. (66) Und dann gab es [die Behauptung dass er inkompetent sei], womit niemand

   and then give it the claim that he incompetent be where with no one

   gerechnet hatte.

   reckoned has

   ‘And then there was the claim that he was incompetent, which no one had expected.’

2. (67)

   1. a. *?Gestern gab es [den Jungen] in der Mensa, womit niemand

      yesterday give it the boy in the cafeteria where with no one

      gerechnet hatte.

      reckoned has

   2. b. Gestern gab es [einen Jungen] in der Mensa, womit niemand

      yesterday give it a boy in the cafeteria where with no one

      gerechnet hatte.

      reckoned has

      ‘Yesterday, there was a boy in the canteen, which no one had expected.’

So NCC-DPs in German do exhibit certain parallels to English process nouns, but these are not consistent but confined to existential contexts. The definite article is clearly not optional for NCC-DPs, unlike what was observed for process nouns in English.

Considering determiners other than articles, NCC-extraction is licit if the associated DP is introduced by a quantifier (68).

1. (68) Dass er das Handout noch schaffen würde, hatte er alle/?jede Hoffnung

   that he the handout still finish would had he all/every hope

   schon aufgegeben.

   already up-given

   ‘He had already given up all hope that he would still finish the handout on time.’

And again, this is quite expected, given that these quantifiers can figure in NCC-context when no extraction takes place. The emerging generalization is that NCC-extraction is not subject to the Specificity Condition. Instead, it seems as if NCC-extraction is possible with DPs that can also figure in non-extraction NCC-contexts. The observations complicate the already intricate picture of extraction from DP-phenomena. However, within the confines of NCC they appear to be quite consistent and expected, given the in-situ counterparts. Let me next move on to reconstruction effects.

4.4 Reconstruction effects

Connectivity effects commonly serve as a diagnostics for movement, that is, they indicate the base-generated position of a phrase; the absence of connectivity effects can be taken as an indication of the absence of movement. Reconstruction effects are instances of such connectivity effects. The well-established contrast between Left Dislocation (69a) and Hanging Topic Left Dislocation (69b) in German may serve as an example (taken from Grohmann Reference Grohmann2000: 75–76).

1. (69)

   1. a. Diesen  Satz,  den  mag  ich besonders.

      this.acc  sentence,  that.acc  like  I especially

      ‘This sentence, I like (it) especially.’

   2. b. Dieser  Satz,  den  mag  ich besonders.

      this.nom  sentence,  that.acc  like  I especially

      ‘This sentence, I like it especially.’

Both constructions involve a left-peripheral XP, the DPs diesen Satz or dieser Satz respectively. This XP is resumed by a demonstrative pronoun, den (‘that’) in (69a)/(69b). The displaced DP matches in case with the demonstrative in Left Dislocation, while in Hanging Topic Left Dislocation, the DP in the left-peripheral position invariably bears nominative. Case betrays the DPs clause-internal base-position in Left Dislocation, because the verb mögen (‘like’) governs accusative. This is a connectivity effect. No such effect is observable in Hanging Topic Left Dislocation. Semantic effects (not shown here) like reconstruction for Binding theoretic Principles (Chomsky Reference Chomsky1981) and for variable binding further corroborate the conclusion, namely that in Left Dislocation, the XP is syntactically associated with a clause-internal gap. In Hanging Topic Left Dislocation, no such reconstruction is observed, which commonly is taken to indicate the absence of a movement dependency between the XP and a clause-internal position.

I take these diagnostics as given and apply them to NCC-extraction. If NCC-extraction exhibits reconstruction effects, the conclusion receives support that the left-peripheral NCC is associated with the postnominal position by a movement dependency. These diagnostics do not conclusively show that movement is involved; however, it needs to be shown how they can be made to follow from an analysis involving base generation of the NCC in the left-peripheral position.

As it turns out, there are effects of reconstruction for Principle C of the Binding Theory, as (70a) shows;Footnote ¹⁶ (70b) is the corresponding in-situ case. In both cases, the pronominal and the R-expression Peter cannot receive a co-construal reading.

1. (70)

   1. a. *Dass  Peter_i  so  verfahren  sollte,  äusserte  er_i  die  Meinung.

      that  P.  so  proceed  should  uttered  he  the  opinion

   2. b. *Er_i  äusserte  die  Meinung,  dass  Peter_i  so  verfahren  sollte.

      he  uttered  the  opinion  that  P.  so  proceed  should

      * ‘He_i was of the opinion that Peter_i should proceed this way.’

If the left-peripheral CP in (70a) is base-generated in its surface position or a hanging topic, respectively, these cases are difficult to explain.

Instances of reconstruction for quantificational binding support a similar conclusion. The example (71a) allows for a construal in which the pronoun embedded in the left peripheral clause covaries with the universal quantifier in the main clause.

1. (71)

   1. a. Dass  er_i  so  verfahren  sollte,  war  jeder_i  der  Ansicht.

      that  he  so  proceed  should    was  everyone  the  opinion

   2. b. Jeder_i  war  der  Ansicht,  dass  er_i  so  verfahren  sollte.

      everyone  was  the  opinion  that  he  so  proceed  should

      ‘Everyone_i was of the opinion that he_i should proceed this way.’

Reconstruction for Principle C and facts from variable binding suggest that dislocated NCCs involve syntactic movement.

Scrambling of NCCs will be considered next. There is a distinct effect for reconstruction of Principle C in contexts where the NCC is scrambled; see (72a), where the in-situ counterpart is given as (72b) for comparison.

1. (72)

   1. a. *Dass [Peter_i fein anzuziehen] er_i [die Absicht] gehabt hatte, war

      that P finely to-dress he the intention had had was

      niemandem aufgefallen.

      no one noticed

   2. b. *Dass er_i [die Absicht [Peter_i fein anzuziehen]] gehabt hatte, war

      that he the intention P finely to-dress had had was

      niemandem aufgefallen.

      no one noticed

      *‘No one had noticed that he_i had had the intention to dress Peter_i nicely.’

To repeat, this suggests that the scrambled NCC is not base-generated in its surface position, because under base-generation the Condition C-effect is puzzling: No c-command would hold between er and Peter at any point in the derivation. Moreover, it shows that scrambled NCCs are not subject to anti-reconstruction effects, their presumed syntactic status of an adjunct notwithstanding.

Comparing reconstruction of Principle C with the state of affairs of reconstructing for bound variable readings corroborates the previous observation. As (73b) shows, reconstruction of a scrambled NCC for a bound variable construal is a possibility.

1. (73)

   1. a. Dass  [seine_i  Schwester mitzunehmen] jeder_i [die Absicht] hatte,

      that  his  sister with to take everyone the intention had

      war klar.

      was clear

   2. b. Dass  jeder_i  [die  Absicht  [seine_i  Schwester mitzunehmen]] hatte,

      that  everyone  the  intention  his  sister with to take had

      war klar.

      was clear

      ‘It was clear that everyone_i had the intention to bring along his_i sister.’

Again, these observations support the previous conclusion that NCCs reconstruct in a DP-internal position.

As a final reconstruction diagnostic, Negative Polarity Items (NPI) can show up in displaced NCCs, with a negative element structurally intervening between the NCC's surface position and the DP-internal gap. I use the NPI jemals (‘ever’).

1. (74)

   1. a. ?Dass  Fritz  jemals  wiederkommen  würde,  hat  niemand  die

      that  Fritz  ever  back come  would  has  no-one  the

      Auffassung vertreten.

      opinion represented

   2. b. Niemand  hat  die  Auffassung  vertreten,  dass  Fritz  jemals

      no-one  has  the  opinion  represented  that  Fritz  ever

      wiederkommen würde.

      back come would

      ‘No-one was of the opinion that Fritz would ever return again.’

If the NCC is base-generated within the DP, the negative subject c-commands and licenses the NPI. If the NCC is base-generated in the left-periphery, (74a) should be bad.

To conclude, facts from reconstruction for Principle C, variable binding and NPI-licensing are not expected if the NCC is base-generated in its surface position, while they fall in place if movement is posited to characterize the dependency between the NCC and the DP-internal gap.

4.5 A tentative conclusion

I take both the possibility of scrambling and the reconstruction effects to tentatively indicate that the NCC-extraction does involve movement of the actual clause from a DP-internal position. That is, for now I disregard the possibility that the left-peripheral clauses are, for example, hanging topics, maybe associated with a silent operator in SPEC-CP of the host-clause. I would like to emphasize and repeat, though, that the conclusion is tentative, because it hinges on the delicate judgments involved in the construction more generally.

5.1 Explicative vs. implicative complement to nouns

Based on a number of criteria, Fabricius-Hansen and von Stechow (Reference Fabricius-Hansen and von Stechow1989) establish a distinction between two types of NCCs, which they name implicative (75a) and explicative (75b) nominal complementation respectively.

1. (75)

   1. a. Die  Fähigkeit,  schön  zu  singen,  ist beneidenswert.

      the  skill  beautifully  to  sing  is enviable

      ‘The skill of singing beautifully is enviable.’

   2. b. Wladimirs Ärger (darüber), seinen Fahrradhelm verloren zu haben,

      Wladimir's anger there-over his bike-helmet lost to have

      war gross.

      was great

      ‘Wladimir's anger over losing his cycling helmet was great.’

Explicit NCCs are characterized by their possibility of coming with a pronominal element which functions as a correlate to the NCC. Example (75b) shows this with the possibility of a postnominal PP, which has an R-pronominal correlate in it.Footnote ¹⁷ Implicit NCCs, by contrast, do not straightforwardly have this option. Another important difference between the two is the following: The head noun can function as a predicate in explicative NCCs (76a), while this is impossible in implicative NCCs (76b) (Fabricius-Hansen and von Stechow Reference Fabricius-Hansen and von Stechow1989: 175).

1. (76)

   1. a. Schön  zu singen ist eine Fähigkeit, die nicht jeder hat.

      beautifully  to sing is a skill which not everybody has

      ‘Not everyone has the skill to sing beautifully.’

   2. b. *Seinen Fahrradhelm verloren zu haben ist ein Ärger,  der     gross

      his   cycling helmet lost   to have  is  an  anger   which  great

      war.

      was

      Intended: ‘The anger over losing his cycling helmet was great.’

Is there a correlation between the distinction between implicative and explicative NCCs and the possibility of extracting the sentential complement? Sticking to the relevant examples, there is no discernible difference.

1. (77)

   1. a. *Schön  zu  singen  hat  sie  die  Fähigkeit.

      beautifully  to  sing  has  she  the  skill

      Intended: ‘She has the skill to sing beautifully.’

   2. b. *Seinen  Fahrradhelm  verloren  zu  haben  hatte  er  den  Ärger.

      his  cycling helmet  lost  to  have  had  he  the  anger

      Intended: ‘He had the trouble of having lost his cycling helmet.’

These observations do not allow for strong conclusions, but deserve further investigation.Footnote ¹⁸

5.2 A remark on an analysis

This section touches on aspects that an analysis of NCC-extraction from DPs should take into consideration. That is, in lieu of an analysis it offers hints at an analysis. Pending a better understanding of the factors involved, I believe that a full-fledged formal analysis is premature. Yet, I would like to pursue an intuition that might hint at something like the common denominator of the phenomena described in this article, or at least capture a property most of the phenomena have in common.

There is a recurring theme when it comes to analyzing extraction from DP. The general idea is that the islandhood of a constituent can be lifted when a bound pronominal occupies a left peripheral position of the island. Ingredients of the approach by Davies and Dubinsky (Reference Davies and Dubinsky2003: 32) are given in (78).

1. (78) NCC hatte sie₁ [_DP die [_FP PRO₁ F [_NP  Hoffnung t_NCC]]]  geäußert.

   NCC had  she   the        hope      uttered

The subject of the verb controls the attitude or mental state of the subject of the embedded (noun-containing) clause (result nouns for Davies and Dubinsky Reference Davies and Dubinsky2003: 32, but deverbal nominals for the cases at hand): One way of encoding this is to use PRO. PRO would then occupy the specifier position of an FP, which is sandwiched between DP and NP. Moreover, PRO would be bound by the clausal subject. Next, one could state that extraction from DP is licit iff PRO is locally bound.

Similar island-ameliorating effects of bound pronouns in island-edge positions have recently been reported by Grano and Lasnik (Reference Grano and Lasnik2018) (see Huang Reference Huang2018 on DPs). Grano and Lasnik's (Reference Grano and Lasnik2018) assumptions include the following:

• • Bound pronouns optionally enter the derivation with unvalued ϕ-features.

• • Phases are in part defined by convergence, so that unvalued features may void a phase and in effect extend the locality domain of syntactic operations.

There is an undesirable empirical prediction that follows from the analysis if one adopts Grano and Lasnik's (Reference Grano and Lasnik2018) approach to account for NCC-extraction from DPs. Remember that passives improve NCC-extraction and arguably also unaccusatives. However, these do not obviously involve external arguments which can function as binders. For the analysis to work, one would have to assume that PRO must be bound by a structurally represented silent by-phrase or some counterpart in unaccusatives. While this might work technically, it neither promises to shed light on the reasons why extraction is possible in some cases but not in others, nor is such an approach plausible.