On the Nature of Proper Government

  • Howard Lasnik
  • Mamoru Saito
Part of the Studies in Natural Language and Linguistic Theory book series (SNLT, volume 20)


In the last few years, there has been extensive investigation of the Empty Category Principle (ECP) — the requirement that an empty category must be properly governed. Attention has particularly focused on the ECP as a constraint on the representations resulting from Wh Movement (see for example Jaeggli (1980a), Chomsky (1981a), Kayne (1981a), Stowell (1981), Huang (1982)). In this article we will address such issues as (a) the definition of “proper government,” (b) the level(s) of representation to which the ECP applies, and (c) the relationship between the ECP and other principles. We will integrate this discussion into a treatment of Wh Movement in general and wh-in-situ in particular.


Matrix Comp Embed Clause Original Trace Initial Trace Embed Question 
These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.


Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.


  1. 1.
    See, for example, Baker (1970), Chomsky (1973), May (1977), Aoun, Hornstein, and Sportiche (1981), Kayne (1981b). Baker and Chomsky did not explicitly make use of an LF level, but their observations and analyses can easily be translated into a framework with such a level.Google Scholar
  2. 2.
    We will return to the issue of whether this movement is substitution or adjunction. As (3) illustrates, Polish seems to lack such LF movement.Google Scholar
  3. 3.
    We have slightly altered the example for the sake of clarity. The example presented in AHS had certain complications irrelevant to the immediate question.Google Scholar
  4. 4.
    See Wachowicz (1974), Giejgo (1981), and the last section of Toman (1982) for arguments that all Wh Movement is to a Comp in Polish. Note that our argument is not affected even if co in (11) is not in a Comp but instead is adjoined to the embedded S, a Qossibility considered by Toman. Here, we will simply assume the every wh must be in an A-position at S-structure in Polish.Google Scholar
  5. 5.
    The Comp feature [+wh] is essentially indistinguishable from the Q morpheme of Baker (1970), as far as we can tell.Google Scholar
  6. 6.
    Co cannot syntactically move to the Comp immediately under zastanawia,since Polish lacks syntactic movement out of indicative complements. *Co Maria powiedzia{a, ze Piotr kupit t what Maria said that Piotr bought What did Maria say that Piotr bought?’Google Scholar
  7. 7.
    This contrast, along with others similarly involving wh-in-situ that we will discuss later, seems to disappear in echo questions. That is, (i) is fully acceptable as an echo question. (i) What did who see? This suggests that the focus wh in such constructions does not undergo LF movement.Google Scholar
  8. 8.
    These definitions are taken from Aoun and Sportiche (1981). See also Chomsky (1981a).Google Scholar
  9. 9.
    We will see that in (21) what 2 properly governs t2 as well.Google Scholar
  10. 10.
    Hendrick and Rochemont (1982) point out that certain superiority effects do not follow from the ECP. Examples (i) and (ii) illustrate such a “pure” superiority phenomenon. (1) Who did you tell to read what? (ii) ?*What did you tell who to read? In (ii) both the syntactic trace of what and the LF trace of who are lexically governed (by read and tell,respectively). Hence, there is no ECP violation, yet the example is worse than (i), where the “superior” wh-phrase moved in the syntax. See Pesetsky (1982) for further discussion. Another phenomenon, which we will not address here, is illustrated by the contrast between (iii) and (iv). (iii) *What did who buy at the store? (iv) ?What did who buy where? As noted by Chomsky (1981a, 238), “when a wh-phrase appears throughout,” as in (iv), there is substantial improvement. See Kayne (1983) and Pesetsky (1982) for discussion.Google Scholar
  11. 11.
    Here we ignore a number of potentially relevant details, for example, the D-structure position of why. We tentatively assume that why is immediately dominated by S in D-structure.Google Scholar
  12. 12.
    Locatives and temporals are superficially a problem for this analysis. In general, these phrases are not complements; yet, unlike, why, they are grammatical in situ. Contrast (i) and (ii) with (27b). (i) What did you buy where? (ii) What did you buy when? See Huang (1982, chap. 7) for a possible solution to this problem.Google Scholar
  13. 13.
    Since the matrix clause lacks a question marker, nani cannot take matrix scope. If no `Q’ is added at the end of the sentence, nani can take matrix scope and the sentence is grammatical.Google Scholar
  14. 14.
    The well-formedness of (35) also shows that LF Wh Movement is not subject to the Crossing Constraint. For recent discussion of this constraint, see Pesetsky (1982).Google Scholar
  15. 15.
    Since overt complementizers appear on the right in Japanese, we assume that LF Wh Movement operates rightward.Google Scholar
  16. 16.
    As noted by Huang (1982), adjuncts do not display that-trace effects. This is true of both LF movement and syntactic movement. For example, (i) sharply contrasts with (ii), whereas (iii) and (iv) do not contrast. This is unexpected since lexical government does not obtain in either paradigm. (i) Who, [do you think [t,[t, left early]]] (ii) *Who, [do you think [t, that [t, left early]]] (iii) Why2 [do you think [t2 [he left early t21]] (iv) Why2 [do you think [t2 that [he left early t2]]] We return to this issue in section 4.Google Scholar
  17. 17.
    It has been proposed (for example, by Kayne (1980) and Stowell (1981)) that bridge verbs can govern traces in Comp.Google Scholar
  18. 18.
    (41a) is not a fully natural Japanese sentence. But we believe that this is due to the processing difficulty inherent in center-embedded constructions rather than to a grammatical principle.Google Scholar
  19. 19.
    Although it is occasionally claimed that the status of (51) does depend on the presence or absence of that, we will assume that AHS are correct on this point. See Kayne (1981b, 324) for the contrary position.Google Scholar
  20. 20.
    Note that we have not provided any evidence against a parametrized version of (10), and in fact we are aware of no compelling evidence. That is, one could specify that LF Comp-to Comp movement is possible in Chinese, Japanese, and Polish, but not in English. Though English does generally lack such movement, this property can be deduced as a theorem from filter (13), the ECP, and the requirement that even the matrix Comp be specified as [± wh] in English.Google Scholar
  21. 21.
    Noam Chomsky (personal communication) has suggested as an alternative that the complement S’ is extraposed out of the VP in LF. We will not pursue the implications of this proposal here.Google Scholar
  22. 22.
    Another possibility, which we will not pursue, is that VP is not a maximal projection, in particular, that S’ is a projection of V. See Kayne (1981a) for relevant discussion.Google Scholar
  23. 23.
    The assumption that the head can be governed from outside its maximal projection is now a rather standard one. See, for example, Kayne (1980), Belletti and Rizzi (1981). It is argued in Elliott (1982) that this holds for antecedent government as well as for lexical government.Google Scholar
  24. 24.
    We assume that the objective Case-marker is stranded by the LF movement of the object and functions as a proper governor for the trace.Google Scholar
  25. 25.
    James Huang (personal communication) informs us that the Chinese counterpart of (64a) is almost fully acceptable and that, contrary to his judgement in Huang (1982), the Chinese counterpart of (63a), though not perfect, does not have the entirely unacceptable status of an ECP violation.Google Scholar
  26. 26.
    For our purposes, it is irrelevant whether co was in the embedded Comp or in some other sentence-initial position at S-structure.Google Scholar
  27. 27.
    Chomsky (1981a) and Stowell (1981) have made a proposal with a similar effect. They suggest that under certain circumstances a trace in Comp may be deleted. Though this proposal may have the same salutary effect on the examples we are considering, our proposal eliminates a stipulation.Google Scholar
  28. 28.
    See May (1977a) and references cited there for discussion of this constraint.Google Scholar
  29. 29.
    See section 4 for detailed discussion of alternatives to this conclusion.Google Scholar
  30. 30.
    As mentioned earlier, there is no that-trace effect with adjuncts, even in English. In section 4 we will offer an explanation of this phenomenon.Google Scholar
  31. 31.
    Below, we will see that at S-structure English has no Comp’s of the form (87). Hence, the question of the trace in (87) allowing the Comp to be a proper governor will not even arise.Google Scholar
  32. 32.
    Thus, in addition to the optionality of trace, we are at this point adopting the intermediate trace deletion of Chomsky (1981a) and Stowell (1981), but as an instance of a more general rule.Google Scholar
  33. 33.
    The potential implications of this fact for the analysis of long-distance Wh Movement of adjuncts were pointed out by Noam Chomsky (class lectures, Fall 1982).Google Scholar
  34. 34.
    This assumes that when adjunction is to S, that S-node does not count as a Subjacency bounding node for that particular operation. This follows immediately if Subjacency is a constraint on movement rather than on representation.Google Scholar
  35. 35.
    It might be thought that movement of who in this case violates Subjacency, thus causing the ungrammaticality of (106). However, (106) seems far worse than normal Subjacency violations. Further, (i) is just as bad, even though whether does not move. (i) *Why, do you wonder [whether [John left t1]]Google Scholar
  36. 36.
    An alternative, which we will not explore here, is that clause-internal adjunct movement leaves no trace.Google Scholar
  37. 37.
    Given (109), (106) is ruled out exactly like its Chinese structural counterpart in (45). Regardless of whether Wh Movement of an adjunct takes place in syntax or LF, the trace must be properly governed at LF.Google Scholar
  38. 38.
    This follows immediately in a formalization of phrase structure such as that in Chomsky (1955) or Lasnik and Kupin (1977), since in those formalizations the following equivalence holds: S S’Google Scholar
  39. 39.
    We noted that when Comp is deleted, the S’ loses its head. It is possible that this results in S’ losing its categorial status entirely, a form of S’-Deletion. Since S’ is a barrier to antecedent government, the reduced S’ might be permeable to such government. This suggests an alternative treatment of wh-adjunct movement. In the following LF configuration, the trace of why is directly antecedent-governed by why itself, since there is no intervening S’. No adjunction is necessary. As in the analysis in the text, an adjunct cannot be extracted out of an embedded question, since deletion of a Comp with semantic content (a whoperator) is impossible.Google Scholar
  40. 40.
    Kayne (1981b) discusses this phenomenon in terms of the Nominative Island Condition. This aspect of the NIC was later incorporated into the ECP.Google Scholar
  41. 41.
    A contrast similar to the one in (120) obtains in the case of extraction from an adjunct: (i) a. ?*What did you leave [before buying t] b. *Why did you leave [before buying it t] We speculate that the adjunct in (ib) has the following structure: (ii) [pp before [Np Es, Is PRO buying it t]]]] Then, even if why moves from Comp to Comp in (ib), the intermediate trace violates the ECP since an NP node intervenes between it and its antecedent. Thus, we predict the contrast in (i): both examples violate Subjacency, but (ib) additionally violates the ECP. Although the contrast between (ia) and (ib) may not be completely clear in English, the contrast is quite sharp in the case of LF Wh Movement in Japanese. (iii) a. Dare-ni atte kara, uti-ni kaetta no? who-to meet after house-to went-back Q Lit.: `Who did you go home after meeting t?’ b. *John-ni naze atte kara, uti-ni kaetta no? why Lit.: `Why did you go home [after meeting John t]?“ Since the Subjacency Condition does not apply in LF, the ungrammaticality of (iiib) cannot be attributed to this condition. The contrast in (iii) thus indicates that an extraction of a noncomplement out of a noncomplement results in an ECP violation.Google Scholar
  42. 42.
    In fact, Spanish seems to lack adjunct preposing entirely. Thus, such a derivation is independently excluded. See Torrego (1983) for relevant discussion.Google Scholar
  43. 43.
    Huang regards this as a serious problem for the ECP. As we will show, under our revision, the problem disappears.Google Scholar
  44. 44.
    The fact that (130) is not as bad as expected becomes even clearer when we compare it with (i). (i) *Who thinks [that John won the election why]Google Scholar
  45. 45.
    See den Besten (1978), Stowell (1981), and Pesetsky (1982) for relevant discussion.Google Scholar
  46. 46.
    If this approach is correct, then c-command as defined in (56) is a necessary condition for antecedent government but not for government. That is, even in (134) Intl must govern the subject for nominative Case assignment to take place, yet c-command does not obtain.Google Scholar
  47. 47.
    Jaeggli (1980b) accounts for this phenomenon by stating that a non-Case-marked trace does not block contraction. This is approximately correct, descriptively, but we will show that to the extent that it is correct, it follows from deeper principles.Google Scholar
  48. 48.
    Note that this requires that Subjacency be a constraint on movement rather than on representation. Pesetsky (1982, chap. 3) thus assumes that Subjacency is a condition on movement. We will return to this issue.Google Scholar
  49. 49.
    One way to resolve this paradox might be to allow the Delete a case of Affect a to apply in the PF component, prior to contraction. This possibility seems dubious to us, however, since even a subject trace could then presumably be deleted, incorrectly allowing (139b). This is so since neither the Projection Principle nor the prohibition against vacuous quantification seems a plausible PF constraint.Google Scholar
  50. 50.
    This is the only type of case we are aware of where no principle prevents LF Comp-toComp movement in English.Google Scholar
  51. 51.
    As is well known, there are additional constraints on contraction. That is, adjacency is a necessary but not sufficient condition. For recent discussion, see Postal and Pullum (1982), Bouchard (1982), Pesetsky (1982), and Milsark and Safir (1983).Google Scholar
  52. 52.
    For discussion of the adjacency requirement on Case assignment, see Stowell (1981) and Chomsky (1981a).Google Scholar
  53. 53.
    Note that the S’-Deletion analysis of wh-adjuncts suggested in footnote 39 faces a potential problem with respect to wanna contraction (cf. (140)). For contraction to take place, there must not be an intermediate trace at S-structure. The S’-Deletion analysis would then require that how directly antecedent-govern its trace, by virtue of the application of S’-Deletion. But if, as argued in Chomsky (1982), condition B of the binding theory applies at the level of LF, (140) will represent a violation since want will govern PRO and PRO will be bound in its governing category, an illicit situation for a pronominal. If, on the other hand, S’-Deletion does not apply, an intermediate trace will be needed at LF. Hence, lowering or trace-insertion will be needed in any event.Google Scholar
  54. 54.
    The idea that antecedent government might be involved in Raising constructions was first suggested by Bouchard (1982. 196).Google Scholar
  55. 55.
    The problem does not arise if sentential adjuncts are adjoined to VP, as assumed in Huang (1982).Google Scholar
  56. 56.
    Our theory entails that the following example involves an ECP violation, as pointed out to us by Noam Chomsky: (i) *John, seems that it2 is like [s t, to win] 2 The trace is neither lexically governed nor antecedent-governed. This is a desirable result, since this trace satisfies the binding theory as formulated in Chomsky (1981a), yet the sentence is worse than would be expected of a simple Subjacency violation.Google Scholar
  57. 57.
    We do not know what accounts for this difference between indicatives and subjunctives.Google Scholar
  58. 58.
    Interestingly, adjuncts in Polish behave as if they were lexically governed. Thus, (i) is fully grammatical even though the initial trace of jak will not be antecedent-governed (or an intermediate trace antecedent-governing the initial trace will not be). (i) Spotkates mgzczyznç, ktôry jak rozwiazat zadanie? you met the man who how solved the problem Lit.: `How did you meet [the man who solved the problem t]?’ Perhaps jak is not actually an adjunct but rather is some kind of complement to the verb.Google Scholar
  59. 59.
    Toman (1982) discusses multiple wh-questions in Polish, arguing on independent grounds that the sentence-initial wh’s are not in the same Comp. He also suggests that there are multiple Comp-nodes in such sentences, but that only the left most one is introduced under S’. This hypothesis is consistent with our theory.Google Scholar
  60. 60.
    We assume, then, contrary to what Freidin (1978) has argued, that this principle does in fact exist. Freidin’s reduction of the strict cycle to other principles relied crucially on Subjacency’s being a constraint on representation rather than on movement. The arguments in section 5.2 that Subjacency must be a constraint on movement are thus indirect arguments for the cyclic principle.Google Scholar
  61. 61.
    As Noam Chomsky has pointed out to us, there is one case that may be problematic for this claim. Consider the following structure: (i) I bought this book, [s, Oi for Is John to read t,]] In (i) for and the empty operator are both in the embedded Comp. Although this reflects standard assumptions, we know of no compelling argument that the operator must be in this position. If it is actually adjoined to S’ instead, the problem disappears. Another possibility, considered in a different context by Fiengo (1980), is that the for is not in Comp but is actually adjoined to the subject.Google Scholar
  62. 62.
    Note that (173) apparently cannot be strengthened to (i): (i) A Comp is [awh] if it is headed by an [awh] element. This is so since a [—wh] Comp need have no head, as in (ii). (ii) I think John left.Google Scholar
  63. 63.
    Actually, all we require is that (179) be permitted to apply at this level. There is no need to restrict it to LF by stipulation. No problems arise if the filter applies at S-structure as well, since any t marked [—y] at S-structure will still be present, and still be marked [—y], at LF. No LF principle can ever salvage a derivation in which [—y] appears at S-structure.Google Scholar

Copyright information

© Springer Science+Business Media Dordrecht 1990

Authors and Affiliations

  • Howard Lasnik
  • Mamoru Saito

There are no affiliations available

Personalised recommendations