Filters and Control

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


The questions that we want to consider here have arisen in a number of different contexts in recent work on the nature and use of language. Among these are the following:
  1. (1)a.

    Restricting the options for transformational grammar (TG) (discussed in section 1.1 below);

  2. b.

    Perceptual strategies and syntactic rules (section 1.2);

  3. c.

    Problems of obligatory control (section 1.3);

  4. d.

    Properties of the complementizer system (section 1.4).



Relative Clause Surface Filter Perceptual Strategy Movement Rule Deletion Rule 
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.
    Cf. as well Ross (1973); also Hudson (1972), for a proposal dealing with one of the phenomena we will discuss. Lasnik has proposed a surface filter requiring that a complement sentence with a subject begin with an overt complementizer if it is separated from its governing verb. Postal (1974, 128) discusses a version of this filter. Below, we consider various filters related to this one.Google Scholar
  2. 2.
    Note that these notions are not to be confused with the notion “complement” as when that John left is called a complement to the verb think in I think that John left. Google Scholar
  3. 3.
    Cf. Bresnan (1972), Chomsky (1974), for various interpretations of these results.Google Scholar
  4. 4.
    Cf. Jackendoff (1972); Chomsky (1972).Google Scholar
  5. 5.
    Cf. Aronoff (1976).Google Scholar
  6. 6.
    We note that the notion “deep structure” is to be understood as a technical term within the context of EST. For discussion of some confusions about this notion, see Chomsky (1975).Google Scholar
  7. 7.
    We will also not consider here the question of whether phonological rules are interspersed among transformations, as proposed by Bresnan (1971). If this conclusion is correct, then some modifications (though not essential ones) are required in the formulation presented here.Google Scholar
  8. 8.
    Cf. Chomsky (1977a, chapter 4).Google Scholar
  9. 9.
    Cf. Chomsky (1975; 1977a). We adopt here the form of trace theory presented in Chomsky (1977b). Cf. also the references of note 14.Google Scholar
  10. 10.
    See references in Chomsky (1975); also Jackendoff(1977) and several papers in Akmajian, Culicover, and Wasow (1977).Google Scholar
  11. 11.
    Among the rules of construal are the rules of control, the rules assigning antecedents to bound anaphors (reflexives, reciprocals, etc.), and the rule of disjoint reference. Interpretive rules include also rules of focus, rules assigning quantifier scope and interpretation, etc. Cf. Chomsky (1977a, b) for some indication of what we have in mind. We assume that the rules governing quantifiers are clause-bound in the unmarked case. Cf. May (1977b).Google Scholar
  12. 12.
    Cf. Lasnik (1976), Reinhart (1976).Google Scholar
  13. 13.
    Cf. Chomsky (1975; 1977a, b). It is quite natural to regard these conditions as, in effect, late rules of interpretation that mark certain positions within cyclic categories as “opaque” to anaphoric connections outside these categories; specifically, positions in the domain of a subject (the subject condition) and tense (the propositional island condition). We omit further elaboration here. See below, section 1.3.Google Scholar
  14. 14.
    . Cf. Fiengo (1974; 1977), and the references of note 9.Google Scholar
  15. 15.
    For some indication of the scope, for one rule of core grammar, cf. Chomsky (1977b). See also Quicoli (1976a; 1976b; 1980), extending the very important work of Kayne (1975a).Google Scholar
  16. 16.
    In more or less traditional terms, we may think of [+N] as “substantive” and [+V] as “predicable”. For a different theory of category features, cf. Jackendoff (1977).Google Scholar
  17. 17.
    The assumption is not without empirical consequences, and alternatives are easily imagined, but we will not pursue this interesting question here.Google Scholar
  18. 18.
    In fact, there is little reason to suppose that lexical items are inserted in base structures, in this theory. For some arguments to the contrary, see Otero (1976), den Besten (1976).Google Scholar
  19. We will continue to accept this assumption here for ease of exposition, but everything we say can be translated into an alternative theory in which lexical insertion takes place in surface structure and only abstract features are generated in the base (which is now limited to the categorial component) in positions to be filled by lexical items.Google Scholar
  20. 19.
    For an account of control and thematic relations, cf. Jackendoff (1972). For some critical discussion, cf. Freidin (1975) and Hust and Brame (1976).Google Scholar
  21. 20.
    There are stylistic differences between expressions of the forms (17b) and (17c). We ignore this matter here.Google Scholar
  22. 21.
    This is the filter (203) of Chomsky (1973).Google Scholar
  23. We are assuming here that relative clause constructions have the structure NP-S (the man _ who is here) and that the basic structure of S is NP-tense-VP. These are quite natural assumptions; cf. Vergnaud (1974), Emonds (1976). Alternatives that have been proposed require some modifications of the rules presented below.Google Scholar
  24. 22.
    A somewhat different analysis will be presented below.Google Scholar
  25. 23.
    The latter construction is somewhat marginal, a matter to which we return. Note that in case (23a), we are now assuming that the embedded structure he left is the NP subject. Thus in the strategy (22), we restrict attention to construction C properly including the structure in question. Later, we will suggest that he left is not the NP subject, following Koster (1978a).Google Scholar
  26. 24.
    By a “true universal” we mean a principle that holds as a matter of biological necessity and therefore belongs to UG, as contrasted with a principle that holds generally as a matter of historical accident in attested languages. The distinction may be difficult to establish, but it is fundamental.Google Scholar
  27. 25.
    This is in fact the case, as we see below, but we are now concerned simply with the logic of the situation.Google Scholar
  28. 26.
    Similar remarks apply in the case of a theory that permits language-specific ordering restrictions or conditions on rules, or distinctions between obligatory and optional rules.Google Scholar
  29. 27.
    Bever and Langendoen (1971) suggest a historical argument relating disappearance of inflections to restrictions such as (20). As noted by Fodor, Bever, and Garrett (1974), the account at best provides only partial “correspondence between structural facts and perceptual heuristics” (p. 360). Expressions such as (24) provide another illustration of the limits of the correspondence. We owe example (24) to Wilson Gray.Google Scholar
  30. 28.
    Or matrix indirect object, as in They appealed to John to leave. We might argue that in (28b) the object is actually indirect, as in John’s promise to Bill to leave. Google Scholar
  31. 29.
    More specifically, an arbitrary agent. Working within a somewhat different framework, Lasnik and Fiengo (1974) discuss this aspect of predicates requiring obligatory control. They suggest that the understood missing subject of a complement of want can bear any subject relation to the complement VP, while the understood subject of the complement of (e.g.) force must bear the agent relation.Google Scholar
  32. 30.
    This property of control follows from the Tensed-S Condition (which blocks any anaphoric relation between an anaphor in a tensed sentence and an antecedent outside it) and the Specified Subject Condition (which permits only the subject of an embedded sentence or NP to be related anaphorically to an antecedent outside). Cf. note 13. Therefore, from these conditions, which are quite independently motivated, it follows that only the subject of an infinitive or gerund is susceptible to control.Google Scholar
  33. 31.
    Note that the sentence (35c) is grammatical but not the structure indicated.Google Scholar
  34. 32.
    Given other properties of the grammar, it also follows, though not from the considerations presented here alone, that they are in complementary distribution in deep structures, though the class of relevant contexts is more complex in structure.Google Scholar
  35. 33.
    E.g. Chomsky (1977a, chapter 4).Google Scholar
  36. 34.
    Unless, of course, we were to assume a dual selectional classification for all of the want-type verbs, assigning them as complement both null-complementizer + PRO + VP and for + lexical — NP + VP. This would require a redundancy rule in the lexicon, a complication of the grammar that calls for empirical motivation, lacking in this case, since there is a simpler explanation, as we see directly. Furthermore, under this alternative analysis we would be forced to complicate the theory of control. Verbs would no longer be categorized as assigning control (presumably, by virtue of their semantics), since there would be nothing then to prevent the control rule from applying to want in cases (38a, b), blocking the sentence as ill-formed for the same reasons that blocked (35).Google Scholar
  37. 35.
    We cannot have John believes sincerely himself to be incompetent, analogous to (38d), for independent reasons to which we return below.Google Scholar
  38. 36.
    Cf. Chomsky (1974; 1977a, chapter 4).Google Scholar
  39. 37.
    Careful analyses restricted to Equi have characteristically noted that there are two quite different processes of Equi NP Deletion, corresponding to our distinction between Equi and control. Cf. Kayne (1975).Google Scholar
  40. 38.
    See Lightfoot (1977).Google Scholar
  41. 39.
    This is loosely put. As noted earlier, the rules of “semantic interpretation” applied to surface structures relate to the syntax of logical form.Google Scholar
  42. 40.
    The increase in the class of grammars itself imposes considerable burdens on the task of the language learner. Furthermore, these new devices appear to raise substantial questions regarding learnability. Cf. note 26.Google Scholar
  43. 41.
    Although the system we present here is both simpler and more adequate than that one, the basic rules of interpretation for ± WH can be carried over, in essence.Google Scholar
  44. 42.
    Note that by “declarative”, we mean to imply nothing more than “noninterrogative”. We will not explicitly treat subjunctive complements here; they appear to pattern like finite clauses.Google Scholar
  45. 43.
    There is no particular difficulty in formulating the A/A Condition to guarantee this result. In fact, there are quite independent conditions that exclude the result of this unwanted deletion, having to do with dangling prepositions. Cf. Chomsky (1973) for some discussion; also Koster (1976).Google Scholar
  46. 44.
    Deletion under identity, if this process exists as part of sentence grammar, does not fall under this constraint, but that is irrelevant here. On these phenomena, see Sag (1976), Williams (1976), and (for some general remarks on deletion that seem to us appropriate) Koster (1976).Google Scholar
  47. 45.
    See Bresnan (1972), for an illuminating discussion of the semantics of complementizer constructions. There seems no reason, however, to associate the semantic properties with the complementizer itself rather than with the construction.Google Scholar
  48. 46.
    For an analysis of wh-of the sort we have in mind, see Chomsky (1973; 1977a, chapter 4; 1977b). There is a residual semantic content in relative pronouns, namely the feature Human (the man who I saw, *the book who I saw). But this (or any) residual semantic content is redundant, since it is expressed in the head of the construction. We assume that an appropriate concept of recoverability will be restricted to lexical items, ignoring such redundant features. Note that if we accept a raising analysis of relatives, as in Vergnaud (1974), then the lexical elements within the wh-phrase have been moved by a transformational rule to the position of the head, leaving only wh and a trace along with redundant features. The recoverability condition will not apply to the wh-word, now devoid of semantic content.Google Scholar
  49. These remarks apply only to restrictive relatives. In appositives, a raising analysis is inappropriate and a full NP with lexical content remains in the wh-phrase. This explains the fact that deletion of the wh-phrase is impossible in appositives. Note that it then follows, given filter (53), that the complementizer that will never appear with appositives.Google Scholar
  50. 47.
    Cf. Chomsky (1973), note 56.Google Scholar
  51. 48.
    As in the case of complementizers, we may think of the rule (59) as inserting certain abstract features, realized as it in English. If we assume lexical insertion to be at surface structure (cf. note 18), then this alternative is obligatory.Google Scholar
  52. 49.
    Similarly, abstract feature (see preceding note) block movement rules, given that the latter are restricted to [NP e]. The rule (59) has other consequences. Thus, it blocks hypothetical derivations that might yield such ungrammatical sentences as *who is it certain to be present. The necessity to exclude such sentences is noted in Jenkins (1977).Google Scholar
  53. 50.
    Taking (60) to be a lexical insertion rule, it can be simplified by exploiting the conventions on selection. There is no need to specify explicitly the complements to V* apart from S.Google Scholar
  54. 51.
    His analysis depends on a variety of assumptions and devices that we hope to be able to dispense with in the theory of grammar. Cf. Bresnan (1972) for a critical discussion.Google Scholar
  55. 52.
    As noted above, we will eventually take that to be simply the realization of —WH. In that case, the rule (75) can simply refer to —WH and no problem arises from the range of possibilities for that. Google Scholar
  56. 53.
    As noted above, we assume that English that is the language-specific realization of a set of universal features. We may assume that Spanish and French que realize the same features.Google Scholar
  57. As remarked earlier, we are omitting any consideration of subjunctives here. Note however that with an analysis such as (68), we will regard if as a variant of that before subjunctives, so as to account for *a problem that I would be surprised if t were solved as compared with a problem that I would be surprised if he were to solve t (in both case, t the trace of the moved wh-phrase).Google Scholar
  58. 54.
    Note that where the quequirule is inapplicable, the sentences in question are ungrammatical because of (68). Thus *plus de gens sont partis que Pierre a dit (qui, que) étaient restés.Kayne (1976) cites the examples *la fille que je tiens à ce (que, qui) l’épouse, *la fille qu’il est évident (que, qui) _ t’admire. We are indebted to Jean-Roger Vergnaud for bringing these facts and their significance to our attention.Google Scholar
  59. 55.
    Perlmutter actually puts the matter differently. Noting that (64) is not universal, he restricts his attention to languages that observe (64). These languages do not have a rule of Subject Deletion.Google Scholar
  60. Perlmutter cites Dutch as an apparent exception to (71), on the basis of data that are in fact restricted to certain dialects. But even for these dialects there seem to be reasonable alternative analyses, under which they too fall under the generalization. We are indebted to Hans den Besten and Jan Koster for pointing this out to us. See also Zaenen (1977).Google Scholar
  61. 56.
    Under a raising analysis such as that of Vergnaud (1974), the head of the relative actually is the subject of the embedded verb; on alternative analyses, the head of the relative indicates what is to be taken as the subject under a rule of interpretation.Google Scholar
  62. 57.
    But Subject Deletion is not always permitted in languages with this property, as French indicates. The situation is different in languages that permit pronouns to be absent freely. In these languages, we might assume that an abstract feature [+ pro] can be generated in the base and simply not filled by lexical insertion; thus there is no deletion of pronouns. These properties should follow from an appropriate formulation of the recoverability condition on deletion rules.Google Scholar
  63. 58.
    Bresnan (1972) suggests that if a language allows Subject-Pronoun Deletion, it would be difficult for the language learner to find evidence that the FSC holds; hence the FSC should not hold of such languages. But it would be no less difficult for the language learner to find evidence for the language-specific principle FSC even without subject pronoun deletion. In fact, considerations of this sort cast serious doubt on the possibility of language-specific principles such as the FSC; cf. note 26.Google Scholar
  64. Bresnan (1977) extends the FSC to nonsubject cases, referring to it now as the “complementizer constraint on variables”. We will continue to refer to the constraint as theGoogle Scholar
  65. FSC for ease of exposition. Note that the extensions she discusses fall under the surface filter, which is not restricted to subjects.Google Scholar
  66. 59.
    The following examples were provided by Joseph F. Foster, to whom we are much indebted, along with Suzette Elgin and James Stalker, for information about Ozark English. See appendix 3 for further discussion, supporting the conclusion we tentatively reach here.Google Scholar
  67. 60.
    THE trace of who. With dialectal variation, a resumptive pronoun may appear in the position of trace, under various circumstances that are not relevant here. There are many extremely interesting properties of these dialects that deserve much more extensive study.Google Scholar
  68. 61.
    We are indebted to Henk van Riemsdijk for the following observations.Google Scholar
  69. 62.
    It has been suggested in the literature that for-Deletion is restricted to immediately postverbal contexts. We have assumed, instead, the simpler and more general rule (52) that permits free deletion in COMP, covering wh-phrases as well as all complementizers. The former proposal, apart from requiring a complication of the grammar, is descriptively inadequate in that it fails to account for the very similar distributional properties of Ø-complementizer verbs, to which we return directly.Google Scholar
  70. 63.
    By N’ we mean N with any number of bars, in the bar system. By the notation *(...) we mean that the sentence is excluded if ... is deleted.Google Scholar
  71. 64.
    Sentences of the form (ii) are marginal for many speakers even if for remains. Note that we are assuming here that V + en is of the category Adjective. We return to the passive construction below, in section 2.2.2.Google Scholar
  72. 65.
    We have not exhausted all possible constructions. The notion “sentence-initial” in case (89e) must be more broadly construed, to include such structures as Presumably, *(for) John to be successful would be unlikely, As for Mary, *(for) John to take part would bother her etc. A proper characterization of this concept awaits a fuller analysis of the notion “root sentence”. Cf. Emonds (1976). Or consider pseudoclefts such as What he prefers is *(for) John to leave. Here, John to leave appears to be postverbal and thus permitted by (88); but if, for example, the correct analysis of pseudoclefts is along the lines suggested in Higgins (1973), as a type of “list structure”, then they might be considered not to fall under (88). We will have to put these and other questions aside.Google Scholar
  73. 66.
    Verb or COMP, respectively. We will continue to assume that for is in COMP but not in —WH; cf. (84) above. If not, the rules given must be complicated. Thus we have a slight additional motivation for the suggested reanalysis.Google Scholar
  74. 67.
    Cf. Lasnik and Kupin (1977) and earlier work, dating back to Chomsky (1955).Google Scholar
  75. 68.
    We would have a sharper test for this assumption if there were an intransitive verb to replace bother me in it bothers me for (that) S etc., preserving the structure; but there seems to be no such verb. Note that the concept of adjacency assumed here, consistent with the conventional theory, has also been tacitly assumed in our discussion of the *[for-to] filter (61), which we have taken to be operative even if for and to are separated by trace.Google Scholar
  76. 69.
    Cf. Reinhart (1976). This is a generalization of the notion defined in Lasnik (1976). As Reinhart notes, this is essentially the notion “in construction with” of Klima (1964).Google Scholar
  77. 70.
    This was the notion of “domain” used informally with reference to conditions on rules, in note 13. Assuming the theory presented in Lasnik and Kupin (1977), we need not refer to branching in the definition of c-commandGoogle Scholar
  78. 71.
    There is some reason to believe that the complementizer for shares properties with the homophonous preposition (cf. the discussion of (56d), above), but this evidence is hardly compelling.Google Scholar
  79. 72.
    Given the parallelism between ing and to one might ask whether the structures (95), or some of them, are analogous to such NP-ing + VP constuctions as I found John studying in the library etc. (or to one of the interpretations of the ambiguous I found the boy studying in the library). Google Scholar
  80. Indefinite noun phrases are characteristic in constructions such as (95) (e.g. *1 found “Ode to a Grecian Urn” to memorize etc.). The distribution of the constructions seems to match that of NPs and we will assume that they are relative clauses. Actually, the assumption is not particularly crucial to the discussion. What is crucial is the assumption that they involve Wh Movement. For a general discussion of this question, see Chomsky (1977b).Google Scholar
  81. 73.
    We continue to assume that the derived constituent structure of an ordinary finite relative is [NP NP S], as, e.g., in Vergnaud (1974). Cf. note 21.Google Scholar
  82. 74.
    Assuming a raising analysis of relatives, as in Vergnaud (1974), the underlying embedded S would actually be something like Bill to work on a topic. Placement of wh-on the NP a topic would then be followed by Wh Movement to COMP and raising of a topic to the head position. The details of relativization are not pertinent here. For ease of exposition we will give the underlying structures, here and below, in the form (97).Google Scholar
  83. 75.
    Recall that no subphrase of the wh-phrase can be deleted; cf. the discussion following (53) above.Google Scholar
  84. 76.
    Or, the trace of wh-the man. Cf. note 74.Google Scholar
  85. 77.
    Actually, case (105c) reveals the same ambiguity, but its resolution is irrelevant in case (105c).Google Scholar
  86. 78.
    Or, “the domain of verb or for” if we take (93) rather than (93’). If we assume (107), then to exclude such sentences as *John to win would surprise me we must not take the phrase John to win to be the subject NP. We will later suggest that this is the case. Cf. note 87.Google Scholar
  87. 79.
    More accurately, there are eight possibilities if the element relativized is in a PP and four possibilities if it is the object of the VP.Google Scholar
  88. 88.
    A further suggestion is that PRO-self be taken as the underlying form for Equi. Because of the structures involved, there will always be antecedent control in the case of Equi in verb complements.Google Scholar
  89. 81.
    Cf. note 74.Google Scholar
  90. 82.
    It should be the case, then, that in dialects such as Ozark English that do not have the filter (61), cases analogous to (129c) should be possible, unless some other property of the grammar exclude them. We have not been able to satisfy ourselves as to the facts, since it has, so far, been difficult to find clear cases.Google Scholar
  91. 83.
    Again, we would assume that in dialects lacking (61), such structures as It is illegal for to leave would be permissible, with appropriate choice of Adjective.Google Scholar
  92. 84.
    Note that the sentence (134) is grammatical under a different analysis, with the pronoun it raised from the embedded subject position, referring, say, to a horse that is certain to win the race. But we are considering here the Equi structure, which will have no NP (specifically, no trace) in the embedded subject position in surface structure.Google Scholar
  93. 85.
    Or alternatively, in the context t. Cf. p. 92, below. Note that (128c) is now also permitted.Google Scholar
  94. 86.
    As noted before, the choice of complementizer with Adjectives is partly idiosyncratic. Cf. (47)—(48), above. But it is also fairly systematic. In general, sentence modifiers such as seem and certain take O-complementizers.Google Scholar
  95. 87.
    There is, perhaps, a third way, namely, as in For John to leave is illegal. It may be, however, that the infinitival is not the NP subject in this case. Cf. Koster (1978a).Google Scholar
  96. 88.
    Note that the pseudocleft variant of (127a), namely, What we believe is John to be incompetent is ungrammatical. Therefore, we cannot take this to be a position in which 0 takes the feature [—N]. The result follows if we accept the variant of (135) proposed in note 85. But in any event, pseudocleft constructions behave quite differently from normal predication, so we would presumably want to exclude this possibility for “list structures” (cf. Higgins (1973)), even under the variant (135) as given. Note that there is no expressible pseudocleft associated with (127a). Taking pseudocleft to be in effect a stylistic variant, the propositional content of (127a) is nonetheless expressible. The situation is different in the case of (127b) and (133b). Were it not for some such rule as (135), the propositional content itself would be inexpressible in this case.Google Scholar
  97. In Ozark English, the variants what I look for is (for) Jimmy to win the election both appear to be acceptable. If so, either pseudoclefts are differently analyzed or the filter is slightly different. Again, we do not have adequate evidence to decide.Google Scholar
  98. 89.
    For some discussion, see the references of note 9.Google Scholar
  99. 90.
    Cf. Chomsky (1965) and many more recent studies.Google Scholar
  100. 91.
    Emonds (1970), Bresnan (1972). Some arguments for this conclusion are given in Dresher and Hornstein (1979), within the framework of trace theory.Google Scholar
  101. 92.
    Note further that this conclusion is the best case. Trace is simply an NP; the grammar is complicated significantly if some rule must ignore NPs with specific content (in this case, null content). Bresnan (1976b; 1977) proposes that rules “may not refer to traces and trace-binding”, arguing that this proposal “would restrict idiosyncratic, language-particular rules from access to `global’ information, greatly simplifying the task of the language-learner”. We agree with the conclusion (with a qualification to be noted), but on different grounds from hers. Her argument is misleading. In the first place, in the specific case she is discussing (namely, (85)), the filter appears to be universal. it is not the surface condition (85) but rather the proposed alternative, the FSC, which is idiosyncratic and language-particular, and thus imposes a nontrivial (we suspect, empirically unacceptable) task on the language learner. Second, we agree that language-particular rules should not refer to trace, but on more general grounds: rules should not refer to any specific contents of NP, e.g. the man or e. This option is “highly marked” and excluded from core grammar quite generally. Finally, the issue of “globality” does not arise. The methodological objections to global rules do not apply to trace theory. Cf. Chomsky (1975, 117f.).Google Scholar
  102. The qualification is that some extensions of the grammar that are “highly marked” when regarded in isolation may still be the “unmarked case” for broader reasons, as we have noted several times in this discussion; for example, if they permit the satisfaction of more general conditions such as those relating to expressibility. The question, which is an interesting and important one, arises again in the considerations of this section.Google Scholar
  103. 93.
    Note that while the surface structure (149d) underlies no grammatical sentence, nevertheless the sentence Bill is eager to win is grammatical, under Equi, which leaves no subject trace in the embedded sentence.Google Scholar
  104. 94.
    (150c) is independently excluded by the fact that it would have an unfilled matrix subject, the rule of It Insertion being inapplicable before a 0-complementizer.Google Scholar
  105. 95.
    We leave the notion “adjunct” without specific explanation in (151) and below, pending further study. The examples illustrate cases to be included: adverbial modifiers and indirect objects.Google Scholar
  106. 96.
    Note that under the alternative analysis for empty complementizers sketched just above, we replace 0 by F in (151’b).Google Scholar
  107. 97.
    Note that we now must define “adjunct” so as to exclude the first NP of a double NP complement. Cf. John gave Bill a book, We elected John president etc. Cf. note 95.Google Scholar
  108. 98.
    We assume here the variants (93’) and (151’b). Obvious changes are required if other variants discussed are adopted.Google Scholar
  109. 99.
    Bresnan observes that “the preposition, which deletes in front of the complementizer, nevertheless seems to protect the complementizer itself from deletion” (1972, 98). Thus we need a rule blocking the normal optional deletion of for in postverbal position just in case the for associated with the verb deletes. Bresnan nevertheless assumes that it is the for of hope for rather than the complementizer that deletes, on the basis of such examples as (i)—(iii):Google Scholar
  110. (i)John would be ashamed*of for *0 us to see him.Google Scholar
  111. (ii) What John would be ashamed of would be for us to see him.Google Scholar
  112. (iii)*What John would be ashamed would be for us to see him.Google Scholar
  113. But we may just as well assume that ashamed like the verb believe takes NP object or sentential (either finite or infinitival) complements — like other adjectives, it takes for-clause rather than 0-clause complements, for reasons already discussed. The subject of (iii) now has no source, exactly as in the case of * What John was sorry was that Bill left. Note that the of in ashamed of NP is really a mark of the construction rather than a true component of the adjective; it may well be introduced by a rule of Of Insertion that applies in [[+N] _ NP] (e.g. one of the men, the destruction of the city, proud of John etc.). Where of really is inherent to the lexical item and is not a mark of the construction, as in the case of the verb approve of, we do not have the comparable paired expressions I approve for.... I approve that.... Cf. Chomsky (1972, chapter 1, note 27).Google Scholar
  114. 100.
    Whether X = PRO or my need not detain us here. Cf. note 80.Google Scholar
  115. 101.
    Note that this rule would have to be added to the system even if we were to assume that it is the for of hope for that deletes to give (162). Cf. Bresnan (1972, 104). Therefore rule (167) lends no support to this assumption, with its undesirable consequences noted above.Google Scholar
  116. 102.
    Recall that in the case of filters, as for transformations, we assume that string adjacency suffices.Google Scholar
  117. 103.
    This analysis of (171) is presented in Chomsky (1977a, chapter 4). An alternative, proposed in Chomsky (1977b), is excluded by the organization of grammar given in (6), since it requires that rule (170) apply after the rule of interpretation for wh-. That alternative was motivated by observations of Lightfoot (1976) and Pollock (1976) concerning the appearance of trace in COMP with unbounded movement, which should then block such cases as (171a). But the latter objection does not apply to the system presented here, since the trace in COMP can delete by the free deletion rule. Thus the simplest analysis can be retained.Google Scholar
  118. 104.
    Case (e) of (174) is perhaps marginal, but deletion of that renders it completely ungrammatical. Judgments vary with regard to case (178c). We return to this matter below.Google Scholar
  119. 105.
    We continue to assume that relatives have the structure NP-S. Furthermore, we assume that in a that-complement such as the fact that John was here (as distinguished from the fact that you mentioned to me) the structure is not NP-S but rather Determiner-N, where N = N-S. The latter assumption is implicit in the X-bar theory. Cf. Chomsky (1972, chapter 1), and subsequent work.Google Scholar
  120. 106.
    Consider in contrast I suggested (indicated, etc.) to Bill that we would win. These seem to us possibly more acceptable than (174e) with that deleted. The reason may be that in an indirect object following a verb, the NP generally behaves as if it were not in a PP, with regard to c-command. Compare (i), (ii), and (iii):Google Scholar
  121. (i)I spoke to the men about each other.Google Scholar
  122. (ii)I spoke about the men to each other.Google Scholar
  123. (iii)I argued with the men about each other.Google Scholar
  124. Case (i) is grammatical, but not (ii) or (iii). In general, the reciprocal of a phrase must be in the (c-command) domain of a preceding antecedent, suggesting that the men c-commands each other in (i), but not in (ii) or (iii), where it is in a prepositional phrase. To put it diffierently, the to of the indirect object phrase behaves as if it were a case-marking, not a preposition. Or, from another point of view, it behaves as if it were part of the preceding verb.Google Scholar
  125. Assuming the latter interpretation, consider the examples (iv) and (v):Google Scholar
  126. (iv) Which men did Bill speak to t about each other? (t the trace of which men) Google Scholar
  127. (v) *To which men did Bill speak t about each other? (t the trace of to which men) Google Scholar
  128. In case (iv), each other is in the c-command domain of t which is the NP directly following the complex verb speak-to; therefore, the antecedent-anaphor relation can hold between t and each other. This will give a properly formed representation in LF, when variables are inserted by the appropriate conventions, which we will not discuss here. But in case (v), the only possible antecedent for each other is which men in the prepositional phrase to which men. Since to cannot be associated with a preceding verb in this case, the required c-command relation does not hold and (v) is ungrammatical.Google Scholar
  129. 107.
    This would follow, for example, on the analysis of topicalization in Chomsky (1977b). If one were to assume that the that-clause is, say, c-commanded by the subject of the sentential clause, then we would have to modify (175), replacing “adjacent to” by “following and adjacent to”. _Google Scholar
  130. 108.
    Note that (183d) is grammatical, because the tensed S (a of (178)) is in the domain of the adjacent NP you. In comparison, deletion of that is impossible in (174e), which is analogous to (183d) except that in (174e), the that clause is not in the domain of the NP Bill. The filter (178) thus covers a variety of cases.Google Scholar
  131. 109.
    Bresnan (1977) argues that this theory may not be more restrictive than familiar alternatives, but her argument is based on several incorrect assumptions. She assumes that the “trade-off” is between (1) permitting unbounded rules (deletions and transformations), and (2) permitting a richer class of filters. She opts for (1) and assumes that we choose (2) over (1). But note first that here, as in earlier work cited, the question of boundedness of rules has been only a subsidiary concern. A much more important point is that the filters seem to eliminate any need for the elaborate enrichment of linguistic theory noted in the text (ordering, obligatoriness, etc.), while accounting for quite a range of previously unexamined and unexplained data. Second, consider the question of unbounded rules, which arose only in section 2.1, in the discussion of the surface condition (85) and the FSC. It is true that we see no argument for permitting unbounded transformations and deletions in linguistic theory; thus we restrict ourselves to a narrower theory that also provides an explanation for the Complex NP and Wh-Island Constraints. Cf. the references of note 9. Bresnan discusses only this one issue, but she fails to note that the filter in question, (85), in contrast to the FSC, is exceptionless (so far as is known), and provides an explanation for the basic generalization (71). In contrast to the FSC, a universal filter such as (85) does not extend the expressive power of the theory and, of course, places no burden on language acquisition. Third, Bresnan seems to agree that some filters are needed, specifically, the one described in note 1 above; cf. Bresnan (1976c, 493). A theory of filters rich enough to permit that one would allow all those we have discussed, it appears. Thus there is no reason to believe that the alternative (2), with resort to filters, allows any extension of the expressive power of grammar.Google Scholar
  132. The correct conclusion seems to be that the trade-off is between (1) a theory that permits rule ordering, obligatoriness, multiple factorings, rich contextual dependencies, unbounded deletions and transformations, etc. — and a certain class of filters; and (2) a theory that permits just that class of filters and none of the other devices.Google Scholar
  133. 110.
    Cf. Bresnan (1977), which discusses a proposal virtually identical to (85) suggested in discussion at a conference in June 1976.Google Scholar
  134. 111.
    That is, if we reject the suggestion that the movement rule be replaced by an interpretive rule (cf. Chomsky (1973, §17)); cf. final remarks of Chomsky (1977a, chapter 4), for some comments on this issue.Google Scholar
  135. 112.
    Bresnan asserts that a surface filter of the sort we have in mind could not distinguish the ungrammatical (188e) from the grammatical sentence This food is still cooked in the same way that is prescribed in ancient books since in both case there is a structure NPthat-trace. If (85) were the appropriate device to exclude (188e), this observation would require a stipulation in (85) that the NP cited be the antecedent of the trace — i.e. coindexed with it, in our framework. But the question is academic, since neither (85) nor the FSC is, in our opinion, the appropriate device.Google Scholar
  136. Notice that if that is deleted in (188e) the sentence becomes, if anything, even worse. Again, neither the FSC nor (85) accounts for this fact, which follows from our filter (178).Google Scholar
  137. 113.
    Case (191d) may be less acceptable than the others, with the pronoun remaining, but the matter is of little importance, having to do with the preference for application or nonapplication of the stylistic rule (190), a peripheral issue at best.Google Scholar
  138. 114.
    On this matter, see Koster (1976).Google Scholar
  139. 115.
    Or “nondistinctness”. Cf. Chomsky (1965).Google Scholar
  140. 116.
    Since the arguments in favor of the filter (85) as against the FSC or other proposed alternatives seem to us compelling, we have not gone on to discuss particular problems that arise in the case of these alternatives. But there are problems. Consider the FSC. Bresnan (1977) formulates a deletion rule for relativization in the context: NP [S X rel Y]. In her theory, the specific context cited is required to invoke the FSC. But evidently it is necessary to add some conditions — call them C — to guarantee that rel (pronoun, whword, empty element, or whatever) appears in the relative clause complement to NP. But given C, the context appears superfluous: thus the rule can be simplified to: “delete rel under C”. But now the FSC will not apply. Many other questions of this sort arise in connection with the deletion rules proposed.Google Scholar
  141. 117.
    Bresnan suggests that the FSC might be generalized to include subordinating conjunctions as well as complementizers, that and as now being taken as subordinating conjunctions. But since these seem to be the only relevant examples of subordinating conjunctions in this sense, there appears to be no generalization here but rather an ad hoc extension of the FSC to include than and as if they are not complementizers. The condition does not apply to a natural class in any sense that we can see. As noted in the text, we see no reason to depart from the simplest assumption; namely, that than and as are not individually categorized at all, but are simply elements of the items er ... than, as ... as. There seems no more reason to assign a category to the second element of these compound forms than to the first element. On the element if see note 53.Google Scholar
  142. 118.
    Note that this is not true of the 0-complementizer. But since the principles that exclude 0 in the position of that, for in (197) (etc.) are required independently to bar the results of free deletion in COMP, this exception is irrelevant to the point at issue.Google Scholar
  143. 119.
    Note that arguments one, two, and four also hold against the proposal that than and as are subordinating conjunctions, assuming that the latter is the category that includes after, while, if because etc. Thus compare (199) with *John spoke to Bill after Tom to Mary, *They left while John spoke to Bill and while Tom to Mary, *John surely spoke to Bill if Tom to Mary, *John hit Bill because Tom Mary etc.Google Scholar
  144. 120.
    An interesting case of a language that forms comparatives in this way is Afrikaans, which even permits “pied piging” in comparative constructions. Cf. Koster (1976).Google Scholar
  145. 121.
    For further discussion of the status of than and as see Huang (1977). In Chomsky (1973, note 32, and references cited there), it is argued that considerations derived from the theory of the base support assignment of than and as to COMP, but these arguments seem to have little force, and they collapse entirely under the analysis of comparatives in Chomsky (1977b).Google Scholar
  146. 122.
    For the sake of discussion, we assume their version of the data. However, the standard source, to which Bresnan refers for corroboration of her assumptions about Old English, states: “When the relative pronoun is whom or (the) which the preposition is placed either before the pronoun or at the end of the clause. Both types are represented in English from the beginning of the Middle English period on” (Visser (1963, 400)). Visser also cites data in support of this conclusion (p. 401).Google Scholar
  147. In the case of Old English, Visser’s presentation is also at variance with Bresnan (1976b). Cf. Visser (1963, 397). He indicates that the preposition can be stranded both after the complementizer be and after the relative pronoun bæt or bœr. Bresnan cites onlyGoogle Scholar
  148. Visser’s examples with ¢æt suggesting that it may be “an invariant relative marker.” However this may be, some additional attention should be paid to the examples Visser cites with ¢ær which is clearly a relative pronoun.Google Scholar
  149. There may well be additional hypotheses concerning Old and Middle English that might permit one to maintain the assumptions concerning the relevant data as presented in (201). But it seems that they would require additional devices. We can take no stand here on these further possibilities. We hope to explore the further details of the historical development of the relative clause construction elsewhere. As we will see directly, these questions, however resolved, seem to have no bearing on the topics that concern us here. We are grateful to Phoebe Huang for bringing the facts cited here to our attention.Google Scholar
  150. 123.
    Bresnan (1976b) claims that the system of (201) was stable for a long period in Old and Middle English, concluding that this argues against the assumption that (201) is “highly marked”. She directs this argument against the analysis in terms of filters, but if it were valid, it would apply as well to her assumption that the grammars of these languages included (208a). Clearly, Modern English has none of the complications of (208). According to Visser, the same was true of Middle English and apparently Old English as well.Google Scholar
  151. 124.
    Bresnan argues (1976b) that the constructions that she calls “comparative subdeletion” demonstrate that these constraints hold for certain deletion rules, on the grounds that the Wh Movement analysis for these constructions proposed in Chomsky (1977b) is descriptively inadequate. But that is incorrect. The proposed Wh Movement analysis accounts for all of the facts she postulates, and does so within a very limited theoretical framework, with no recourse to multiple factorizations, as under the deletion analysis. We know of no valid arguments for unbounded deletion observing these constraints. Cf. Koster (1976) for general discussion of the issues.Google Scholar
  152. 125.
    Also, presumably, (214), though we have not as yet been able to check this specific case.Google Scholar

Copyright information

© Springer Science+Business Media Dordrecht 1990

Authors and Affiliations

  • Noam Chomsky
  • Howard Lasnik

There are no affiliations available

Personalised recommendations