A phenomenon that has received much attention in the recent Minimalist literature is what Chomsky has dubbed ‘‘defective intervention’’ (Chomsky 2000). This is a phenomenon in which some head (usually finite T) seeks a matching NP to agree with or attract, but some other NP intervenes. In simple intervention cases, this other NP is itself eligible for agreement or movement. In cases of defective intervention, in contrast, the intervening NP is not itself eligible for agreement or movement, since it has already had its features checked by some other element, typically a preposition. The result is ungrammaticality: the lower NP is unable to check its features with the head, because the other NP is in the way. A small sample of the numerous publications discussing defective intervention includes McGinnis 1998, Boeckx 1999, 2008, Holmberg and Hróarsdóttir 2003, and Hartman 2012. The pattern was first identified for Italian by Rizzi (1986), for Spanish by Torrego (1996), and for Icelandic by Sigurðsson (1996).The paradigm case of defective intervention is subject-to-subject raising constructions in Romance languages. However, recent work by Hartman (2009, 2012) has also identified much more robust defective intervention in tough-movement in Romance languages and in English, and so I begin with that phenomenon. Consider the following triplet (from Hartman 2012:125, (16a–b), (20)):In (1a), an experiencer PP can follow the tough-predicate and precede the nonfinite clause, if the surface subject of the tough-predicate is an expletive. If an argument NP occupies this position instead in an instance of tough-movement, as in (1b), the PP is not allowed in this same position. According to Hartman (2009, 2012), the NP in the PP is a defective intervener and blocks A-movement of cholesterol from a lower position (for Hartman, tough-movement involves improper movement: a step of Ā-movement to the edge of the nonfinite clause, followed by A-movement to the surface subject position). Finally, in (1c), the PP is allowed if it appears in a position where it does not interfere with the putative A-movement. Here it is not in a position between the surface position of the raised NP and the position it is hypothesized to have raised from.This squib points out two problems for defective intervention as it has been characterized in the literature. First, placing the PP in a slightly different position, but where it should still interfere with Amovement or agreement, is grammatical. Second, adjunct phrases, even ones that demonstrably do not interfere with A-movement (as in the passive), are ungrammatical in the same position as experiencer PPs. I illustrate these problems in sections 2 and 3, and then show in section 4 that the same problems arise in subject-to-subject raising in French and Italian.The following examples (from Hartman 2012:125, (16), (18), (19)) illustrate his claimed defective intervention with tough-movement:Again, the PP may not follow the tough-predicate just when toughmovement takes place.The problem is that experiencer PPs can occur between the copula and the tough-predicate with complete acceptability (and without the pauses that characterize parentheticals).The PP still comes between the surface position of the subject and its putative base position (for Hartman, the edge of the nonfinite clause). As a result, it should still count as a defective intervener, and block A-movement, or the agreement that accompanies it. The fact that it does not is problematic for the view of defective intervention found in the literature, and, in particular, for the application of it to toughmovement in English (Hartman 2009, 2012).The same problem arises in other languages, as well, although the effect is weaker. As Hartman (2012) shows, French and Italian also seem to have defective intervention in tough-movement. I illustrate with French data (from Hartman 2012:123–124, (4a–b), (8b)).In French, as in English, an experiencer PP is not allowed between the tough-predicate and the nonfinite clause just when tough-movement takes place. I have confirmed these data with a grammaticality survey of nine speakers of French. Speakers were asked to rate sentences on a scale of 1 to 5, with 1 being ‘‘sounds awful’’ and 5 being ‘‘sounds perfect.’’ Sentence (6a) received an average rating of 5, and sentence (6c) an average of 4.6. Sentence (6b), in contrast, got an average rating of 1.9.As in English, however, simply putting the experiencer to the left of the tough-predicate, but still to the right of the surface subject position, leads to improvement. The following sentence was rated 2.9, still degraded but significantly improved:If experiencers intervened in A-movement (or agreement), sentence (7) should be just as bad as (6b). In both sentences, the experiencer comes between the surface A-position of the subject and the position it has supposedly moved from. (One can of course argue that a rating of 2.9 is still ungrammatical, but still, the statistically significant improvement is unexpected on the defective intervention account.)The actual generalization seems to be that no PP can occur between the tough-predicate and the nonfinite clause when tough-movement takes place. It is this position in particular where PPs may not occur. Putting the PP in any other position, even a position that should still cause defective intervention, is allowed, fully in English and marginally in French. (Expressing the experiencer as a preverbal clitic in the Romance languages falls under this generalization as well; see the works cited.) If defective intervention were what was at work here, the PP should be ungrammatical in any position along the path of the putative A-movement, not just one. This makes it doubtful that the defective intervention account is the right one for this phenomenon.The second problem for defective intervention is that nothing else can occur in the problematic position, either. Even adjuncts that can be shown not to interfere with A-movement or agreement are banned in this same position. Consider English tough-movement again. Adjuncts are also banned from the position between the tough-predicate and the nonfinite clause.As with Hartman’s experiencer cases, like those in (12), moving the ‘‘intervener’’ to the front of the sentence fixes the problem, as shown in (13). ((12a) is from Hartman 2012:125, (20).)As in section 2, putting the adjunct between the copula and the toughpredicate is also allowed (again without the pauses that characterize parentheticals).The issue for defective intervention is that adjuncts do not generally block A-movement or agreement. For instance, they do not disrupt movement to subject position in the passive.They also do not disrupt agreement in expletive constructions.This means that the problem in (8)–(11) could not be the disruption of A-movement or agreement, since adjuncts do not disrupt A-movement or agreement (see Rizzi 1990).French and Italian again behave exactly like English. Adjuncts are banned from the same position as experiencer PPs. (17) illustrates adjuncts in French, with the average rating from the survey of nine speakers reported for each.Note that intervening adjuncts are rated even worse than experiencer PPs (even in the expletive version), although this difference is probably not significant (1.4 vs. 1.9).Italian is illustrated in (18). The judgments here come from a survey of eight speakers, using the same rating scale.Once again, putting the adjunct between the auxiliary and the tough-predicate is rated higher by speakers of both languages. In French, that word order is rated 2.7, a significant improvement over 1.4 (see (17b)). In Italian, the rating jumps from 1.6 to 2.9. The best position for an adjunct seems to be initial position, but the position between the tough-predicate and the nonfinite clause is uniformly judged to be worse than any other position.Since adjuncts do not block A-movement or agreement, it will not do to say that even they give rise to defective intervention in toughmovement. Once again, the real generalization is about word order: nothing is allowed to occupy the position between the tough-predicate and the nonfinite clause when tough-movement takes place. Any other position is fine (in English) or greatly improved (in Italian and French), even positions that should interfere, and the problem with the banned position is clearly not the blocking of A-movement or agreement.As mentioned above, the paradigm case of defective intervention is subject-to-subject raising in French and Italian. However, the literature has reported a significant amount of interspeaker variation on this paradigm.1 Nevertheless, I have found exactly the same pattern: wherever an experiencer PP is degraded, an adjunct is too. (19) illustrates the standard defective intervention paradigm in French (examples based on McGinnis 1998:90–91, (55)–(56)).I tested these judgments using the same grammaticality survey reported above. The average ratings are reported with the examples in (19). Note that the French speakers I surveyed did not find initial position for the experiencer to be fully grammatical (19c). Additionally, while seven of the respondents rated (19b) very low (1 or 2), two gave it a 4.The French speakers that I surveyed find an adjunct degraded in the same position.The adjunct is fine between the raising predicate and the lower clause when no raising takes place (20a), but it is rated significantly lower when raising takes place (20b).The same holds in Italian, although the judgments are not as clearcut. The following examples illustrate the standard defective intervention caused by an experiencer PP, with the average rating from the same grammaticality survey (examples based on McGinnis 1998: 92–93, (57)–(58)). Note that the speakers that I surveyed did not find the order with no raising and a finite lower clause (21a) fully grammatical, in contrast to what has been reported in the literature (McGinnis 1998). Raising plus an intervening experiencer (21b) was also rated quite low, whereas McGinnis (1998) reports mild deviance (two question marks).Adjunct PPs are also degraded between the raising verb and the lower clause when raising takes place, although the Italian survey ran into a complication here. Respondents seemed to be reading the sentence with significant pauses around the adjunct, as a parenthetical (one respondent noted this explicitly). Once they were instructed not to do that, most of the respondents revised their rating, but a few respondents did not reply a second time. I report the ratings here, but note that they are probably not fully accurate. In particular, I believe the rating for (22b) is high.(Note that this issue did not arise in French, because French-speaking respondents were asked from the beginning not to read the sentences with pauses.)Even with these issues, it appears that the real generalization is about word order: many speakers prefer or require that the nonfinite clause immediately follow the raising predicate when subject-to-subject raising takes place. This does not seem to be about A-movement or agreement, however, since adjuncts demonstrably interfere with neither.Finally, in contrast with the Romance languages, English permits an experiencer PP to come between a raising verb and its nonfinite complement. As predicted, it also permits an adjunct PP in the same position.This is consistent with the overall generalization that we have found here: experiencer PPs pattern with adjunct PPs in the positions they can occupy.2The two problems described here indicate that defective intervention has been mischaracterized in the literature, at least in the domains of tough-movement and subject-to-subject raising in Romance languages. The effect is not about intervention in A-movement or agreement; it is about permissible positions for adjuncts. Adjuncts of all types are banned in the same position as experiencer PPs, even though they do not interfere with other cases of A-movement. It will also not do to say that all of these elements are defective interveners, including the adjuncts, because of the first problem: moving the adjunct slightly leftward, but still on the path of the A-movement, ameliorates the effect (completely in English, to a lesser extent in French and Italian).I conclude that it is not the case that already-licensed elements that occur between the surface position of some NP and its putative base position block A-movement. Rather, in the constructions and languages analyzed, there is one particular position where adjuncts and PPs are banned. This position is that between a tough-predicate and its nonfinite complement when tough-movement takes place (in all languages), and between a raising predicate and its nonfinite complement when raising takes place (in the Romance languages). Notice that this is very much like the fact that there is one particular position where adjuncts are banned in English—namely, the position between the main verb and its object.No one has ever claimed that this word order restriction in English is due to intervention in A-movement. Nor should we conclude that a similar restriction on raising predicates in Romance languages and tough-predicates in all languages has such a source. They are both similar in that what is banned is something intervening between a particular kind of predicate and its internal argument of a particular kind (in English, NPs but not PPs or CPs; in tough-movement and raising, nonfinite clauses but not finite ones).Despite many attempts, the word order restriction in English illustrated in (24) still lacks a nonstipulative account (for one attempt, see Johnson 1991). It is beyond the scope of this squib to explain either this or the word order restriction on tough- and raising predicates, but a few speculative remarks are in order.First, one might try to account for the restriction on raising predicates in the following way: the only way to get a PP in between the raising predicate and its nonfinite complement would be to move the nonfinite complement to the right. This might then make it an island to extraction, since it is often suggested that moved phrases are islands (Wexler and Culicover 1980). What would then be problematic is raising: the subject of the infinitive would be unable to move out of the nonfinite clause. This account might relate the violation here to the fact, noted by Rizzi (1990), that raising clauses, unlike nonfinite control clauses, cannot be extraposed or pseudoclefted. A finite complement of a raising predicate could move, because no raising takes place out of it. The chief difficulty with this line of analysis is the fact that English does not show the same effect. If moved phrases are islands, they ought to be islands in every language.A slightly different account might claim that only full CPs can ever undergo movement. Raising clauses, being reduced clauses (typically, IP or TP), simply cannot move. Control clauses and finite complements of raising predicates, being full CPs, can. Since raising clauses cannot move, there is simply no way to get an adjunct in between them and the predicate that selects them. Again, the difficulty with this analysis is English, where adjuncts are fine. (Note that English was also a problem in the defective intervention account: it was never clear why experiencer PPs would be defective interveners in Romance languages but not in English.)Turning to tough-movement, one possible avenue for exploration would build on the null operator account of tough-movement (Chomsky 1977). As noted above, Hartman (2012) analyzes tough-movement as improper movement. However, his evidence for the A-movement step is exactly the claimed defective intervention effect. If this is not real, there is no reason to think that A-movement takes place in toughmovement (putative reconstruction effects are also not real; see http://lingcomm.blogspot.com/2012/05/two-step-tough-movement.html). If there is instead a null operator at the left edge of the nonfinite CP, one might relate the restriction in tough-movement to the fact that relative clauses with null operators may not extrapose.Perhaps there is a general restriction requiring null operators to be adjacent to elements they are in a syntactic and semantic relation with (the head N in a relative clause, the tough-predicate in tough-movement). Another possibility is that the null operator is a phonological clitic that must cliticize onto the tough-predicate at PF, as has been suggested for other null C elements by Pesetsky (1992) and Bošković and Lasnik (2003).3Returning to defective intervention, this squib has shown that two cases of claimed defective intervention, in tough-movement and Romance subject-to-subject raising, are not best analyzed that way. The question arises, then, whether there is anything that should be analyzed as defective intervention. That is, should defective intervention be expunged from the theory of grammar?Again, it is beyond the scope of this squib to definitively answer this question, but I believe there are indications that the answer is yes. For instance, cases of defective intervention in Spanish have been shown to be a different phenomenon entirely (the issue is selection; see Torrego 1996, Anagnostopoulou 2003:237–238). Cases of defective intervention in Icelandic have turned out to be much more complicated, and much less prevalent, than originally thought (Sigurðsson and Holmberg 2008). The Icelandic case has also been argued to involve simple intervention rather than defective intervention (Broekhuis 2007). That is, where dative NPs intervene, they are fully active, in the sense that they have unchecked features.4 The same might be true for cases of blocked passivization across datives and blocked object shift across datives (see Vikner 2006 and references there). This also might be true in cases of the Person Case Constraint, or that constraint may have a source entirely different from defective intervention (see Bonet 1994, Adger and Harbour 2007, and much other literature). If true cases of intervention in A-movement and A-agreement are actually of the active sort, defective intervention disappears as a phenomenon. This should be a welcome result, since it would lead to a much simpler model of grammar, and one where only active elements intervene (which is the expected situation).
Stored only when both papers are in the corpus — 69 citations exist upstream.