Proponents of Neg-transportation (NT) in Spanish disagree with respect to how the transformation comes to interact with the choice of mood in verbal complements. NT is the rule which raises the negation of a verbal complement to the governing verb:

(1b) is apparently the result of applying NT to (1a). In both sentences in (1) the embedded verb is in the subjunctive, required in the context of a governing matrix expressing volition. Where the disagreement arises is in cases like (2), in which the application of NT might result in either (2b) with a complement in the indicative or (2c) with a complement in the subjunctive. Bolinger (1968) claims that (2b) (but not [2c]) has undergone NT in its derivation. On the other hand, Rivero (1971) argues that the application of NT always yields a complement verb in the subjunctive, as in (2c). Finally, Luján (1978) hypothesizes that both (2b) and (2c) can be derived by NT, the choice of mood depending on whether another syntactic rule has also been applied.

On the opposite side of the fence, Takagaki (1984) rejects the whole idea that NT is a rule in Spanish, claiming that it is applicable only to a limited group of verbs like creer «to believe» and parecer «to seem» (256). [He does not mention volition verbs like querer «to want».] Compare these verbs, he remarks, with the case of asegurar «to assure»,

where the meaning is «obviously not identical» (256).

To strengthen Takagaki's argument, one could also point to Luján's examples in which a factive verb like lamentar `to regret' occurs in the matrix:

More striking than the case of (3), the two sentences in (4) have virtually opposite meanings, their respective complements expressing presuppositions that cannot both be true (Luján, 62-63). Thus, Takagaki's position that the NT rule is «untenable because it is very low in productivity» (256) appears to be well supported.

It will be shown in this paper, however, that, contrary to what Takagaki claims, the application of NT does not need to be restricted to any particular set of verbs. In the analysis proposed here, NT can, in fact, be applied wherever its structural description is met, its output subject to general semantic restrictions which, it will be argued, are needed irrespective of the sentences in (3) and (4). If this analysis is correct, then Takagaki's only argument against including NT in the grammar of Spanish cannot be maintained.

It will also be demonstrated that, unlike previous analyses involving NT, this solution provides a reasonable explanation as to why the complement of some negated matrix verbs can be expressed in either the indicative or the subjunctive, as in (2b-c). As it turns out, the disagreement mentioned earlier will be resolved in favor of Rivero, to the extent that I will also maintain -albeit for entirely different reasons- that the application of NT necessarily results in a complement for which only the subjunctive mood is possible. It will be shown, though, that none of the previous solutions, including Rivero's, can account for all of the relevant data concerning negated matrices and the choice of mood in Spanish. On the other hand, these facts are easily explained in the semantically-based analysis

––––––––   213   ––––––––

which will be proposed here.

Significantly, this solution utilizes the semantic features which Takagaki (1984) introduced to characterize the semantic interpretation of mood in Spanish¹¹⁸. Thus, while this analysis is used to argue against Takagaki's claim that NT is not a rule, at the same time it provides additional motivation for his treatment of the subjunctive.

Finally, as the title of this paper indicates, the present analysis also entails that there is a «neg-trace» which, it will be argued, occurs in any derived structure to which NT has been applied. This proposal, based on Chomsky's (1981, 1982, 1986) assumption that all movement rules leave a trace, will be crucial to the formulation of the Neg-trace Condition, according to which the semantic interpretation of a verbal complement is affected by the presence of this negative element.

In his article comparing postposed main phrases (PMP) in English with the choice of mood in Spanish, Bolinger (1968) observes that there is a correlation between sentences in English that allow the matrix to be placed after its complement and sentences in Spanish whose complement must be in the indicative. Where the PMP is not permissible in English, then the corresponding sentence in Spanish requires the subjunctive in its complement:

However, as Bolinger notes, the correlation does not appear to hold in the case of negated main verbs expressing belief. If the matrix of (5a) is negated, for example, then in Spanish the complement can be expressed in either the indicative or the subjunctive, as in (7).

Since a PMP in English is excluded in (7a), the indicative form viene in (7b) should also be ungrammatical.

To account for the acceptability of the indicative in (7b), Bolinger proposes that the sentence I don't think he's coming should be viewed as equivalent to I think he isn't coming, the former resulting from the application of NT. Of course, the sentence with the affirmative matrix allows the PMP:

Thus, Bolinger argues, the fact that Spanish admits the embedded indicative in (7b) is not a counterexample to his proposed correlation, despite the ungrammaticality of the PMP in (7a), as long as the indicative version of (7b) can be associated with the PMP in (8).

According to Bolinger, I don't think he's coming has what he calls «mixed status ... the sharing of the negation between two verbs» (23), as evidenced by the fact that in English the negation can be doubled when the matrix is postposed:

This sharing of the negation, Bolinger suggests, plays an important role in the semantic interpretation of sentences which have undergone NT:

If Bolinger is correct on this point, and I believe that he is, then the best account of NT must explain why the negative element which has been raised to the matrix and now negates the main verb is still connected with the complement from which it was extracted.

In Bolinger's view, the application of NT in Spanish results in a sentence whose complement verb is in the indicative. This is so, he contends, because native speakers judge (10a) to be closer in meaning to (10b) than (10c) is [Bolinger, 23]:

(The difference in meaning between [10a] and [10c], too slight to indicate in the gloss, will be discussed in section 3.) Moreover, if (10a) and (10b) have a common underlying structure, as Bolinger suggests, then just like the indicative version of (7b), (10a) fits the parallel between the indicative in Spanish and the PMP in English:

Rivero (1971) rejects Bolinger's analysis of the sentences in (10) in favor of a solution in which the application of NT results in a complement in the subjunctive. (In an earlier paper, Rivero 1970 states that, contrary to what Bolinger's informants report, sentences like [10b] and [10c] can be interpreted as synonymous [20]). According to Rivero, there are two kinds of facts which Bolinger's hypothesis fails to explain. Examples of the first type involve idiomatic expressions like palabra de «a word of» and gota de «a drop of» which are grammatical, she claims, only when the clause in which they occur is negated in underlying structure:

In (12b) and (13b), which both contain a subjunctive complement, palabra de and gota de are acceptable even though the negation is in the matrix. Rivero argues that the grammaticality of these sentences leads to the conclusion that the negation originated in the complement (as in [12a-13a]) and was raised to the matrix by NT. On the other hand, the unacceptability of these idioms in (12c) and (13c) suggests that when the complement is expressed in the indicative, then the negation in the matrix cannot be the result of NT; instead, the negative element must be located in the matrix to begin with in underlying structure. Since the underlying complement clauses in (12c) and (13c) are not negated, she concludes, expressions like palabra de and gota de are therefore disallowed¹¹⁹.

Rivero finds further support for her analysis in sentences which she claims have undergone Sentence-pronominalization. Her examples are given below (310-11):

According to Rivero, (14a) «indicates that John and I are of the same opinion and agree on the fact that Mary will probably not come», whereas (14b) «indicates two opposite beliefs: John doesn't believe that Mary is coming but I do believe it» (311). The derivations of (14a) and (14b) in Rivero's analysis are like those shown in (15a) and (15b) respectively¹²⁰:

As illustrated in (15a), (14a) is derived by pronominalizing the embedded clause María no viene under identity with the embedded clause in the first conjunct; subsequently, NT is applied to this first conjunct, moving no to the matrix. By contrast, (15b) shows that lo «it» in (14b) is pronominalizing the non-negative clause María viene, which is identical to the complement in the first conjunct. NT is not applicable in (15b), the negation originating in the matrix in the first conjunct. Rivero argues that this analysis of (14a) and (14b) thus provides an explanation for the difference in meaning between the two sentences. Moreover, she adds, these sentences support her claim that, contrary to what Bolinger believes, the application of NT results in a complement verb in the subjunctive -never in the indicative.

However, Luján (1978) points out an additional fact about negation in verbal complements which seems to be a clear counterexample to Rivero's hypothesis. A sample of Luján's data is given below (107):

The sentences in (16) show that if a verb like creer «to believe» is negated in the matrix, then

––––––––   215   ––––––––

its indicative complement cannot also be negated. However, if the embedded verb is in the subjunctive, then both the matrix and the complement can be negated:

To account for the restriction on indicative complements, Luján proposes that in (16), the negation occurring in the matrix is originally the negation in the complement clause (107). In the derivation of these sentences, no is raised from the complement to the matrix by NT, she argues; and consequently, no does not appear in the complement in surface structure. Luján states that «this interpretation in turn presupposes the existence of an operation raising the negation of the indicative complement» (107).

Luján also suggests (contrary to what Bolinger and Rivero claim) that there is no semantic difference between the indicative and the subjunctive complements of negated matrices (98). For example, according to Luján, the sentences in (18), derived in this analysis from the deep structure shown in (19), can be interpreted as synonymous (97-98):

In Luján's analysis, indicative and subjunctive clauses differ structurally in that the former (but not the latter) contain the element Tense¹²¹. If no rules are applied to the structure in (19), then the sentence in (20) is obtained:

On the other hand, if NT is applied, then either (18a) or (18b) is derived, depending on whether an optional rule of Tense Deletion (TD) is also applied. Luján assumes that TD is triggered by the negation in the matrix and that it figures in the derivation of (18b): once the complement in (18b) becomes tenseless by TD, then it is expressed in the subjunctive. However, if TD is not applied, then the complement is in the indicative, as in (18a).

When all of the facts presented so far are considered, each of the three analyses sketched above turns out to be inadequate.

For instance, Bolinger cannot account for Rivero's data in (12-13) in which the complement of a negated matrix contains an idiomatic expression like palabra de and gota de. His analysis predicts that if (12a-13a) undergo NT, then the resulting complements will be expressed in the indicative; however, the ungrammaticality of (12c-13c) does not bear out his prediction. Furthermore, given Rivero's assumption that these idiomatic expressions can occur only in a negated clause in underlying structure, Bolinger's solution also predicts, incorrectly, that the sentences in (12b-13b) should be ungrammatical. This is so because in his analysis the subjunctive complements in (12b13b) are claimed not to have contained the negation in deep structure (the negation originating in the matrix instead); therefore, the occurrence of palabra de and gota de in these complements should be unacceptable, contrary to the facts.

Bolinger's solution is also inadequate when the sentences in (14) containing a sentential pro-form (lo) are taken into account. According to his analysis, (14a) is not the result of applying NT; thus, lo in (14a) should pronominalize María venga rather than María no viene (Rivero's interpretation), with también «too» presumably ungrammatical since John and the speaker would be of opposite beliefs. This is completely contrary to the reading which Rivero assigns to (14a). By the same token, Bolinger's claim that the first conjunct in (14b) undergoes NT cannot be maintained; otherwise, the antecedent of to should be interpreted as María no viene (instead of the correct reading María viene), with también possible since both John and the speaker would agree that Mary is not coming. Again, Rivero's interpretation of the sentence is entirely different. It must be concluded, then, that Bolinger cannot account for any of Rivero's data presented above.

Luján data in (16-17) in turn pose a problem for Rivero's analysis. For example, consider (16a-17a), repeated as in (21):

In (21), both the matrix and the complement are negated, indicating that both occurrences of no are «in situ» (unmoved from their original positions in deep structure). Rivero claims that where the negation in the matrix is not the result of NT, then the complement is expressed

––––––––   216   ––––––––

in the indicative. That is, in her solution (22a) undergoes NT but (22b) does not:

By Rivero's own admission, NT cannot have applied in (21)¹²²; yet, unlike (22b), the indicative complement in (21) is ungrammatical. Conversely, the subjunctive in (21) is acceptable even though NT has not taken place. These facts are just the opposite of what Rivero's analysis predicts.

Finally, Luján has no better luck accounting for Rivero's data than Bolinger does. Consider, for example, how the sentence in (12b) would be derived in Luján's analysis:

As shown in (23), sentence (12b) is derived in Luján's solution by first transporting the negation to the matrix via NT and then deleting Tense in the complement clause. Recall, however, that Luján's proposed rule of TD must be optional in order to account for pairs of sentences like (18a) and (18b) in which the indicative and the subjunctive alternate in the complement. This being the case, there is nothing to prevent TD from being skipped in the derivation shown in (23), resulting in the ungrammatical (12c) containing an indicative complement. Thus, there is no way to block (12c) in this analysis.

Luján's solution also fails to account for the difference in meaning between sentences (14a) and (14b). Luján would presumably derive both sentences from the underlying structure in (24):

(24) /Juan cree que María no Tense viene y yo creo que María no Tense viene (también)/

Assuming that Sentence-pronominalization is applied before NT (Rivero 1971, 311), the complement in the second conjunct in (24) changes to lo under identity with María no Tense viene in the first conjunct. Subsequently, no in the first conjunct is raised to the matrix by NT. If TD is later applied, then the sentence in (14a) is derived. However, if TD is skipped, then (14b) is obtained. Since (14a) and (14b) are derived from the same deep structure in Luján's analysis, one would expect these sentences to be semantically equivalent (cf. [18a-b]). However, the ungrammaticality of también «too» in (14b) [but not (14a)] and the different readings which Rivero assigns the two sentences clearly indicate that they are not equivalent, contrary to what Luján's analysis suggests.

For the above reasons, it must be concluded that the analyses which have been proposed to account for facts involving negated matrices in Spanish each fall short of explaining all of the data presented here and must therefore be rejected. To replace these solutions, an analysis is developed in section 3 which is based on two recent proposals, trace theory (Chomsky 1981, 1982, 1986) and independency (Takagaki 1984), notions which until now have appeared to be totally unrelated. Specifically, it will be argued that if a verbal complement from which no has been extracted by NT still retains a trace of the negation in derived structure, then that complement cannot achieve independency (for reasons which will be discussed below) and will therefore be expressed in the subjunctive. Furthermore, it will be shown that in this analysis, NT can be applied freely wherever its structural description is met, its output subject to general restrictions on the assignment of independency features. Thus, matrix verbs do not need to be specially marked in the lexicon to permit the application of NT, as authors have previously assumed. Finally, it will be demonstrated that the present analysis not only accounts for all of Rivero's and Luján's «problem» sentences, but it also provides a formal way of accommodating Bolinger's insight concerning the sharing of the negation between the matrix verb and the complement verb in sentences which have undergone NT Bolinger appears to have anticipated trace theory years before its inception.

In the current literature (for instance, Chomsky 1981, 1982, 1986) it is assumed that when a constituent is moved from one position to another in the derivation of a sentence, it leaves a trace (t) in its original position which, although phonetically null, plays a crucial role in determining whether the resulting sentence is grammatical¹²³. To give a frequently cited example from English, who has been moved to sentence-initial position by WH-movement

––––––––   217   ––––––––

in both (25a) and (25b):

Want to can usually be contracted to wanna in colloquial English: I want to go I wanna go. Notice, however, that while contraction is possible in (25a), as in (26a), the ungrammaticality of (26b) shows that contraction is disallowed in (25b):

These facts can easily be accounted for if «something real» intervenes between want and to in (25b), preventing contraction from taking place (Lasnik and Uriagereka 1988, 22-23). The assumption that t is left behind when movement occurs is thus supported by (26b), in which the trace of who resulting from WH-movement (hence, labeled wh-trace) blocks the contraction.

In keeping with trace theory, I assume that one effect of applying Neg-transportation, as in the application of any other movement rule, is that a trace of the moved element is left behind. This means that a sentence which has undergone NT contains a trace of no -a «neg-trace»- in the verbal complement from which no has been extracted, as illustrated in (27):

The application of NT in the derivation of (27b) moves no to the matrix, leaving behind a neg-trace which, like other traces in this theory, is present in derived structure (and, as I will show, has implications for the semantic interpretation of derived structure). If NT is not applied, then (27a) is obtained.

To understand how the presence of neg-trace can affect the choice of mood in Spanish verbal complements, let us consider Takagaki's (1984) recent proposal concerning the subjunctive as the marker of subordination¹²⁴.

Takagaki hypothesizes that a proposition which is expressed in the indicative has been assigned what he terms «independency» at the level of surface semantic interpretation. He defines the concept as follows (251):

A complement proposition P is «affirmatively evaluated», according to Takagaki, if the main clause expresses the subject's commitment to the truth of P. Being «affirmed», he notes, is a necessary condition for being «stated»; the speaker must «state» the true P, «if he wishes to convey it to the hearer as a normal sentence» (251). The proposition [p the girl is pretty] in (29), for example, is both affirmed and stated.

P is affirmed, because when the speaker utters (29), he is expressing his belief that P is true. At the same time, P is stated: the very utterance of P essentially has the «stating force» (Takagaki's term) of the simple declarative sentence La muchacha es bonita. P achieves independency, Takagaki claims, only if it possesses the stating force. If it does, then P is expressed in the indicative. However, if P is lacking the stating force, then it is not assigned independency but instead remains as the subordinated complement and is expressed in the subjunctive.

According to Takagaki, P remains subordinated if it meets either of the following conditions (251):

He offers the examples in (31) and (32) as corresponding to (30a) and (30b) respectively.

The main verbs in (31) expressing doubt and hope do not affirm that P is true; and since in this analysis affirmation is a prerequisite for stating, P does not come to be stated. In (32), the complement expresses a presupposed P which gets affirmed, Takagaki claims, because the speaker assumes «in advance» that P is true and proceeds to express his reaction to P in the main clause. However, Takagaki continues, P in (32) «remains unstated as a kind of "background material"», the speaker focusing his attention on what he wishes to convey in the main clause (252).

––––––––   218   ––––––––

Lacking the stating force in the sentences in (31-32), P remains subordinated and is thus expressed in the subjunctive.

In what follows I adopt both Chomsky's trace theory and Takagaki's independency hypothesis. In particular, I assume that 1) a proposition P from which no has been extracted by NT contains a neg-trace, and 2) P remains subordinated if it does not get affirmed.

In light of these assumptions, consider again sentences containing a negated matrix whose complement can be expressed in either the indicative or the subjunctive:

The examples in (33-34) are used by Bell (1980, 382-83) to illustrate what he calls the «semantic complexities» of expressing the complement in the indicative in a sentence that «could be expected to take the subjunctive» (382). In the case of the negated first-person matrix in (33a), the indicative complement states someone else's belief that the embedded proposition [P the Prophet will return] is true. According to Bell, (33a) «represents the speaker making an explicit denial of a statement (el profeta vuelve) made by another person, or otherwise implied by the context» (382). On the other hand, in (33b), the speaker is simply expressing his belief that P is false, without implying that some one else believes P to be true. A similar contrast can be seen in (34), in which the negated matrix is in the third person. Bell observes that in (34a), «Isidro holds a certain belief, which is contrary to that held by the speaker; i. e., the speaker believes that the Prophet is returning» (383). In (34b), however, only Isidro's belief is reported, the speaker remaining uninvolved in the context.

In keeping with Takagaki's analysis, I assume that at the level of surface semantic interpretation, that is, after the syntactic rules in the grammar have had a chance to apply, P in (33a-34a) gets affirmed and also comes to be stated; in (33b-34b), however, P does not get affirmed and therefore cannot be stated. The question remains, though, as to how sentences (a) and (b) are distinguished at the level of surface structure such that P can be interpreted as affirmed in (a) but not in (b). I believe that the correct answer to this question lies in trace theory: one of the sentences in each pair has undergone NT and thus contains a neg-trace in the complement. The other sentence contains no such trace, the negation originating in the matrix.

I propose that the sentences in (33b-34b) containing a subjunctive complement undergo NT in their derivations and that they are subject to the following restriction on the semantic interpretation of propositions:

The NTC states, in effect, that the speaker (or the higher subject if distinct from the speaker) cannot be committed to the truth of P, if P contains a neg-trace -a trace of raised no which, although phonetically null, is interpreted as a denial of the truth of P Since P containing a neg-trace does not get affirmed, therefore it remains subordinated and is expressed in the subjunctive. For example, the derivation of (33b) in this analysis proceeds as in (36):

In contrast, the rule of NT (hence, the NTC) is not applicable in the derivation of (33a);

therefore, the embedded proposition gets affirmed (by someone other than the speaker, as Bell observes) and comes to be stated in the indicative.

In an earlier paper (Reider, 1989), I concurred with Takagaki's (1984) statement that whether a complement is assigned independency depends on the meaning of the main clause. For example, in discussing the occurrence of the subjunctive in the sentence in (38),

Takagaki remarks, «It is usually claimed that certain doubt in the complement triggers the subjunctive, but in my opinion, it is attributed to the negative meaning in the main clause,

––––––––   219   ––––––––

which deprives the complement of the independency» (251). While I still agree that the negative meaning of the main clause weighs heavily in the decision to use a subjunctive complement, the facts in (33-34) clearly indicate that some other factor must also be involved, at least as far as negated verbs of belief are concerned. In the solution advocated here, that second factor is the presence of neg-trace in the complement which prevents P from getting affirmed.

Significantly, by virtue of the fact that it is a trace of no, neg-trace always co-occurs with a negated main clause. It might be suggested, then, that it is really the combination of a negated matrix and a negated complement that triggers the subjunctive. Evidence that this is so is provided by Luján's data in (16-17), in which no occurs in both the matrix and the complement, requiring an embedded verb in the subjunctive. Consider, for example, the close similarity of the derived structures in (39): (39a = 17a)

The above sentences are identical except that the negative element in the complement is no in (a), neg-t in (b). That the subjunctive is required in both sentences suggests that at the level of surface semantic interpretation, the combined force of the negated matrix and the negated complement prevents P from getting affirmed, in (a) as well as (b). Thus, Luján's data are not only consistent with the present analysis, but they also support the assumption that neg-t behaves exactly like (embedded) no regarding the manner in which it collaborates with the negated matrix to affect the semantic interpretation of P.

This analysis also accounts for Rivero's examples, and it actually lends credence to her claim that the application of NT always results in a complement verb in the subjunctive. First, sentences involving idiomatic expressions like palabra de and gota de are derived here by NT, just as in Rivero's solution; but in this analysis no leaves a trace,

which along with the NTC ensures that the resulting complement is expressed in the subjunctive, as required. Moreover, this solution accounts for the difference in meaning between sentences (14a) and (14b), whose structures at the level of surface semantic interpretation can be represented as in (41a) and (41b) respectively:

Recall that, according to Rivero, sentence (a) means that John and the speaker agree on the fact that Mary will not come; on the other hand, (b) indicates disagreement: John doesn't believe that Mary will come but the speaker does believe it. The antecedent of lo «it», that is, the embedded proposition in the first conjunct, is shown in square brackets in each sentence. In sentence (a), P contains a neg-trace and, as expected, lo is interpreted as Mary will not come. However, in (b), lo refers to the affirmative proposition Mary will come and is so interpreted, again exactly as expected. Thus, the presence of neg-trace in (a) not only accounts for the difference in mood between the complements in (a) and (b), but it also explains the difference in the meaning of lo in these sentences -without resort to a rule of Sentence-pronominalization (as in Rivero's solution), a transformation which has lost favor in recent years¹²⁵.

Finally, this analysis provides a formal way of characterizing Bolinger's proposed «sharing of the negation» between the main verb and the embedded verb in sentences which have undergone NT Raised no in the higher clause is associated with its trace in the lower clause; and, as I have argued, the combined presence of these negative elements has implications for the semantic interpretation of P (in accordance with the NTC) and ultimately for the choice of the subjunctive in the verbal complement.

Thus, unlike any of the previous solutions involving NT, the analysis proposed here accounts for all of the facts presented in section 2; and it provides a unified treatment of sentences containing both a negated matrix and a negated complement, whether the negative element in the latter is no (as in Luján's examples in [17]) or neg-trace. It must be concluded, therefore, that this analysis is to be preferred over any other solution proposed

––––––––   220   ––––––––

to date.

Having adopted the trace theory approach to describing the behavior of no in Spanish, in section 4 I discuss the possibility that NT can be applied freely wherever its structural description is met, in view of Takagaki's claim that the rule does not exist because it is applicable only to a few verbs.

Recall that in order to support his claim that NT is low in productivity, Takagaki points to the fact that the sentences in (3), repeated in (42), are not identical in meaning.

Takagaki does not state what he believes the sentences in (42) show with respect to NT; however, his calling attention to the semantic difference between (a) and (b) suggests that his reasoning is based on the hypothesis that transformations cannot affect meaning. Originally formulated by Katz and Postal (1964) [hence, known as the «Katz-Postal Hypothesis»], this proposal entails that in the derivation of a sentence, it is the level of underlying structure (rather than derived structure) which is most suited to semantic interpretation (Newmeyer 1980, 79). Under this approach, transformations are claimed to be «meaning-preserving», in the sense that they preserve (i. e., do not change) the semantic interpretation which has been assigned to the underlying P-marker.

According to the Katz-Postal Hypothesis, then, if the two sentences in (42) are derived from the same underlying structure, their derivations differing only in that NT has applied in (b) but not in (a), then they must have the same meaning. The fact that they are not synonymous thus leads Takagaki to conclude that 1) sentences (a) and (b) are not derived from the same underlying structure, and 2) sentence (b) cannot have undergone NT. (I am inferring that this is the thrust of Takagaki's argument, based only on his observation that [a] and [b] are not identical in meaning).

To say the least, the Katz-Postal Hypothesis has not received widespread acceptance among linguists. Those who have rejected the proposal, including Chomsky and his followers, have constructed numerous counter examples, arguing that the application of transformational rules can affect meaning and that therefore surface structure is relevant to semantic interpretation (Newmeyer, 124, 136). Others have maintained that the Katz-Postal Hypothesis is correct and that all interpretation takes place at deep structure (Newmeyer, 96).

My purpose in pointing out this controversy is not to argue for or against the Katz-Postal Hypothesis, but to show that Takagaki's preoccupation with the difference in meaning in (42) is not consistent with his main assumption that semantic interpretation (and consequently the choice of mood in Spanish) is determined at the level of surface structure. Indeed, if the rule of NT must be meaning-preserving, as Takagaki seems to be advocating, then in strict accordance with the Katz-Postal Hypothesis, surface structure is irrelevant to semantic interpretation. This being the case, either Takagaki's assessment of the significance of (42) is incorrect or his entire approach to describing the choice of mood in Spanish must be rejected. The position taken here is that the application of NT can, in fact, affect the semantic interpretation of the resulting surface structure and that, moreover, as I have argued in section 3, the trace which this rule produces plays a role in determining whether a proposition gets affirmed-hence, whether it can be assigned independency at the surface level, in accordance with Takagaki's main proposal. Thus, in the present analysis, the two sentences in (42) can be derived from the same underlying structure, with NT having applied in the derivation of (42b).

Although the application of NT is not limited to any particular class of verbs in this analysis, there are sentences containing a negated matrix which, for purely semantic reasons, cannot have been derived by NT. To see why, consider again Luján's examples in (4), repeated in (43), in which a factive verb is involved.

As in (42), the fact that the sentences in (43) do not have the same meaning does not in itself block the application of NT. Suppose, then, that the rule is applied, as illustrated in (44):

As in previous examples, since neg-t is contained in the embedded proposition P, the NTC requires that P not get affirmed. Notice, however, that unlike any other negated matrix presented so far, the negated factive no lamento «I don't regret» is an inherently «affirming» matrix, its complement expressing a proposition which is presupposed to be true. Indeed, the sentence in (43b) [also (43a)] falls under Takagaki's condition in (30b): P is affirmed, but does not come to be stated.

I assume that if P is embedded in an affirming matrix like no lamento, then it must get affirmed. If so, then (43b) cannot be derived as shown in (44), since the NTC would prevent P from getting affirmed. In effect, the derivation in (44) results in an irresolvable semantic paradox (P must be both affirmed and unaffirmed); therefore, the surface structure in (44) is simply uninterpretable. The only possible derivation of (43b), then, is that the negation originates in the matrix, with no neg-trace in the complement. In that way, P can get affirmed, as required by the meaning in the main clause.

To the best of my knowledge, there are only two types of negated matrices which are inherently affirming, that is, which require that embedded P gets affirmed: negated factive verbs, as in (43b) [also no deplorar «not to deplore», no alegrarse de «not to be glad», no sentir «not to regret», etc.); and negated verbs of doubt, especially no dudar «not to doubt» and no negar «not to deny». Because these matrices govern a P which must get affirmed, their negation must therefore not have originated in P; otherwise, if P contains a neg-trace owing to the application of NT, then, as in (44), the derivation is ruled out.

Thus, unlike previous analyses, in which verbs are claimed to be specially marked in the lexicon to allow the application of NT, the present solution claims that NT can be applied freely wherever its structural description is met, its output subject to the interaction of the NTC and the assigned meaning of the negated matrix. Contrary to what Takagaki suggests, then, NT is viewed here as a highly productive rule which can affect the meaning of surface structure, as long as it does not result in a derivation which is uninterpretable.

In this paper an analysis has been proposed in which the application of Neg-transportation results in an embedded proposition P containing a trace of the moved negative element; and, as prescribed by the Neg-trace Condition, P does not get affirmed. In accordance with Takagaki's hypothesis, then, P remains subordinated to the main clause and is expressed in the subjunctive. Additionally, though, I have argued that in the case of inherently affirming matrices, the presence of neg-trace in the complement is semantically anomalous (P is unable to get affirmed as required) and that therefore the derivation is ruled out. Thus, without imposing any restrictions on the NT rule itself, the Neg-trace Condition limits the interpretation of surface structures which have undergone the transformation; and it thereby serves as a crucial link between the syntactic and the semantic components in the grammar of Spanish.

The semantically-based analysis developed in this study utilizes both Takagaki's notion of «independency» and Chomsky's trace theory, two quite different proposals which, when carried to their logical conclusions, have lead to the formulation of the Neg-trace Condition. If this analysis is correct, then it provides strong additional support for both of these already well-founded proposals.

Bell, Anthony. «Mood in Spanish: A Discussion of Some Recent Proposals». Hispania 63 (1980): 377-90.

Bolinger, Dwight. «Postposed Main Phrases: An English Rule for the Romance Subjunctive». The Canadian Journal of Linguistics 14 (1968): 3-30.

Chomsky, Noam. «Conditions on Transformations». A Festschrift for Morris Halle. Ed. S. Anderson and E Kiparsky. New York: Holt, Rinehart and Winston, 1973. 232-86.

___. «Conditions on Rules of Grammar». Linguistic Analysis 2 (1976): 303-51.

___. Lectures on Government and Binding. Dordrecht: Foris, 1981.

___. Some Concepts and Consequences of the Theory of Government and Binding. Cambridge: MIT P, 1982.

___. Knowledge of Language: Its Nature, Origin, and Use. New York: Praeger, 1986.

Contreras, Heles. «Clause Reduction, the Saturation Constraint, and Clitic Promotion in Spanish». Linguistic Analysis 5 (1979): 161-82.

Horn, Laurence R. «Remarks on Neg-Raising». Syntax and Semantics IX. Ed. Peter Cole. New York: Academic, 1978, 129-220.

Katz, Jerrold and Paul Postal. An Integrated Theory of Linguistic Descriptions. Cambridge: MIT P, 1964.

Lasnik, Howard and Juan Uriagereka. A Course in GB Syntax: Lectures on Binding and Empty Categories. Cambridge: MIT P, 1988.

Luján, Marta. «Clitic Promotion and Mood in Spanish Verbal Complements». Montreal Working Papers in Linguistics 10 (1978): 103-90. Rpt. in Linguistics 18 (1980): 381-484.

Newmeyer, Frederick J. Linguistic Theory in America: The First Quarter-century of Transformational Generative Grammar. New York: Academic, 1980.

Prince, Ellen F. «The Syntax and Semantics of Neg-Raising, with Evidence from French». Language 52 (1976): 404-26.

Reider, Michael. «Clitic Promotion, the Evaluated Proposition Constraint, and Mood in Spanish Verbal Complement». Hispania 72 (1989): 283-94.

Rivero, María-Luisa. «Estudio de una transformación en la gramática generativa del español». Español actual 17 (1970): 14-22.

___. «Mood and Presupposition in Spanish». Foundations of Language 7 (1971): 305-36.

Suñer, Margarita. «Clitic Promotion in Spanish Revisited». Contemporary Studies in Romance Languages. Ed. Frank Neussel, Jr. Bloomington: Indiana U Linguistics Club, 19 80. 300-30.

Takagaki, Toshihiro. «Subjunctive as the Marker of Subordination». Hispania 67 (1984): 248-56.

Hispania [Publicaciones periódicas]. Volume 73, Number 1, March 1990