Syntactic Structures reformed
This section explores an alternative to Lasnik’s hybrid analysis. In contrast to the hybrid analysis, in which an inflectional element can be introduced into a derivation (as an affix or a set of features) independent of the lexical form it inflects, this alternative is based on the assumption that lexical items enter the derivation fully inflected, hence affixes will not be disassociated from the stems that they attach to at PF. Furthermore, there will be no putative distinction between an inflectional affix and corresponding inflectional features. One immediate consequence of this lexicalist analysis is that there is no basis for positing rules of affix hopping and do-support.
Let us set aside the issue of the form in which main V enters the derivation (which was considered at the end of the previous section) and turn instead to the question of the linear order of verbal elements, which has a direct bearing on it. If the linear order of verbal elements cannot be stipulated via phrase structure rules, then one possibility is that the order follows from lexical properties of the verbal elements themselves. Selection is an obvious candidate (see Freidin 1992 for a detailed analysis).30 Modals and infinitival to select a bare V (auxiliary or main); perfective have selects the past participial form and progressive be, the present participial form. Selection will account for the distribution of affixes without recourse to affix hopping. In this way it also accounts for the order of auxiliaries as they occur. And it may also provide the full account of the order of auxiliaries—that is, why the verbal elements can only occur in the order in which we find them. For example, the order perfective+modal would be prohibited because modals only occur as tensed forms and therefore cannot satisfy the selectional requirement of perfective have.
That modals and infinitival to both select a bare form demonstrates a common property. Let us assume that this is actually a categorial property: both are instances of T. Thus when a modal enters a derivation it is, like infinitival to, merged as T (not V as in Lasnik’s analysis). The fact that modals and infinitival to are in complementary distribution lends further support to this analysis. Whether we assume bare phrase structure theory or not, we need to account of this complementarity.
Periphrastic do also has the property of selecting the bare form. Moreover, it too is in complementary distribution with modals and infinitival to. Let us also consider it to be an instance of T—perhaps uncontroversially because in other analyses it always adjoins to the tense inflection T. However, controversially, let us assume that periphrastic do exists in the lexicon in its various finite forms do/does/did33 and is merged fully inflected as T. This creates several problems, some of which I attempt to solve in what follows.
First, there is the question of why, if it exists as an item in the lexicon, periphrastic do does not occur more generally. For example, why isn’t (1b) allowed along side of (1a) where both have the same interpretation?
(1)
One possible answer is that this is a case of blocking where left blocks did leave. However, it isn’t exactly like the standard cases of morphological blocking where one form in the lexicon blocks the existence or use of another—for example, as went blocks *goed. Rather, this has the flavor of an economy constraint, which for lack of a better term I will call Morphological Economy (henceforth ME). ME requires that the fewest grammatical formatives are used to express a grammatical concept. In this case, ME prefers one formative left to the two in did leave. Thus, (1b) is only permitted when did receives emphatic stress, thereby distinguishing did leave from left.
ME has some further appealing consequences. It accounts for the nonoccurrence of periphrastic do with aspectual auxiliaries.
(2)
(3)
ME works as well for the cases with emphatic stress, where the comparison will be between the stressed auxiliary alone and the stressed periphrastic plus the aspectual auxiliary. Furthermore, it accounts for the nonoccurrence of periphrastic do (stressed or unstressed) with the copula, as in (4).
(4)
Additionally, ME accounts for the nonoccurrence of unstressed periphrastic do in wh questions where the wh-phrase is in subject position.
(5)
If we assume that the derivation of (5b) involves both head movement from T to C and wh-movement to Spec,CP, then while there will be traces in both T and Spec,IP (italicized as in (6)), its PF representation would presumably have did in C adjacent to leave in VP because the traces have been eliminated.
(6)
(7)
If we allow ME to compare the morphological strings does not be and is not, ruling out (7a) as a violation, then ME should in principle prefer the deviant (8b) to the well formed (8a), all things being equal.
(8)
However, (7) and (8) are not equivalent. We assume that (8b) is deviant for some reason not connected to ME, whereas the only explanation for the deviance of (7a) is ME. Thus, ME compares morphological strings that are otherwise allowed by the grammar. It therefore could not compare (7a) to (7b).
Without ME, examples like (7b) demonstrate the appeal of the SS analysis over alternatives. Under the Chomsky/Lasnik analysis, (7b) cannot be generated at all because not will block affix hopping. If (7b) is generated by merging the inflected verb with not as a first step, then some other explanation of deviance is required. Let us assume that (7b) is a TP and therefore contains a covert T. The uninterpretable of leave are on the verb itself, so that in order to be checked, they must be moved to T where spec-head agreement with the subject will eliminate them. The deviance of (7b) could therefore be a result of the failure of these features to be checked. Presumably, the presence of the negative will block covert movement from V to T. Notice that this could not be a consequence of the Head Movement Constraint, taking not to be a head. That is because in (7a) did selects the bare form leave. If not were a head, then this selection would be blocked.
Before we turn the issue of V-to-T movement for main verbs, it will be useful to consider how bare phrase structure theory might affect the analysis of auxiliary verbs. Given that a finite auxiliary verb will always wind up in T position, is there any reason to merge it as a V first and then raise it to T? If not, then on simple economy of derivation considerations there would be no V-to-T raising for finite auxiliaries. As far as I can tell, the raising analysis is just an artifact of the old top-down perspective of phrase structure rules. That is, given that aspectual auxiliaries sometimes co-occur with an independent inflectional element (e.g., a modal), they can occur in a non-T position. Phrase structure rules force this to be the general and underlying case. Building phrase structure bottom up—hence without phrase structure rules—there may be no reason not to merge a finite aspectual auxiliary as T. In contrast, there is a reason to merge a main verb as a V and then move it to T if that is allowed—namely, V constructs with its (internal) arguments to form a predicate in which θ-roles are assigned.
If this perspective is correct, then the only difference between languages like English and French is that main V can raise to T in the latter but not the former. It is not the case that the process of V-raising is more general in French than in English because there is no raising of finite auxiliaries in either language.
This analysis suggests that predicate adjectives and nominals in finite sentences are just that—that is, an AP or NP complement of T, not a complement of a V Thus, (9) would have the structural analysis (10).
(9)
(10)
As noted by Len Babby (p.c.), this structure would explain why predicate nominals and adjectives do not occur in the structural accusative Case in languages with rich overt morphological Case.
On the potentially questionable side, it may require that the corresponding infinitival forms of predicate NPs and APs have a different structure from their finite forms. Consider, for example, (11).
(11)
In (11a), given that T is filled by to, be will have to be the head of an independent projection, in contrast to (11b) where the finite form is is merged as T. Whether this lack of parallelism is a problem remains to be determined.
So far, the basic assumption of this alternative lexicalist account—that all verbal elements come from the lexicon fully inflected—leads to no serious empirical problems or obvious conceptual defects. We have added to it the further assumption that the “tensed” verbal element, with the exception of main verbs, enters the derivation as T.
Under this analysis, the order of verbal elements is determined by natural selectional properties of the verbal elements. Furthermore, the distribution of periphrastic do is governed by a principle of morphological economy. Of the three empirical issues for the analysis of the verbal system (6a-c), what remains is an account of the one true case of V raising, the movement of a main verb to T and the parametric variation associated with it.
Given that the uninterpretable on English finite main verbs must be checked before the LF interface, then one plausible way to eliminate them is via covert raising of V to T (e.g., Chomsky 1993). English and French finite clauses would then have the same structure at LF. The difference between the two languages is limited to the movement of finite main verbs to T: overt in French and covert in English. From the perspective of bare phrase structure theory as developed above, it’s a tossup which one is the special case. Looking at this difference in terms the contrasting metaphors of “procrastinate” versus “earliness” does not add anything useful. We would still have to account for why the other language is able to escape the general strategy. Thus, it seems that we have not progressed much beyond the stipulation in SS via transformational rules that English finite main verbs cannot occur to the left of either negation or a subject NP. A principled account that distinguishes between PF positioning of finite auxiliary versus main verbs in English remains a desideratum for some future account.
Nonetheless, we can sharpen our focus on the residual problems a bit more by scrutinizing an implication of the analysis based on ME. Given that ME operates on the PF side of the derivation, then whatever rules out (8b) should also apply on the PF side. Otherwise, the application of ME would require that information from the LF side be available to the PF side. Let us continue to maintain what has been more or less tacitly assumed since Chomsky and Lasnik 1977 first proposed the derivational split for PF and LF—namely, that when the derivation splits into two separate tracks, neither one has access to information from the other. Given this, the explanation for the deviance of (8b) cannot be that negation blocks covert V-to-I raising, because ME at PF would not have access to that information. Instead, the deviance of (8b) must follow from some problem on the PF side.
There are prima facie two ways to account for the deviance of (8b) on the PF side: either the derivation is blocked by some condition on the application of rules, or the structure that is derived violates some general output condition (filter) on representations at PF. To determine which of these options is correct (possibly both), we need a comparison case whose derivation yields a well-formed result. Therefore, a prerequisite to an account of (8b) is an understanding of the derivation of the well-formed (12).
(12)
Given that the clause (12) is structurally an TP, the derivation of such clauses, where T is covert because V must not raise to T overtly, raises several fundamental questions.
Perhaps the most fundamental question for this analysis concerns the origin of T. If the V left comes from the lexicon fully inflected, then what could T be? Minimally it will at some point in the derivation contain the tense and agreement features of V so that the agreement relation with the subject in Spec,IP can be checked. However, there is no need to postulate in the lexicon the independent existence of a T with tense and agreement features, because this element duplicates information already specified in the lexical entries of the various verbal elements when they enter the derivation. In the case of finite auxiliaries, the TP projection is created by merging the finite form with its VP complement. The question, then, is what happens when the finite form is the main verb. Given that the verb is prevented from moving to the T position—that is, from projecting T (unlike auxiliary elements), but nonetheless T is required and must project to form a clause structure—a minimal solution would be to extract just the tense feature of the verb to merge with maximal projection of the verb while projecting T. Thus, the structure of (12) would be (13), where left also contains the same set of features.
(13)
Note that we are assuming that the of left must still be moved to T covertly at LF to be checked.
Given the representation (13) for (12), the corresponding representation for (8b) would be (14).
(27)
Both (13) and (14) contain a chain, the tense feature [+past] in both V and T. Given that a chain is always an illegitimate object at PF (and in morphological structure; MS), we must assume that the chain in (13) is eliminated at SpellOut. Presumably, the head of the chain, the tense feature in T, remains, whereas its trace in V is deleted. At MS/PF, where the hierarchical structure has been eliminated as well, the feature set will be adjacent to the V that now lacks this feature. It is plausible that MS simply reads the adjacent tense feature as part of V In contrast, the elimination of the chain in (14) creates a representation at MS/PF where the tense feature that belongs with the V is separated by a phonetically realized morphological unit not. The presence of not blocks the interpretation of [+past] as part of V. The fact that [+ past] in this configuration cannot be interpreted as part of an adjacent lexical item renders the representation deviant— presumably a violation of Full Interpretation. This analysis is analogous to Lasnik’s Stranded Affix Filter analysis but not identical because there is no need to identify the tense feature as an affix or postulate such a filter. Alternatively, it could be that V, which now lacks tense, constitutes the deviant element.
The alternative analysis presented here is obviously a radical departure from that of SS and its successors. Some of this analysis is motivated by relatively recent developments in syntactic theory (e.g., bare phrase structure theory) and the rest is informed by analytic concepts developed over the past almost four decades that have deepened our understanding of the structure of human language. It is offered as another way to think about the verbal morphology system, following Feynman’s observation that having multiple theories to compare gives us a better chance at guessing at the truth. The case of the verbal morphology system remains central to a theory of syntax because it is primarily through the analysis of this system that we derive an explicit theory of the structure of the clause.
