«I. Introduction Consider sentences like (1): 1. Null Complement Containing Sentences a. Aryn followed b. Marie-Odile promised c. Corinne left d. ...»
IV. Problems with Fillmore’s Taxonomy, and Our Alternative We said there were two problems with Fillmore’s account. The first had to do with the meaning effects he suggests for null complements. The second, which is our focus now, has to do with the taxonomy he provides. That problem comes into focus when we inquire why there are verbs (or, more precisely, verb-senses) marked as INC at all. Since, as just argued, what had been called DNC verbs can be either free or bound, why also have null complement verbs which can only be (existentially) bound? Or again, why can verbs like ‘eat’ never be read referentially, if they genuinely belong to the same syntactic class as ‘follow’? And why can INC verbs never be read as bound by some contextually supplied non-existential quantifier, if they share the same syntax as ‘leave’? (E.g., ‘Whenever Moonisah cooks lasagna, John eats’, does not have a reading on which it means Whenever Moonisah cooks lasagna, John eats it.) This might just be a brute semantic difference not reflected in the sentential syntax: something which must be stipulated in the lexical entry of INC verbs. But we think a more insightful treatment is available – a treatment which reconceives things rather radically.
We will defend the unorthodox idea that all genuine null complements are interpreted as variables. If their syntax is really that of a null complement, with the requisite lower number of realized non-subjects compared to what the semantic frame prescribes, their semantics is that of a variable. All can thus be read as referential or quantified, depending upon context. And none is restricted to existential quantification.
That is the single semantic rule that governs all sentences which contain null complements.5 Verbs which have obstinately existential readings, the so-called INC verbs, we reclassify as not really null complement verbs after all. Verbs like ‘eat’ -- which, when used without a complement, cannot mean ‘V-ed it’, with ‘it’ either free or bound by a non-existential quantifier -- are marked in the lexicon as intransitives. Hence not, in fact, as verbs which allow null complements. This preserves the generalization regarding semantic effect, does away with the apparent exceptions (i.e., the ones which are always quasi-existential), while still accommodating the existence of so-called INC verbs: the latter, it turns out, are plain old intransitives.
But, it will be objected, this introduces an unacceptable ambiguity. There is now the intransitive ‘eat’, which is what we see when the verb appears alone, and in addition the transitive verb, which is what appears with an overt complement. Continues the objection, this is unacceptable on two fronts. First, specifically with respect to introducing the intransitive ‘eat’, whenever someone eats, they eat something; but, goes the worry, treating ‘eat’ as intransitive we will miss this fact. Second, with respect to both entries, we are positing an ambiguity without necessity here. Our reply to the first worry is that this fact, i.e., about eating requiring that something be eaten, is real enough, but is not reflected in the semantic frame of the intransitive verb. The thing-eaten is not a linguistically inherent participant, the verb being intransitive, but it is a “metaphysically” inherent one – and that’s enough to capture the fact in question. (To see the reasonableness of this suggestion, compare: whenever someone eats, they eat somewhere;
that, however, is patently a poor reason for thinking that ‘eat’ is linguistically marked as expressing a three-place relation between an agent, a thing-eaten, and a location.) Our reply to the second worry is that there is no ambiguity posited without necessity, because we do not believe that there is a transitive verb ‘eat’! Instead, the sole semantic frame for ‘eat’ is (29):
29. a. Verb: ‘eat’ b. Semantic Frame: [Participant = Agent] This, in turn, raises an obvious question: How, then, can ‘eat’ take a complement, if it’s always intransitive? ‘Die’ and ‘fall’, genuine intransitives, cannot. The answer is that ‘eat’ is an intransitive verb which is marked for optional addition of an argument.
Whereas ‘promise’ is marked [+ omit complement] (and ‘lock’ is marked [- omit complement]), ‘eat’ is marked [+ add complement]. (And ‘die’ is marked [- add complement].6) Sentences which have previously been classed as transitive occurrences of ‘eat’, such as (30) and (31), we now classify as occurrences of the intransitive ‘eat’, but with an added argument.
30. Renald ate [a sandwich]
31. Tracy ate [that apple you bought] Thus ‘eat’ is not ambiguous after all: it is univocally intransitive, though marked [+ add complement]. In sum, in contrast to Fillmore’s taxonomy, given above, our taxonomy
looks like this:
This approach has two further advantages. First, it explains why it is difficult to find a complement-containing paraphrase of ‘No one ate’ (and similar sentences), a paraphrase which shares all of the logico-semantic properties of (19), including scoping possibilities: this is so difficult because ‘No one ate’ is the underived form. ‘No one ate’ is not derived by the phonological omission of ‘something’ from (18). It’s not derived by the phonological omission of any term. It’s not surprising, then, that (19) doesn’t exactly mean ‘No one ate something’, or even ‘No one ate stuff’. ‘No one ate’ means, rather, that no one ate. Second, it explains a usage-based contrast between ‘eat’ and true null complement verbs. Informal questioning by the authors revealed that, in the vast majority of cases, untrained speakers, when instructed to give examples of sentences containing words like ‘follow’ and ‘promise’ – which we class as genuinely exhibiting null complements – provided sentences in which a complement is phonologically realized.
Examples offered included ‘You promised me I could have a hamburger’, ‘You promised me you would clean your room’, ‘I promised to finish my homework tonight’, ‘She promised to marry me. I was thrilled’, ‘They promised the world but never delivered’, ‘He follows in his father foosteps, ‘My favorite program follows the news’, ‘They followed me right into the trap’ and ‘I told him to follow me’. Not attested were things like ‘He followed’ and ‘They promised’ tout court. In contrast, when asked to give examples of sentences containing ‘eat’, ‘sing’ and ‘sew’ -- which we believe do not genuinely exhibit null complements, but are rather [+ add complement] -- untrained speakers strongly tended to provide sentences without complements. Though certainly not the only one, one obvious explanation for this usage pattern is that, in giving example sentences, speakers tend to opt for the “default” specified in the semantic frame – neither adding nor omitting complements, even though this is optionally permitted. If that’s what speakers are doing, in giving example sentences, the result we predict is that ‘promise’ and ‘follow’ would appear in their “basic form”, with a complement, while ‘eat’ and such would equally appear in their “basic” form, without a complement. And this is just what our initial testing found.
Conceiving things this way, we save our elegant generalization about what null complements mean. But, turning to another possible objection, don’t we do so at the cost of introducing a bizarre unattested feature? Have we not avoided the mystery that Fillmore faces – viz., why there are two kinds of meaning effects for a single syntactic construction – by fabricating another? The point is well taken: at first glance, [+ add complement] may seem a very curious feature for a verb to have. In fact, however, something of this kind is attested cross-linguistically. East Cree, for example, has a class of verbs that are morphologically intransitive, but which can nevertheless take third person objects. For instance, ‘minihkweu’, meaning s/he drinks, is morphologically marked as intransitive: the verb does not carry the usual affixes which appear on transitive verbs. Nevertheless, unlike other intransitive verbs, this intransitive verb can optionally take animate (e.g. ‘milk’) or inanimate (e.g., ‘tea’) objects. In short, ‘minihkweu’ is overtly [+ add complement].7 Interestingly, such verbs in East Cree pass
Fillmore’s test for (so-called) INC verbs. Thus (32) is perfectly felicitous:
32. Chii minihkwe-u David. Eishi chekwaayuu chii minihkwe-u?
past drink-3 David. Wonder what past drink-3 “David drank. I wonder what he drank.” Something quite similar occurs in Blackfoot: some intransitives are marked [+ add complement], and semantically they correspond to what traditionally would be treated as (obstinately existentially quantified) “null complements”. What’s more, the class of intransitive verbs that take an object can, in Blackfoot, only take nouns that are inflected with a non-referential suffix. (See Frantz 1991: 40-41.) So, our semantic generalization – it is actually intransitives that are always heard as existentially quantified -- is partly mirrored in overt syntax, in Blackfoot: when verbs are intransitive, but are used with an object, they can only take morphologically non-referential objects. This suggests that the property of being non-referential/quantificational is, as hypothesized, associated with (morphological) intransitivity. In sum, Algonquian overt verbal morphology independently suggests an “add complement” approach: intransitive verbs to which an object can be added. (This in contrast to the more traditional “omit complement” approach: a transitive verb from which an object can be omitted.) Nor is the Algonquin data just an isolated case. “Valency increase” (i.e., addition of a core argument to a verb) is a common derivational procedure, employed across languages to create new words (or constructions). In English, a beneficiary can be added to a transitive to yield a ditransitive applicative: ‘sings’ gives rise to ‘sing me a song’.
(Something similar seems to occur with causatives derived from intransitives: ‘walk’ gives rise to ‘walk the dog’.) And in some languages, even an intransitive can have a beneficiary added, to become an applicative verb. (See Dixon and Aikhenvald 2000 for discussion.) Given this cross-linguistic evidence, we can, without introducing anything especially odd, hold to our generalization that all genuine null complements are understood as if they had a variable – which can be either deictic/free or bound. We can sustain the generalization precisely by denying that ‘eat’-type verbs are genuine null complement verbs after all. Their normal form is without a complement. It is ‘eat’ with a complement that is special.
V. Our Syntax for Null Complements We have said that genuine null complement constructions – of which ‘Anita ate’ is not an example, being instead an intransitive – are read “as if the empty spot was a variable”.
That variable can be free, in which case the sentence is understood as if it had a referring context-sensitive term; or it can be read as if bound by a quantifier, supplied either by prior linguistic material or by non-linguistic context. This is our positive view about the semantics of null complements. But what syntax would we propose for the resulting null complement containing sentences?8 In particular, taking ‘The press followed’ as our example, is there a special never-pronounced empty element where an overt complement would typically sit, as in (33)? Is there an ordinary phrase present in the syntax, but simply not pronounced when speaking, as captured by the syntactic structure in (34)? Or is there nothing there at all, as in (35)?
33. [S The press [I’ [INFL past, 3rd person][VP [V follow][NP e]]]] 34. [S The press [I’ [INFL past, 3rd person][VP [V follow][NP him]]]] 35. [S The press [I’ [INFL past, 3rd person][VP [V follow]]]] Some evidence that there isn’t covert ordinary syntax – run of the mill material which is simply not pronounced in this instance -- is provided by Grimshaw (1979). She notes that the null complement spot can be semantically controlled by a syntactic item which cannot itself occur in that spot. For example, as Grimshaw notes (1979: 308-309), the null complement containing sentences in (36)-(38) are fine, though their fully spelled
out counterparts are quite bad:
36. Bill asked the time, so I inquired. [Versus *‘...so I inquired the time’]
37. Bill claimed to want to know the reasons for my decision, but he didn’t really care.
[Versus *‘...but he didn’t really care the reasons for my decision’]
38. Bill desperately tried to discover the name of the person who had abducted him, but the police didn’t give a damn. [Versus *‘...but the police didn’t give a damn the name of the person who had abducted him]
39. Bill wanted to know the height of the building, but I wasn’t sure. [Versus *‘...but I wasn’t sure the height of the building’] Nor can an ordinary indexical be what appears in the syntax, though unpronounced. The
continuations in (36)-(43) are bad as well:
40....so I inquired it 41....but he didn’t really care it 42....but the police didn’t give a damn it 43. …but I wasn’t sure it So, at the very least, it seems that it cannot be phonological deletion that is going on, since the “fully pronounced” version, supposedly produced but partly unspoken, would be ungrammatical.
One might reply to Grimshaw’s worry by maintaining that the unpronounced ordinary material in (36)-(39) should not be an NP, but a semantically related S-bar. Thus the material which would go unpronounced in (37) ought to be not ‘the name of the
person who had abducted him’, nor ‘it’, as suggested above, but rather (44):
37. Bill claimed to want to know the reasons for my decision, but he didn’t really care 44. [CP what the reasons for my decision were] This suggestion of a CP continuation, however, is not workable; indeed, it brings up another general reason for abandoning the “unpronounced ordinary material” approach.