«I. Introduction Consider sentences like (1): 1. Null Complement Containing Sentences a. Aryn followed b. Marie-Odile promised c. Corinne left d. ...»
As Clapp (2002) and Elugardo and Stainton (2001) have both stressed, in a slightly different context, there will often be no single candidate for what the ordinary but unpronounced material would be, consistent with the content asserted: either there are too many candidates, or there are none available to the language users. Applied to the case at hand, there is no reason to choose precisely ‘what the reasons for my decision were’ as the material omitted in (37). Equally good (and hence equally bad) would be ‘what reasons I had for my decision’, ‘what those reasons were’, etc. Put crudely, the content of the null complement slot is often more vague than any ordinary item of structure, supposedly unpronounced, would be: each paraphrase is too precise to be the thing left out.
A second piece of evidence against taking the null complement to result from surface phonological deletion comes from facts about how the complement may (and may not) serve as antecedent. The view that there is unpronounced ordinary material makes the wrong predictions about this. For instance, suppose the question is asked, ‘How do we know that Jim robbed a bank?’. If the answer in (46) was syntactically and semantically just like (45), underlyingly, as per the view being discussed, then one would expect (45) with the material spoken and (46) with the null complement to have precisely
the same meaning potential, in that discourse context:
45. He confessed that he robbed a bank, though we still don’t know when
46. He confessed, though we still don’t know when But, in fact, even as a response to ‘How do we know that Jim robbed a bank?’, (46) is apt to be heard as just meaning that we don’t know when he confessed, while (45) is apt to be heard as ambiguous between not knowing that, and not know when he robbed the bank.
The explanation of this semantic divergence, we think, is that the abbreviated answer to ‘How do we know that Jim robbed a bank’, viz. (46), does not contain the ordinary syntactic material [CP that he robbed a bank] at any level of representation, hence this material is not available as an antecedent to the sluicing construction ‘we don’t know when’.
Similarly, (47) is grammatically better than the abbreviated (48), because – contra the phonological omission account of null complements -- only the former
actually contains an appropriate antecedent for the anaphor ‘so’:
47. Juan testified that Anabelle was guilty, but in his heart he still didn’t think so 48. ??Juan testified, but in his heart he still didn’t think so None of this refutes definitively the idea that null complements involve unpronounced ordinary syntactic material. But enough has been said for present purposes. Still possible, for all we have said above, is the idea that what appears in the null complement position is a special element of syntax, which never has a pronunciation.
Or better, what appears are several such elements: one for each syntactic category of null non-subject.
At a minimum, something along the lines of null DPs, null CPs, null (infinitival) IPs, and null PPs would be required, as in:
8. Hershad followed [DP e] quietly 1b. Marie-Odile promised [CP [C e] [IP e]] 1g. He volunteered [IP [I e] [VP e]] 12a. She arrived [PP [P e][DP e]] (Recall that (12a) was used to mean that she arrived at the summit.) Our reasons for rejecting this latter idea are methodological, and theory internal -- they have rather less to do with data coverage. True, we have been at pains to argue that null complements are interpreted as if there were a variable present. But that is far from a sufficient reason for positing a variable present, but unheard, in the syntax. We object in general to positing hidden structure solely on the basis of what an expression means; and we think that, at present, this is the only positive evidence for such empty elements, in the case of null complements. For this reason, we presently opt for (35) as the syntax for ‘The press followed’. That said, we fully recognize that there is much more to be said on the issue.
A remaining question is how an expression with the syntax in (35) could end up with the meaning it has. Specifically, where is the content “as of” a variable coming from, if it is not in the syntax? Addressing this is detail would require a whole other paper. The short answer, however, is that the semantic component can recognize that a null complement construction is in play, can introduce a variable on that basis, and this variable can then be bound, or free in the semantics. More exactly, the semantics need only be able to recognize that a verb with an n-participant semantic frame is occurring with too few overt participants, recognize on the basis of surface syntax and the semantic frame which participant is unrealized, and then, at the level of content, treat the verb as having a variable in that slot. The “binder” too, if there is one, can then be provided either by prior syntax, or by context. This requires, of course, that the semantic component do more than (i) assign a content to each element of syntactic structure, and (ii) apply several iterations of function-argument application. It also implies that compositionality, in Richard Montague’s very strict sense of isomorphism between syntactic structure and semantic content, does not hold for null complements. But these implications don’t bother us, since we take them to be independently established. Indeed, natural languages are simply rife with constructions that impact the semantics directly. (See Goldberg 1995 for numerous examples, and discussion. Note that we do not endorse Goldberg’s larger theoretical framework. But we do emphatically agree that constructions are crucial for semantics.) Null complements is just one more example.
To sum up, then. The fundamental mark of the null complement phenomenon is a mismatch between the number of linguistically inherent participants noted in a verb’s semantic frame, and the number of “phonologically realized” non-subjects in fully grammatical sentences containing that verb. That’s what it is to exhibit a null complement -- what “omission” amounts to. In terms of semantics, the meaning effect in all such cases was as if there were a variable. Sometimes the variable occurs free, in which case the utterance reads as if it had an extra deictic. Sometimes the variable occurs bound, in which case the utterance sounds quantificational. Cases which seem to violate this generalization – e.g., ‘Anita ate’ – we reclassify as not really involving omitted complements: they are actually intransitives which are marked for the addition of a nonsubject. Regarding the syntactic structure of sentences exhibiting this “mismatch”, we argued that there is not unpronounced ordinary material, and suggested briefly that there are not unpronounced special null elements either. (Admittedly, the latter claim is based on methodological orientation, rather than on data.)
REFERENCESBach, Kent (1994a). "Semantic Slack: What is said and more". In S.L. Tsohatzidis (ed.) Foundations of Speech Act Theory. London: Routledge.
---- (1994b). “Conversational Impliciture”. Mind & Language 9: 124-162.
Carston, Robyn (1988). Implicature, Explicature and Truth-Theoretic Semantics. In R.
Kempson (ed.) Mental Representations: The Interface between Language and Reality. Cambridge: Cambridge University Press, pp. 155-181.
Clapp, Lenny (2002). What Unarticulated Constituents Could Not Be. In J.C. Campbell, M. O'Rourke, and D. Shier (eds.) Meaning and Truth: Investigations in Philosophical Semantics. New York: Seven Bridges Press, pp. 231-256.
Cummins, Sarah & Yves Roberge (2003). Null objects in French and English. Linguistic Symposium on Romance Languages 33, Indiana University, April 2003.
---- (2004). Grammaire, sémantique et pragmatique de l'objet nul. Congrès de l’ACL
2004. University of Manitoba, May 31, 2004.
Dixon, R.M.W. and Aikhenvald, A. (2000). Changing valency: case studies in transitivity. Cambridge: Cambridge University Press.
Elugardo, Ray & Robert J. Stainton (2001). Logical Form and the Vernacular. Mind and Language 16(4): 393-424.
Fillmore, Charles (1986). Pragmatically controlled zero anaphora. In V. Nikiforidou, M.
Vanllay, M. Niepokuj and D. Felder (eds.) Proceedings of the XII Annual Meeting of the Berkeley Linguistics Society. Berkeley, CA: BLS.
Frantz, Donald G. (1991). Blackfoot Grammar. Toronto: University of Toronto Press.
Goldberg, Adele (1995). Constructions. Chicago: University of Chicago Press.
Grimshaw, Jane (1979). Complement Selection and the Lexicon. Linguistic Inquiry 10(2): 279-326.
Groefsema, Marjolein (1995). Understood arguments: A semantic/pragmatic approach.
Lingua 96: 139-161.
Iten, Corinne, Marie-Odile Junker, Aryn Pyke, Robert Stainton & Catherine Wearing.
(2005) Null Complements: Licensed by Syntax or by Semantics-Pragmatics? In Proceedings of the 2004 Canadian Linguistics Association Annual Conference, ed. Marie-Odile Junker, Martha McGinnis and Yves Roberge, 15 pages.
http://www.carleton.ca/~mojunker/ACL-CLA Junker, Marie-Odile, Marguerite MacKenzie and Louise Blacksmith (2005). East Cree Verbs (Southern dialect). In Junker, Marie-Odile (ed.) (2000-2006). The Interactive East Cree Reference Grammar. www.eastcree.org Partee, Barbara (1989). Binding Implicit Variables in Quantified Contexts. CLS 25: 342Récanati, François (1989). The Pragmatics of What Is Said. Mind and Language 4: 294Searle, John R. (1980). The Background of Meaning. In J.R. Searle, F. Kiefer & M.
Bierwisch (eds.) Speech Act Theory and Pragmatics. Dordrecht: Reidel.
---- (1978). “Literal Meaning”. Erkenntnis 13: 207-224.
Sperber, Dan & Deirdre Wilson (1986). Relevance. Oxford: Blackwell.
Stainton, Robert J. (1997). Utterance Meaning and Syntactic Ellipsis. Pragmatics and Cognition 5(1): 49-76.
Stanley, Jason (2000). Context and Logical Form. Linguistics & Philosophy 23(4): 391Travis, Charles (1985). On What Is Strictly Speaking True. Canadian Journal of Philosophy 15(2): 187-229.
Some authors would argue that this overstates the case. On their view, there isn’t a binary divide between verbs which do and verbs which do not grammatically allow null complements. Instead, goes the idea, some verbs allow omission very easily, some less easily, and some only under quite special circumstances. (For discussion, see Cummins & Roberge 2003, 2004.) Two kinds of arguments are typically given for preferring this graded scale. First, corpus studies consistently find null complement involving uses of verbs like ‘lock’ and ‘devour’ which, on a Fillmore-type account, simply cannot occur grammatically without complements. Second, corpus examples aside, there are constructions which intuitively seem to allow null complements fairly freely, even for those which Fillmore would label as not subject to omission. For instance, take the construction: “There are those who merely __ (close/promise/leave) and those who __ (lock/guarantee/ vacate)”. Our own view is that these arguments don’t really show that there is gradation. But we won’t pursue the point here, resting content with the idea that some verbs allow null complements freely, which others resist them.
One might balk at the terminology of ‘verbs on a sense’. In particular, some of these examples might be better treated as homophony: wholly distinct verbs which happen to sound alike, rather than one verb with a variety of senses. The crucial point for us, however, is that it is not a mere sound pattern that does or does not allow null complement – it is, at a minimum, a sound pattern plus a content.
Binding of null complement slots was noted independently in Partee (1989).
Groefsema (1995) points out that ‘won’, which Fillmore labels DNC, can actually appear without a specific, contextually-given competition. Her example is ‘Martina Navratilova has won again’, uttered during a discussion of the achievements of older sports people. This is another example of the kind of problem we have in mind, of purported DNC verbs shifting from reference to quantification.
M. Kawai (p.c.) rightly noted that our taxonomy is not exhaustive. For instance, reflexive verbs like ‘shave’ don’t seem to fit anywhere. They are not obstinately existential, so they aren’t intransitive. And they can’t be read as having a free variable as complement content either -- being reflexive, the complement content is always the agent -- so they aren’t transitive null complements either. (Cf. ‘John shaved’ cannot be used to mean that John shaved the salient man with all the nasty razor burn.) In using the notation of features, we are by no means committed to the idea that they can apply freely to verbs, regardless of the content of the verb in question. There undoubtedly are semantic constraints to be met as well. Thus, to take an obvious example, it is no accident that ‘die’ is an intransitive that cannot add a complement, for what would that added complement be?
Cree also has verbs which, like English ‘die’, are strictly intransitive; and it has verbs, which, like English ‘buy’, are strictly transitive. Cree verbs are marked for transitivity in several ways, first by an affix called the verb final which combines semantic content as well as transitivity marking, then by a direction morpheme indicating subject and object person combinations, and finally by person agreement suffixes. See Junker et al. (2005). Our thanks to Louise Blacksmith for help with the East Cree examples.
This question about syntax is an interesting and important issue in its own right. But, though we haven’t space to address them here, it also has important implications for the on-going debate about whether pragmatics plays a part in determining what is strictly and literally asserted. Specifically, the question arises whether pragmatics plays this part when complements are omitted in talk exchanges.
We hope to return to this larger issue in a separate paper. See Bach (1994a, 1994b), Carston (1988), Recanati (1989, 2002), Searle (1978, 1980), Sperber & Wilson (1986), Stainton (1997) and Travis (1985) for arguments in favor of pragmatic determinants of what is said. See Stanley (2000) for an opposing view.