There are several different situations where an argument that might be expected to be present is not present. They have been given a wide variety of terms: “null arguments”, “omission”, “ellipsis”, “deletion”, “implicit arguments”, and so on. The question is important, because some central grammatical notions such as “transitive verb” depend on this. For example is a sentence such as (1b) an intransitive sentence, or is it just as transitive as (1a) except that the direct object is not realized? And does (2) exhibit “valency-reduction”, because the agent argument was “removed”?
Here I propose that four rather different subtypes of argument omission should be distinguished (this follows Haspelmath 2022: §13):
definite zero anaphora (or latency)
In the literature on valency, distinctions such as these have long been discussed (e.g. Vater 1978; Koch 1981: 89-95; Matthews 1981: 123-126; Mosel 1984), because obligatoriness was often thought to be an important criterion for argument (vs. adjunct) status. This may be so in some European languages, but as the earlier literature notes, there are a lot of different reasons for “optionality” or omissibility of arguments.
The simplest case is zero anaphora, the omission of a definite argument whose reference can be inferred from the context. This is common in many languages, for all kinds of arguments. For example, zero expression of anaphoric objects is common in colloquial Brazilian Portuguese (but not in other Romance or Germanic languages, which generally require object pronouns):
In Baule (a Kwa language of Côte d’Ivoire), object omission is obligatorily interpreted as anaphoric (4), and cannot have an unspecified-indefinite interpretation (for which a generic object noun is required, as in 4c).
Zero anaphora is not widespread in English, but it does occur with some verbs (e.g. What did they say about the offer? They accepted Ø; Fillmore 1986: 101). A felicitous term for zero anaphora is latency (Matthews 1981: 38, 125; Croft 2001: 273): we can say that an argument position is latent when there is no overt expression (though overt expression would be possible) but it is interpreted as definite anaphoric.
Like zero anaphora, languages frequently allow argument withholding, i.e. the omission of an argument that is intepreted as unspecified and indefinite, as in They ate, or She is writing. Here it is clear that the argument referent exists, but it need not be recoverable from the context. Object withholding in Italian was recently studied by Cennamo (2017), who provides example (5). The understood object need not have been mentioned earlier.
For this phenomenon, the term implicit object has often been used (e.g. Glass 2014), and we find indefinite object deletion in the realier literature.
Some languages also allow withholding of subjects, e.g. Lezgian (a language I studied over 30 years ago), where a meaning such as ‘the articles have been published’ is simply rendered as ‘Ø published the articles’ (Haspelmath 1993: 287-289), with no subject stated (Creissels 2014 reports the same for the related language Akhvakh). Argument withholding has not been studied in many languages, so it is not well understood. Object withholding in English seems to be subject to a variety of conditions (e.g. Goldberg 2001; Cornish 2007; Glass 2014).
Zero anaphora and argument withholding have often been treated under the heading of “null subjects/objects”, or “null instantiation”. Some authors such as Croft (2001: §7.7.3) and Lambrecht & Lemoine (2005) distinguish between definite null instantiation and indefinite null instantiation, for zero anaphora and argument withholding, respectively (following Fillmore 1986). This terminology has gained some currency in the Construction Grammar literature, but not more widely. An earlier author who made the same distinction (but was not cited by Fillmore) is Thomas (1979), who distinguished between ellipsis (= zero anaphora) and non-realization (= argument withholding).
I must admit that I never understood the difference between “zero” and “null”, and my own preference for “zero anaphora” is probably influenced by Givón (1983; 2017) and Kibrik (2011) (who was himself strongly influenced by the Givón tradition). Instead of zero or null (argument), many authors also say implicit, non-overt, covert, understood, deleted (argument), but it appears that all these terms have been used both for zero anaphora and for argument withholding, so they do not make the crucial distinction. (The term ellipsed argument is perhaps restricted to latent/zero-anaphoric arguments.)
I think that these phenomena should be kept terminologically distinct and the “null/zero” terminology should be avoided for argument withholding, because it is semantically rather different from zero anaphora. Only argument withholding could possibly be regarded as a kind of intransitivizing valency alternation (clauses with zero anaphora like (3) from Portuguese and (4b) from Baule are never regarded as intransitive.
But should sentences with argment withholding like They ate and the Italian sentence in (5) be regarded as showing a valency alternation (a kind of object removal)? According to Dixon (in various works), both English break and English eat are “ambitransitive”: break is O-ambitransitive, and eat is A-ambitransitive
However, I do not think that argument withholding should be regarded as a valency alternation, because the conditions on object withholding do not seem to be clearly verb-specific in most languages. Thus, Creissels (2014: 923) notes that for English, we would not want to set up two different constructions for They ate and They ate soup, but rather a single construction [subject verb (object)].
It is true that in some languages, the omission of objects with an unspecified interpretation is possible only with specific verbs and under specific grammatical conditions, as in Mandinka, a Mande language (Creissels 2014: 914; 2015: §2.5). When the object of ‘cross’ is not present, as in (6), the verb shows intransitive tense marking (completive suffix -ta) rather than the transitive marker ye.
Here it does make sense to posit a flexivalency alternation 〈V, X[subj], Y[obj]〉 ≈ 〈V, X[subj]〉 ‘X acts (on Y)’, but it is unclear whether this is a common cross-linguistic alternation type. We need more systematic cross-linguistic research on argument withholding (making sure to distinguish it from zero anaphora), but for the time being, we should treat this phenomenon separately from the phenomenon of flexivalency unless we have specific reasons (as we do in Mandinka).
Let us now get to the third type of argument omission: Argument suppression is well-known from agent suppression in passive constructions in some languages which cannot express the passive agent at all. A suppressed argument is interpreted as indefinite-unspecified, as in argument withholding, but it cannot be overt. In (7), we see a patient-suppressing antipassive alternation. (Heaton 2020 notes that most antipassives are patient-suppressing, and that the patient-demoting type is much less common.)
The fourth type is called role removal here. This refers to a situation in which a role that is present in one valency construction is not present in another alternant. The best-known case of this is in anticausative alternations like (8), from the Bantu language Luragooli, where the agent role is removed.
It seems that role removal applies only to agent removal, and that patiient removal is not a possible type of argument omission.
While it is not controversial that these four situations are distinct in principle, distinguishing in practice between argument withholding, argument suppression, and role removal is not always straightforward, as is illustrated by the discussion in Creissels (2014). For a long time, I was confused by the different terms in the literature, and it was only more recently that I made up my mind on what to think about the different subtypes of argument omission and how they relate to valency alternations (see Haspelmath 2022: §13). I am not claiming that what I have proposed here is the only way to cut up the pie, and as usual, the concepts I discuss here are general (or comparative) concepts, and particular languages may require different concepts that do not map clearly onto those discussed here.
However, at the moment I am quite happy with the current conceptual system, and I look forward to more discussion.
Cennamo, Michela. 2017. Object omission and the semantics of predicates in Italian in a comparative perspective. In Hellan, Lars & Malchukov, Andrej L. & Cennamo, Michela (eds.), Contrastive studies in verbal valency, 251–274. Amsterdam: Benjamins.
Cornish, Francis. 2007. Implicit internal arguments, event structure, predication and anaphoric reference. In Hedberg, Nancy & Zacharski, Ron (eds.), The grammar-pragmatics interface, 189–216. Amsterdam: Benjamins.
Creissels, Denis. 2006. Syntaxe générale: Une introduction typologique, 2 vols. Paris: Lavoisier-Hermès science.
Creissels, Denis. 2014. P-lability and radical P-alignment. Linguistics 52(4). 911–944.
Creissels, Denis. 2015. Valency properties of Mandinka verbs. In Malchukov, Andrej L. & Comrie, Bernard (eds.), Valency classes in the world’s languages: A comparative handbook, vol. 1, 221–259. Berlin: De Gruyter Mouton.
Croft, William. 2001. Radical Construction Grammar: Syntactic theory in typological perspective. Oxford: Oxford University Press.
Farrell, Patrick. 1990. Null objects in Brazilian Portuguese. Natural Language & Linguistic Theory 8(3). 325–346. (doi:10.1007/BF00135617)
Fillmore, Charles J. 1986. Pragmatically controlled zero anaphora. Annual Meeting of the Berkeley Linguistics Society 12. 95–107.
Givón, T. 2017. The story of zero. Amsterdam: Benjamins.
Givón, Talmy (ed.). 1983. Topic continuity in discourse: A quantitative cross-language study (Typological Studies in Language). Amsterdam: Benjamins.
Glass, Lelia. 2014. What does it mean for an implicit object to be recoverable? University of Pennsylvania Working Papers in Linguistics 20(1). 121–130.
Gluckman, John & Bowler, Margit. 2016. The anticausative alternation in Luragooli. In Payne, Doris L. & Pacchiarotti, Sara & Bosire, Mokaya (eds.), Diversity in African languages: Selected papers from the 46th Annual Conference on African Linguistics (Contemporary African Linguistics 1), 271–289. Berlin: Language Science Press.
Goldberg, Adele E. 2001. Patient arguments of causative verbs can be omitted: The role of information structure in argument distribution. Language Sciences 23(4). 503–524. (doi:10.1016/S0388-0001(00)00034-6)
Haspelmath, Martin. 1993a. A grammar of Lezgian (Mouton Grammar Library 9). Berlin: Mouton de Gruyter.
Haspelmath, Martin. 2022. Valency and voice constructions. Leipzig, MPI-EVA, ms. (https://ling.auf.net/lingbuzz/005941)
Kibrik, Andrej A. 2011. Reference in discourse. Oxford: Oxford University Press.
Koch, Peter. 1981. Verb, Valenz, Verfügung: Zur Satzsemantik und Valenz französischer Verben am Beispiel der Verfügungs-Verben. Heidelberg: Winter.
Lambrecht, Knud & Lemoine, Kevin. 2005. Definite null objects in (spoken) French. In Fried, Mirjam & Boas, Hans C. (eds.), Grammatical constructions: Back to the roots, 13–55. Amsterdam: Benjamins.
Maslova, Elena. 2003. A grammar of Kolyma Yukaghir (Mouton Grammar Library 27). Berlin: Mouton de Gruyter.
Matthews, Peter H. 1981. Syntax. Cambridge: Cambridge University Press.
Mosel, Ulrike. 1984. Towards a typology of valency. Arbeiten des Kölner Univeralienprojekts (akup) 58. 1–19.
Thomas, Andrew L. 1979. Ellipsis: The interplay of sentence structure and context. Lingua 47(1). 43–68. (doi:10.1016/0024-3841(79)90066-4)
Vater, Heinz. 1978. On the possibility of distinguishing between complements and adjuncts. In Abraham, Werner (ed.), Valence, semantic case and grammatical relations, 21–45. Amsterdam: Benjamins.)
What about the languages like Niuean, where the non-realization (or ‘non-expression’) of P shitfs the ergative case of A into absolutive:
Massam (2020: 48, 115)
Ne kai e Mele e apala.
PST eat ERG Mele ABS apple
‘Mele ate the apple.’
Ne kai a lautolu.
pst eat abs 3.pl
‘They were eating.’
If I understand you well, (b) illustrates the argument withholding and it should not be considered to illustrate valency alternation?
The language also has a few verbs (laugh a laugh, dance a dance, pray a prayer), where the cognate P can be omitted and this omission shifts the case of A from ERG to ABS. (1) Is not the P withholding in such a case clearly verb-specific? (2) Cannot the shift of case from ERG to ABS of P be viewed as a signal of valency alternation?
Thanks for these challenging questions! Indeed, when there is a change in the coding of the arguments, I would say that there is a valency alternation – the verbs are flexivalent (to use Creissels’s term). The various types of “omission” are relevant only when there is no change in the coding of the non-omitted arguments (I should make this clear in the next version of the valency and voice paper). Of course, this means that the types that I distinguish are not only functional types, but also formal types (so to speak). I guess one would need a general term that covers both your example (b) (Niuean ‘they were eating’) and English “They were eating” – I hadn’t thought of that. It’s really hard to get the terminology right…
Regarding the following:
And does (2) exhibit “valency-reduction”, because the agent argument was “removed”?
(2) “the soup was eaten”
Why would we say that the agent argument was “removed”, rather than merely not expressed?
It’s possible to say “the soup was eaten by the children”, in which the agent argument is expressed.
Is the fact that you have to add “by” to make it grammatical relevant to why we would say that the argument is “removed”? (Since you can’t say *the soup was eaten the children) Like, the passive construction “removes” the agent slot, and “by” adds it back?
Yes, there is no “removal” in “The soup was eaten” – here the agent argument merely becomes optional. In my 2022 paper, I say that the agent is *demoted*, but that’s all. We find *role removal* in cases like Russian lomat’-sja ‘break (intr.)’, where the breaker argument is no longer even implied. So I used scare quotes here in order to highlight the problematic nature of these metaphrical expressions.