Skip to navigation – Site map

A compositional analysis of VP anaphors

Gabriel Flambard


The VP anaphors do it and do this/do that have been little studied in the literature, although they are mentioned in passing in a number of more general works such as Hankamer and Sag (1976), Culicover and Jackendoff (2005) or descriptive grammars by Quirk et al. (1985) or Huddleston and Pullum (2002). With the exception of Souesme (1985), very few extant studies are available and there are almost none that explore the internal structure of VPAs in detail. This paper argues for a compositional analysis of the structure of VPAs, proposing that they are ordinary transitive constructions of the form ‘do + object pronoun’ and, following Simner (2001) and Stroik (2001), that it is the pronoun, rather than the VP expression as a whole, which serves as the anaphoric element. The compositional nature of VPAs is evidenced by a range of syntactic and semantic facts which point to the relative independence of the object pronoun in such constructions. Foremost among these is the very possibility of alternating between it and demonstrative pronouns, which is itself highly suggestive of a compositional structure. Further evidence is supplied by passivisation (it/this/that was done) or pseudo-clefting with do this/that (This/That is what I did). Secondly, we will address the question of how both parts of the VP contribute to its interpretation. Object pronouns in VPAs are distinguished by having a VP as their antecedent, or ‘antecedent-trigger’, as Cornish (1992, 1996) calls the segment of discourse through which an appropriate antecedent is retrieved. The anaphoric relation between the object pronoun and the VP trigger holds regardless of the fact that the pronoun cannot be replaced with any VP expression (e.g. * John did {mowing/ (to) mow} the lawn), as replaceability is not a necessary condition for antecedenthood.

Top of page

Full text

I would like to thank the two anonymous Anglophonia reviewers for helpful comments on the analysis.

1. The compositionality problem

1The present study is concerned with the internal structure of the English VP anaphors do it and the demonstrative variants do this/do that (the variant do so is excluded from consideration, since it is clearly not transitive. However, we will make some further comments on this construction later on). Their use is exemplified below:


John mowed the lawn.

a. He did it in under an hour.

b. He did this/that in under an hour.

2In each of these examples, the do + pronoun combination is interpreted in terms of the underlined VP, which could be substituted for it. Thus, at first glance, such constructions appear to be anaphoric ‘as a whole’ for VP predicates, an approach adopted (albeit implicitly) by Hankamer and Sag (1976), who call do it a ‘pro-form’ (i.e. an anaphor) or Culicover and Jackendoff (2005). Alternatively, these constructions may be viewed simply as the combination of main verb do and an object pronoun, similar to any other transitive constructions. This conception of VPAs is implicit in the discussion of attested examples of VPA by Cornish (1992, 1999) and is also explicitly the position of Schuster (1988) and more recently Huddleston and Pullum (2002) as well as Miller (2011, 2013).

3Regardless of the proposed analysis, the various authors cited generally provide little or no evidence in support of their claims, so that there are few that actually make a case either for or against compositional structure in VPAs.

4This paper will present a range of arguments in support of a compositional analysis, treating do as main verb and it/this/that as the only anaphoric component, ruling out the hypothesis that do is also anaphoric. We begin by reviewing a number of studies which implicitly adopt either a compositional or non-compositional analysis of VPAs, but do not make any specific claims in this respect. We then discuss a small number of proposals that support treating VPAs as transitive constructions, and bring some further evidence for compositional structure, whilst also reviewing potential counter-evidence. In particular, it will be argued that the proposed structure can explain a number of aspects of the usage of VPAs, and as such provides a good basis for further analysis of VPA alternation.

2. The internal structure of VPAs

5The VPAs do it/this/that have received little attention in the otherwise extant literature on anaphora (the substantial work of Souesme (1985) being an exception). Although, as discussed in section 3, several authors have made more or less detailed remarks on the structure of VPAs, usually in the context of more general works, the structural properties of these constructions have rarely been studied in any great detail. This is in stark contrast to the considerable amount of work on the syntax of VP ellipsis (or post-auxiliary ellipsis (PAE), to use the term proposed by Sag (1976)), for instance by Hankamer and Sag (1976), Hardt (1993), Merchant (2001, 2004), amongst many others. The structure of do so has also been a matter of some debate, albeit to a lesser extent —see Bolinger (1970), Lakoff and Ross (1976), Higgins (1992), Cornish (1996), Houser (2010), Miller (2013). Although do it/this/that are occasionally mentioned alongside these constructions, they are usually described more briefly. The relatively minor place of VPA in the literature is in part attributable to the widely-held assumption that they are largely interchangeable with each other as well as with PAE and do so. As Miller and Hemforth (2014, 3) explain:

Generative studies of PAE and verbal anaphors such as do so or do it/this/that have never addressed the discourse conditions on these constructions. In fact, it has been tacitly assumed that there is no difference in their uses beyond those resulting from some version of the deep/surface distinction of Hankamer and Sag 1976 and the fact that do in do so and do it/this/that is the main verb.

6Thus, there is much left to explain about the structure and usage of VPAs. This paper will concentrate on the first of these issues, but in so doing will also shed light on various aspects of how VPAs are used in discourse.

2.1 The non-compositional view: VPAs as pro-VPs

  • 1 For instance, Kertz points out that the attested example (46) reported by Kehler (2000) is more acc (...)

7A number of studies in the psycho-linguistic literature have compared do it to PAE in order to establish whether the latter is a case of ‘surface anaphora’ or ‘deep anaphora’ as understood by Hankamer and Sag (1976) —a surface anaphor being one that requires a linguistic antecedent (and specifically one that is syntactically parallel to the anaphoric VP), while deep anaphors, such as do it, do not. Studies of this kind (in which do this/that are systematically left out of consideration) have obtained mixed results, either supporting the surface status of PAE (Tanenhaus and Carlson, 1990) or finding no evidence for it (Murphy, 1985), and in some cases questioning the validity of Hankamer and Sag’s proposed distinction. More recently, Kertz (2010, 2013) argues that syntactically ‘mismatched’ antecedents of PAE are acceptable under certain discourse conditions also applicable to non-mismatched uses, although they induce processing difficulties which may decrease the acceptability of mismatch cases where the appropriate conditions are not respected.1

8Regardless of their conclusions, studies which examine do it only as a prototypical example of deep anaphora have little to say about its internal structure. However, there is no indication that it is considered the sole anaphoric component of the VPA, making it doubtful that a compositional analysis is assumed. As already noted, there are very few straightforward accounts of the structure of VPAs, but it is usually possible to infer one or the other analysis from the author’s treatment of VPAs. Culicover and Jackendoff (2005, 126) offer a good example of an implicitly non-compositional account, treating do it/this/that and do so as pro-VPs (collectively referred to as ‘do X anaphora’) that stand in for ‘a subpart of a VP —including the verb itself and the non-contiguous portions of the VP’ (‘non-contiguous’ parts being any remnant complements or adjuncts, e.g. Robin broke the window with a hammer and Mary did it to the table top (=broke the table top with a hammer)).

2.2 The compositional view: VPAs as do + pronoun

9Other accounts have adopted a compositional analysis of VPAs, often only implicitly, though some authors openly defend this approach. Among those who appear to treat VPAs are compositional (though without stating it as such) is Cornish (1992), who discusses do it and do so as merely a subcase of the alternation between so and it also observed with a range of other host verbs such as say, think, believe, etc. In his 1999 book on anaphora, he discusses the following attested example of do it, claiming that it refers to an action, and that its antecedent may be expressed as ‘a specific, definite act of shooting down at least one ICBM in space’ (p. 84):

10(2) The Americans have developed a new method of shooting down intercontinental ballistic missiles [ICBMs] in space: they did it on Sunday. (BBC Radio 4, 12 June 1984)

11A similar approach is adopted by Schuster (1988), in her discussion of it and this/that used with action referents, of which do it/this/that are taken to be just one instance. This is apparent in her notation of anaphoric relations, reproduced in (3) (the exchange occurs between an Emacs instructor and a novice user learning to perform basic tasks with the software; interactions are made via a computer terminal):

12(3) U: Tell me how to [define a region]i again. I forgot how to do it i.
E: [You set the mark with <esc>-M at one extreme of the text and you
move the cursor to the other extreme]j . That defines a region. (ex. 2, p. 1)

13Here, as in Cornish’s example above, it in do it is clearly assumed to be the only anaphoric element, although again no particular justification is provided for this description. Also, neither author explicitly claims that this is the correct analysis of VPAs (the term is never used at all in their accounts).

14Simner et al. (2003), commenting on the ambiguity of do it in examples like the following, explain that it may receive either a ‘nominal’ interpretation, where it is anaphoric for a previously-occurring NP object, e.g. the puzzle here, and a ‘verbal’ one where do it as a whole is interpreted in terms of the previous VP. In this case, the nominal reading would be did (= solved) the puzzle, and the verbal one, borrowed the puzzle. Since the nominal interpretation (did the puzzle) involves an NP object which can be directly substituted for the pronoun, it is a case of NP anaphora, rather than VP anaphora.

15(4) Sam borrowed a jigsaw puzzle, and he did it while the others were out. (ex. 2, p. 2)

16Simner et al. are chiefly concerned with the contextual factors which influence the choice of one or the other interpretation, but do not go into much detail regarding the syntactic differences between them. However, their account of the structure of do it, which is based on the earlier proposals of Simner (2001), suggests that a compositional approach is adopted: it is treated either as a ‘deverbalized NP’, in the verbal interpretation for instance here (that thing that is) the borrowing of a puzzle, as opposed to a simple NP (e.g. the puzzle) in the nominal interpretation.

17We now turn to the proposals of authors who have made a more explicit case for the compositional structure of VPAs do it, do this and do that. An early such account is Souesme (1985, 35), who simply explains that do it is interpreted as do something mentioned before and later on suggests (79) that the choice between do this/that is parallel to that between demonstrative this/that. Miller (2013, 126) further claims that compositionality sets apart do it/this/that from do so, which in his view can be argued to be non-compositional, citing as evidence the resistance of the do NP forms to stative antecedents (see (5)–(6) below), which is indicative of the main verb status of do:

18(5) [W]henever the relation of complementary distribution holds between phones belonging to a common phoneme, #it does it/this/that because. . .

19(6) We assume that logical thought, syllogistic analytical reason, is the necessary, right thought—and #we do it/this/that because. . . (ex. 6a-b, p. 123)

20Simner (2001, 58) is one of the most explicit proponents of a compositional approach. Criticising the ‘traditional’ analysis, as she calls it, which views do it as a pro-VP, she instead advocates a structure in which the pronoun alone is anaphoric and do is main verb. Simner (58) insists that in this latter structure “although the ‘do it’ expression is syntactically a VP, it is anaphoric only in part (i.e. only the pronoun is an anaphor)”. In support of this claim, she cites independent evidence that it can derive its meaning from a previous VP without the verb + it sequence being treated as anaphoric. For instance, the second clause of (7) would not be regarded as involving ‘try it anaphora’:

21(7) Mary dived into the pool. John wanted to try it too. (ex. 17, p. 63)

22Likewise, Stroik (2001, 364) openly rejects a non-compositional view of the structure and insists that in do it and do so, “it is the object proform it/so, not the do so/it constituent, that replaces the main VP” (though as discussed above, do so does not lend itself to a similar analysis).

3. Evidence for compositionality

23In what follows we explore the syntactic and semantic properties of do it/this/that in greater detail, showing that these point to the relative independence of main verb do and the object pronoun. Further evidence for this comes from certain pragmatic factors which appear to be at play in the alternation between VPAs. We will also present arguments in favour of treating it/this/that as the sole anaphoric components of VPAs.

3.1 Syntactic evidence: independence of main verb do and the object pronoun

24Quirk et al. (1985, 866) initially describe do it and do that as ‘complex pro-forms which are substitutes for predicates or predications’ but later explain that ‘they combine the substitute function of do and the coreferential function of it/that ’ (876). Although do this is not mentioned, it certainly lends itself to a similar description. In support of their claims, Quirk et al. note the possibility for the object pronoun to become the subject of a passive sentence with a VP antecedent and do as main verb, and it is also possible for the VP antecedent to serve as the antecedent of a relative pronoun heading a clause that contains main verb do. This is illustrated in the following from Huddleston and Pullum (2002) (that has been added to the original example for completion):

25(8) They climbed all four peaks in one day — it/ this [/that] had never been done before. (p. 1532)

26(9) They climbed all four peaks in one day, which had never been done before. (50iv, p. 1533)

27And below are attested examples from the British National Corpus (BNC) of passive do this/that:


a. In practice it is necessary to classify these soils. This is done by relating the soil to the type of cleaning agent. . . (BNC)

b. I shall not go into details about his speeches; that was done at great length on 12 June (BNC)

  • 2 Although (12) is an example of cataphora, in which the antecedent-trigger is introduced after the V (...)

28A further argument in favour of compositionality is the possibility for do this/that to receive prosodic stress on the object pronoun, whereas the it of do it is of course always unstressed. This, together with other properties described above, shows that the object pronouns in do it/this/that retain their usual properties, which is good evidence of the compositional nature of VPAs. Quirk et al. only provide the example of do that shown in (11), while in (12), from Schuster, this would most likely be stressed2:

29(11) Are you trying to light the stove with a match? I wouldn’t do that.

30(12) E: Do this: set a “mark” at some point (any old point) by typing
<esc>-M. It will say “mark set” try it. (Schuster, 1988, ex. 17, p. 9)

31In addition to the evidence from passives and relatives just discussed, Stroik points out the possibility for a VP constituent to be targeted by interrogative or fused relative what in interrogatives (13) or pseudo-clefts (14) containing main verb do:


a. What are you doing?

b. Eating the leftovers. [ex. 6, p. 364]

32(14) What Pat is doing is fixing the car. [ex. (iia), fn. 6, p. 364]

33We may also add to these data the existence of inverted pseudo-clefts with do this/that of the type shown below (these are primarily encountered with this/that as the focused element; examples with it are much rarer):

34(15) In other words I was free to speculate and this is what I did. (BNC)

35(16) So what do I do? I use my brains, laddie, that’s what I do. (BNC)

36(17) I DJed at Clink Street the infamous after-hours warehouse just round the corner from the original Shoom club. I’ve been doing it for years. It’s what I do. I know how to mix. (BNC)

37These data highlight two essential facts about VPAs: first, that do is main verb and it/this/that are pro-VP object pronouns, and secondly, that the two are relatively independent from one another syntactically. This is further shown by the possibility of making subsequent references to the antecedent (after the VPA) with either it or a demonstrative.

38(18) —Well we’re asking people to enter in teams of four to come and plant as many trees a they can in the space of an hour. So we’ll have groups of teams coming along for an hour at a time and hopefully we’ll, you know, keep the impetus going through the day by doing that. (BNC)
It sounds absolutely exhausting. How many trees do you reckon a team of four can actually plant in an hour? (BNC)

39As is well known, the pro-VP use of it/this/that is also attested outside of VP anaphora. We discussed earlier an example with it from Simner (2001); below are further examples from Schuster (1988) also illustrating the pro-VP use of this/that:

40(19) U: How do I center the comment headers?
E: Type <esc>-C. That will center them. (ex. 13, p. 7)

41(20) E: To kill a region, the whole thing at one time, you should set the mark <esc>-M at the top of the region, then move the cursor down to the bottom of the region and type <esc>-control-K. This will kill the region ok?? (ex. 16, p. 8)

42(21) E: Oh. Well a simple way around that this time is to search for DEBUG, then do a down-line, front-of-line, and you’re in the same position. Right?
U: Yes but that won’t work for the application in mind. It would lead to recursive multiple (unwanted) changes. I tried it. (ex. 8, p. 4)

43A final, more fundamental piece of evidence comes from the fact that pronoun alternation is possible in VPAs, which would not be expected if they were somehow idiom-like or non-compositional constructions. Moreover, the pronouns retain their usual distribution, as shown by the possibility of do this/that but not do it receiving stress on the object pronoun.

44In short, do + pronoun VPAs exhibit all of the syntactic properties of regular transitive constructions, leading us to conclude that the use of it/this/that in VPA is analogous to their use with any other transitive verb. However, while the description given so far has focused on the syntactic compositionality of VPAs, it is also valid semantically, and several of the syntactic arguments presented above have a bearing on the semantics of VPAs. The next section explores the semantically compositional nature of VPAs in greater detail.

3.1 Semantic evidence: agentivity and pronoun choice

45If we assume that VPAs are simply the combination of main verb do and it/this/that, then we expect their usage to be ultimately determined by the semantic properties of main verb do and the object pronouns involved. Two main facts support this view: first, do it/this/that generally prefer an agentive, non-stative antecedent, and secondly, the choice between it and demonstrative this/that appears to operate under much the same conditions as it does with other transitive verbs, or in various other contexts. We review these in turn.

3.1.1. Agentivity constraints on VPAs

46It has been claimed that VPAs require an agentive antecedent, by which is meant an action carried out by a volitional animate agent (see Gruber (1967)’s definition, cited in Cruse (1973)). The following example from Huddleston and Pullum (2002) illustrates this constraint with do it (arguably do this/that are also infelicitous here):

47(22) When the tree fell, it did so/#did it with a loud crash. (ex. 49, p. 1532)

48The exact statement of this constraint varies, with most authors seeing it as a true requirement —Souesme (1985), Huddleston and Pullum (2002), Culicover and Jackendoff (2005)— and some others simply as a semantic preference —Quirk et al. (1985) claim that the requirement holds ‘for some speakers’, and Miller (2011, 83) discusses attested examples of non-agentive do it such as (23) (taken from the COCA corpus), where the subject is a non-volitional cause.

49(23) Although straight narrative can advance a plot faster and less circuitously, dialogue often does it more interestingly. (ex. 1b, p. 83)

50In many such accounts, non-agentive is in fact a byword for ‘non-stative’, reflecting the fact that do it/this/that do not license stative antecedents, as shown in the following example adapted from Culicover and Jackendoff (the original sentence is with do so, but the authors consider that do it/this/that are subject to the same semantic constraints):

51(24) *Robin dislikes Ozzie, but Leslie doesn’t do it/this/that. (adapted from ex. 2a, p. 284)

52The idea that VPA antecedents must be agentive is implicitly motivated by the fact that main verb do is itself agentive and non-stative in PAE, where do is an auxiliary rather than a lexical verb, it shows no such requirement, as the following, also from Culicover and Jackendoff, illustrates:

53(25) Robin dislikes Ozzie, but Leslie doesn’t. (ex. 4a, p. 284)

54The notion of an agentivity requirement on VPAs is, however, problematic. The claims against non-agentive or stative do so have been refuted by Michiels (1978) and more recently Houser (2010) and Miller (2013), based on experimental data. As for do it/this/that, while stative antecedents are clearly unacceptable, it is possible to find very natural-sounding occurrences of do it/this/that in which the antecedent is not agentive according to the above definition, i.e. the action is not carried out by a volitional agent.

55(26) New plantings should always be watered in after planting; don’t assume the rain will do it. (

56(27) Victor Wilcox lies awake, in the dark bedroom, waiting for his quartz alarm clock to bleep. It is set to do this at 6.45. (BNC)

57(28) The name of the game is how can we reduce the amount of offending in this country. If cautioning could do that we’d all be right behind it. (BNC)

58Examples such as these are fairly rare, but they are nevertheless attested, showing that VPAs admit non-agentive antecedents at least in certain cases. A detailed explanation of the relevant constraints is beyond the scope of this paper, since the exact conditions under which non-agentive uses are acceptable remain largely unclear at present. As regards the contrast between the above sentences and (22), it may be noted that the subject is a non-causal theme, undergoing a change of position or location, while in (26) through (28) it is always a non-volitional cause.

59Whatever the constraints on non-agentive VPAs, their acceptability is not surprising if we consider that main verb do itself is not restricted to agentive predicates, and can readily express simple causation:

60(29) The religious wars of the later sixteenth century (…) did much damage to people and to places in the Pyrenean region... (BNC)

61The behaviour of do so with respect to agentivity is quite different from do it/this/that. As mentioned above, it is not subject to the same constraints, as it allows not only non-agentive antecedents (and this to a much greater extent than do it/this/that), but also stative ones. Attested examples of either type provided by Houser (2010) are shown below:

62(30) If benefits accrue, but do so later during follow up, it would be difficult to discriminate between the effects of more aggressive vs. earlier lipid lowering therapy. (ex. 36h, p. 52)

63(31) The three clusters with fewer matches seem to exhibit a mixture of expression patterns, while the first cluster does so to a much lower extent. (ex. 38a, p. 53)

64Miller (2013) argues that these facts make it difficult to maintain a compositional analysis of do so: if it were indeed compositional, one would expect it to resist non-agentive and stative antecedents in the same way as do it/this/that.

3.1.2 Discourse properties of the object pronoun

65It was argued earlier that alternation between it and demonstratives in VPA is one of the central arguments for compositionality. In many respects, this alternation appears to be determined by the usual properties of the object pronouns involved.

66Quirk et al. explain that do that signals information that is ‘treated to some extent as new or contrastive’ (877) whereas do it conveys ‘entirely given’ information. Likewise, Huddleston and Pullum explain the contrast between do it /do that in the following pair by saying that do it is expected if A’s resignation is ‘already in the air’, while do that would be preferred if A’s resignation is new.

67(32) A: I’ve sent in my resignation. B: Why did you do it?

68(33) A: I’ve sent in my resignation. B: Why did you do that? (ex. 54i-ii, p. 1534)

69Corpus data provide good evidence that information status of the antecedent is an important factor of VPA choice, at least form the point of view of discourse newness: in a sample of 900 occurrences of VPAs from the BNC corpus, containing 500 occurrences of do it and 200 each of do this/that, it was found that do this/that have a discourse-new antecedent (not previously mentioned prior to the antecedent clause) in 80% and 73% of cases, as opposed to just 40% with do it. There is also a tendency for do this to signal ‘new information’ in the more general sense of ‘unfamiliar’ or ‘previously unknown’ (rather than simply ‘new to the discourse’):

70(34) While Christians in Britain lapse, in Germany (…) they formally notify the local authorities of their Austritt (resignation). Every baptised German who has not done this and who pays income tax must also pay the additional church tax. (BNC)

71The choice between do this and do that also appears to be partly subject to the criterion of ‘distance from the speaker’ often employed to account for the meaning of demonstrative pronouns, with proximal this signalling referents that are closer to the speaker in various senses, and do that ones further away. Distance here is not to be understood exclusively in a spatial sense (otherwise the above example would make little sense) but in more abstract ways, such as in terms of temporal proximity or remoteness, of things more or less central to the speaker’s concerns, or more or less pleasant, likeable, or other. Do this, for instance commonly refers to actions that are ongoing at the time of speaking, or have recently happened; do that, on the other hand, can express temporal remoteness as well as actions deemed impossible, unlikely, or even reprehensible:

72(35) Rodney cracked two eggs into the frying pan. ‘I’m not doing this every morning,’ he said. ‘With so many of us we ought to set a rota for cooking.’ (BNC)

73(36) I don’t know anyone who enjoys the criticism. You would have to be a masochist to do that. (BNC)

74Although not all occurrences of do this/that display the proximal or distal meaning to such a degree, those that do are relatively common, providing further evidence that VPAs are semantically compositional. As previously noted, this property sets do it/this/that apart from do so, but it also distinguishes anaphoric do it from its various idiomatic uses, some of which are illustrated below.

75(37) What he says is that if anything goes badly, he tells the team, ‘I did it’. If anything goes reasonably well, he says, ‘We did it!’ But when everything goes really well, he says to them, ‘You did it!’ (BNC)

76(38) I became so engrossed with the eclipse that I watched it in an almost hypnotic trance. And that did it. I’ve been hooked on the sky ever since. (Corpus of Contemporary American English [COCA])

77(39) Just got three bullets.., but that ought to do it. (I. Blasingame, Dakota Cowboy, cited in OED)

78The behaviour of do it in such occurrences differs significantly from its VPA usage. Being an idiom, it is not truly anaphoric insofar as it is not interpreted through the retrieval of an antecedent, but instead has a variety of more or less fixed meanings, which the OED glosses as ‘succeed’ (37), ‘provoke a change or reaction’ (38) or ‘achieve the required result’; ‘be sufficient or enough’) as in (39) where it is roughly synonymous with work or other idioms like do the trick. Alternation with do this/that is in all cases impossible. Note that also that in such uses do it is frequently non-agentive, as it is in (38)–(39), and as such readily allows inanimate pronouns as its subject, in contrast to its usual VP-anaphoric use. The exact range of idiomatic meanings, and the conditions under which they arise, remains to be determined, but it is apparent that they are not straightforwardly analysable as being simply the combination of main verb do and it.

3.2. Is main verb do anaphoric?

79The previous sections have provided evidence that main verb do and the object pronoun of VPAs are both syntactically and semantically independent of each other. Yet the question remains how both parts of the VPA contribute to its anaphoric meaning. Although our analysis leads us to reject the hypothesis that VPAs are anaphoric VPs, it may still be argued that both parts of the VPA may be viewed as anaphoric —do taking up the idea of an action, and the object pronoun specifying the action which is performed.

80While this seems to us a valid description of the meaning, the label ‘anaphoric’ appears unwarranted, as it would imply that VPAs involve a special use of do which is somehow distinct from its use with full NP complements, as in e.g. do the job or do the puzzle. However, the semantic facts discussed earlier show that the two uses are in fact very similar —in both cases, do is an action verb taking a complement which itself denotes an action (either a full NP such as the job, or in the case of VPAs, a pronoun having an action antecedent) or one that is reinterpreted as such, as with the puzzle for instance. Given this, one might suggest that do is also anaphoric in ‘do +NP’ constructions other than VPAs.

81A further difficulty with this analysis is that an anaphoric expression is by definition a semantically non-autonomous one. Thus if do were truly anaphoric, on the same level as the object pronoun, one would expect it to be interpreted via the retrieval of a suitable antecedent. The antecedent would vary depending on the context, just as the interpretation of it/this/that does. Yet this appears to be at odds with the actual behaviour of do, since the action meaning remains constant regardless of context and does not depend on accessing an antecedent. These facts lead us to consider do not as anaphoric, but simply as a generic action verb, as proposed by Huddleston and Pullum (2002):

[T]here is no general deictic verb and no anaphoric verb (except for the dialect-restricted pro-form do), and doga[general agentive] combines with appropriate NPs to form deictic and anaphoric VPs. It also combines with various comparative expressions such as the same thing, likewise, or otherwise, and these retain the semantic and syntactic properties that they exhibit elsewhere.

82Saying that do is generic does not make it an anaphor in its own right, but simply reflects the fact that it refers to the general idea of acting, without more. Thus there is every reason to maintain that VPAs do not involve anaphoric do, but action verb do and an anaphoric complement.

4. The problem of non-finite VP complements with main verb do

83The crux of the hypothesis proposed here has been first that VPAs are ordinary transitive constructions rather than non-compositional, idiom-like chunks, and secondly that only the object pronoun is anaphoric, not the main verb or the VPA as a whole. Although this approach is appealing for the reasons laid out above, a potential objection is that VP antecedent can never be substituted for it/this/that, as the resulting sentence is ungrammatical. Specifically, the semantic paraphrase of (40a) given in (40b) cannot be realised felicitously by the structure in (40c):


a. John mowed the lawn. He did it/this/that in under an hour.

b. Semantics: He carried out the action of mowing the lawn in under an hour.

c. Syntax: *He did mowing/(to) mow the lawn in under an hour.

84Although non-finite VP complements are encountered with other verbs, as shown below, they are ungrammatical with main verb do. This is all the more surprising if we consider that they unambiguously denote actions, and are therefore of the appropriate semantic type for the complement of do.


a. He contemplated/considered passing the exam.

b. He started revising/to revise for the exam.

c. I helped him revise for his exam.

85It may be noted that the do + VP sequence is somewhat redundant; after all, there is no obvious difference between the ungrammatical variants in (40c) and the simpler equivalent He mowed the lawn. Yet although redundancy clearly is an issue here, it is unclear how it can lead to the resulting sentence being ungrammatical, rather than just unnatural to some degree.

86Observing the impossibility of substituting a VP for it, Simner (2001, 73), elaborating on an earlier proposal from Bresnan (1971), argues that it in do it should be regarded as an NP anaphor, coreferent with a previous NP. Under this approach, the antecedent (or rather, the linguistic representation of the antecedent) is “some type of nominalised version of the preceding VP”; for instance, in (42) it would be interpreted roughly as ‘(that thing that is) the diving into the pool’.

87(42) Mary dived into the pool. John did it too. (ex. 16, p. 60)

88Even though the proposed paraphrase is a valid representation of the antecedent (insofar as it accurately reflects the meaning of the pronoun), this does not prove that this particular sort of NP (or indeed any other NP expression) is involved in the processing of do it (or do this/that). The characterisation of it as an NP anaphor seems motivated primarily by the assumption that the antecedent of it must be an NP. Such an assumption is, however, unnecessary if we clearly separate the syntax of the antecedent from its referential dimension. Cornish (1992, 1996, 1999) proposes to capture this distinction through the use of two different terms, ‘antecedent’ referring to the discourse model representation of an entity, and ‘antecedent-trigger’ being the linguistic expression through which this representation is accessed (see also Schuster (1988)’s analogous distinction between ‘antecedent’ and ‘referent’). In other words, interpreting anaphoric expressions always involves accessing an antecedent, in the conceptual sense, but this is not necessarily achieved by means of an antecedent-trigger. Under this conception, there is no requirement that the anaphor should be replaceable by its antecedent-trigger, even though this is often the case.

89Thus, the VP which is traditionally described as the antecedent of VPAs can be said to serve instead as the antecedent-trigger of the object pronoun, which is interpreted in terms of the referent of that VP. The fact that it cannot serve as the complement of main verb do is therefore not an argument against the compositional structure of VPAs.

90It is in fact possible to find other situations in which an element that clearly functions as the antecedent-trigger to an anaphoric expression cannot be substituted for that expression. An example of this is the case of subject-auxiliary inversion (SAI) with pronouns that have clausal antecedent-triggers: although the underlined clause in A’s turn is unquestionably the antecedent-trigger of it/this/that in (43), it cannot be substituted for the pronoun, since no finite that-clause can undergo SAI. Similarly, a that-S is excluded as the complement of a preposition and therefore cannot replace the pronoun in (44). In spite of this, the coreferential relation between the anaphor and antecedent-trigger is in both cases beyond doubt.


a. A–I missed my train.
B–How did it/this/that happen?

b. *How did that you miss your train happen?


a. I will come, you can depend on it/this/that.

b. *You can depend on that I will come.

91The concept of antecedent-trigger provides a way of dealing with such cases by correctly identifying the clause as the element which aids in the interpretation of the anaphor, without being viewed as its structural counterpart.

5. Conclusion

92English ‘do + pronoun’ anaphors do it, do this and do that are transitive constructions of the form ‘main verb do + object pronoun’. Structurally, they are analogous to any other transitive verb-pronoun combination, and display the usual syntactic behaviour of such structures. In addition to a range of properties which demonstrate the syntactically compositional nature of VPAs, there is evidence that they are also semantically compositional, as they are subject to the same semantic constraints as those governing the use of main verb do, and the choice between it and demonstratives is driven by the same factors as in other contexts.

93It or this/that are the only anaphoric elements in do NP VPAs. The object pronouns retain their usual discourse properties, such as the proximal/distal opposition with this/that; a generic action verb do is not anaphoric since its interpretation does not require the hearer to access an antecedent.

94These features distinguish do + pronoun VPAs from do so, which is clearly non-compositional, as it does not license passivisation and allows stative antecedents at least in certain cases. They also allow us to separate anaphoric do it from its idiomatic uses, which are non-compositional by nature since it is used non-referentially and cannot alternate with this/that.

95Given the conclusions of this paper, one may wonder whether do it/this/that can still be referred to as VP anaphors. Though it is arguable that ‘do + pronoun anaphors ’ now seems a more accurate description; the term ‘VP anaphor’ remains in widespread usage in the literature, and makes more obvious the relation between these constructions and do so as well as ellipsis, which are also (in the typical case) interpreted in terms of a previous VP.

96The analysis of VPAs defended here paves the way for further investigation into the factors governing their alternation in discourse. Specifically, the aim is to understand how the choice between it and this/that operates when these pronouns are used with action referents. Discourse-newness of the antecedent was mentioned above as one important difference between do it and do this/that, but since it makes similar predictions for the demonstrative variants, it leaves open the question of what other differences exist between them beyond the proximal/distal distinction which appears to be involved in some but not all of their uses. More generally, a central task for a corpus-based study of VPAs will be to account for the fact that VPA alternation is relatively free in some cases, but very restricted in others.

Top of page


Arregui, Ana, Charles Clifton Jr, Lyn Frazier, and Keir Moulton. “Processing Elided Verb Phrases with Flawed Antecedents: The Recycling Hypothesis.” Journal of Memory and Language 55 (2006): 232–246.

Bolinger, D. “The Meaning of Do So.” Linguistic Inquiry 1 (1970): 140–144. Bresnan, J. “A Note on the Notion ‘Identity of Sense Anaphora’.” Linguistic

Inquiry 2 (1971): 589–597.

Cornish, Francis. “So Be It: The Discourse–Semantic Roles of So and It.”

Journal of Semantics 9 (1992): 163–178.

Cornish, Francis. “‘Antecedentless’ Anaphors: Deixis, Anaphora, or What? Some Evidence from English and French.” Journal of Linguistics 32 (March 1996): 19–41.

Cornish, Francis. Anaphora, Discourse, and Understanding: Evidence from English and French. Oxford University Press, 1999.

Cruse, D. A. “Some Thoughts on Agentivity.” Journal of Linguistics 9 (February 1973): 11–23.

Culicover, Peter W. and Ray Jackendoff. Simpler Syntax. Oxford: Oxford University Press, 2005.

Gruber, Jeffrey S. “Look and See.” Language 43 (1967): 937–947.

Hankamer, Jorge and Ivan Sag. “Deep and Surface Anaphora.” Linguistic Inquiry 7 (July 1976): 391–428.

Hardt, Daniel. Verb Phrase Ellipsis: Form, Meaning, and Processing. PhD thesis, University of Pennsylvania, 1993.

Higgins, Francis R. “Vicarious Do and the Auxiliary.” Ms., University of Massachusetts, Amherst, 1992.

Houser, Michael John. The Syntax and Semantics of Do So Anaphora. PhD thesis, University of California, Berkeley, 2010.

Huddleston, R. and G. K. Pullum. The Cambridge Grammar of English. Cambridge: Cambridge University Press, 2002.

Kehler, Andrew. “Coherence and the Resolution of Ellipsis.” Linguistics and Philosophy 23 (2000): 533–575.

Kertz, Laura. Ellipsis Reconsidered. PhD thesis, University of California at San Diego, Unpublished PhD dissertation, 2010.

Kertz, Laura. “Verb Phrase Ellipsis: the View from Information Structure.” Language 89 (2013): 390–428.

Lakoff, George and John Robert Ross. “Why You Can’t Do So Into The Sink.” Syntax and Semantics, Volume 7: Notes from the Linguistic Underground. Ed. James D. McCawley. Academic Press, 1976.

Merchant, Jason. The Syntax of Silence: Sluicing, Islands, and the Theory of Ellipsis. Oxford University Press, 2001.

Merchant, Jason. “Fragments and Ellipsis.” Linguistics and Philosophy 27 (2004): 661–738.

Michiels, Archibald. “A Note on the Relation between Agent and Stativity.”

Neophilologus Groningen 62(2) (1978): 172–177.

Miller, Philip. “The Choice between Verbal Anaphors in Discourse.” Anaphora Processing and Applications: 8th Discourse Anaphora and Anaphor Resolution Colloquium, DAARC 2011, Faro Portugal, October 6-7, 2011. Revised Selected Papers. Ed. Iris Hendrickx, Sobha Lalitha Devi, Antnio Branco, and Ruslan Mitkov. Springer, January 2011.

Miller, Philip. “Usage Preferences: The Case of the English Verbal Anaphor do so.” Proceedings of the 20th International Conference on Head-Driven Phrase Structure Grammar. Ed. Stefan Müller 2013.

Miller, Philip and Barbara Hemforth. Verb Phrase Ellipsis with Nominal Antecedents. Ms., Université Paris Diderot (submitted), 2014.

Murphy, Gregory L. “Processes of Understanding Anaphora.” Journal of Memory and Language 24 (1985): 290–303.

Quirk, R., S. Greenbaum, G. Leech, J. Svartvik, and D. Crystal. A Comprehensive Grammar of the English Language. Cambridge: Cambridge University Press, 1985.

Sag, Ivan. Deletion and Logical Form. PhD thesis, Massachusetts Institute of Technology, 1976.

Schuster, Ethel. “Pronominal Reference to Events and Actions: Evidence from Naturally-occurring Data”. Technical report, University of Pennsylvania, 1988.

Simner, Julia Claire. Engaging Long and Short Term Memory During Anaphor Comprehension. PhD thesis, University of Sussex, 2001.

Simner, Julia, Martin Pickering, and Alan Garnham. “Discourse Cues to Ambiguity Resolution: Evidence from “Do It” Comprehension.” Discourse Processes 36 (2003): 1–17.

Souesme, Jean-Claude. Do something et ses diverses réalisations en anglais contemporain. PhD thesis, Université de Nice, 1985.

Stroik, Thomas. “On the Light Verb Hypothesis.” Linguistic Inquiry 32 (2001): 362–369.

Tanenhaus, Michael K. and Greg N. Carlson. “Comprehension of Deep and Surface Verb Phrase Anaphors.” Language and Cognitive Processes 5 (1990): 257–280.

Top of page


1 For instance, Kertz points out that the attested example (46) reported by Kehler (2000) is more acceptable than the constructed one from Arregui et al. (2006) in (45) since it bears focus on the auxiliary verb rather than the subject: (45) #The student was praised by the old schoolmaster, and the advisor did too. (Arregui et al., 2006, p. 241, ex. 17). (46) This problem was to have been looked into, but obviously nobody did. (Kehler, 2000, p. 548, ex. 22)

2 Although (12) is an example of cataphora, in which the antecedent-trigger is introduced after the VPA (in contrast to endophoric ones where the antecedent-trigger precedes the VPA), it is just as much a case of anaphora, since interpreting the pronoun also requires accessing an antecedent.

Top of page


Electronic reference

Gabriel Flambard, « A compositional analysis of VP anaphors », Anglophonia [Online], 23 | 2017, Online since 16 January 2017, connection on 20 June 2018. URL : ; DOI : 10.4000/anglophonia.1192

Top of page

About the author

Gabriel Flambard

Université Paris Diderot, Sorbonne Paris Cité

Top of page


Licence Creative Commons
Anglophonia – French Journal of English Linguistics est mis à disposition selon les termes de la licence Creative Commons Attribution - Pas d'Utilisation Commerciale - Pas de Modification 4.0 International.

Top of page
  • Logo Presses universitaires du Midi
  • OpenEdition Journals