This study investigates variability in second language (L2) learning, namely the discrepancy between a person’s first (L1) and second language abilities, and whether the knowledge gap between these languages can ever be closed. Knowing whether an L2 learner can ultimately attain a native-like representation of the L2 they are learning has practical implications for L2 teaching methodology and targets: decisions between the amount of time spent attempting to instil knowledge of principles versus time focused on compensatory strategies are easier to make if a teacher understands what a student has recourse to, and the setting of achievable goals also becomes possible and expectations more manageable. This question – the nature of the linguistic inventory an L2 learner can tap into once their L1 grammar is acquired – has been addressed in several theoretical accounts. According to some accounts, an L2 learner can exploit the very same linguistic inventory they used for their L1 (see Schwartz & Sprouse 1996; White 2003). Differences in learners’ L1 and L2 attainment are explained by appealing to concepts such as deficient input and interference from L1. A more restricted approach is put forward in the Feature Assembly Hypothesis (Choi & Lardiere 2006; Lardiere 2009), which states that the available inventory is characterised by features that are set according to L1 specifications. The learnability problem is stated as a challenge to reconfigure these features when L1 and L2 specifications are at odds with one another. A still more restricted view of ultimate attainment is that the computational system is in effect ‘closed off’ after a certain period (e.g. Hawkins & Chan 1997), and, more specifically, that features of an L2 that are not present in that learner’s L1 will never be acquired, as in Tsimpli and Dimitrakopoulou’s (2007) Interpretability Hypothesis. The prognosis for a learner restricted in this way is that if they are learning a language that has constructions that are absent from their L1, their ultimate L2 attainment will be deficient. A number of studies has tested the empirical reach of these theories (see Domínguez et al. 2011 on Spanish perfective/imperfective aspect; Belletti et al. 2007 on subjects in Italian; Méndez & Slabakova 2014 on resumptive pronouns; Schmid et al. 2014 on voice onset time, vowel discrimination and vpe) and our study contributes to this literature.

Our focus in this paper is on the latter approach, namely the Interpretability Hypothesis, which proposes that features which contribute to the interpretation of a sentence string (interpretable features), are universally available to L2 learners but those that are merely markers of agreement and so do not contribute to interpretation (uninterpretable features) are not (Tsimpli & Dimitrakopoulou 2007). We assessed this claim by focusing on Greek speakers’ knowledge of English verb phrase ellipsis (vpe) as exemplified in (1) below.

As a construction that is subject to lexical, syntactic and discourse constraints, vpe allows us to distinguish between aspects of grammar that L2 language learners can acquire from those they cannot (see Al-Thubaiti 2019). Greek does not allow vpe in the same way that English does, despite allowing arguments to be dropped in specific circumstances (Merchant 2018; Paparounas & Sitaridou 2019) so by concentrating on Greek speakers’ knowledge of this construction in English, it is possible to assess whether they can acquire a grammatical phenomenon that is absent from their L1, that is not taught explicitly, and, as we shall illustrate, whose subtle variations are not entirely deducible from the linguistic data surrounding them. A further reason to investigate vpe is that the elided material in English vpe can contain interpretable or uninterpretable features, so by manipulating these clauses according to this variable, we can test if Greek L2 learners of English distinguish between them when making acceptability judgements. Their judgement patterns can help us gauge whether syntactic features not instantiated in Greek can nevertheless be recognised and integrated into their developing grammar. As such, our study feeds into the aforementioned debate about what restrictions there are on the grammatical properties an L2 learner can acquire and what the source of these restrictions is (White 2003).

The next section illustrates the properties of vpe that are relevant to the variability under consideration and sets out the trigger for our study and its precursor, namely Hawkins (2012), which first examined this phenomenon. We then set out our research questions and explain our methodology.

English vpe is a well-studied phenomenon. It refers to constructions in which the main predicate of a clause is elided, together with any internal arguments it has, as in (2). Despite the absence of the material in the ellipsis site (illustrated by ‘[e]’), this elided material can only be understood as a consideration of whether Javier will attend the party, which implies that the elided site ‘recovers’ its meaning from the antecedent clause. For this reason, the elided material is better described as unpronounced, rather than missing.

English vpe can be licensed by infl, such as the modal verb in (2) above, an auxiliary verb, as demonstrated in (3), or the infinitival marker to, as seen in (4). However, as (5) demonstrates, it is not permissible with a lexical verb.

In Greek, neither auxiliary verbs (6) nor lexical verbs with definite objects (7) license VP ellipsis.

However, Greek does occasionally allow instances of ellipsis with copula verbs and some lexical verbs with indefinite objects.

Different analyses have been offered for these examples. Whereas in Merchant (2018) they are described as verb-stranding ellipsis, Paparounas & Sitaridou (2019), claim that they are best viewed as argument ellipsis. For present purposes, what is important is that the constructions given to our participants did not have these implicit arguments.

For example, (10) below, would be ungrammatical in Greek, as (11 a and b) illustrate.

Only when the entire TP is elided, as in (12), leaving just the remnant – which is usually the subject or object – and apolarity marker stranded, do these sentences improve (Kolokonte 2008).

Since Modern Greek does not license the type of vpe available in English (Kolokonte 2008, Paparounas & Sitaridou 2019), ascertaining whether Greek learners of English can learn this construction contributes well to the question of whether L2 learners can acquire a linguistic phenomenon that is absent from their L1. However, to understand the acquisition hurdles facing a learner of this construction, we should first illustrate the constraints that regulate vpe. With that in mind, we turn first to what are conventionally termed as the ‘parallelism constraints’ that regulate vpe.

In the PF-deletion approach to vpe adopted here, the elided material is represented syntactically but not pronounced (see Chomsky & Lasnik 1993; Lasnik 1999; Kennedy & Merchant 2000; Merchant 2001; Kennedy 2003; Merchant 2004). An additional morphosyntactic and semantic parallelism is often argued to hold between the elided material and the antecedent clause. The examples in (13) illustrate this point. (13a) is acceptable because both the elided material and the antecedent sentence are in the passive voice. A mismatch in voice between the two sentences in (13b), however, renders the result less felicitous.1

However, not every morphosyntactic mismatch between the antecedent sentence and elided clause leads to ill-formedness, as (14) and (15) illustrate, which runs counter to the aforementioned parallelism constraint. Here, the antecedent sentences contain a simple past and present perfect respectively, neither of which match the uninflected elided verb. Despite this mismatch, the sentences are felicitous:

Lasnik (1995) dealt with this anomaly by appealing to regulations that operate at different levels of a sentence’s derivation. He proposed that the parallelism constraint regulating the dependency between the two sites held prior to the application that combined bare stems with their affixes. To illustrate, prior to this application, the representations of (14) and (15) would be as in (16) and (17), respectively. These examples show that at this point in the derivation parallelism between the verbs holds; this allows meaning to be recovered, and in turn predicts their acceptability.

In contrast, examples like (18) are expected to be bad because the progressive affix in the VP marked for deletion is at odds with the affix in the antecedent sentence.

Lasnik’s account extends to another infelicitous example of vpe involving copula be:

L1 speakers generally dislike these examples (see Hawkins 2012; Al-Thubaiti 2019), a judgement that falls out from a proposed lexical difference between inflected main verbs and inflected copula forms. The former are inflected in the syntax (through affix hopping) whereas the latter are stored inflected in the lexicon so when entering the derivation, they merge directly with a non-affixal infl category. Since the copula verb is inserted as is, there is no level at which syntactic identity holds between the antecedent and the elided clause, thereby breaching the parallelism constraint.

A sticking point in this account is that one would expect vpe with a perfective form in the elided clause, (20), to be judged on a par with vpe with a progressive form in the elided clause (21), contrary to fact.

L1 English speakers consistently dislike (21) and find (20) substantially better. This is strange as in neither example is parallelism upheld, which should predict both to be rejected. In this case then, the account underpredicts. However, a principle that anticipates these different judgements is formulated in Rouveret (2012), which introduces the notion of interpretability as a pivotal factor governing recoverability of elided material and so whether or not the elided material is licit:

Rouveret (2012) suggests that the criterion that feature interpretability is based upon is the meaning contribution of these two affixes. In English, the -en affix does not convey the perfective meaning, as it occurs both in perfective verb forms (e.g. have eaten) and passive forms (e.g. be eaten). The perfective meaning can be found on the auxiliary verb and not the affix -en. In contrast, the -ing morpheme only occurs in progressive verb forms (e.g. be eating) so it is assumed that -ing rather than be carries the progressive interpretation, and as a result, it bears an interpretable feature.

Thus, according to his line of thought, in (20) the perfective feature on the unpronounced participle is purely an agreement feature (valued by the interpretable feature on the main verb has). Since -en does not contribute to meaning, it is uninterpretable so irrelevant to (22), which is only concerned with interpretable features. However, example (21) does not escape the principle because the progressive feature makes a semantic contribution, signalling the progressive reading. As an interpretable feature, it is needed to compute the material but since its meaning cannot be recovered from the antecedent clause, it violates (22) and is correctly barred. This distinction between uninterpretable and interpretable features neatly captures native speakers’ graded judgements, who rate examples like (21) as worse than (20). If it is the case that L2 speakers also follow this constraint, they should be able to discern the subtle distinction between (20) and (21). This will depend not only on their ability to acquire vpe per se but also on whether or not the progressive and perfect features in their L1 are interpretable. If the interpretability values of these features in their L1 differ from those in English, this distinction will serve as a good testing ground for ascertaining which features can be acquired and which not.

So far, we have illustrated some key restrictions on vpe, focusing particularly on the constructions that the present study will assess. They are summarised in Table 1. We have seen that English vpe is permissible when strict morphosyntactic parallelism holds between the antecedent and elided clause, as in (A) and (E), but it is illicit when there is a finite lexical verb (B) and infelicitous with an inflected form of the copula verb, as in (F). We have also attributed the subtle contrast between (C) and (D) to the interpretability of the progressive marker versus that of the perfective.

Clearly, English vpe is a complex phenomenon with specific parallelism and recoverability conditions involving the syntax-lexicon interface and the syntax-discourse interface, which makes it an exciting topic for investigation in sla. Our interest is twofold: to examine if Greek speakers acquire this phenomenon at all, and if they do, to find out whether this acquisition is complete or restricted by feature interpretability. Note that the linguistic evidence available to them is scarce. Firstly, the complicated syntactic constraints outlined above are not taught in classrooms and second, these are not generalisations a typical learner could draw from linguistic data they encounter. An example will corroborate this second point. A vpe sentence like (B) in Table 1 is ungrammatical. However, some transitive verbs in English seem to permit being stranded without their arguments.

Such examples, in fact, involve so-called ‘null arbitrary objects’, indicated by their permissible readings: the elided clause cannot mean ‘Mary wrote Jill a letter too’ only that ‘Mary wrote something else’, as Hawkins illustrates (Hawkins 2012: 406). There is nothing to help learners distinguish between licit sentences with null objects as in (23) and ungrammatical vpe sentences in (B) of Table 1. If the learners recognise the ungrammaticality of (B), this demonstrates that they have acquired the basic licensing conditions of ellipsis. In addition, if their judgements on (D) and (F) are also target-like, it would indicate that they follow the recoverability principle, judging them as ill-formed because in both instances, the elided phrase’s content cannot be recovered from the antecedent clause.

A further issue concerns the distinction between (C) and (D) in Table 1. Understanding the precise conditions of vpe licensing and recoverability is not enough to distinguish these two examples, which we have seen depends upon the interpretability of the aspect in question. If both languages share the same interpretable features, one might expect knowledge to transfer across but if not, the question arises as to whether these feature mismatches can be reconciled in the learner’s L2 grammar.

Regarding the progressive feature, Greek does not have a specific affix that marks the progressive. The progressive interpretation is conveyed by the context, as in (24), and/or adverbials as in (25).

With respect to the perfective, the present perfect is formed by the verb ɛχω ‘have’ in the present tense, followed by the third person singular of the past subjunctive form of the verb. This is a dependent form that carries the features of [+perfective] [–past] (see, for example, Holton et al. 2012). The past subjunctive is formed by the aorist stem plus the suffix -ɛι, as in (26). In contrast to English, this form does carry an interpretable feature: it distinguishes between the active (26) and passive voice (27) (Varlokosta et al. 2006).

If Greek speakers distinguish between examples such as (C) and (D) in Table 1, it would indicate that they recognise the uninterpretability of the perfective aspectual feature in English and consequently, that they can acquire uninterpretable features not instantiated in their L1. In contrast, if these uninterpretable features are no longer accessible to them, as the IH predicts, then we expect these sentences to cause learnability problems, even at advanced stages of L2 acquisition, what Al-Thubaiti (2019) refers to as ‘selective vulnerability’.

Hawkins (2012) examined the acquisition of English vpe by collecting L1 Mandarin and L1 Saudi Arabic speakers’ grammaticality judgements of vpe sentences. Aligning with analyses that classify Mandarin and Arabic as lacking verb-stranding ellipsis (Aoun and Li 2008),2 Hawkins asked whether the learners could acquire this grammatical phenomenon at all, and if so, whether they showed evidence of having acquired the more subtle contrasts, as discussed above. Results were mixed. First, both groups demonstrated knowledge of vpe constructions generally; this was evident from their acceptance of examples similar to (A) in Table 1 above and rejection of those represented by (B). However, on probing them with different vpe constructions, for example, where the material in the elided clause did not parallel that in the preceding sentence, their judgements departed from L1 English speakers in important ways. Specifically, whereas L1 speakers drew the predictable contrast between examples similar to (C) and (D) in Table 1, the Mandarin speaking and Saudi Arabic speaking groups did not, judging them as equally bad.

If neither Saudi Arabic nor Mandarin have an uninterpretable perfective feature, a ready explanation for why the L2 groups did not distinguish between these examples was that the uninterpretable feature in question had not been instantiated in their grammar. In other words, the Saudi Arabic and Mandarin speakers were treating -en and -ing similarly, namely as having interpretable features. Appealing to the Interpretability Hypothesis, Hawkins (2012) suggested that their performance was consistent with the uninterpretable feature of the perfective -en remaining inaccessible to them. A caveat posited was that a similar pattern being exhibited by highly advanced learners of English would be necessary to support this claim. Al-Thubaiti (2019) did just this with 34 highly proficient Saudi Arabic learners of English. Using a bimodal timed acceptability judgement task, she found her participants also judged sentences with -en and -ing as equally bad. This latter study, therefore, provides stronger support for the selective vulnerability of L2 speakers with respect to the grammar-internal property of interpretability.

The ability of L2 speakers to acquire an elliptical construction not present in their language has been investigated in other languages as well. Duffield et al. (2009) examined advanced learners of English, with various L1s that are argued not to instantiate English type vpe either, such as Dutch, Spanish and Japanese. Their participants undertook a sentence completion task consisting of vpe sentences (and other structures not directly relevant to the current discussion) where parallelism had or had not been disrupted. Amongst other questions, they asked whether the learners’ judgements implied that they had grasped the vpe construction, and whether they exhibited parallelism effects, namely preferring undisrupted examples of vpe to disrupted ones. Results were mixed: Dutch speakers showed a high acceptance of vpe as well as strong parallelism effects, whereas the Spanish and Japanese speakers gave much lower acceptance ratings for vpe yet still demonstrated parallelism effects. Of particular relevance to our study was that the Dutch speakers’ high performance on English vpe shows that it is possible for a construction that is entirely absent in a learner’s L1 to be acquired. Koyama (2016), who also studied Japanese learners’ knowledge of English vpe (using a sentence acceptability judgement task), found that her participants accepted grammatical vpe constructions, too. However, what was unusual in her study was that the participants did not reject vpe with a finite lexical verb, which is ungrammatical in English. Koyama pointed to the availability of object drop in Japanese, which might have been influencing their judgments. However, as the study did not refer to the level of these learners’ English, it is difficult to ascertain whether this was a passing phenomenon or indicative of a more permanent impasse. In the next section, we turn to our research questions, which built on this literature.

This study extends the research on the topic of selective vulnerability by examining Greek advanced learners of English. Modern Greek not only lacks vpe but also marks aspect differently from English, thus contributing to the cross-linguistic data available thus far. Testing speakers who have reached a higher level of proficiency is also important. Unlike Hawkins (2012), incorporating speakers of higher proficiency will enable us to see if the relevant features are problematic even at advanced stages of language acquisition. We also test a larger range of vpe sentence types and increase the number of tokens, thereby assessing the empirical reach of previous findings. Our research questions are as below.

Our study was interested in the extent to which advanced learners of an L2 can acquire grammatical phenomena that are absent from their L1. The test we chose was the learning of English vpe by Greek learners of English. This was because Greek is argued to lack vpe and its feature constellation – in terms of the interpretable/uninterpretable feature distinction – is different to that of English. These differences allowed us to test whether (a) Greek speakers accepted vpe at all, (b) whether they followed the principles that regulate it, and (c) whether they drew a distinction between interpretable and uninterpretable features in the elided clause. Our results for the former two questions were unequivocally affirmative but our results for the latter remain inconclusive. We discuss these in turn, relating them to the literature that provided the impetus for our investigation.

Our first research question focused on whether Greek speakers demonstrated knowledge of English vpe despite this construction being absent in Greek. Recall that Hawkins (2012) had asked this question of Mandarin and Saudi Arabic speakers, which were languages reported to lack this kind of vpe,4 and had found that both groups accepted it. Our study took Hawkins’ study as a starting point but constructed a novel and broader set of sentence sets, examined a different L2 language, and focused on highly advanced learners of English. Our Greek and English participants gave very high ratings to the vpe constructions and to the vpe control sentences but both groups significantly preferred the vpe constructions over the control ones. This is different from Hawkins (2012), whose participants gave similar sentence sets equally high ratings. With respect to our data, we did not find it surprising that our participants preferred the vpe sentences, which were shorter and more typical than the rather long-winded control sentences. Hawkins’ test items differed from ours in a way which might explain the contrast between our findings: his target sentences were preceded by two contextualising sentences (as opposed to ours, which only used one), and this might have provided a better balance between the antecedent and the elided clause. However, what is relevant to us is that the Greek speakers found English vpe to be perfectly acceptable. These results not only corroborate Hawkins (2012) but also the study reported in Al-Thubaiti (2019), which tested Saudi Arabic highly advanced learners of English on vpe. Al-Thubaiti did not directly compare vpe constructions with non-vpe sentences. Instead, her study contrasted strict vpe identity cases with modals and finite lexical verbs in the elided clause. However, her results were comparable to ours, as her Saudi Arabic speakers also rejected examples with finite lexical verbs. Our results add another group of L2 learners who have proven able to acquire a construction not instantiated in their L1, thus aligning with the aforementioned studies as well as Duffield et al (2009). On the other hand, they contrast with those of Koyama (2016) on Japanese speakers, who accepted ungrammatical sentences with finite lexical verbs. We suspect that the speakers’ lower proficiency might answer for this discrepancy.

Our second research question turned from strictly parallel cases of English vpe to those where parallelism had been disrupted. This set of vpe examples included different types of mismatches. Some of them are accepted by L1 speakers, even if graded, while others are rejected by them outright. Rouveret’s principle of recoverability (repeated below) had proved successful in capturing the contrast in L1 speaker judgements of such sentences. The question we asked was whether Greek speakers’ ratings would pattern in accordance with this principle, too.

The first set of non-parallel examples compared vpe with a stranded lexical verb in the elided clause with those that had modal verbs in the elided clause (Table 6). Recall that these cases, which are accepted by L1 speakers, align with the principle of recoverability, because despite the mismatch, the elided clause does not contain a non-recoverable interpretable feature. If judgments on vpe with ‘did’ were a reliable reflection of participants’ more general vpe representations, then these judgments should extend to vpe with a variety of modals. We found that both L1 and L2 speakers drew a distinction between these sentence sets. This showed that Greek speakers not only accepted vpe but also understood one of the restrictions on it, namely that only certain heads – inflection and modals – permit vpe. Of note, however, is that for neither group were the vpe judgments with modals as high as they were for vpe with ‘did’. This can be explained, however, by the parallelism having been disturbed. Al-Thubaiti (2019) drew similar comparisons and her results were in sync with ours. Across all her groups (L1 and L2), vpe with non-parallel antecedents received lower ratings than vpe with parallel ones.

The next test of non-parallelism also involved use of a modal verb in the elided clause but a finite copula in the antecedent sentence. We saw in section 2 that affix hopping does not apply to the copula, meaning that the copula is already inflected upon insertion. Thus, this example violates Recoverability so was predicted to have been severely disliked. This sentence type was tested against vpe examples which had a stranded lexical verb in the elided clause. The expectation was that all participants should rate both very low, and this was indeed the case. However, both language groups gave slightly higher ratings to the vpe sentences with the stranded verb. A potential reason for this discrepancy was put forth by Hawkins, namely that some transitive verbs (e.g. read, write) optionally allow arbitrary null objects. For example, the elided clause of Jack wrote Jill a letter. Mary wrote ___ too, could be understood as containing a null object (i.e. Mary wrote something else). The possibility of a different reading might have interfered momentarily with participants’ judgements, a suggestion that could be probed by subsequent testing.

To further examine participants’ adherence to recoverability, we compared vpe examples which had the past progressive in the elided clause but not in the antecedent sentence against those with the opposite pattern (see Table 8). Since an interpretable feature in the elided clause violates Recoverability in this instance, participants were predicted to reject the progressive in the elided clause, which is exactly what they did. Our results were again in accordance with Hawkins (2012). However, to ensure that this result did not simply reflect a dislike for the progressive per se, our study also compared vpe control sentences with the progressive against vpe sentences with a progressive in the elided clause. Our finding that participants accepted vpe control sentences with the progressive lends further support to their rejection of the relevant vpe sentences being due to Recoverability rather than a general dislike for the progressive in this environment. To further increase the empirical range of this finding, we also tested the present progressive and found the same pattern. All participants gave much lower scores to sentences with the interpretable aspect in the elided clause than when it occurred in the antecedent clause (Table 10). Such a result is compatible with the conclusion that our Greek speakers operate in accordance with the principle of Recoverability.

Research question III focused on un/interpretability and whether Greek speakers would treat the English perfective feature as uninterpretable or interpretable. The first test sentences inserted the uninterpretable perfect aspect feature in the elided clause or in the antecedent clause. If the Greek speakers treated this English feature as uninterpretable, they should not have exhibited much difference in their preference for which clause the perfective was in. If, however, they were treating the feature as interpretable (as it is in Greek), they should have given lower ratings to the sentences with the perfective aspect in the elided clause. The results showed that Greek participants gave the sentences with perfective aspect in the elided clause significantly lower ratings, with a median of 9 and 7, respectively. At first sight, this appears to support their treating the perfective feature as interpretable, but the validity of this test is scuppered by the English speakers’ results, which also exhibited the same asymmetry. Although English speakers gave high ratings to both – medians of 11 and 8, respectively – the lower rating of the perfective feature in the elided clause was unexpected and affects the validity of the test. In our last crucial comparison, however, between sentences with the present perfect in the elided clause against those with past progressive in the elided clause, both groups showed a significant preference for the present perfect, and this was despite the extra tense mismatch in the former. At first glance, this pattern indicates that Greek speakers are treating the perfective feature as uninterpretable, thus going against the Interpretability Hypothesis, which argues that later learners can no longer acquire these features. On closer inspection, however, we notice that while both L1 and L2 speakers were quite homogenous in their judgments for the progressive, they exhibited a greater individual variation in their judgment of the perfective, perhaps due to the further mismatch in tense (see Table 12). A future study should incorporate more examples that can disentangle the effect of tense and other factors. At this point, the question of whether interpretability lies at the source of these judgements can only be answered tentatively; the contrast drawn between the progressive and the perfective is suggestive of this particular uninterpretable feature having been identified as such by some of our Greek speakers.

This study asked what happens when an L2 learner encounters a construction that does not exist in their mother tongue, and whose regulation is complex and cannot be deduced solely from their linguistic environment, be that a taught environment or not. We examined whether a learner could acquire such a construction at all, whether they conformed to the principles that regulate its use, and whether they showed evidence of discerning between features whose L1/L2 settings differ. Building on key papers that have addressed these questions, particularly Hawkins (2012) and Al-Thubaiti (2019), we focused on an additional language and learner group, testing highly-advanced Greek learners of English on their knowledge of English vpe. Using a different task, and novel and expanded materials, we found that Greek speakers did accept English vpe and their judgements also indicated that they conformed to its regulatory principles (Rouveret 2012). However, our final question on feature settings needs further work before any conclusions can be drawn reliably. With respect to feature interpretability (Tsimpli & Dimitrakopoulou 2007), previous research on vpe (Al-Thubaiti, 2009; 2019) suggested that even advanced learners struggle with features that have a different specification from that in their L1. The variability displayed by the participants in the current study on similar examples which tampered with the interpretability of the features in the elided clause prevent any strong conclusions about this final research question. The Greek speakers’ strong dislike for the progressive in the elided clause came through clearly but the heterogeneity evident in their judgements of the perfective in the elided clause might suggest that whereas some advanced speakers’ judgements are compatible with their having recognised it as uninterpretable, others are not. In addition, even though English speakers rated both sentence types highly, their judgements on the perfective also exhibited a variation that was not expected and differed from that reported in Hawkins (2012). Our paper has thus also brought into focus the need to consider carefully the subtle distinctions that arise when creating example pairs and their possible repercussions for people’s judgements. Looking at ellipsis in this broader context will help move this question forward.