1.1 Relevant properties of Jamaican Creole

While JC is a non-null subject language, children acquiring JC display a null-subject phase (De Lisser, Durrleman, Rizzi & Shlonsky 2016), as is typically observed in the development of non-null subject languages. It is an SVO language with various free pre-verbal morphemes that accompany the verb to express Tense, Modality and Aspect. Such markers form a functional sequence, which conforms to the general cartographic hierarchy identified by Cinque (1999), as shown in detail in Durrleman (2008). Unmarked non-stative verbs are normally interpreted with a past reference (example 1a).² In the case where the past tense marker did (or its variants) is inserted before a non-stative verb, it tends to yield an anterior past interpretation (example 1b). Unmarked stative verbs tend to be interpreted as non-past (example 2a), for which the insertion of the past tense marker is necessary in order to yield a past interpretation (example 2b).

Other factors, such as context, may impact the temporal reading of the verb (Durrleman 2008). An unmarked verb alone cannot express progressive aspect, although it can express habitual aspect (example 3a). The progressive aspect marker a, da or de, and the prospective aspect marker a_(g)o, gaa, or gwain, are obligatory, as their omission yields differences in temporal/aspectual interpretation (examples 3b and 3c).³

Unmarked verbs can also be non-finite verbs which need no pre-verbal infinitival marker. However, fi, go, or both fi go would generally occur before non-finite verbs (example 4). According to Patrick (2007; p.7) “Fi is common but optional with a number of verb types licensed for zero-infinitives in other languages,” however, in some cases infinitival fi cannot be deleted, nor can it be replaced by its counterparts go or fi go (example 5); proving infinitives to be a complex phenomenon in JC.

Child productions in our corpus contain both utterances with adult-like specifications of aspect markers (examples 6 and 7), and target-inconsistent utterances with bare lexical verbs, but with contextually inferred interpretations of progressive or prospective aspect (examples 8a and 9a). The examples in (8b) and (9b) illustrate the adult grammar meanings of child productions of aspectual interpretation with bare verbs.

De Lisser, Durrleman, Shlonsky and Rizzi (2017) provide evidence that JC children at an early age have a clear grasp of the functional projections of the IP of their target grammar. Specifically, they produce utterances consistent with the cartographic structure of the entire TMA domain and do not entertain target-inconsistent TMA combinations. In the current work, we turn our attention to instances of TMA omissions where a particular temporal and aspectual meaning is intended, a target-inconsistent production typically found in development.

1.2 Root Infinitives

Much work has been devoted to the RIs acquisition stage over the years, and different models to interpret it theoretically have been proposed: the Agreement/Tense Omission Model (ATOM) (Schuetze & Wexler 1996); the Unique Checking Constraint Model (UCC) (Wexler 1998); the Truncation Approach (Rizzi 1993/4); the Aspectual Anchoring Hypothesis (Hyams 2007, 2011); the Variational Learning Model (Legate & Yang 2007); and models stipulating a feature-based developmental pattern, with uninterpretable (inflectional) features emerging later than interpretable features (Tsimpli 2005), etc. Most of these approaches are united in arguing that, at an early stage of development, children’s syntactic representation and computation is, in some sense, impoverished as compared with adult grammar. According to the ATOM, for instance, RIs (or Optional Infinitives/OI) arise when tense or agreement features are left underspecified in the child’s clausal representation, causing the verb to surface in its bare form. The ATOM later developed into the UCC, which states that only one uninterpretable feature can be checked in the derivation, i.e. either T or Agr can be checked, but not both. Another view is that the parametric values of such inflectional features, being uninterpretable at LF, are acquired later than interpretable features such as wh (see Tsimpli 2005). The Aspectual Anchoring Hypothesis holds that RIs are a result of children exploring the parametric option of encoding temporal meanings via aspectual instead of temporal marking, an idea also found in the Variational Learning Model (Legate & Yang 2007). The latter further explains why the temporal option is more delayed in some instances than in others, i.e. this variation would be due to lack of sufficient evidence. More specifically, as UG allows tense to be active morphosyntactically (as in English or Spanish) or not (as in Mandarin), children would require sufficient evidence to set the parameter one way or another, and this setting would be delayed in instances where the input provides less unambiguously tensed verbs (as for English) than when the input is more consistent in this respect (as in Spanish), yielding an RIs phase in the former case.

Despite the merits of these approaches, they do not explain various structural restrictions associated with RI, which follow straightforwardly from the Truncation model (Rizzi 1993/4), to which we now turn. This approach assumes that child systems can truncate the structural tree at different layers: the tree will then exclude all the positions above the truncation site and include all the positions below it.

So a truncated configuration can only have missing layers which are external and higher than the truncation site (F1, F2, F3 in (10)), but not internal missing layers. In (10), F5 could not be missing if it is an obligatory layer while F4 is present. Classical truncation understood the process as involving the structural absence of the truncated material (Rizzi 1993/4): in these terms, F1, F2, F3 are simply not present in representation (10), and the root of the structure is the projection of F4. De Lisser et al. (2016) discussed a weaker version of truncation, looking at it as a spell-out phenomenon. In this view, the representation in (10) is complete and projects up to F1, but spell-out stops at F4, so that F1, F2, F3 are structurally present but unpronounced. The two views make different predictions in certain cases, but we will not go into the intricate details of the empirical evidence distinguishing them.

The truncation hypothesis predicts that if RIs involve truncation below the T node (hence missing the tense specification and requiring a non-finite verbal form), they should be unavailable in sentences requiring access to structural layers from TP upwards. We will focus in the current work on the subset of properties that we can immediately test in JC, namely:⁴

1. RIs should be incompatible with wh-questions, which require the presence of the CP layer to host the wh-element (Crisma 1992 on French). This prediction is to a large degree confirmed: For instance, in Child Dutch, Haegeman (1995a) found 16% RIs among non-interrogative clauses and 2.5% among wh-interrogatives. Similarly, for French, Crisma (1992) reported 15% nonfinite main verbs in declaratives and 0% in wh-questions.⁵

2. RIs may be consistent with yes-no questions, if, in the particular language under consideration, these do not have to be signaled through the activation of the left periphery (as in English, with I to C movement). See below for discussion.

1.3 Research questions

Based on the above, we can address the issue of the existence and manifestation of RIs in child JC. What would RIs look like in a Creole language, in which verbs are not morphologically inflected, and in the absence of a distinctive morphology for infinitival verbs? Building on Pratas and Hyams’ (2009) analysis of Capeverdean Creole, we assume that a root infinitive involves a clause with a bare verb in contexts in which the adult grammar requires the appropriate tense and/or aspect marker. In this article we focus on progressive and prospective aspect, the most common aspectual specifications which are mastered early on by language learners (De Lisser, et al. 2017). So, the following questions arise:

1. Do we find, in child JC natural productions, a robust attestation of target-inconsistent cases of bare verbs expressing interpretations of progressive or prospective aspect?

2. Do such target-inconsistent uses manifest the structural properties of RIs in other languages, i.e. are they incompatible with wh-questions, while being consistent with yes-no questions?

3.1 Bare verbs in JC

The data reveals a total 12,592 bare verbs in the corpus. The temporal and aspectual interpretation of each example was evaluated by a native speaker in the context in which the example occurred, in order to tease apart target consistent and target inconsistent uses by children. Out of this total, 5765 non-stative bare verbs occurred in contexts with a past time interpretation (example 11) and 4404 with a present interpretation with a stative verb (example 12), in accordance with adult grammar. However, 2423 bare verbs were used target inconsistently in that they occurred in contexts which expressed progressive and prospective aspectual interpretations (example 13): such cases would require overt progressive or prospective aspectual marking in adult grammar.

Why do children overextend the use of bare verbs to utterances reflecting aspectual interpretations? Do children also show adult-like behavior in their production of aspectual markers? An investigation of the production of progressive and prospective aspects will be presented in the following section.

3.2 Progressive and prospective aspects in early Jamaican Creole

As examples (8) and (9) demonstrate, the omission of the aspectual markers in JC leads to a difference in the temporal interpretation of the utterance. Though these differences are subtle, based on the context of utterance, we are able to determine where they should have been present but were omitted. Such an analysis allows us to map more clearly the development of the aspectual zone. We present in Table 1 a detailed analysis of the development of the progressive and prospective aspect, mapping the production of the overt progressive and prospective aspectual markers compared to their respective omissions, as deduced from the context of the utterance.

The data shows that children abundantly produced both progressive utterances (4745) and prospective utterances (3007), whether overtly marked or not. The total number of correctly marked progressive and prospective aspects, reflecting adult-like usage, is 3432 and 1897, respectively. While the total number of unmarked structures with progressive and prospective interpretations is lower, 1313 and 1110 respectively, their attestation in the data is substantial. The progressive marker shows a 28% omission rate while the prospective aspect marker was omitted in 37% of the utterances under consideration.

We would like to note that in adult JC, the use of the progressive marker is limited to non-stative verbs. The data was therefore analyzed to determine whether other errors, besides omissions, were prevalent in the children’s production. The examination of the data reveals that children do not incorrectly overextend progressive markings to stative verbs (in line with Bickerton 1981; Erbaugh 1992; Shirai and Andersen 1995; among others).⁶ This suggests that children master the use of the progressive marker from early on.

Additionally, we see where the progressive marker is used target-consistently with the following arguably stative verbs (verbs of bodily sensation) but with a dynamic reading: pien ‘pain’, fiil ‘feel’, si ‘see’, luk ‘look’ and ier ‘hear’. The data also reveals utterances where the progressive marker was incorrectly extended to a prospective utterance, as in (14). The number of these utterances was very small as only 15 cases were identified.

The analysis of the aspectual markers therefore reveals that errors of commission are virtually unattested, while errors of omission are robust.

Examples (15)–(20) are cases in point of utterances with aspectual interpretations, but for which there are grammatical errors of omission. These are child-specific constructs that, based on the contexts of utterance and intended interpretation, clearly omit the overt aspectual distinctions, and as such are not in line with adult JC grammar.

Recall that in adult JC, bare non-stative verbs are interpreted as past tense. However, as these examples demonstrate, children employ bare verbs also in contexts of present progressive (15–17) and prospective (18–20) interpretations. These utterances, which express aspectual reading in the absence of the appropriate aspectual morphemes, are naturally interpreted as the language-specific variants of the RIs stage in JC.

RIs in JC, understood as target-inconsistent omissions of aspectual markers, are at their peak at the beginning of the recordings (1;08). The phenomenon decreases in the course of the third year of life and drops to around 10% at age 3:

Figure 1 demonstrates the production of RIs with aspectual interpretations in the corpus (see Appendix 1 for details).

Figure 1

Production of RIs with aspectual interpretations.

The interpretation of utterances lacking an aspectual marker as a language-specific variant of RIs is made plausible by the roughly parallel loss of bare verbs with aspectual interpretation, as evidenced from Figure 1, and the loss of root null subjects, as discussed in De Lisser et al. (2016). Figure 1 shows that bare verbs with aspectual interpretation steadily decline in the second half of the third year of life, and fluctuate around 10% at age 3. Similarly, as figure 7 of De Lisser et al 2016 shows, the production of sentences with null subjects drops to 20% at age 2;08, and fluctuates around 10% at age 3. A concomitant loss of subject drop and RIs is expected if the two are traced back to the same underlying factor, namely the availability of truncation; see §1.2.

If the interpretations of aspect drop as the language specific manifestation of RIs is made plausible by these considerations, what remains to be done to consolidate the analysis is to determine whether the child productions with missing aspect respect the other structural constraints which typically characterize RIs across languages, as discussed in § 1.2. We address this question in the next sections.⁷

3.3 Root Infinitives and Overt/Null Subjects

The data in our corpus contains both structures with target-inconsistent aspect omission which have a null subject, as in examples (21-30), as well as such structures with an overt subject, as in examples (31–40):⁸

The two cases are not equally frequent, though. Overt subjects are only attested in 16% (395 out of 2423) of RIs utterances. So RIs are much more frequent in environments where the subject is dropped (84%), compared to environments where the subject is overtly pronounced.

Further analysis of the data reveals that null subjects are much more frequent with RIs than with finite clauses. As demonstrated in Table 2, the total proportion of null subjects in finite declarative clauses in JC is a mere 18.9% when compared to 88.9% null subjects with root infinitival declarative clauses.⁹

Figure 2 offers a visual representation of the findings detailed above. There is a striking asymmetry between the environments where null subjects occur. While they are preponderant in root infinitival clauses, they are much less frequently attested in finite clauses over the whole period taken into consideration (they are robustly attested till around age 30 months, (see De Lisser et al. 2016 for discussion).

Figure 2

Subject Omission in Finite and Root Infinitival clauses in JC.

The findings, as illustrated in Figure 2, are in line with findings from other languages as detailed in Table 3.

The proportion of null and overt subjects in JC sentences with a missing aspectual marker is thus fully comparable to the proportion of null and overt subjects with RIs in other languages. This provides comparative support for the hypothesis that aspectless bare verb clauses in child JC are the language-specific expression of RI.

If we now go back to the truncation approach, an analytic question arises. If overt subjects require case (Chomsky 1981) and if nominative is dependent on the presence of T, then structures truncated below T would be predicted to never permit overt subjects. The data show that, while clearly less frequent than in finite environments, overt subjects are attested. Why is it so? Notice that the prediction of the truncation approach is fulfilled by weak pronominal subjects, which are virtually unattested in RI (see Haegeman 1995/6 for Dutch and Rasetti 2000 for French).¹⁰ So, why is the distribution of strong subject DP’s less strictly constrained?

Pursuing the case analysis, we observe that there may be other ways to case mark a strong DP, for instance by assigning it a default case in an otherwise caseless environment. This may be the mechanism involved in case marking DP’s in hanging topic position (Cinque 1999 and much related literature), and possibly in special non-finite constructions such as English Me do that?! Never! So, in cases with overt subjects with RIs the child may be exploring such a default case assignment mechanism, permitting overt subjects in some cases. Such a mechanism would not suffice to satisfy the stricter licensing requirement of weak subjects (which in fact are not legitimate in Hanging Topic and similar special positions). Moreover, it would give rise to a clearly distinct, and weaker, proportion of overt subjects than in finite environments, which can rely on the normal nominative assignment mechanism, hence do not require any special assumption by the language learner on default case assignment.

Whatever the merits of these speculations, it seems to us that the sharp difference in the proportion of overt and null subjects in full clauses and clauses with missing aspectual marker, as illustrated in Figure 2, supports the hypothesis that structures with missing aspectual marking are the language specific manifestation of RIs in JC. This interpretation is reinforced by the fact that the observed difference in subject drop is in line with similar differences between finite and RIs environments in other child languages.

3.4 Root infinitives, declaratives and interrogatives

As pointed out in section 1.2, the truncation approach predicts a sharp contrast between declaratives and wh questions: the former, not involving any overt marker in the higher layers of the clause, should be consistent with RIs, whereas the latter, necessarily involving the C-system to host the wh-phrase, should be inconsistent with RIs. This selective compatibility of RIs has been observed in Child Dutch (Haegeman 1995a), Child French (Crisma 1992), Child German (Clahsen, Kursawe & Penke 1996), Child Swedish (Josefsson 2002), among others. As detailed in Table 4, a similar strong asymmetry between declaratives and wh-questions is observed in child JC. While there are 39.9 % of structures with target inconsistent missing aspect among declaratives, only 6.5% structures with missing aspect were produced with overt wh-interrogatives.¹¹ This sharp difference is expected if structures with missing aspect are to be assimilated to RIs, and under the truncation approach.

Can the truncation hypothesis explain the fact that the proportion of yes-no questions with missing aspect roughly corresponds to the proportion of declaratives with missing aspect, namely, 42.5%? On the face of it, these data present a problem. Although main yes-no questions in JC are solely marked by an intonational contour and do not seem to overtly engage the left periphery, their interpretation as questions (and the licensing of the intonational contour) need to be structurally anchored. One possibility that comes to mind is that children can merge a (null) yes-no question operator within the structure of the IP. Such structures would then resemble truncated or fragment yes-no questions in adult language. However, while adult sentences like “sleeping?” or “coming?” are interpretively restricted, (for example, they all require their null subject to be the addressee), children may have recourse to the formal strategy of marking fragments as questions in interpretively less restricted environments. That the marker of yes-no questions may be expressed within the IP gains additional plausibility in view of the fact that some adult languages morphologically mark yes-no questions in an IP-internal position. For instance, the –ti interrogative marker found in Valdôtain dialects in sentences like Tu pékè-ti? ‘Are you eating?’ (Fénis dialect; see Ermacora (in progress) for discussion) seems to have such characteristics.

These data, as mapped out in Figure 3, clearly indicate that there is a structural difference between the productions of RIs in contexts requiring the overt projection of the CP as opposed to contexts where the CP can be truncated.

Figure 3

% RIs with Declaratives and Interrogatives.

The following data provides examples of root infinitives in yes/no interrogatives (41–46), and wh-questions with overt subjects (47–52).

The 11 root infinitives found in wh-questions with overt wh-elements are provided in examples (53–63).

4.1 Growing trees and truncation

In recent work, Friedmann, Belletti and Rizzi (this volume) have worked out a “growing trees” approach to language development by which learners’ productions start with minimal structures in a bottom up fashion, and then higher zones get added on top of the structure as development proceeds, following the hierarchical organization uncovered by cartographic work. This approach is implemented in Friedmann, Belletti and Rizzi (this volume) to capture the developmental stages of the acquisition of different A’-constructions in Hebrew. The growing trees approach is distinct from the truncation idea. The fundamental difference has to do with optionality. In the growing trees approach, at a given stage a higher layer (for instance a certain zone of the CP-system) is systematically missing in the child’s representation, whereas under truncation, a higher layer is accessible to the child, it may be present in a representation, but it may optionally be truncated.

The two ideas are clearly distinct, but not incompatible. If we try to put them together, the possibility which suggests itself is a three-stage developmental sequence. In the first stage, the upper layer under consideration is always absent, the relevant subtree has not “grown” yet. In the second stage the layer becomes available to the child, but it can be truncated: we thus have the period of optionality which is typically observed in the course of the third year of life. In the third stage, adult-like, the layer becomes obligatory. In this way, the two ideas can coexist, and in fact, if properly integrated, can become a useful tool to look at developmental effects.

Throughout the third year of life, we have observed in our corpus study of child JC that aspectually marked forms alternate with bare verbal forms in contexts in which special aspectual interpretations are required. This alternation is expected under the truncation approach which we have adopted. The optional truncation analysis is also consistent with the findings in Friedmann, Belletti and Rizzi, which assumes an initial stage in which the complete IP structure is available to the child, whereas different left-peripheral zones must grow.

We will not undertake here a more detailed analysis of the possible relevance of a division of labor between growing trees and truncation. Let us simply observe, in closing, that the very first relevant cases of verbal forms in contexts requiring progressive or prospective interpretations are bare forms (Table 1), and a robust alternation between bare and marked forms starts only later. This may suggest the possibility that the initial recordings capture a phase in which the relevant aspectual zones have not grown yet. At later points, such zones become available to the child, and can be expressed in some structures and undergo truncation in other structures. We will not try to develop this hint here, and will leave for further work a systematic assessment of the integration of growing trees and truncation into a single coherent model of the development of the clausal structure.

In conclusion, the acquisition of JC manifests a well-identifiable RIs phase, in which bare verbal forms can occur without the aspectual markers which would be required by the adult grammar. Our conclusion thus concurs with Pratas and Hyams’ (2009) analysis of Capeverdean Creole in showing that specific varieties of the RIs construction can be found in the development of Creole languages. We have shown that certain basic properties characterizing children’s use of verbal forms with omitted aspect markers are naturally explainable through the truncation approach, applied here to the language specific variety of the RI construction.