2.1 Empirical motivation

Instances of i-language – like vocatives, tags, and response particles – occur early on and often precede the acquisition of complex p-language. A brief look at cross-linguistic data clearly demonstrates that the acquisition of i-language displays interesting acquisitional patterns that require explanation, and which suggest an underlying system that regulates these patterns. It is thus essential to include i-language in the study of L1A.

For example, Japanese sentence-peripheral particles that belong to i-language are acquired before particles that belong to p-language. Shirai et al. (2000) report that sentence-final particles which encode the speaker’s attitude toward the proposition are acquired earlier than particles encoding propositional content, like past tense or clause-typing. Shirai et al. (2000) provide a fictional dialogue that exemplifies the use conditions of the particles under investigation.

In a comparison of four toddlers, who are aged 1;03 to 1;09 at the beginning of their corpus data, Sihrai et al. then show that these clause-peripheral particles all appear early (around MLU 1.1), and crucially they appear before the onset of tense markers. Other particles, which do not reflect speaker attitudes, have a later onset.

Similarly, in Northern Mandarin, sentence-peripheral particles that encode the speaker’s attitude precede the acquisition of modal particles and particles that encode the illocutionary force of the sentence (Zhang et al. 2019). Moreover, Paul & Yan (2022) show that three Mandarin sentence-peripheral particles which are associated with IP and CP are acquired nearly simultaneously. Attitude markers begin to appear at the same time as force markers as well as particles that anchor an event to the utterance time.

Sentence-final particles have been the subject of investigation in Mandarin and Japanese for quite some time, precisely because they are so widespread and paradigmatic in these languages English sentence-final particles along with other instances of i-language (e.g., vocatives) appear to show similar acquisitional patterns as their Mandarin and Japanese counterparts, in that they are used in early childhood utterances. Examples (7) and (8) show that i-language co-occurs with p-language even at early stages of syntactic development.

Crucially, combining i-language with p-language is not trivial because its distribution is constrained and therefore reasonably assumed to be syntactically mediated. The examples in (7) and (8) establish that some sentence-peripheral UoLs are acquired before inflection or auxiliaries. This is a pattern that needs to be accounted for in theories of L1A.

2.2 Theoretical motivation

There are several theoretical arguments for why i-language should be considered in the study of L1A. One argument is that language in interaction is constrained by grammar (Ginzburg & Poesio 2016; Kempson et al. 2016; Wiltschko 2021). Here, we review some key pieces of evidence discussed in Wiltschko (2022) that imply that i-language displays all the hallmarks of grammar.

First, the form-meaning relation of vocatives is mediated by their syntactic distribution. Specifically, Zwicky (1974) establishes that there are two types of vocatives (calls and addresses), which differ both in terms of their interactional function and in terms of their distribution. Call vocatives serve to activate the addressee’s attention and can be accompanied by phrases like hey or excuse me. In contrast, address vocatives are employed to maintain the interlocutor’s attention and can be accompanied by phrases like listen or I’m afraid. Crucially, calls only occur turn-initially, while addresses can occur at either end of a sentence, as well as in some sentence-medial positions. Since these different interactional roles correlate with different distributional patterns, we can conclude that they are associated with different syntactic positions just as different grammatical roles are. In other words, just as subject and object roles are grammatically conditioned, so are the interactional roles of call vs. address (Wiltschko 2022).

Second, the interpretation of invariant tags is structure-dependent: sentence-final eh is a root phenomenon such that it will always be interpreted with scope over the whole clause, as in (9).

There is one exception to this generalization, however, which shows that scope is not merely a matter of linear order. If the clause is embedded within a bleached, interactional clause (you know, I guess), the scope of eh will not include this interactional clause but instead scope over the embedded propositional clause only, as shown in (10).³

Thus, i-language has to be sensitive to syntactic structure rather than to linear order.

Finally, i-language displays patterns of contrast and paradigmaticity: for example, invariant tags in Mandarin form paradigms defined by speaker- vs. addressee-orientation, on the one hand, and old vs. new information, on the other (see Section 3.1 for details). Arguably, paradigmatic contrasts derive from properties intrinsic to grammatical structure.

Another argument for the grammatical integration of i-language has to do with considerations of the architecture of grammar. Units of i-language are frequently part of utterances, in children and adults alike. Crucially, many of these units are prosodically integrated into the utterance as a whole. For example, in the presence of an invariant tag, the rising intonation typically associated with inquisitiveness is realized on the tag, not on the host clause (Heim & Wiltschko 2020). Thus, whatever system is responsible for assigning intonational contours must be sensitive to both p-language and i-language. If i-language were to be regulated by a system distinct from p-language (for example if it were purely pragmatic), one would have to assume that intonation is assigned after the utterance is assigned its pragmatic properties. But this requires the postulation of a system that unites syntactic and pragmatic aspects of an utterance – at least some of them. Thus, the more parsimonious assumption is that grammatical architecture includes i-language. And if it does, then i-language will have to be part of the empirical domain of L1A studies.

3.1 Evidence that i-language is regulated by the topmost syntactic structure

As mentioned above, there is an emerging consensus that classic sentence structure (i.e., p-language) is embedded into structure dedicated to regulating speech acts and interaction. In a nutshell, this consensus rests on new evidence for Ross’ (1970) ‘performative hypothesis’ according to which a proposition is introduced by a performative phrase (e.g., I tell you that…) which is deleted before spell-out. While the mechanics have changed, the spirit of the proposal remains the same: speech acts are syntactically represented. Support centres around the distribution of units of i-language, agreement in or with units of i-language, and sensitivity of units of i-language to clause-typing. These are all classic arguments for syntactic integration, as is evident in the literature on gapping and ellipsis (Lobeck 1995; Johnson 2001; and many others). We discuss each of these properties in turn.

Distributional support for a syntactic integration of (Indo-European) units of i-language comes from strict ordering restrictions of sentence-peripheral discourse particles, as for example reported in Haegeman (2014: 131) for West Flemish né and wè. The vocative Valère exposes the underlying order at the clausal periphery: independently of linearization, the particle wè must be directly adjacent to the sentence radical (in CP) while né can be separated from it by a vocative.

Similar arguments have been brought forth for non-European languages (Heim et al. 2016), which suggests that the sequence of addressee-oriented units of i-language has to be associated with the syntactic spine in a position higher than speaker-oriented units of i-language, contra the original order in Ross (1970). The data in (12) from Cantonese shows that these ordering restrictions are reminiscent of the data from West Flemish. The clausal periphery can host two discourse particles, which are strictly ordered, with the addressee-oriented one (ho2) mandatorily following the speaker-oriented one (me1).

The marking of agreement within i-language is evidenced in Southern dialects of German, for example. The sentence-peripheral particle ge(ll), geu (and its variants) serve to confirm the interlocutors’ beliefs, similarly to what we have seen in the Cantonese data above. Both variants can inflect for the second person plural or in formality (Heim 2019a; Wiltschko 2021). Crucially, agreement on this particle links to the addressee, rather than the subject of the clause. This is shown based on the Swiss German example in (13).⁴

Another case of agreement triggered by i-language is discussed in Miyagawa (2022), based in part on data and observations by Yamada (2019). Japanese, a language where typical phi-feature agreement is absent, presents evidence for honorific agreement. Specifically, -mas is used for the honorification of the addressee by the speaker. In the absence of -mas the utterance is classified as colloquial, as in (14a), while the presence of -mas renders the utterance formal, in (14b).

The sensitivity of -mas to the context of interaction suggests that its use is somehow dependent on the topmost structure, i.e., the syntax of i-language. However, on the surface, this does not seem to be the case as -mas occurs too low in the structure (i.e., below tense) to be associated with the treetop, as Miyagawa (2022) observes. Thus -mas appears to be linearized in a position different from where it is interpreted. This type of dissociation between linear position and interpretation has long been established in syntactic theory and can be analyzed as a form of (abstract) agreement or (covert) movement. Significantly, Miyagawa (2022) suggests that there is an overt reflex of this dependency between -mas and the topmost structure, namely in the form of sequential allomorphy. That is, when -mas is used, all higher heads must reflect the formality of -mas, including negation, tense, and certain higher interjections (daroo vs. desyoo). This is shown in (15).

According to Miyagawa (2022), this sequential allomorphy can be analysed as a form of agreement (Yamada 2019) and hence provides support for the syntactic integration of i-language.

A final argument comes from the sensitivity of i-language to clause-typing. In Wiltschko & Heim (2020), we show that Canadian English eh has a strong preference for being combined with declarative clauses, which in turn can be analyzed as a form of selectional restriction. The only exception to this generalization concerns instances where the particle seeks confirmation of the speech act, rather than the interlocutor’s belief. Here, the particles readily combine with interrogatives and exclamatives and mark a request for confirmation of the appropriateness of the speech act, just as in the Cantonese data above.⁵

For further evidence supporting the postulation of a dedicated interactional structure, we refer to Wiltschko (2021), where the interactional spine is introduced in detail. For the present purpose, the data above sufficiently support the assumption that CP (which we refer to as the domain of Linking, following Wiltschko 2014) is embedded in the interactional structure, i.e., projections that can host speaker- and addressee-oriented discourse particles. These projections have been labelled Grounding projections in Wiltschko & Heim 2016, to reflect their relevance to updating the Common Ground (Stalnaker 1978; 2002). Each of these grounding projections hosts an abstract argument which can be viewed as the grammatical manifestation of the epistemic state of its holder (i.e., the speaker or the addressee).

Because particles hosted in these projections also combine with a call-on-Addressee (in the sense of Beyssade & Marandin 2006), often in the shape of rising intonation, we further follow Wiltschko & Heim (2016) in assuming a third projection at the very top of the spine, namely Responding. The full structure we assume is schematized in Figure 1.

Figure 1: The interactional spine (Wiltschko 2021).

3.2 An analysis of adult huh

To show the value of the interactional spine for analyzing i-language, we now turn to the analysis of huh in adult language. In North American (adult) English, huh has two functions: it serves as an other-initiated repair strategy, as well as a sentence-final (confirmational) particle. We discuss each of these functions in turn.

When used in isolation, huh functions as an other-initiated repair strategy (Dingemanse et al. 2013). In this use, huh serves to request a response from the addressee. Requesting a response without specifying what the target of the response is leads to the interpretation that it is the previous move of the interlocutor that requires a response, as this is the most salient target. As such, it marks a problem in the current interaction that needs to be addressed (Wiltschko 2021). Arguably, the contribution of the particle huh in this context is to provide a host for the rising intonation, which otherwise, too, serves to request a response (Heim 2019b), and which is associated with RespP. Thus, huh as an other-initiated repair strategy serves as a bare RespP, as illustrated in (16).

The interpretation of huh when used as a sentence-final particle is more complex, and this complexity is reflected in the syntactic analysis developed in Wiltschko (2021). To see this, consider the examples in (17). The bare declarative in (17a) asserts that the addressee has a new dog. This would be appropriate if the addressee does not know that they have a new dog (e.g., if the speaker gives the addressee a dog as a present). In contrast, when the particle huh is added, as in (17b), the speaker wishes to confirm their suspicion that the addressee has a new dog (e.g., if the speaker sees the addressee walking a dog they have not seen before). Crucially, the addition of huh indicates that the speaker does not know if the proposition (p) encoded in the host clause is true. In other words, the speaker is not committed to the truth of p.

According to Wiltschko (2021), this interpretation derives from the syntactic configuration schematized in (18). huh occupies Ground_Adr and combines with the rise (↑) in RespP. Crucially, Ground_Spkr remains empty reflecting the speaker’s lack of commitment to the truth of p.⁶

Assuming that huh associates with Ground_Adr (but not with Ground_Spkr) captures the contrast between sentence-final huh and sentence-final eh. Unlike huh, eh is compatible with an interpretation where the speaker is committed to the truth of p, as shown in (19) and (20).

eh is compatible with an interpretation where the speaker confirms that the addressee knows p, as well as with an interpretation where the speaker wishes to confirm that p is true. Crucially, huh is only compatible with the former interpretation. This contrast between eh and huh follows on the assumption that only eh can be associated with Ground_Adr or Ground_Spkr, as in (21a). If it associates with Ground_Spkr it marks the speaker’s commitment to the truth of p. The restriction on the interpretation of huh follows if huh can only associate with Ground_Adr but not with Ground_Spkr, as schematized in (21b).

Note for completeness that if huh (and eh) were associated with RespP only, the difference between them would be mysterious. Moreover, if huh in its use as a sentence-final tag would only be inserted to host the sentence-final rise associated with RespP, as is the case for free-standing huh, we would expect there to be no difference between interrogatives realized with and without huh, contrary to fact. As shown in (22a), a polar question (with rising intonation) is used when the speaker requests an answer (i.e., the addressee is expected to provide either a positive or negative answer). In contrast, a polar question with sentence-final huh, as in (22b), can be used to express one’s suspicion that the addressee has the same question. In this case, the speaker does not necessarily expect an answer but rather wants the addressee to agree that the question is the right one to ask (Heim & Wiltschko 2022).

In what follows, we adopt this analysis of adult huh when interpreting the patterns of use of huh during its acquisition. We further assume that the acquisition of the interactional spine is an essential task during L1A, which has so far received no attention.

7.1 Foundational assumptions

According to the bridge model (Hinzen & Wiltschko 2023), which underlies the Inward Growing Spine Hypothesis, language bridges two pre-linguistic, domain-general capacities: categorizing and social interaction. These capacities precede linguistic abilities as evidenced by the fact that they are present in animals as well as in neonates. Human linguistic abilities emerge when the spine bridges these two capacities. We hypothesize that it is the linking layer of the spine that first surfaces to generate this bridge, and that it is first realized in the form of the pointing gesture. Subsequently, the other layers of the spine unfold, as outlined in section 4. Adopting the bridge model and integrating i-language as part of the research agenda for L1A implicates some foundational assumptions, which we now discuss.

First, considering i-language to be part of grammatical knowledge requires us to reconsider the classic generative distinction between competence and performance. This is because these interactional aspects of language are part of the capacity for language, and they are acquired early. This conceptualization of language is in the tradition of proposals that postulate a communicative competence (Campbell & Wales 1970; Hymes 1972; Keenan 1974; Mittwoch 1976; Chomsky & Lasnik 1978; Chomsky 1980; Ginzburg 2012; Wiltschko 2021).¹² It is likely the classic generative division of language competence and performance that led to the neglect of i-language in generative studies of L1A. There, the study of both adult and child language has typically focussed on the expression of propositional thought. Once we integrate i-language into our models of grammatical knowledge, language acquisition must also be viewed through the lens of communication. And indeed, from its earliest stages, language links thought and communication, as reflected in our case study above. Huh is used interactionally even before it is fully integrated with p-language, and its use-conditions expand in parallel with functional growth in representations of p-language.

This further implies that the acquisition of i-language is to be studied on par with p-language and is therefore part of the syntactic acquisition process. This requires a conceptualization of syntax that differs from the classic one. Specifically, syntax has traditionally been viewed as the linguistic domain that is responsible for the combination of words to create sentences. In turn, this implies that syntactic acquisition starts with multi-word utterances and that only words that belong to p-language are relevant. Accordingly, most aspects of i-language might appear irrelevant as they do not belong to the classic domain of the sentence. In contrast, if we include i-language in this domain, then multi-modal forms of language (e.g., intonation, gestures, and eye gaze) will have to be considered as well, since these belong to i-language. And if so, syntactic acquisition starts much earlier than the classic multi-word stage (cf. Radford 1996; Friedmann et al. 2021). On this view, syntactic combination is not defined by the combination of words with each other, but rather by the systematic combination of any linguistically significant form with the spine. Crucially, systematicity here must be defined based on the child’s output rather than on specific conventions of the language community the child is born into. From the start, verbal behavior is entrenched in communicative contexts. Just like there is evidence that a child’s linguistic competence of p-language cannot be deduced from production data alone, we submit that the competence for i-language, too, is present before it is produced. Consequently, we need to look at preverbal behavior for evidence of the structural development of the interactional spine. Early on, there is evidence for synchronization between infants and caregivers in preverbal communicative behavior. Of particular interest are vocalizations from newborns that are directed toward their caregivers. These are often contingent on the caregiver’s turn and occur with little delay (Dominguez et al. 2016). At 5 months, turn-taking assumes a language-specific format with very few gaps and overlaps between turns (Hilbrink et al. 2015). Only when joint attention and pointing begin to surface, does the synchronization of turns begin to include noticeable pauses, which likely reflects the planning of an utterance by the infant. According to the Inward Growing Spine Hypothesis, Linking (arguably first instantiated by the pointing gesture) allows the infant to communicate its observation about the world. Categorizing the world is now linked to interacting with other individuals. Thus, the first concrete evidence for the presence of Linking consists of combinations of early (pre-)verbal utterances and gestures, which anticipate the earliest word-gesture combinations. Carter (1978) shows that speech acts initially consist of sound-gesture combinations, such as headshakes and nasalized glottal stop sequences for rejections and [d]- or [l]-initial sounds and pointing for drawing attention to an object.

As a case in point for early systematic and complex communicative behavior, consider an example from the Forrester corpus (Forrester 2002). In (31), the father readily engages with Ella’s combination of hand gestures, rising intonation, and monosyllables by offering an adult variant of the utterance in his response. This response interprets the child’s utterance as a question (with whaa strongly resembling adult what) and the pointing to an object as demonstrative (that in the father’s words).

We suggest that the child’s raised arm and rising intonation serve as a request for a response and thus as being associated with RespP. We further suggest that yehh↑ links the response layer with the categorizing layer by communicating Ella’s interest in an object (the target of the pointing gesture) to her interactant.

Cases like this indicate that there is no discrete step from which non-syntactic single-word utterances turn into syntactically integrated multi-word utterances, just as there is no discrete step from where variegated (non-repetitive) babbling turns into a one-word stage (Messick 1984). Syntax begins where turn-taking (in Response) and first syllables (in Categorization) are linked in meaningful, dialogical communication well before the arrival of multiword utterances.

In sum, integrating i-language into the study of L1A requires the study of language well before the use of words, hence the research agenda has to include the study of the child’s preverbal competence. The exploration of i-language implicates that we do not just study the content of children’s utterances but also the way this content is communicated (e.g., joint attention seeking, pointing, drawing attention, etc.). Crucially, pre-verbal communicative interactions are primarily facilitated by multi-modal cues, such as in example (31). Gestures and gaze serve as substitutes for referring to words yet unknown or not yet pronounceable. Because these interactions are combining turn-taking and categorizing, we assume a syntactic representation.

7.2 A Novel Research Agenda

Given that i-language has largely been ignored in previous studies of L1A, there are currently no models available that look at the way early i-language and p-language relate to each other. It is essential to establish the sequencing of acquisition of p-language relative to i-language. The Inward Growing Spine Hypothesis is the only framework that makes clear predictions about this trajectory and initiates the exploration of i-language as a systematic research agenda. Specifically, it invites the following new research questions:

While the study of i-language in the context of L1A is new, the study of the acquisition of p-language has a long tradition. However, the Inward Growing Spine Hypothesis leads to a novel set of predictions that run counter to models that assume an upward trajectory of maturation. Specifically, it predicts the following trajectory of acquisition for p-language (parallel to i-language, see Figures 3, 4, 5, 6).

By hypothesis, the Inward Growing Spine Hypothesis applies to both the clausal and the nominal spine. This follows Wiltschko’s (2014) Universal Spine Hypothesis, according to which the spine is intrinsically category-neutral (i.e., not specified for being verbal or nominal). In turn, Wiltschko’s claim is based on the well-documented parallelism between the clausal and the nominal spine (Chomsky 1970, Abney 1987, a.o.). If indeed the spine is not intrinsically nominal or verbal, it follows that the Inward Growing Spine Hypothesis should equally apply to both instantiations. We submit that this provides a fruitful avenue of analysis for the acquisition of determiners, for example. We consider it likely that the acquisition of articles mirrors the development of i-language, as exemplified by our case study of huh. Indeed, early articles lack a clear distinction between definite and indefinite uses. Schaeffer & Matthewson (2005) show that the in/definite distinction in St’at’imcets (Lillooet Salish) resembles the usage of early articles in English. In contrast to adult English, both St’at’imcets and child English allow a definite article in contexts where the addressee is unfamiliar with the referent. In adult English, these contexts require an indefinite article. Schaeffer & Matthewson (2005) propose for English children that a concept of non-shared assumptions is responsible for this overextension. This pattern is amenable to an analysis according to which the Grounding layers in the nominal spine are not yet articulated. Interestingly, Guasti et al. (2008) make a similar proposal for Dutch children omitting articles. Spoken adult Dutch allows bare singular nouns if speaker and addressee have shared knowledge of the referent. Thus, both Schaeffer & Matthewson (2005) and Guasti et al. (2008) independently draw on an ability to distinguish interlocutor beliefs to explain the acquisition of articles. This is consistent with the hypothesis that the child starts with a generalized Grounding layer where speaker and addressee ground cannot be distinguished.

7.3 The Inward Growing Spine Hypothesis and p-language

Finally, we briefly discuss how the Inward Growing Spine Hypothesis addresses classic problems in research on the acquisition of the syntax of p-language. Specifically, we briefly discuss the acquisition of Anchoring, whose trajectory is difficult to analyse on the assumption that trees grow upwards but falls into place assuming the Inward Growing Spine Hypothesis.

In adult English, the Anchoring domain corresponds to what is known as IP, the syntactic layer of structure that hosts tense and agreement morphology, as well as grammatical subjects. IP is located above AspP, which hosts aspectual morphology and grammatical objects, and below CP, which hosts complementizers, wh-pronouns, and auxiliaries in questions and negation. According to the upward growing tree hypothesis, AspP and IP should be acquired before CP. However, the empirical facts do not bear out this prediction.

To see this, consider again the data in (23) and (24) repeated from above. At the age of 2;07, Adam uses wh-questions (marked by an utterance initial wh-word) but the rest of the clause differs significantly from its adult counterpart. What is missing is precisely the UoLs that occupy IP: subjects, auxiliaries, and inflection for tense and agreement.

Note that the absence of IP-internal material is independent of the presence of huh. That is, Adam also produces wh-questions without subjects, auxiliaries, or inflection when he does not use the sentence-final huh, as shown in (37)–(39). The presence of the wh-pronoun is key for our claim here since the growing spine hypothesis also assumes IP to come after an initial vP/VP stage where subjects stay low. Postverbal subjects of various kinds are in fact well attested in early, tenseless multiword utterances (see e.g., Tsimpli 1992 citing examples from Bowerman 1990: “ride Christy”, “fall mommy” or “write Sissy”). Yet any material above IP before the arrival of subject movement is not expected in a strictly upward growing tree.

This pattern follows straightforwardly from the assumption that CP (i.e., Linking) is present before IP (Anchoring). Specifically, it appears that VP-internal material (the verb with its arguments) can be realized in combination with an utterance-initial wh-word (by our hypothesis, occupying Spec, CP). However, auxiliaries and inflection (associated with IP in adult language) are conspicuously absent.¹³

To account for this pattern, the upward growing tree hypothesis has to be enriched with auxiliary assumptions. Specifically, it is often assumed that children truncate structure (Rizzi 1993) which is (in principle) available to them (e.g., the CP). Significantly, the presence of some finite wh-questions in English is challenging for this approach (see Guasti 2017 for an overview). The data in (37)–(39) would then require the wh-pronoun to associate with the specifier of whatever root node is available after truncation. This option is, however, not considered by Friedmann et al. (2021) because the use of wh-questions will always require an adult-like structure in the lower IP domain.

The truncation account is typically also invoked for another property that characterizes English child language, namely the frequent absence of subjects (Hyams 1986), as in (37) above. That is, subjects may be absent in the context of wh-questions, which differs from adult English, where subject drop is ungrammatical (apart from fragments, diary style, and imperatives). However, subject drop is consistent with the assumption that the spine grows inwards. Accordingly, the CP will first embed VP thus providing no position for grammatical subjects. Note that the presence of subjects in early child language does not contradict the claim that Anchoring is acquired late. That is, assuming the VP-internal subject hypothesis it follows that subjects may remain VP-internally and thus may occur early on. Thus, the optional presence of subjects (as seen in (37) and (38)) can be analyzed as the realization of a VP-internal subject. Arguably, the overt realization of subjects becomes mandatory only when the Anchoring layer unfolds completely.¹⁴ This hypothesis is consistent with the fact that expletive subjects are acquired significantly later than lexical subjects (Yang 2002). It follows if early subjects are VP-internal and hence necessarily thematic. Consistent with this idea is the fact that agreement surfaces around the same time as expletive subjects, namely at the age of 3;0 (Lukyanenko & Miller 2023).

Finally, according to the Inward Growing Spine Hypothesis, the Anchoring category is predicted to first emerge as a category that subsumes both IP and AspP of adult English (see Vainikka & Young-Sholten 1996 for a similar proposal in second language acquisition). This is consistent with the well-known observation that in the course of L1A, tense and aspect inflections are often applied inconsistently (see Guasti 2017 citing Valian 1991; 1992; Wexler 1998). Moreover, according to Valian (2006), 2.5-year-olds perform better with tense comprehension than with aspect comprehension against the prediction of the Aspect First Hypothesis (Antinucci & Miller 1976), which assumes the opposite order. This is consistent with the Inward Growing Spine Hypothesis but is not expected under the classic upward growing tree trajectory according to which aspect should be acquired before tense.

7.4 Possible alternatives to the Inward Growing Spine Hypothesis

Assuming that i-language is part of grammar, and thus part of the spine (Section 3), the fact that i-language is acquired early casts doubt on the assumption that the spine matures in an upward trajectory. In this paper, we have proposed an alternative, according to which the spine matures in an inward fashion (Section 5).

An anonymous reviewer suggests an alternative, which would save the traditional view that the spine matures in an upward fashion. According to this reviewer, “the early emergence of interactional phenomena could simply be an artefact of such phenomena being a system separate to the syntax (or at least not integrated into the syntactic spine). If interactional phenomena aren’t part of the syntactic spine beyond some interface condition, their early emergence could be quite independent of the growth of the syntactic spine, and there would be no need to propose the inward growing spine hypothesis.” In other words, the reviewer suggests that the acquisition of i-language fundamentally differs from the acquisition of p-language and hence need not be taken into consideration when modelling the acquisition of (p-)language, where i-language is typically ignored, as illustrated in Figure 8 where '*' signifies recursive input.

Figure 8: Classic assumptions regarding language acquisition.

On this view, what is modelled is the acquisition of p-language alone: only units of p-language (Uop-L) are considered, and grammatical development is viewed as the development of the propositional spine. This is in fact the implicit assumption that underlies all work on the acquisition of syntax and has led to the status quo where the question as to how i-language is acquired is not addressed. It is, of course, an empirical fact that the input for the child contains units of interactional language (Uoi-L). Hence, a comprehensive theory of language acquisition will have to take into consideration how these are acquired alongside propositional language. What the reviewer suggests is something along the lines schematized in Figure 9, where the acquisition of i-language is qualitatively different from the acquisition of p-language.

Figure 9: Acquisition of i-language differs qualitatively from acquisition of p-language.

This model, however, faces some challenges which our proposal does not. Specifically, the assumption that i-language is different exacerbates the classic word discovery problem (Brown et al. 1969), i.e., the problem as to how children are able to identify words even though there are no reliable signals in speech. Assuming that i-language is acquired in ways that are fundamentally different from the way units of p-language are acquired only amplifies this problem. This is because the child does not only have to identify words, she will also have to determine whether a given unit “counts” as a word belonging to p-language and hence will need to be integrated into the spine or whether it does not count as such and will have to be ignored.¹⁵ What would make this task particularly difficult is the fact that many units of i-language are identical to units of p-language, with their difference in use correlating with a difference in syntactic distribution. For example, in English right can be used as a propositional predicate (We might be right.) or as an interactional confirmational (I know, right?). In fact, it is well-documented that propositional language can develop into interactional language (Haegeman & Hill 2013), akin to processes of grammaticalization (Heine 1997; Shukla et al. 2022).

Moreover, recall that we assume that grammatical knowledge in adults includes i-language (see Sections 3 and 7.1). If so, the proposal sketched in Figure 9, will have to be augmented with a mechanism which – at some point during language acquisition – will integrate i-language and p-language to arrive at the adult system, as in Figure 10, where grammatical development is first restricted to p-language before i-language is integrated.

Figure 10: Integrating i-language during the course of acquisition.

Crucially, the hypothesized separation of the acquisition of i- and p-language could only be maintained until we see the development of i-language that unequivocally requires syntactic integration. Variant tag questions like It’s raining, isn’t it? are a case in point. They require agreement and movement processes as well as an awareness of speaker knowledge for licensing their use. Traditionally, these tags have been assumed to be acquired late due to their syntactic complexity (Bellugi 1965). Yet, novel data from British English (Woods & Roeper 2021; Heim 2023) clearly show that children produce adult-like tag questions as early as 2;02 – an age where even conservative estimates assume the presence of syntactic representations (Diessel 1984). If syntactically integrated i-language is available so early, the requirement to separate the acquisition path of i- and p-language would have to end here. Thus, ignoring early i-language in modelling language acquisition comes at a price: it requires us to enrich the model of the language faculty with a component that is responsible for the acquisition of i-language and one that will then allow for its integration. We thus submit that the assumption that i-language is acquired with p-language at the outset, as in Figure 11, is more parsimonious than relegating it to a different system.

Figure 11: Acquisition of i-language is part of the grammatical development.

But if i-language is part of the syntactic acquisition path then the classic assumption that the syntactic spine matures in an upward fashion cannot be upheld. Hence, the goal of this paper was to introduce a viable alternative.