One interesting observation here concerning such (single argument) VS structures has to do with Case assignment. Most verbs found within VS utterances seem to have Accusative Subjects. This observation becomes rather significant when coupled with other data showing correct Nominative Case usage at the same stages of development (see (9) and (11) below). In other words, the data seem to demonstrate that when VS order is used, Subjects get Accusative Case and never Nominative Case. (NB. The notion that such orders are the result of Dislocations has been largely discredited by negation initial construction: e.g., Neg-V-S/Neg-O-V (see §1.3.1)). In fact, not one VS construction yielded a Nominative Subject. (The fact that Nominative case appears at roughly the same stage may indicate some sort of optional Nominative/Infl stage). However, regarding Case assignment to DP, there does seem to be strong evidence that Genitive Case is unaffected by variable word order. In fact, the Determiner system emerges fairly early on in the data and is found in a variety of environments (not to mention within Language Mixing). The following examples in (9) show correct Case assignment and Word Order patterns for the same developmental periods as was shown in examples (7) and (8) above (ex. a, b are arguably formulaic):

(Objects)           Examples of OV/VO structures are also found throughout the range of the data. (Again, recalling our discussion above that word order errors seem to manifest up until the very last file--dependent only on the SAS vs. DAS distinction. Token examples are presented here in (10):

(10)

I think what is crucial to note here is that no overt (INFL)ectional verbal morphology (excluding the copula Is (e.g., / Iz_ka: /> Is a car) found in file 5) has yet to surface: viz., all of the VO/OV data suggest that the verb is unspecified for Person. Nominative Case however, also an indicator of INFL, does seem to be established in early SV(X) structures, but an asymmetry is found between the use of Nominative Case in SV and VS structures. The analysis of OVs and token counts (cf. 10) are given below in (10' & 11):

(WH-Questions)            Another interesting observation is found concerning Wh-Questions. As alluded to earlier with respect to Kayne, no instances of target word order violations were found among early Wh-initial (+/-Agr/-Tns) constructions. Such results might also be used as argumentation in favor of Kayne's linearization analysis. The arguments might proceed as follows:

(12)

(a). In the event that a (base-generated) Wh-element raises, occupying a fronted position, the environment for free word order collapses--recalling earlier that the structure in (5) is restricted to involving a thematic sisterhood relation prohibiting any A-bar or Adjunction positions. Thus, either a full-blown VP or CP>VP structure would project.
(b). Since there would, in principle, be no room for a second argument (an element other than the Subject and Verb) to position itself, a raised Wh-element would then be forced to occupy either an external Spec-CP or possible Adjunction position: both being A-bar positions--in accordance with the Q-criterion--hence, violating stipulations placed on (5) (further assuming that any Adjunction to VP would be driven by some sort of movement operation creating the chain _h...._t ). Again, either a full-blown VP or CP>VP would project.

(One possible analysis for the following Wh-constructions is to suggest that they are Truncated CP>VP structures as proposed by Radford 1990, 1994, and Roeper and Rohrbacher 1994.)

(* C'=Adjunction p² as in Kayne's structure in (4). Simplified by not showing the trace of the Wh-Operator).

An important implication here is that no variable ordering is allowed within the VP of a CP>VP structure for reasons discussed in (12)--the Data bear this out. [22] Moreover, a crucial link becomes available now between the Functional stage (made apparent via O-movement) and stable word order (within the VP). However noting that INFL (Agr/Tns) errors still could appear within such structures--as is evident from (13a'')--since IP is altogether missing from the structure (as indicated by the absence of Nom case: him doing?).

(Early DP-Projections)            A further observation has to do with the fact that word order errors seem also to correspond with the absence of the specification associated with the Head of DP: (possessive 'S). It is observed in the data that the first attempts to project a Possessive construction (DP) (within a non-head DAS) fail to realize the target order. Consider the following examples in (14), where the Head Possessive 'S is unspecified:

(14)

The variable word orders cited in (14) could be schematized in the following ways:

(14')

In (14) above, Genitive Case assignment is assumed to involve a Spec-Head [+Agr(eement)] relation within DP (Radford: ms1997). The universal potential orderings of Spec-Comp/Comp-Spec however still manifest depending on which of the two universal XP schemata is being utilized.) Again, the crucial claim here relates to the above observation that variable word orders seem to appear within SAS or non-Headed DAS projections: it may be that all categories/(phrases) (Noun (NP), Det (DP), and Verb (VP) alike) are effected in the same way by the Comp parameter--in this sense, even though we are dealing with an apparent (Spec/Comp) DAS construct, the crucial point is that there is no overt Head projection in which to trigger the Comp parameter (cf. Kayne; ex. (6)). It is interesting to note that once the Head of DP overtly projects, word order becomes stable (there are no instances of mixed orders found among specified DP structures). Consider the counterpart structures in (15):

(15)

(15')

The structure in (15') draws attention to a possible link between the Functional Parameterization Hypothesis (FPH) via INFL, and the advent of stable word order (at the DAS phase). Namely, as briefly mentioned above, there seems to be some correlation between the onset of unstable word order at the SAS phase and the onset of fixed order at the (FPH) DAS phase. Namely, it is the use of English Possessive INFL(ection) 'S which signals that the child has now developed a specifically ENGLISH type of possessive structure (via INFL) (cf. 15), whereas earlier the child had a 'language-neutral' type of possessive structure (via Non-INFL) (cf. 14).

What the above data suggest is that it is the acquisition of the (language-specific) Spec-Head Agreement morphology that fixes word order. In other words, a Directionality/Linearisation Principle might say something along the following lines:

Linearisation Principle for Spec-Initial Languages: A Specifier is positioned to the left of a Head it agrees with. Non-Agreeing Specifiers (i.e., Adjuncts) may freely insert either before or after a given Head.

(Bare Infinitives)             Finally, some interesting questions can be raised regarding a very small set of observed "contradictory" patterns within Bare Infinitive clauses. These utterances contradict both proposals I and II in the sense that they involve the overall Functional IP stage (contradicting the former proposal) and within (DAS) constructions (contradicting the latter). Consider the distribution of the following VO/OV word order examples found among (matrix clause) Nominative-Subjects with Bare Infinitives:

(16)

An interesting observation about the distribution of such word orders is that strict correlations seem to hold between: (i) Bare Infinitives and variable word orderings, and (ii) language-specific To-Infinitives and fixed orderings. In fact, I have no instances of word order errors found within To-Infinitives: namely, sentences such as e.g., *I want to [mama see (=OV)], or *I want to [come mama], are not found in the data. Such apparent overall IP constructions that permit variable word orderings (within the V-bar) might be accounted for in the following (highly speculative) manner.
            Suppose that when language specific T(ense) and/or Agr(eement) remain non/under-specified, the properties relating to how they restrict Directionality become inert (cf. Tsimpli).
Suppose that such properties regulate directionalities only within their contained maximal (XP) projections.
            Suppose that the Infinitive particle "to" carries some sort of (language-specific) anaphoric +Tense (binding) feature--similar to how to is anaphoric in e.g., He wants to go where the particle (to) is bound and controlled by the Verb (wants) in the main clause. This might predict then that when (non-language-specific) Bare Infinitives appear (i.e., without the infinitive particle to), even within Nominative [+Agr] contexts, their Maximal projection VP contains no anaphoric-bound +Tense properties from INFL for determining Directionality. In such a case, this amounts to saying that there exists a split Functional/Lexical tree--with AGR, T projections involving the top of the tree only (cf. 17a). Under such a model, we might envisage utterances as (16) with erroneous OV orders as having the representation below:

What this binding between (a,b) attempts to illustrate is that the structures of the (lower) VP seem to share some sort of anaphoric binding relation with the Tense property of IP: e.g., (iff a is [+T], then b is [+T] (situated in Spec-VP)) and the correct directionality of V/N within V' is set (cf 17b)). It remains unclear exactly what type of mechanism we are talking about here regarding how the word order eventually becomes fixed. I can only add at this juncture that the spirit behind this proposal coincides with the notion discussed above--namely, that non-language-specific Bare Infinitives represent an earlier 'single system' (of complementation), and that the eventual use of the specific English morpheme TO signals the development of a separate English structure yielding English word order.

3.4 Final Remarks and Residual Problems

(An Additional Grammatical Stage?)                  In light of the discussion in the sections above, a slight refinement might now be proposed for how Developmental Linguists describe and assess the stages of child language acquisition. Before the emergence of the VP-stage, commonly referred to as the Small Clause stage (cf. Radford 1988b), an SAS-phase may exist permitting a maximum of two words (e.g., Proto-Argument/Predicate) with an unstable word ordering. The term SAS-phase here doesn't necessarily mean that three or more words can't project, but rather, that the majority of utterances found at this phase seem to comprise of predicate+argument strings only. Furthermore, there is no reason why such a defined SAS-phase could not, in fact, exist alongside a DAS-phase--not to mention the possibility that the phase may be skipped by the child altogether. (In the former case, the child immediately enters the DAS-stage.) The crucial difference then with our proposed SAS-phase, as opposed to a VP-stage, would be the added stipulation that all positions are exclusively A-positions which are in turn directly Q-marked by the Verb-Head within V-bar. Such a stipulation renders movement operations invalid (in adherence to the Theta-criterion--which states that a given argument can receive only one theta-role). The full range of XP expressions (viz., the internal argument V' position), though biologically predetermined to come on-line at the VP-stage, are somewhat suspended at the SAS-phase until specific properties of lexical items having to do with principles of X-bar syntax mature.. [23](In this broader sense, we are talking about the Specifier-Complement asymmetric opposition at the (DAS) phase which triggers the Complement Directionality Parameter as mentioned earlier in (§1.2)). The above stated differences between SASs and DASs amounts to leaving open the question whether or not movement operations are allowed at the VP-stage.

(Residual Problems under Minimalist Assumptions)                 In recapping, one thing the two proposed models seem to have in common (cf. (i) Tsimpli, and (ii) a 'weakened' Kayne) is the fact that they both rely on traditional notions of X-bar syntax: notions that have generally been dispensed with within Bare Phrase Structure (Chomsky 1995: Ch.4). The following cited residual problems take into account more recent interpretations of Bare Phrase Structure regarding Word Order.

(Merger)                  Before moving on to the residual problems, let's first briefly examine how the operation Merge itself might be broken down and analyzed into more fundamental components. Recall that the structure in (5) amounted to a Proto-Argument+Predicate construction. In order to derive the reduced tree in (5), aspects of Merge must also be capable of yielding a reduced tree. It is in this light that we conclude Merge to be formed by two rather independent operations: [24]

(18)
(i) Primary Merger--Merge Head with Comp (zero level category takes complement)

(ii) Secondary Merger--Merge Spec with Non-Maximal Projection (X or X')--
[Spec + X/X'] => XP

The notion behind (5) is presented above in the form of the secondary merger operation--viz., merge Head with Spec. However, the redefinition of Spec in (5) as a Proto-Argument, and not a true Specifier, is crucial to maintaining the reduced tree in (5). In other words, the child begins her linguistic career (SAS phase) by engaging the second merger operation first--as it is the only operation available--whereby the "Spec" becomes reorganized as the Proto-Argument slot into which Subjects and Objects alike may enter:

The above observations regarding Merge can be summarized in the following manner:
(1) The child begins projecting simple phrasal SASs by first merging the non-maximal projection of a Head with the external argument Specifier (noting that this Specifier encompasses both Subject and Objects). (NB. The need for the child to identify early-on the external argument hierarchy might be closely linked to inherent and innate principles of X-bar syntax. [25]That is, XP [=external Spec+Head] may be hard-wired-in from the beginning and not prone to any maturational factors or parameterizations, etc.)

(2) Once the child begins to project Double Argument Strings (DAS), as witnessed by her usage of both Spec and Comp positions, the XP schema is seen as being fully-fledged--thus deriving the potential VP-Shell (cf. Larson) where both Specifiers and Complements are highly active (for reasons having to do with movement operations, etc.) VP-shells, in this sense, are naturally derived from DAS constructions. Target Word Order is set via a coupling of (i) Kayne's Head-Medial Principle and (ii) the eventually setting of the Comp Parameter that triggers correct word order via an agreeing Specifier relation with the Head. Once the VP-shell projects, the functional non-thematic stage has commenced, allowing movement operations and A-bar positions to flourish.

(VP-Shell)               The Larsonian VP-shell, I believe, contains two essential ingredients for supporting the proposal at hand. They are as follows:
First, the original definition of Subjects and Objects as symmetric and mnemonic place holders (i.e., Agr-S/O respectively) stems directly from Pollock's (1989) theory which takes the Agreement relation--be it Subject or Objects--to be of a Spec-Head nature. This subsequently allows the Specifier position to be non-prejudicial in permitting an Object to project from out of a Spec position. This holds important consequences for how we, in turn, account for the structure in (5) restated above in terms of Merger Theory. Specifically, since the Spec position may now be seen, theoretically, to host the Object, we can now dispense (for the time being) with the Comp position since it needn't project at all for the early SAS phase. Moreover, the traditional idea claiming Head-Comp to be the "most local-relation" (i.e., thematic) needn't be jeopardized, since all arguments at the SAS phase are thematic, regardless of their relationships: e.g., the external Spec being Q-marked by its sister X'.

Second, the Larsonian VP-Shell further supports this notion by addressing the problems associated with the Theta-markings of multiple internal arguments, etc. Consider the underlying structure of a 3-place predicate sentence (e.g., John put the book on the shelf) where the Object the book is placed in Spec VP:

(20)

(Proposal I))            One notable problem for Tsimpli's model has to do with the assumption of a fully-fledged VP structure at the two-word stage. For instance, consider the child's typical Single Argument String utterance in (I) where only a Specifier and Verb projects:

Regarding current notions of Merger, the idea that a 'traditional' V-bar configuration should inherently take the initial role of merging two syntactic objects [a₂, b] forming a new combined syntactic object [a₁] is, I think, an intuitive assumption (cf. Chomsky 1995:226, 245). (I fully recognize that some traditional arguments for XP are no longer maintained within Bare Phrase Structure; however, Chomsky does continue to recognize the Head-Complement relation as the most "fundamental" and "local" relation.) However, in (I) above, it is the Specifier and Head which would seem to Merge leaving this fundamental sister Complement position of the Head empty. (This would be counterintuitive, suggesting that children (at the SAS phase), in fact, do recognize that the full XP hierarchy projects). Under Kayne's revised model presented in proposal II, the (SAS) utterance "Baby cry" would be derived by the Merging of [a b] forming [a'] (order irrelevant) via the only relation possible (local or otherwise) between a Proto-Argument (Spec) and Predicate (Head):

(II)

In-note I acknowledge that Kayne's model is inconsistent with Chomsky 1995-- where merger theory excludes non- branching single-bar categories. In light of Chomsky's discussion of Baby cry Kayne (p.338), such a reduced two segment structure might be looked on as having an internal Adjunct structure (i)

The reduced structure in (II), I believe, captures both the essence of what is innately given for the child--i.e., her innate ability to realize that X-bar syntax must contain an external branch for the proto-argument--while, at the same time, offering us an account of how the child strings such SASs together.

Secondly, a pre-minimalist claim that movement operations might be inoperative at the VP stage, being restricted to Functional categories, has recently been challenged by the notion that lexical Verbs may raise into the light-verb position within double VP-shell structures (Larsonian Shell)--assigning respective q-roles to object, indirect object along the way. Such movement operations would undoubtedly require a trace: in this respect. In light of this, it now becomes rather questionable why the VP-stage should be incapable of (i) movement operations (per se) and/or (ii) the setting of correct word order.
Recall Tsimpli's account that argued what initially impeded the setting of word order at the VP-stage was the total lack of non-thematic/Functional categories. Well, Atkinson notes (1992:295 fn1) that the relative ordering of the Head Verb and its Complement ought not be parameterized by the FPH whatever the case--since Theta-role assignment is a property of lexical heads and Objective Case is typically assigned by Verbs, a lexical category. Atkinson's argument hints that a fixed (VP-stage) word order does not seem too unreasonable--since word order may not be dependent on functional categories, but rather on a relational property of Spec>Head>Comp. The obvious problem however with this is that (based on acquisition literature) we indeed find DAS word order errors at the proper VP-stage (albeit rarely). It is not clear what the reasons would be under Kayne's original account. I suppose however, that under Kayne's 'weakened' version (modified here), which takes a two-pronged Directionality model--coupling the Head-Medial Principle with that of the Complement Parameter--a feasible reason having to do with the nature of parameterization itself might be given. Namely, once the two arguments project, a certain amount of time is still required to correctly set the [Complement Directionality] parameter. Hence, a variety of SVO, OVS might briefly appear before the actual setting of correct word order (see note 6). Moreover, recapping the notion put forward regarding the Larsonian Shell, a rather different story could likewise be run suggesting that all word order variations--other than SVO, OVS--found at the DAS VP-stage are indeed actually derivations involving movement.

(Proposal II)           The most outstanding problem for Kayne's approach to word order lies in his dependence on category-terminal distinctions. In Chomsky's Bare Phrase Structure, there is no such distinction. Thus a typical counterpart of (3) above would be the following:

(III)

The problem in (III) above is that (m, p) break asymmetrical c-command (ACC) since there is no hierarchical maximal projecting terminal node in the Complement position (p). Chomsky however finds a way to save (ACC) by making a somewhat strong stipulation that either all Complement categories must be a Complex category (with both terminal/non-terminal nodes), or that all right branching structures must end in a trace. (I will not go into the complicated details here, suffice it to say here that such strong stipulations would undoubtedly have ramifications for language acquisition)

A second and perhaps more fundamental problem regards the final result--as predicted by Epstein (ibid)--that (LCA) would likewise make available OVS as a (UG) base-generated pattern. There is little evidence to suggest that this pattern actually occurs at all cross-linguistically. Joseph Greenberg (1963) notes regarding Universals that OVS occupies roughly (2%) of the world's languages--a major empirical hurdle to get around (see note 11). At best, it would have to be claimed that OVS is a base structure that simply doesn't survive without immediate morphologically driven movement (at PF)--for one reason or another. (Japanese as a potential OVS language might fall into this category: cf., Tonoike 1995).

In sum, this chapter on Word Order overall has demonstrated that indeed a Phase does exist in the acquisition of language where variable word orders are permitted: including both SV/VS, VO/OV structures. The availability of such free orderings was seen to be predominately linked to the early two-word VP-stage (described as the SAS phase, although variant ordered SASs partially continued into the otherwise IP-stage). Utilizing a 'weakened' model of Kayne's LCA (which, in its original form, called for a Spec-Head-Comp universal hierarchy), we showed how the Comp Parameter could establish the correct word order. This finding supported Bowerman's claim that it is the emergence of both overt arguments (viz., the Subject's vs. Object's competitive drive to secure a Spec-position) within a syntactic DAS structure that contributes to the correct setting of the Word Order Parameter. The findings in the data were also made consistent with Tsimpli's work that likewise gave evidence of free word order at the early VP stage. However, we differed with Tsimpli in a number of respects--namely, that free word order was not seen to involve Double Arguments Strings (DAS), and that the correct setting of word order was not seen as a sole consequence of Functional Parameterization, but rather as a result of the two-pronged condition placed on (i) the Head Medial Principle, as well as (ii) the Comp Parameter--in this sense, it was the placement of the Complement in opposition to the Specifier (creating an agreeing Spec) which brought about either universal word order setting: SVO or OVS. The notion that clauses start-off with a minimal VP-structure (a reduced XP), and thenceforth expand as required (once newly acquired structures project), is consistent with the overall Structure-Building model assumed here.

<< Back to Index

Chapter 3 Notes