42.

Looking at the data (Galasso: 1999), we indeed find a strong correlation between SAS strings and mixed word order alongside DAS strings and fixed order.

Table 5            Word Order

I.  Some token examples include:

(a) SV: Daddy cooking. Him go
(b) OV: Dog kick (= I kick dog). A egg cook. (= I cook egg).
(c) VS: Open me (= I open). Work bike (=Bike works)

In terms of structure, before on the onset of DASs, a Proto XP could be assigned to our SAS stage providing the variable word orderings:

43.

In addition to general word order variability, Wh-word order patterns emerge in our early files (age 2;4-3;0) showing semi-formulaic consistencies when examined in light of the general acquisition of complex structure—as mentioned above regarding SAS vs. DAS complexity. Our data evidence a pattern showing Non CSV (Non Comp Subject Verb) ordering which could be interpreted as formulaic in nature.  This stage roughly overlaps with our SAS stage mentioned above. Like Kayne on Word order, Cinque (1990) has formulated a strong universal position claiming that all Wh-elements universally position within the Spec-of-CP. Recall, that CP is a functional category that should have a delayed onset time under any maturational theory (cf. Radford: 1990). Here too we need to weaken the strong position by adding the stipulation that in order for this Spec-CP analysis to hold, the subject must simultaneously surface forcing the Wh-element to raise and preposition in Spec-CP. Otherwise, very early (stage-1) Wh-arguments (e.g., What, Who) seemingly get initially misanalyzed as base-generated 3Person Pronoun/Quantifiers placed in superficial subject Spec-VP position. This miscategorization often results in Agreement errors where the Wh-word, seen as incorrectly taking the thematic-role of the subject, agrees (by default) with the verb. Consider the structures of the two following CP- structures below:

Table 6                        Wh-word order

44.

In sum, arguments could be devised suggesting that early Wh-structures are prime examples of semi-formulaic strings base generated (VP insitu). A later second stage (or even overlapping stage) may thus be seen as converting formulaic processes into rule driven processes whereby syntactic manifestations of Wh-movement occur with or without Auxiliary inversions. (See Stromswold (§6) above for Non-Aux inversions). Regarding formulaicy, Pine & Lieven (1997), Pine et al. (1998) claim that a non-rule based account is what is behind the formation of early correct wh-questions (a U-shape learning take on the data). While adopting a constructivist account in explaining the high rate of correctly inverted Wh + Aux combinations, they go on to predicted that correctly inverted questions in a the child’s stage-1 data would be produced by those wh + aux combinations that had occurred with high frequency in the child’s input. They go on to specify that there is evidence that the earliest wh-questions produced with an Aux. can be explained with reference to three formulaic patterns that begin with a limited range of wh-word + aux. Combinations (e.g., “whydon’t” you/she) (Rowland & Pine, 2000). Such findings on early formulaic structures parallel what Tomasello (1992) and Newport (op. cit.) suggest regarding an initial stage-1 that reflects a processing deficit tied to functional grammar. In other words, child stage-1 processing which shows a bias toward the modeled high frequency lexical input (vs. rule driven analogy) may arise due to constraints imposed by the low memory bottle-neck of distributional learning (Braine 1987, 1988)

Lexical Stage-1: A Recap

In light of the above data, and the collections of data elsewhere, it could be argued for our stage-1 that the child’s utterances involve pure projections of thematic argument relations. In Minimalist terms, the operation ‘Merge’ would directly reflect thematic properties and that this operation is innately given by the Language Faculty: Verbs directly theta-mark their arguments as in predicate logic expressions:

Table 7       Argument/Predicate  Structure  

The above Word Order/Syntax includes (SV) and (SVO) patterns and is structured below:

             [vP[N Dad][[v0][VP[V work]]]     [vP [N Mom][[v0][VP [V see][N dad]]]]

            (vP=light-verb Phrase).

(i) Syntax, as understood in Chomsky’s  Pre-Minimalist’s terms, may for all intents and purposes reduce to specific bundle-of-features that are encoded in ‘parts-of-speech’ words, (rendering a seemingly bottom-up learning mechanism where ‘meaning’ governs not only how words are learned, but how their syntactic properties project).

(ii) Syntax may no longer be considered as a top-down generator of sentence types, and so words have the capacity to emerge in a early stage of language merely encoded with  ‘l-morphology or [+Interp] features. In this way, one may be able to define an early stage-1 word as exhibiting more or less only the phonological shell of the word void of its otherwise embedded syntax. If this is indeed the case, a viable maturational story can likewise hold for the onsets of ‘f’-morphology [-Interp] features for the given word. Much in the manner of Roger Brown’s observation leading to a sequence of morphological development (staring with -ing and ending with the Aux. Clitic etc.), a similar story could likewise hold regarding how certain features mature and then merge in a word—a maturation of features however which would not delay the onset of the word in phonological terms (or ‘l’-morpheme values), but would only delay the relevant selectional properties (or ‘f’-morpheme values, etc.) associated with its functional grammar.

The twin notions above would ultimately buttress any theory which would see language development as a maturational interplay of features—as captured here in our discussion of a Converging Theories Model.

A typical Chomskyan syntactic tree asserts that functional features (features having to do with M(ood), T(ense) and Agr(eement)) are assumed to be projected in a top-down way: these functional features are understood to be what is behind the notion of movement—lexical items move up the tree in order to acquire and check-off these features. The following question certainly could be formulated in Chomskyan terms: ‘why can’t lexical items have such features embedded in their sub-categorical entries, and if they can, what then would motivate movement other than some ad hoc stipulation requiring features to be checked-off in a overall top-down environment’? Chomsky’s interpretations are clear here—some top-down (deductive) measure must be constructed in order to establish a proper rule-driven syntax. Well, this may in all likelihood be correct, but where/what is it in the system that says that the syntax must start out in this way. Consider the tree below (reduced showing only M & T/Agr features):

The tree above positions the T/Agr features, along with their specific phrases, as having a top-down representation. If such a tree is completely available early-on in language acquisition—as the Continuity view would maintain—than there should be no reason why a child would exhibit 100% omission of say a top-down Agr feature in the way that would affect only certain words and not others. (When only certain words show individual residual affects, e.g., regarding subcategorization, syntax etc., then a strong claim can be made that the overarching phrase structure is not what is behind the phenomenon, but rather specific lexical-parameterizations may be involved.) (See J. Fodor 1997, Baker 2002 for a seemingly bottom-up treatment of lexical parameterization). In other words, if the structure is in place (from top-down) to deliver the feature of Agr (as with Case), than it would be hard to explain away the fact, if observed in the data, that some words could maintain Case while others (which should maintain Case in the target language) do not. Guasti and Rizzi (2001) say: ‘When a feature is not checked in the overt syntax UG makes it possible to leave its morphological realization fluctuating’. Fine. But, this is seemingly a bottom-up problem. It seems that such optionality would have nothing to do with the phrase (per se). What do we say when the feature itself (as projected from the tree top-down) seems to select some words over others regarding inflection? Surely, if this is a top-down venture, then the features should project onto all verbs (for the appropriate phrase), and not just a select few. But this is in fact what we find at our stage-2 of language development—some words may (optionally) inflect/project the specific feature while others completely by-pass it (entirely).

This gives us the flavor of specific words (and not word classes) taking on functional features (bottom-up). The problem here is how does one maintain the higher-ordinance structure of functional grammar originating from the latter two upper layers of the tree while selecting the functional projection on only a select handful of words. One way around the dilemma may be to suggest that the lexical word itself has part of the (upper-branching) tree embedded in the very lexical item itself (as in sub-categorization). In this way, a specific word may reflect a specific functional feature or parameter while another word may not (on a specific lexeme by lexeme basis)—in all actuality, what we are talking about here is that (i) the initial process of the acquisition of functional grammar involves one word at a time (in a bottom-up way), and that (ii) only at a later more developed stage does such feature projection extend to the overall class of words (which then extent to phrases). Following in the spirit of Lexical Parameterization (Borer), Janet Dean Fodor in a similar vain has tentatively suggested in some recent work that parameterization may affect certain words (as in lexical feature specificity) and not others (outside of the scope of its word class) (talk presented at the University of Essex, 1997). One outcome of this would assume that children establish parameter values (perhaps piece-meal) and not grammars as wholes. An example of such bottom-up parameterization or say feature specificity (only selecting [+/-Nom] Case marking here) might then be diagrammed in the following manner:

Such an exclusively bottom-up parameterization method would however obscure correlations often found in the data regarding Case and/or Agreement—such as a seemingly top-down holistic correlation which seeks to link (i) [+Nom] Case if in an agreement relation with a Verbal INFL, (ii) [+Gen] Case if in an agreement relation with a nominal INFL, (iii) Default Case otherwise. It may be that such correlations do come on-line after an initial ‘non-phrase’ parameterization stage—hence, an initial and not fully fledged parameterized stage would meagerly work with individual words, delaying class-parameterization to a slightly later stage.

U-Shaped Learning.    One of the most striking features of language acquisition is the apparent so called U-shaped Learning Curve found straddling the two stages of language acquisition. In brief, the U-shaped curve is understood in the following way:

(i)Inflection. Children’s earliest Inflected/Derivational word types are, in fact, initially correct-that is, it appears to be the case amongst very early MLU that children have correct formulation of rules. (It goes without saying that typical early MLU utterances indeed have no tense markings to speak of (cf. Wexler & Radford’s Maturational Theory). The point here is that whenever a small sampling of Tense does appear in early MLU speech, it always appears correctly). An example of this is the early emergence in the data of the past tense and participle affixes [ed] and [en] e.g., talked/gone (respectively). The initial Past Tense and Plural forms are correct, regardless of whether or not these forms are regular (talked/books) or irregular (went/sheep). However, and what is at the heart of this striking development, it also appears that this initially correct performance stage is then followed by a period of attrition during which the children actually regress—that is, at this slightly later stage in development, they do not only lose some forms of affixation, but in addition, produce incorrect over-generalizations in tense forms (go>goed>wented), and plural forms (sheeps), as well as non-inflected tensed forms e.g., talk-ø/go-ø (=past tense). To recap, the first occurrence of inflectional overgeneralization roughly at age 2 years that supports a rule-based grammar is preceded by a phase without any errors at all.

(ii)Phonology. Similar to what one observes regarding a u-shape grammatical/inflectional development, children also appear to follow a u-shape learning curve with regards to phonology. An example of this is the often cited early productions of e.g., (i) slept /slept/, cooked /kƯkt/, played /plae:d/ > to (ii) sleeped /slipId/, cooked /kƯkId/, played /plae:Id/  > and back to (iii) slept /slept/, cooked /kƯkt/, played /plae:d/ (respectively) completing a

U-shaped morpho-phonetic curve yielding ^{/t//d//t//d/.}

What appears to be good examples of ‘rule-based’ inflection and assimilation in (i) and (ii) (above respectively) is in all actuality nothing more than the product of a ‘parrot-like’ imitation sequence—more akin to iconic pattern processing derived from stimulus and response learning. The child can be said to engage in segmental, phonetic-based rules only when s/he appears to process the rules yielding an incorrect overgeneralization of past marker {ed} typically pronounced as the default /Id/ which forms the middle-dip portion of the u-shape curve. Recall, in terms of phonology, the child has three allophonic variations to chose from:

a. {ed} => /t/ “walked” /wa:kt/
b.& {ed} => /d/ “played” /ple:d/
c. {ed} => /Id/ “wanted” /wantId/

It seems that a default setting with regards to phonology (place & manner of articulation) is minus Comp(lex) where [-Comp] denotes one feature distinction over a two or more features (for instance, bilabials /b/ /m/ would have a [-Comp] feature whereas labio-dentals and inter-dental /f/ /q/ (respectively) would have a [+Comp] since both lip and tooth are involved. In addition, it seems that plus voicing [+V] typically wins out over minus voicing [-V]. By using these default settings, we naturally get voiced plosive /b/ d/ /g/, nasals /m/ /n/, as our very first sequence of consonants along with [+V] vowels. By taking this default status, the /Id/ should be the allophone of choice, and it often is. In this manner of speaking, adherence to the default setting suggests at least some formation of the rule: defaults work within rule-based paradigms and so should be considered as a quasi-rule-based generation as opposed to a pure imitation sequence.

The first two stages of development that form this apparent u-shape curve has been interpreted as manifesting the application of qualitatively different processes in the brain—representing different modes or stages in the course of language acquisition. This u-shaped curve arguably provides some support for our stage-1 to be defined in terms of a formulaic stage rather than as a syntactic and true-rule learning stage. The second up-side of the u-shaped curve is found to coincide with an independent syntactic development—the emergence of a Finiteness marker, and that this finiteness marker only emerges at our functional stage-2 (see Clahsen). In sum, the three stages could be described in the following way:

(i) The first period of the first up-side curve (correct production) correlates with a style of rote-learning. This more primitive mode of learning suggests that the mental lexicon is bootstrapped by mere behaviorist-associative means of learning. In such a rote-learning stage, lexical items (either regular or irregular inflections) are stored in an independent mental lexical heavily based on memorization of formulaic chunks and associations and are processed in a different part of the brain. It is of no surprise that irregular verb past inflection (go>went) out number regular verb past inflection (talk>talk-ed): The former being stored in the lexicon as a formulaic chunk, while the latter indicating the morphological rule formation [V+ {ed}]. Hence, our dual converging theories model postulates for a sharp contrast and disassociation between regular vs. irregular inflection. This seemingly early correct production is therefore due to a low-scope, phonological ‘one-to-one & sound-to-meaning’ relationship with no relevance to rules. Hence, our formulaic past tense inflection is not realized as [stem + affix] [talk-{ed}], but rather as one unanalyzable chunk [talked].

(ii) The second stage then marks the onset of a rule process (albeit, not necessarily the mastery of it). Here, the child is seen as letting go with the formulaic lexical representation in favor of rule formations: i.e., patterns of concatenate stems appear along side inflectional affixes. Thus, irregular forms often get over-generalized with the application of the rule resulting in e.g., goed/wented/sheeps. This overgeneralization stage maps onto a chronological functional categorical stage of language acquisition where rule-based mechanisms are becoming operative. Thus, the over-generalized up-swing of the u-shaped curve is linked to children’s syntactic development: over-generalization of inflection appears when the child ceases using bare-stems (as in stage-1) to refer to past events.

(iii) The third and final stage marks the second up-side swing of the u-shaped curve and represents the correct target grammar.

The W&W study suggests that the maturational development of language follows from brain development—and can be summarized below:

a. The lexical stage-1 of language acquisition naturally arises from a disconnect between the more primitive POT (temporal-lobe/lexical-grammar region) and the hierarchical Broca’s area (frontal-lobe/functional grammar).

b.& This disconnect has to do with the biological development of myelination in the bundle of axons that connect the two areas together. Myelination of axons is then said to mature at roughly that chronological stage where we find a lexical (staged) grammar merging with a functional (staged) grammar.

c.   With respect to the brain/language relationships in the child, it is important to recognize that during the period of time typically associated with the initial stages of language acquisition, the brain is still in a relatively immature state.  Neural plasticity begins with the sensory motor-strip temporal area (POT), and then proceeds to move to secondary areas (Broca’s area) related to the frontal lobe region.

---

In the history of all pursuit of science, it has traditionally been the case that science precedes and develops via different methods and theories. Converging approaches always strive to expose inherent weakness in their opposing theories. It goes without saying that convergence methods go far in peeling away biased assumptions which often lead to half-correct assertions. Taking what is good from one theory and throwing away what is not is just common-sense science. For example, on one ‘converging; hand, Chomsky has asserted that syntax is the result of the creative human brain set-up in such a way as to manipulate true-rules. It creates, from nothing external to itself, the structure of language. (See special nativism above). In restricting ourselves to the point at hand, Chomsky has assimilated much of his arguments from the long line of rational philosophy and has converged such reasoning into how he believes an autonomous language structure (internal) might be construed. His belief that syntax is autonomous directly paves a way for him to distinguish between species-specific (human/hominid) language and other modes of cognitive-based primitive communications (animal/pongid). His now famous debates—first between Skinner (Behaviorism) and later with Piaget (Constructivism)—can be readily reduced back to Converging Methodologies between (philosophy and cognition) which sought to return language to seventeenth century nativist assumptions. Later, he would go on to extend such arguments to fight off pure pragmatic/socio-linguistic pursuits of linguistics—saving the study of language from becoming strictly a  ‘humanities’ field of study which emphasized social phenomena with little if any analytical worth: (cf. Quine, Rorty pace Chomsky). Taking his notion of an autonomous syntax further, the natural next step to take would be to say that all other aspects of language (whatever they may be) that can’t fall under this autonomous rule-based syntactic realm might be conversely tethered to both behaviorism and associationism as part of an underlying cognitive mechanism. Chomsky has himself expressed the possibility that general mundane concepts--many of which contain inherent sub-categeorial features that are extremely convoluted and abstract, yet from which we go on to readily attach labels (=words)—may be preconceived and innate: however, he goes on to suggest that such conceptual innateness may be tethered to cognition as a universal ability to get at meaning (Chomsky 2000: p.61-62):

These conceptual structures appear to yield semantic connections of a kind that will, in particular, induce an analytic-synthetic distinction, as a matter of empirical fact.

These elements (he cites concepts such as locational nature, goal, source of action, object moved, etc.) enter widely into lexical structure¼ and are one aspect of cognitive development.

---

On one hand, what Chomsky seems to be saying is that (i) Functional Grammar, or Syntax (par excellence) is autonomous and disassociated from all other aspects of the mind/brain-including meaning and/or cognition. Thus, syntax is created from out of the mind’s creative and independent eye  (with all aforementioned nativist trappings). However, and to the point of this section, Chomsky doesn’t hesitate to attribute those non-syntactic aspects of language, say word learning (based on frequency learning and  associationism, to cognition. This, I believe, goes to the heart of the matter--namely, that a form of converging theories has been evoked here and could be summarized as follows:

Chomsky and Converging Theories

1. Syntax proper (labeled herein as Functional Grammar) is creatively formed by a true-rule process via an innately given Language Acquisition Device (LAD) (more recently called the Language Faculty)—comprising of initial grammatical default settings of which are called Universal Grammar. For example, this is where the more abstract Inflectional rules are housed: the functional features of number/person/case/agreement/tense e.g. Plural [N+ {s}], Past Tense [V+{ed}], etc. Of course, the ‘Wugs-Test’ of Berko goes directly under this category. Meaning is detached from syntax.

2. Word learning (labeled herein as Lexical Grammar) is formed via a one-to-one iconic association between sound and meaning. This process of both word learning on (i) a phonological level, and word learning on (ii) a semantic/conceptual level, is more akin to past behavioristic notions of learning. Very young children (at our stage-1) may exploit and over-extend such processes—this is apparently what we find regarding formulaic type utterances, Irregular Verb/Noun lexical learning and retrieval, as well as Derivational morphology.

---