| | |
Towards a Dual Mechanism Model
of Language Development
Joseph Galasso
California State University, Northridge
Presented at the Child Language Research Forum-2004,
Stanford
Abstract
The study of syntactic development in children, for
all intents and purposes, is reducible to a single minded inquiry into how the
very young child (implicitly) knows to distinguish between lexical stems and functional affixes. Hence, the overriding question burning in the minds
of most developmental linguists is morpho-phonological in nature. For instance,
it would seem that the child must at least know (a priori) the stem before she can then engage in a dual-track
process by which ambient separation of the morpho-phonological distinction
attributive to past tense is carried out, say, between the paradigmatic
representation of the English word play vs. play-ed /ple-d/ (a dual processing which provokes separation of the /play/-stem and
the /d/-affix). Otherwise, it
could be conceivable for the young child that the pair play-played
would represent altogether two different lexical stems, and, stored as such,
reflect two distinct though relatively similar semantic notions (a single
processing): perhaps not unlike what we do find regarding derived words where
an otherwise 'two-morpheme' analysis of [teach]-{er} is
processed (tagged, stored and retrieved) as a 'single-morpheme' stem [teacher], similar to how the word [brother]
is stored.[1]
A
two-point conclusion is reached in this paper: (i) that children have instant
access to and make tacit use of innate syntactic knowledge, allowing them
instinctively to know to separate stem from affix-leading to a Gradual
Development Hypothesis which shows
developmental asymmetry between the acquisition of lexical vs. functional
categories (Radford 1990)-and (ii) that such prima facie knowledge naturally arises from The Dual Mechanism
Model, a processing model that offers
the best of both worlds in that it can account for both how the child comes to
'know' lexical stems in the first place, and subsequently, how such stems come
to be distinguished and project morpho-phonological material leading to stem
vs. affix separation.
Finally,
an interesting and potentially far reaching implication is advanced stating
that there resides not only a dual routing system in the brain for the split
processing of stem+affix material,
but that the proposed dual model can be extended in such a way as to cast an
entirely new dual-typology of language in the sense that (i) 'modular-complex' weak-stem/synthetic languages with a low, medium to high
gradient range of stem modularity (English, Spanish to Hebrew, respectively)
come to use the cerebral rule-based processing area as an additional language
storage capacity, in contrast to (ii) 'modular-simple' strong-stem/analytical, agglutinative languages (Chinese,
Hungarian) which, due to their global modularity, mitigate inflectional
affixation to a much less complex system, thus preserving a more economically
robust single storage capacity based entirely on the frequency-based processing
area of the brain.[2],[3],[4] When
this notion of modular-complexity is raised in conjunction with a previously
established parameter setting dealing with [+/- Bare-Stem] languages, a new and
powerful tool is fashioned allowing us to better describe and explain child
functional category/feature onsets within divergent language groups. It is
therefore argued that many of the cross-linguistic asymmetries found in Early
Child Inflectional Development should be better thought of as reflecting how the
innate Language Faculty provides languages the selection of a single vs. a dual
storage capacity.
Assumptions and Suppositions
We follow Chomsky's (1995) 'The Minimalist Program'
(MP) throughout and assume language has real structural antecedents to the
mind/brain-The I-language is a state of the mind/brain. We then assume there to be real physiological
connections between language, syntactic structure and relevant brain
processing. A 'theory-of-mind' must therefore interest us as we build upon any
language theory. Still, very little is known about how a brain bootstraps
itself and creates a mind, as there is little understanding of the brain/mind
relation. An isolative 'theory-of- brain' and its processing, however, is much
better understood and allows us to be a bit more precise. There is now strong
scientific evidence to suggest that innate architectural principles of the
brain process linguistic information in two fundamentally different ways, thus,
by extension, determining how lexical vs. functional features project into
syntactic structures. We likewise assume that this dual processing, as defined
and expressed (overtly/covertly) in 'syntactic trees' has real linguistic
relevancy and is constrained by certain legibility conditions (at PF/LF interfaces).
We further assume this relevancy captures a natural cut, or a language
divergence, as characterized by a dichotomy between semantic-based language (pertaining to the VP) vs. syntactic-based language
(pertaining to the IP).
Following
the general framework as laid out in Chomsky 1995, along with crucial
assumptions as detailed in 'Distributed Morphology' (DM) (Halle and Marantz
1993) (viz., the notion that functional heads serve as the locus of lexical
insertion, with word formation occurring in the syntax as a result of the
syntactic combination of such heads, and, that due to maturation, functional
heads may go unspecified in the syntax) we assume this cut shows itself in the
lexicon as follows and may have a real brain-language physiological correspondence
to how a higher syntactic position
(IP-dominated) versus a lower syntactic position (VP-dominated) respectively captures the overt vs. covert checking of features:
The
V(erb) P(hrase) contains phono-semantic features. These features are substantive
in some way, and thus interpretable at each relevant interface. Lexical
knowledge associated with the VP is therefore said to be unavoidably trivial
and instinctive to a certain degree, as derived by a set of data-driven 'sound-meaning' relations appropriated to the PF/LF
interface. Features of the VP are labeled +Interp(retable) and, as a rule, do
not require the movement of its head into a higher functional projection. (DM
labels such features as l-morphemes (lexical): they are idiomatic and correspond
to concrete categories of Verb/Noun/Adjective).
The
I(nflectional) P(hrase) contains features that are not substantive in some way
and are therefore -Interp at either phonological or semantic interface.
Functional knowledge associated with the IP is therefore said to be
non-trivial, as derived by a set of rule-based relations which must then be 'checked-off' at the
PF/LF interface. Whenever the typology of a language forces -Interp features to
be checked in the overt syntax, this forces the features to project in the
morphology of the language. Whenever such features can go unchecked in the
overt syntax, the choice between overt vs. covert expression in the language
depends on that language's specific morphological parameterizations. (See ¤35
Typology). These formal features represent for the most part what drives
language variation. The reasoning for their existence in language remains
somewhat of a curiosity (it may be that such non-substantive features
ultimately drive the unique 'displacement property' of human language-in the
very general sense that functional checking motivates movement up the syntactic
tree) (DM labels such features as f-morphemes (functional): they are
non-idiomatic and rely on vocabulary selection).
Suppose
that the VP is exclusively pinned to the sensori-motor component of the brain,
the more primitive part of the brain that controls cognitive motor-skill,
memory and lower-level associative learning. In this sense, the VP presides as
the lexical category (par excellence)
alongside semantic-based [+Interpretable] features. Lexical categories exist prior to any computational numeration.
Suppose,
in addition, that the IP is rather pinned to areas of the brain which house
more abstract levels of thought (perhaps exclusively relating to the
left/frontal lobe region of the brain). Functional categories are the result of the combination of a lexical item
and a functional/inflectional feature (a numeration).
Suppose
further that classic processing distinctions between Derivational vs. Inflectional morphologies may be fuller understood in light of this dichotomy, i.e.,
that derivations ultimately trace their origins back to interpretable features
internal to the lexicon (l-morpheme), and, conversely, that Inflections are
nothing more than 'morpho-phonetic fillers' (expletive in nature), the remnants
of some computational operation of a broader syntactic scope (f-morpheme). It
is this latter observation inflection that will interest us in this paper. More than any other single
linguistic property, it is the nature and projection of inflection that gives a
language its typology-languages differ in inflectional systems. It is suggested
in this paper that the brain may house two separate language storage capacities
(perhaps in some way paralleling PF vs. LF representational systems) which
involve a parameter-setting, reflecting to some degree how a language selects
to show these fillers: a language may opt for fillers as affixes, disjoint from the lexical stem, or opt to incorporate
the fillers somehow as part of the stem. (In generic ways, the former entails
the operation Move, while the
latter entails the operation Merge.
In more concrete ways, the former operation Move may force an entire lexical item to overtly raise with
an affix, as seen in so called pied-piping, since a formal affix unattached to its stem would crash at PF).
Expanding on this argument, a dual-typology of Semantic transparency vs. Grammaticization-based languages is suggested, termed herein as a
'Modular-simple' vs. 'Modular-complex' parameter. The parameter reduces to the
selection of how a given language stores and projects such vacuous filler
material. Nothing hinges on this last supposition: the dichotomy could be
upheld irrespective of whether or not such a dual storage capacity is
ultimately correct. Future developmental research will surely look toward MEG,
fMRI and PET scans to ultimately replace models of syntactic tree diagramming.
Only through better brain-to-language modeling will we eventually transform
language theories into biological certainties.
Introduction
It is now well documented that English children pass
through a (gradual) developmental stage in which they comprehensively fail to
project functional categories and features. At the same time, it is equally
well documented that children of other language groups, relative in age to
their English counterparts, may not proceed in a similar (gradual) fashion in
omitting the same formal categories and/or features. Thus, it would seem any
attempt to tether formal functional deletion to a more general cognitive
deficit, as based on a biologically-driven maturational theory of language
development, would run into several difficulties.
It
is argued in this paper that a unifying heuristic procedure which incorporates
a two-prong analysis of (i) Feature Complexity and (ii) Inflectional
Paradigm Complexity indeed maintains, as the working null hypothesis, a
biologically based maturational account and may help determine the
eventualities of a language's development, as attested in child language
acquisition. The heuristics come to detail a 'Converging Theories Hypothesis'
of sorts which attempts to describe language acquisition along the lines as
being both 'Discontinuous' in nature to that of the adult target language
(regarding top-down paradigmatic/structure analogy, thus saving some elements
of a maturational account), and at the same time 'Continuous' (regarding
type-token frequency learning which is described as a bottom-up cognitive
universal).
The
'Dual Mechanism Model' is then advanced as being best positioned to account for
the attested child language asymmetries found amongst the diverging language
groups discussed. Therefore, it is to be maintained that only insofar as the
Dual Mechanism Model is operative can we then postulate for a maturational
theory of language development.
[0] This
paper constitutes the first segment of 'Twin Working Papers'[5] and is more-or-less a personal exercise in formulating my own ideas on a Research Statement, the purpose of which is exploratory in nature and
mainly designed as a general means of roughly expressing my own thoughts and
understanding in the area of developmental language research. Due to the
'Working' nature of the notes herein, it will appear at times that ideas are
disjoint and surface at will. Having said this, it is my intention that the
framework evolve. The notes come to be centered on a driving notion termed
'Converging Theories', a hypothesis that firmly attempts to pair the divergent cross-linguistic
language data-attested in the differing language groups which show rich vs.
impoverished inflectional morphology-to that of a current unifying hypothesis
termed the 'Dual Mechanism Model' (Marcus et al. 1994, Pinker 1999, Clahsen 1999).
[1] It
is now well documented that English children pass through a (gradual)
developmental stage in which they completely omit functional [categories] and
{features}such as [CP, IP, DP] and {T(ense), AGR(eement) CA(se)}. Following the
developments of research models over two decades, as advanced in Felix (1984),
Borer & Wexler (1987), Guilfoyle & Noonan (1988), Radford (1990),
Wexler (1994), Hyams (1996), and
Radford & Galasso (1998), there is today strong, surmounting
evidence to support some kind of Discontinuity (or, for a lack of a better term, some
'non-adult-like' conceptual linguistic formation) between the English speaking
child and adult, at least with regards to the aforementioned projections of
functional categories and/or their features. (See ¤21 for some data).
[2] After
the attested 'No Functional Categories' stage, children then seem to pass
through a slightly more developed (albeit non-target) stage in which they
confuse much of the functional grammar they have acquired. For example, it is
not uncommon for English speaking children to go through an early stage of
development in which they over-regularize a Gender feature, whereby the pronoun
He is used for both males and
females alike (CHILDES, Brown files: 13). In addition to a gender pronoun
feature non-specificity, a child might confuse the 1/2nd person feature
spell-outs of the pronoun I for You and You
for I, or may simply generalize
the 3rd person possessor features for possessive {'s} (Tom's) onto the first and second person possessor (my's,
your's) (Chiat 1982).
[3] In
addition to some of the feature confusion or non-specificity as mentioned
above, the most salient fact about child language is the systematic 'omission'
of functional material-a 'No Functional Category' stage-both in the forms of
the functional categories themselves, as well as with the associative
parameters that maintain such feature specificity (Spec-Head relations in higher-order functional
phrases).
[4] Notwithstanding
such apparent data which seemingly call for Discontinuity between child and adult grammar, a split has emerged
nevertheless amongst those developmental linguists (i) who espouse for an
initial stage of child syntax which begins in complete absence
of functional material (a Non-Functional/Inflectional stage as suggested in
Radford 1990/Radford & Galasso 1998) contra those (ii) who espouse for an initial stage which
begins with partial absence of
functional material (an O(ptional) I(nfinitive) stage as suggested by Wexler
1994).
[5] Proponents
calling for such Continuity tend
to cite not only English as one potential confirmation source (leading to the
aforementioned rift between Radford vs. Wexler), but cite the early emergence
of functional categories in other language groups, thus complicating Radford's
general claim for an initial No-Functional stage-1. New criteria can be
established in order to make the debates more precise. Questions into whether
or not a language's inflectional system is sufficiently rich to spawn very
early MLU onsets of affix morphology enter into the equation. This then
naturally leads to question regarding whether or not the early presence of
syntactic movement is linked to the checking of functional material (Chomsky
1995). One might ask whether the given language manifests a kind of deviant
inflectional usage (as attested by abnormal Aphasia/Specific Language Impairment
studies)-many such errors in certain language types essentially amount to
errors with wrong inflectional usage of 'commission' as opposed to any complete
inflectional 'omission' (in defiance of the discontinuity theory as mentioned
above). (We'll consider these each in turn in ¤11 below). Hence, the two
differing hypotheses, Discontinuity vs. Continuity, could be said to converge
roughly at the cross-roads of morphological typology: namely, it has been
suggested that when a language has a relatively rich inflection [+INFL], that
language will seemingly exhibit the early projection of functional
categories/features (as attested by child language studies of the relevant
language). However, when a language does not have a sufficiently rich
inflection [-INFL], that language will exhibit a somewhat delayed projection of
functional categories/features (a delay that has been cited as providing
general evidence for maturational based hypotheses). It is in this general
sense that language morphological typology is said to drive (top-down) the
nature of early child syntactic projection (in contrast to, say, a biologically
determined maturational process). Put another way, (i) it is first a priori knowledge of morpho-phonology that triggers the
appropriation of syntax, (ii) the syntax, in turn, then seeks out both the choice
of words and the nature of the
paradigm.[6]
[6] Hence,
following the 'Dual Mechanism Model' (DMM), children 'know' that Inflected
forms are not lexical since such
forms are conceptualized and generated in a separate processing modular. The
DMM maintains that a clean separation takes place between the lexical stem and the affix-a stem is 'meaning based' and thus housed in the temporal-lobe region
of the brain (Wernicke's area), while the affix is 'abstract' and thus housed
in the frontal-lobe region (Broca's area). The lexicon lists only lexical items
(stems) while functional items (affixes) are added at a second stage in the
numeration. (See ¤44 for diagram). The DMM credits the Brain/Mind with having
two fundamentally different cognitive modes of language processing-this dual
mechanism has recently been reported as reflecting inherent qualitative
distinctions found between (i) regular verb inflectional morphology (where
rule-based stem+affixes form a
large contingency) and (ii) irregular verb constructions (where full lexical
forms seem to be stored as associative chunks). The Language Faculty thus provides us with two ways of symbolic
representation.
[7] The
overall approach here is therefore 'top down' since there is no sense in the
word 'Knowing' unless there is first 'Access'. (Though it is generally accepted
in Feature Theory (Chomsky 1995)
that syntax is driven 'bottom-up' by features which peculate up from the lexeme
and enter into a phrase configuration-presumably a spec-head configuration
under a movement analogy termed Merge-the
proposed model here differs only to the extend that we believe the initial
'Access' to such features is 'top-down'). This has the flavor of saying that
the debate over continuity may be misplaced and badly spelled-out, and that the
crucial debate rather hinges on how the maturing, neurodevelopment of the brain
processes (or doesn't process) the 'accessed' paradigmatic/inflectional
material generated by the two diverging language groups. Having said this, I
follow Meisel (1994) and suggest that it is the full-fledge realization of the
INFL-paradigm that should be the real measure of competency, rather than any
language specific piece-meal affixation process by which morpho-particles affix
to stems, since the latter could always be explained away as misanalysis on the
child's part in ways that do not directly speak to a well formulized inflectional
paradigm (e.g., formulaic and/or lexical incorporation, lexical redundancy
rules, un-analyzable chunks,
etc.).[7]
[8] Early
Inflection: Pro-Continuity. Contrary to what some might like in assuming
child-to-adult discontinuity-viz., that all languages start out with deficient
functional categories, say, owing to a general cognitive development of
functional categories-a good amount of data found in the literature calling for
a least some aspect of early functional competence include language groups such
as French (Pierce: 1989, 1992) and German (Clahsen: 1990, 1994).[8] Hence, those clinging to continuity can always claim
that at least one functional category is present from the very earliest MLU of
child speech (hence, quieting any clarion call for a strong discontinuity
hypothesis). For instance, Pierce shows that from the early age of 1;9-2;3
(Nathalie files) +Fin(ite) V(erbs) (hosting T/AGR) show movement via their
raising out of VP above NegP and into a functional IP, whereas -Fin(ite) Verbs
remain VP in-situ. (NegP is traditionally viewed as being an intermediate
phrase situated between the lower VP and the higher IP). Such findings
illustrate syntactic sensitivity on the part of the child by the fact that the
features in the lexeme drive syntactic movement for purposes of checking. (The
checking of formal -Interpretable features such as CA and AGR must ensue under
concord between a functional Spec-Head configuration).
[9] French: IP (Pierce 1992) +/-Finite
Syntax (Pierce 1989)
(i) I NegP [+Finite] [-Finite]
| 
| Neg VP pas verb 11 77
| | 
verb
pas 185 2
V N
(a)
veuxi pas ti lolo => Functional
projection
[1Sg Pres] want not water
(b) Pas manger => Lexical projection
Not
eat [Inf]
[10] German: FP (=Functional Phrase) (Clahsen et al. 1994: p.
9)
(ii) 
Spec F'
[+F] VP
Neg VP
Spec
V'
[+V]
[+Fin]
V2 utterances [-Fin]
V-final utterances
(a)
Kann
er nich (2;9) (c) hase auch auytofahrn (2;4)
(can
he not) (hare also car drive)
(b) ich hab
hier reintecken tasche (2;6) (d) mone auch lump ausziehen
(1;11)
(I
have here put in bag) (Simone also rag take
off)
Clahsen
sites very early MLU German child sentences in which 90% of all Finite Verbs
show raising across negation into a higher host, verb-second position (here,
labeled FP (=Functional Phrase). Non-finite verbs in V2 position are
practically non-existent (op. cit: 13).
[11] SLI:
Pro-Continuity. In addition to the question of the
early (vs. late) projection of inflectional, a second line of inquiry has begun
to look at Specific Language Impairment (SLI) as a means of teasing out just
how the morpho-syntax within divergent language groups is processed. For
example, one factor that arises out of morphological typology is the
classification of a matrix parameter that holds between [+/-INFL(ectional)] to
[+/-Bare Stem]. For instance, languages which have a rich inflection [+INFL]
tend to also have a [-Bare Verb Stem] parameter setting (Hyams 1987) and vice
versa. Languages such as Italian,
Spanish, among others (with perhaps the most sever case being Hebrew) are
non-stem-based languages in the sense that the projection of a bare stem
doesn't constitute a word. (In the case of Hebrew, the radical root of a word
can only be realized and projected in conjunction with the root's
morpho-phonological paradigm). Whereas it is grammatical to project the bare
verb stem speak in English (as in
the infinitive usage I can 'speak' French), it is ungrammatical to do so in Italian or Spanish *parl-,
*habl-. What turns out to be
interesting along these line of inquiry is that with respect to both child
language development and SLI, [+INFL]=>[-Bare Stem] languages tend to show
immediate and systematic functional category/feature projection on the one hand
and only SLI inflectional error usage on the other. (See Grodzinsky (1990) for
a detailed look into Aphasic/SLI studies). For example, Italian SLI subjects never
omit inflections as that would violate word-structure properties. Italian SLI
rather typical involves inflectional errors having to do with misplaced Gender,
Number and Agreement-e.g.,
(i) *Quest-o macchin-a
This
{masc} car {fem}
(This
car)
This
contrasts with both normally developing children as well as SLI children of the
matrix correlate [-INFL]=>[+Bare Stem] (i.e., English) who show a stage in
which functional categories/features are entirely missing. What this seems to
suggest is that before such language specific data can outright challenge the
general notion of discontinuity in child language acquisition, a more detailed
analysis of the language's morpho-syntactic processing must be obtained.
[12] Since
French does constitute as a [-INFL]=>[+Bare Stem] language, how do we square
this with the above observation made by Pierce that there is very early
functional projection? One would expect that since French constitutes a weakly
inflected though synthetically strong language, discontinuity theorists would
assume an initial stage in which inflection is not yet procured. (For further
arguments along this line of assumption, see ¤35 Inflections and Typology).
[13] Well,
as it turns out, there is some evidence to suggest that the picture is somewhat
mixed and that Pierce's French children (perhaps more so than Clahsen's German
Children) do have a stage (albeit short lived) during which they completely
omit functional categories-'a clarion call for discontinuity'. For instance, as
Radford notes (1995) and as pointed out by Atkinson (1995 p. 53)
...'if
we read [Pierce's] work carefully, we uncover the observation that in the very
first recording of Nathalie at age 1;9;3,..., Nathalie uses only nonfinite verb
forms, not finite verbs.'
Atkinson
notes: Radford goes on to cite Pierce's own observation of a very early
stage at which inflected forms are absent. Though it is unclear what Pierce
precisely means by 'uninflected forms', we can at least assume that she has
identified a stage in early French which shows no systematic finite/nonfinite
distinction.
[14] Similar
conclusions could be reached for German and Dutch (Wijnen, Bol 1993, p. 247,
quoted in Atkinson). (Italics
belong to Wijnen & Bol):
'...there
are indications that during the first phase of grammatical development, Dutch
(and German) children only use nonfinite verbs...'. This implies that the 'Optional Infinitive' stage may
in fact not be the very first stage of syntactic speech and that a prior stage
exists which shows No Inflection-a Non-Inflectional stage.
[15] When we turn to Spanish, however, we see a somewhat
more complex picture emerge. In light of the fact that Spanish is a [-Bare
stem] language (and thus must assign some affix material to its stem), other
syntactic considerations must be applied in determining whether or not there is
true functional awareness. For instance, Grinstead (2000) argues that there
exists a stage-one in early child Spanish that exclusively manifests null
subjects and that a correlation holds between this exclusive 'null-subject
stage' and the specific absence of particular affix agreement inflection having
to do with Tense and Number. In brief, Grinstead concludes that it is the
simultaneous onset of tense and number in the inflection paradigm that
ultimately triggers a particular aspect of UG pertaining to the CAse of overt
subjects. The particular onsets of such inflection are complicated and
potentially undermined by the possibility that simply because a young Spanish
child shows affix material (infinitive endings -ir, -ar, -er,) the child may simply be realizing these as parts of
the stem due to the [-Bare stem] parameter (i.e., they are rote-learned
chunks). If indeed this is the case, other factors must be considered in
concluding that functional categories have emerged. A similar conclusion could
be reached regarding early Italian verb+affix constructions. For instance, the
underlying structure of initial inflections (e.g., mett-o, mett-i, mett-e 'put-1sg/2sg/3sg'
(Guasti 1993/1994) may, in fact, be (semi)-rote-learned since there is no
syntactic possibility of a bare stem production. In addition to the bare-stem
factor, it may be that when such inflectional sequences are strong, with high
frequency attached to their production, the clusters may take on stem-like
qualities. (In this sense, weak clusters provide better material for
paradigmatic formation). The fact that subject-verb agreements are consistent
may rather speak to broader morpho-phonetic notions having to do with a kind of
grammaticalized lexical incorporation to the extent that verbs ending with -o
derive first person, etc.) It is
equally important to note that Wexler's OI-stage doesn't manifest in pro-drop
languages such as Italian or Spanish. One might ask why this is. It appears
that the lack of OIs in pro-drop languages can be traced to properties of the
AGR and INFL paradigm systems in the language. For instance, AGR in such pro-drop
languages doesn't have the array of -Interp(retable) features that otherwise
drive DP-subject projection and movement (via the overt checking of the -Interp
subject D-feature, presumably CAse). In this sense, it is understood that AGR
in pro-drop languages rather shows more of the semantic-based +Interp F-features (paralleling to a certain degree what one might expect of
VP/Thematic relations, under the postulation that Interpretable features are
semantic in scope and not part of the formal syntactic numeration). Hence, null
subjects of the Italian, Spanish variety are inextricably linked to the rich
morphologically agreement system. It is therefore not too unreasonable to
assume that aspects of AGR / INFL in pro-drop languages may sustain a certain
amount of subject incorporation (on a scale with subject-verb
grammaticalization). Such inherent inflections could be considered as having
semi-formulaic tendencies, or rather could be viewed as taking on
characteristics of incorporation,
similar to how the derived two-morpheme word teach-er was said to be processed as an inherent chunk. English
children never delete the derivational affix {-er} even when many such true inflectional affixes are
being dropped all around them. English children omit the infinitive 'to'
inflection early-on in stage-1 of their production precisely because it
constitutes a 'true affix', unincorporated marker (as opposed to other early
affix markers such as participle forms {en} and {ing} which seem to
behave as non-affixes).[9] Such participle forms maintain 'strong-stem-like'
properties and thus may share some manner of incorporation. What is interesting
about early Italian data is the fact that AGR inflectional errors are almost
non-existent (a hallmark of rote-learned/lexical productions) despite the fact
that other affix deletions do manifest at the same stage in question. For
instance, what one typically finds at the early stages of Italian syntactic
development are (inter alia)
errors regarding Number on Noun stems (a possible syntactic bare stem
production). What this might suggest is that when the early child data are
examined carefully, distinctions regarding the type of inflection can be made,
leading to discussion of whether or not a production should be considered as a
true stem+affix inflectional production. Italian here may constitute a
periphery case, but when taken in conjunction with more agglutinative type
languages, the findings support the general idea that inflections for [-bare
stem] languages may to varying degrees be incorporating the affix into the
stem. If an inflectional affix is routed and incorporated into the stem, than
processing access of the affix has the same status as access of the stem-with a
kind of lexical incorporation that mimics pie-piping strategies. It would seem
that non-clitic, true rule formations of inflectional morphology would not
trigger such strategies. In brief, developmental studies show that when one has
to use inflection on every form [-Bare-stem parameter], children at early
syntactic stages of development seem to project only one inflected form for
each word and only produce that inflected form (as the default). This default
is presumed to be based on frequency. For instance, it has been reported that
children may (i) mark the most salient affix, say number on nouns, (ii) but
then still confuse how to properly project number on each noun type for e.g.,
gender and case. It is suggested herein that the former affix process (i)
should be thought of as a kind of lexical incorporation, showing no true rule
process, with the latter paradigm process (ii) being viewed as constituting a
true rule-driven process.
[16] Retuning
to general questions of theory
Returning
then to general questions of theory, Chomsky (1999) suggests that there is good
reason to posit a set of universal features along with principles that make up
the Language Faculty. The problem facing the child may be how to build-up
(bottom-up) such universal features into a lexicon (see fn. 8). If indeed a
bottom-up assumption underpinning the processes of lexical construction is correct
(i.e., that the knowledge of morpho-phonology precedes and triggers syntax,
much in the spirit of Distributional Morphology, or the Lexical Learning
Hypothesis), then there is no reason to suppose that the assembling of features
does not follow an incremental process, potentially governed by specific
maturation of cognitive complexity. In other words, some features might be
acquired before others based on their internal conceptual complexity. For
instance, we might hypothesize that the internal complexity of the features [PER(son)]
might be more complex than the feature [DEF(initeness)] since DEF may contain
some amount of cognitive relevance at LF). (This range of complexity can have
large consequences for overt checking. For example, under Guasti and Rizzi's
(2001) model of spell-out, it is assumed that differences between overt vs.
covert checking directly corresponds to where in the position of the syntactic
tree a phrase projects. Those lower positions which are dominated by the VP may
include only +Interpretable features, and so may survive without forcing overt
movement to a higher projection. Higher positions dominated by IP conversely
force such overt movement for the required checking of functional
-Interpretable features.)[10] Hence, the agreement of PER as well as the agreement
of structural CA might burden an otherwise primitive computational capacity for
a young child at our stage-1 of language development. However, this can't be the whole story since some language
groups manifest very early functional categories/features in child development.
It seems the cognitive complexity of features must work in tandem with the
broader implications of morpho-syntactic typology subsumed under the matrix
[+/-INFL]=>[+/-Bare Verb stem]. More concretely, it goes without saying that
not many linguists would wish to posit features as containing different
saliency values from one language to another. Formal features are part and
parcel of a universal language faculty, a C(omputation) built-up on the H(uman)
L(anguage) cognitive template (sometimes labeled as CHL).
Hence, all properties of feature complexity must be in the first instance
universal. It is then in the second instance, where formal features overlap
onto [+/-INFL] language typology, that we can ultimately determine whether or
not the given language will exhibit early functional projections in child
language.
[17] As
a follow up to ¤¤15 and 16 above, one of the many things that remain a puzzle
to my mind is the following question: Why is it that early child English allows
for [-FIN] bare verb stems to occur in Wh-Questions (What he eat?)? This
structure has only been reported for Child English and, I am aware, has never
been attested in other cross-linguistic child studies. It begs the question:
What makes the English bare stem so special in this respect, how does it differ to other language infinitives (e.g., root
infinitives), and how does it function in the syntax? Of course, one way to
react to such anomalies is to simply reconsider it as a [+FIN] bare verb stem,
with only the phonological affix missing.[11] If this is at all possible, then it begs a larger
question: Can functional material be stored and represented in two
fundamentally different ways? It would seem language typology demands it. Surely,
the PF-overt vs. LF-covert movement analogy captures actually this. So, we may
go on to assume that such an 'affix-less' verb nevertheless projects
inflectional material (T/AGR), at least in light of the fact that there is
nominative subject case agreement (He).
One means to create a dialogue about the difference is to set up a heuristic procedure. The idea being if
we can somehow capture the properties behind English bare stems, we might be
better able to understand the overall syntax of (Child) English and throw some
light onto how English morphology fundamentally differs to other language
groups, at least in this one small respect.
[18] The
two-point heuristic procedure takes the following shape (inquiring into
continuity between the child and adult grammars):
Question
#1: Complexity of Features:
Q:
Is there Continuity of Features?
•
If yes, see question #2. (We take this to mean that T/AGR
features project from the very earliest MLU).
•
If no, then a Discontinuity model is appropriated.*
*
(Since English is a [+Bare Stem] Language, it can simply exhibit bare
verbs stripped of its formal features. Nothing hinges on the fact that that it
exists within a Wh-question (an otherwise functional projection).
What then must be looked at is the pending nature of the Wh-Question (noting
the absence of an Auxiliary) not the bare verb, as Wh-operators equally can be
expressed at PF without exhibiting the complete range of formal features
attributive to CP. I understand the assessment fails to deal with optional
projections).
[19] Typically,
the notion of +/-Interp(retable) features comes into play in helping to account
for the asymmetric chronological onsets of the relevant features and functional
projections. It is suggested that [+Interp] features provide semantic material
at LF interface while [-Interp] features must be functionally checked and
erased at LF (Chomsky 1995). In line with the arguments in this paper, it could
also be suggested that the two types of features likewise get processed and stored
in two different regions of the brain-whereby [+Interp] takes on 'strong-stem'
properties and [-Interp] takes on 'weak-stem' properties (See ¤24 below
regarding +/-Interp features in child language acquisition.)
[20] Question
#2: Complexity of Paradigm:
Q:
Is there Continuity of the Inflectional Paradigm?
•
If yes, morpho-syntactic typology must be considered, triggering the
Dual Mechanism Model as a possible means of accounting for the
continuity. (We take it here that some amount of feature projection
may in all actuality be governed by phonological constraints
on the language, such as cliticization, lexical incorporation,
etc.)
•
If no, Then a discontinuity model is appropriated.
In
sum, we take it that Discontinuity is the Null Hypothesis, and that any
evidence to the contrary must be reconsidered in relation to that language's
morpho-syntactic typology. The exact role of the typology here will be made
clear below.
[21] Some
Data
Evidence
for a Discontinuity model is striking ('negative' to Question #1 ¤18). For
instance, Radford and Galasso (1998), Galasso (1999, 2003b) Radford (1999,
2000) provide English data showing that children enter into a 'No Agreement' /
'No Inflection' initial stage-one of acquisition during which they completely
omit functional categories and [-Interp] complex features.
[22] Stage-1:
'No AGReement-No INFLection' (Radford & Galasso 1998)
Possessives:
That Mommy car. Me dolly. No
baby bike. Him name.
Have
me shoe. *Iwant me
bottle. It me.
Question: Where
Daddy car? This you pen? What him doing?
Declarative: Baby have bottle. Car
go. Me wet. Me playing. Him dead
*(Iwant examples
are analyzed as formulaic chunking,
since no other supportive material providing for a functional analysis of
nominative case is found in the relevant stage).
[23] Stage-2:
'OPtional AGRement -INFLection'*
Possessives:
That's Mommy's car. My dolly.
Baby's bike. His name.
Question: Where's
Daddy's car? This is your pen? What (is) he doing?
Declarative: Baby has bottle. Car
goes. I'm wet. I'm playing. He's dead.
*(The OI stage (as suggested by Wexler 1994) would
simultaneously incorporate both data sets as described in his initial Optional
Infinitive stage-1). Radford & Galasso make a clear demarcation between the
two stages, with the complete absence of any optional functional projections
for their stage-1. For complete data/analyses, see Galasso 2003c).
[24] Overall,
children in this initial stage-one of syntactic development are forced into
projecting very limited structure. For instance, (and this is by no means an
exhaustive list):
(i)
Possessive projections, which rely on an AGReement relation with a nominal
INFL, must default to an objective case (e.g. my to me);
(ii)
Verb projections are limited to VPs without INFLection (hence auxiliary-less
question and declarative bare verb stems) (e.g. What him doing?,
Car go.);
(iii)
Subjects, which rely on an AGReement with a verbal INFL, must default
to having an objective case (e.g., Me wet). Consider the syntactic structures
below pairing the two data sets, with stage-one showing no inflectional
phrase (IP) agreement.
[25] Structure:
Stage-One / -AGR Structure:
Stage-Two /+AGR
(i) Possessive: *
[IP Mummy [I {-agr}-ø] car] [IP
Mummy [I {+agr}'s] car]
[IP Me [I {-agr}] dolly] [IP
My [I {+agr}] dolly]
(ii)
Case: [IP Him [I {-agr}] dead] [IP
He [I {+agr}'s] dead]
[IP Me
[I {-agr}] wet] [IP
I [I {+agr} 'm] wet]
(iii)
Verb: [IP Baby [I {-agr} have]]... [IP
Baby [I {+agr}has]]...
[IP Car [I {-agr}go -ø]] [IP
Car [I {+agr} go-es]]
*(Though Radford and Galasso do label their stage-one
structures as IP {-agr} in the 1998 paper, this is mostly notational: see in-note
in ¤26 below).
[26] Similar
findings are born out and widely attested in the literature and are consistent
with the general notion that language acquisition involves some sort of incremental
feature-building (Radford 2000)-viz.,
the notion that if language does proceed in an incremental way, then it should
be of little surprise that the more robust and primitive aspects of a language
should come on-line before more abstract aspects of language-specifically,
default {-agr} feature projections attributed to the VP (by default) come
on-line before {+agr} projections attributed to the IP. (In-note: Though Radford and Galasso do label their stage-one
structures as IP {-agr} in the 1998 paper, this is mostly notational: they are
clear in stating that this in no way should undermine the fact that we are
indeed dealing with a lexical stage-one, with IP remaining Non-specified {¯}
(and for all intents and purposes, not projecting). Galasso (2003) describes
such early stage-one structures as reduced to simple lexical VPs and NPs). This
gives us the flavor of saying that a maturational scheduling is behind the chronological ordering of features (much
in the spirit of the Brown studies (1973) which sought to show a time-line of
affix morpheme development-moving from potentially viable semantic-based
participle forms {en}, {ing} through to true rule-based inflectional forms 3PSg
{s}, Possessive {s}, Past Tense {ed}. Where the opinions of the cited authors
¤¤1,3 tend to diverge, however, is in how to appropriately and accurately
describe such apparent lack of child linguistic knowledge. Given this
divergence, arguments spring up as to whether or not there is really any claim
for discontinuity at all (as based on the research paradigm on offer (Chomsky
1995, 1999)). One might think that the very fact that the child doesn't
produce/generate target 'adult-like' grammar at the initial stage of syntax
should be enough in of itself to say that discontinuity is an accurate
depiction of child language development.
[27] Well,
some of us take maturation more seriously than others. For the sake of
concreteness, let's spell this out. The 'strong maturational' theorist believes
that there is indisputable evidence for some kind of a biologically determined
stage during which functional categories, along with their class of functional
words, parameter settings and formal [-Interp] features, are all together
absent in the child's grammar. In these terms, 'strong maturation' equates to
'discontinuity'. The 'weak maturational' theorist might be more prone to skirt
such biological issues and stake their arguments around interpretations of the
theory. For instance, one way around the strong maturational/discontinuity
account would be to say that all formal categories/features are available to
the child from the very beginning.[12] The apparent discontinuity is then reduced to the
mere phonological spell-out of yet un-set parameter settings of categories
and/or un-specification of features.
[28] Hence,
there are two schools-of-thought of how such omissions of structure should be
explained:
(i)
Full Competence =>
(weak maturation-continuity)
(ii)
Gradual Competence =>
(strong maturation-discontinuity)
Full
Competency in this respect would include any developmental model which credits
the very young child with at least knowing the functional projection
notwithstanding the fact that the features of the projections can go
unspecified or be phonologically null (cf. Wexler's initial OI-stage 1994).
Gradual
Competency in this respect would include any developmental model which holds
off crediting the very young child with knowing any functional projection that
doesn't at least make itself phonological present in the form of functional
categorical words (Case/Pronouns, Auxiliaries) (cf. Radford & Galasso's
initial No-INFL stage 1998).
[29] Full
competence hypotheses state that all functional categories/features are present
in the child's linguistic system (this equates to a child-to-adult continuity).
The child's otherwise non-target grammar is thus the result of un-specification
of features that are nonetheless present in the child's phrase structure. In
other words, the child's otherwise lexical non-functional stage-1 of
grammatical development is thought of as maintaining the full-fledge IP>VP
phrase structure for typical declarative sentences, non-specification of
features notwithstanding, hence, the early German and French data cited above.
In English however, the child may not spell-out the Phonological Form of
articulation (at PF) of such features-for instance, the dual
AGR(eement)/T(ense) feature (which triggers the inflectional affix {-s} for
3person/singular/present) on the main finite verb goes un-specified and
therefore doesn't project. In this sense, a non-inflected bare verb-stem
surfaces as the default: e.g., (The) baby sleep. What this assumption basically claims is that a child
may in fact be producing functional (affix) material all the while despite the
fact that such material is phonologically silent (following assumptions laid
out in Halle and Marantz 1993). Recall, such a treatment was suggested
regarding Bare verb stems which show subject nominative case in embedded
Wh-Questions ¤17. This very real possibility that functional
categories/features can be present though phonologically silent changes the
entire landscape of how maturational theories of language acquisition can be
put to the test. The very idea that +affix material can be present in a grammar without actually being spelled-out in the phonology requires developmental linguists to
revisit their discontinuity theories to the extend that newer models need to be
devised in order to establish whether or not a child at the earliest MLU stage
of development has functional grammar.
[30] A
new condition for a non-functional stage-1
The argument for potential underspecification of features can be naturally expanded to suggest that there ar