137 REFLECTION ON THE AGREEMENT AND TENSE OMISSION

advertisement
REFLECTION ON THE AGREEMENT AND TENSE OMISSION MODEL OF
SPECIFIC LANGUAGE IMPAIRMENT: A CORPUS-BASED STUDY*
YI-AN LIN
Abstract
On the basis of the Leonard corpus in the Child Language Data Exchange System, the
present paper tries to evaluate Wexler, Schütze & Rice’s (1998) two-factor account of
specific language impairment (SLI), the Agreement and Tense Omission Model
(ATOM), and to figure out the nature of syntactic errors made by SLI children. The
result shows that the SLI children in the Leonard corpus did experience difficulties in
the use of the third-person, singular, present tense –s and the preterite verb forms as
predicted by ATOM. However, it is found that these SLI children mark tense better
than agreement. ATOM cannot explain such a discrepancy. In addition, it is
shown that case marking is unimpaired in these children’s grammars. Therefore, an
alternative account which claims that nominative case is assigned by an interpretable
mood feature on T is adopted in order to account for the findings here. Moreover, it
is found that these SLI children do not have any problem with A-movement. This
suggests that their agreement marking errors are not due to the underspecification of
the agreement feature on T as proposed by ATOM. They may instead be merely
spellout errors. By contrast, it was shown that these SLI children’s problems with
auxiliary inversion may be due to the underspecification of the tense feature. Hence,
this indicates that their tense marking errors are caused by the underspecification of
the tense feature on T. However, the proposal that the interpretable tense feature
may sometimes be underspecified requires the functional head T to have another
interpretable feature: mood. Such an amendment prevents the derivation of
sentences from crashing at semantic interface. In summary, the primary deficit in
SLI children is shown to be the specification of the tense feature, and this is arguably
due to the fact that tense is a conceptually complex notion.
1. Introduction
Specific Language Impairment (SLI), also known as developmental dysphasia,
refers to a delayed or deviant language development of children in the absence of
neurological trauma, cognitive impairment, psycho-emotional disturbance, or
motor-articulatory disorders (Eisenbeiss, Bartke, & Clahsen 2005). Due to the fact
that SLI is a heterogeneous disorder, the identification of SLI is usually based on a set
of exclusionary criteria. As Leonard (1998) mentioned, there are some primary
criteria for the diagnosis of SLI. Namely, language test scores are of -1.25 standard
*
This paper is based on the third chapter of my 2005 master’s dissertation submitted to University of
Essex. Therefore, I am grateful to my supervisor, Prof. Andrew Radford, for his guidance on the
analysis presented here. I also thank the audience at the 1st Newcastle Postgraduate Conference in
Theoretical and Applied Linguistics and the anonymous reviewer for the valuable comments and
suggestions. All the remaining errors are my own responsibility.
137
Lin
deviations from the mean, the nonverbal performance IQ is 85 or above, the hearing
passes screening at conventional levels, no recent episodes of otitis media are known,
no neurological dysfunction is shown, oral motor structure and function are both
intact, and the pattern of social interaction is normal.
However, language impairment is the defining feature of SLI. Individuals
with SLI have been reported to experience difficulties with most or all areas of
language, including vocabulary, morphosyntax, phonology, and pragmatics.
Nevertheless, these areas are not equally affected. In English, vocabulary and
pragmatic skills are less deficient than phonology and morphosyntax (Leonard 2004).
For theoretical linguists and psycholinguists, the primary deficit in SLI relates to the
acquisition of morphosyntax. As a result, the aim of linguistics-oriented studies of
SLI is to characterize the morphosyntax of individuals with SLI and the difference
between SLI and typically-developing (TD) children in the framework of current
grammatical theories or theories of grammatical development (Eisenbeiss, Bartke, &
Clahsen 2005).
This paper is especially concerned with the nature of syntactic errors made by
SLI children, which involve core syntactic operations such as movement, agreement
and case-marking. As a result, a two-factor account of SLI, the Agreement and
Tense Omission Model (ATOM), is evaluated on the basis of a corpus of spontaneous
speech production data, the Leonard corpus in the Child Language Data Exchange
System (CHILDES), which was collected before the formulation of the ATOM. In
the next section, the key claims and predictions made by ATOM are presented.
2. Literature Review on the Agreement and Tense Omission Model
The Agreement and Tense Omission Model of SLI originates from Wexler’s
(1994, 1996) studies on clause structure and inflection in young TD children. The
grammar of TD children can be captured in terms of Wexler’s Optional Infinitive (OI)
stage. He claims that TD children undergo a protracted stage during which they
alternate between producing finite and infinitive forms of verbs in contexts where
finite forms are required in adult grammar. Additionally, TD children at the OI
stage tend to omit auxiliaries and copula BE1 in finite contexts as Wexler observed.
Further, Wexler asserts that optional infinitives arise because of the
underspecification or omission of the tense feature in the clause representation.
Adopting Wexler’s idea, Rice, Wexler & Cleave (1995) propose that SLI
children go through an Extended Optional Infinitives (EOI) stage in which a similar
pattern of optional infinitives is found until they are 7 or 8 years of age. Rice and
her collaborators propose that the grammar of SLI children is characterized in terms
of the tense omission model. They predict that SLI children will display limited
proficiency in the use of morphemes marking tense while leaving other inflectional
morphemes unaffected. Additionally, when these children do specify the tense
feature, they respect all its morphosyntactic properties. That is, when SLI children
use tense morphemes or auxiliaries, they use them correctly in the same way as TD
children do. Nevertheless, such a tense-deficit hypothesis wrongly predicts that SLI
children never use accusative subjects with past tense verbs.
1
In this paper, the capitalized word is the label for various forms of that word.
138
Lin
In later work, Schütze & Wexler (1996) claim that optional infinitives can
result from the underspecification or omission of either the tense or agreement
features (or both) in a child’s structural representation of a sentence. As a result,
Wexler, Schütze and Rice (1998) revise their account of SLI with a two-factor theory.
They assume that SLI involves a syntactic feature deficit leading to the optional
specification of tense [TNS]2 and agreement [AGR] features in obligatory contexts.
They refer to their model as ATOM. Further, they argue that the case-marking of
subjects as nominative (Nom) or non-Nom by SLI children directly correlates with
whether or not the agreement is marked on verbs. In other words, the case-marking
difficulties of SLI are a secondary effect of problems with verb agreement.
From their study, Wexler and his collaborators report that English SLI children
(ESLI children) are more likely to use non-Nom subjects and to produce non-Nom
subjects at a much later age than TD children. Additionally, they find that both TD
and the SLI children virtually always correctly mark the case of object. Moreover,
their data show that the optional occurrence of Nom and non-Nom subjects is
correlated with the optional occurrence of finite and non-finite verb forms. More
specifically, they point out that non-Nom subjects almost never appear with agreeing
auxiliary/main verbs that are inflected for both tense and agreement. Therefore,
Wexler and his colleagues take the subject case errors as a reflection of the OI stage.
In summary, Wexler and his collaborators draw the following main
conclusions from their study:
(i) SLI children make more frequent case-marking errors with subjects than
MLU-matched TD children
(ii) SLI children use an accusative (Acc) form of subjects in contexts where adults
require a Nom form
(iii) the case-marking of subjects is determined by whether clauses contain an
auxiliary/main verb3 which agrees in person and number with the subject or not
(iv) SLI children may leave verbs underspecified for [TNS] and/or [AGR] features in
finite contexts, resulting in the production of OI structures with Acc subjects
(v) ATOM makes predictions about the relative frequency of case and
tense/agreement errors made by children in different types of clause structure (e.g.
with subject case-marking errors predicted to be most frequent with uninflected verbs,
less frequent with ambiguous verbs, and least frequent with agreeing verbs).
However, ATOM makes no specific claims about whether the [TNS] feature is
omitted more frequently than the [AGR] feature – or conversely.
3. Results and Discussion
Tense Marking and Agreement Marking
Since ATOM argues that SLI involves an impairment of tense marking, it
predicts that SLI children may sometimes leave verbs underspecified for the [TNS]
feature in finite contexts. It is found that all eleven children in the Leonard corpus
2
3
By convention, features are enclosed in square brackets and often abbreviated.
More precisely, an INFL/Agr constituent
139
Lin
indeed commit tense marking errors. Their mean correctness score for tense
marking on main lexical verbs is 53.87%. In more detail, five of these children
display limited proficiency in the specification of the [TNS] feature on main verbs.
Each of the five children’s percentage specification of the [TNS] feature is at chance
level. Furthermore, it is found that there is a significant positive correlation between
the MLU and the child’s performance on past-tense marking (r=.73, p=.005,
one-tailed 4 ). The children with a higher MLU tend to have higher percentage
specification of the [TNS] feature. The existence of such a positive correlation may
indicate that the SLI children in the Leonard corpus have a delayed acquisition of
tense marking.
Since ATOM also claims that SLI involves an optional specification of the
[AGR] feature in obligatory contexts, it predicts that SLI children will experience
difficulty in the use of the third-person, singular, present tense (3SgPres) –s and the
use of inflected forms of auxiliaries. It is found that all eleven children in the
Leonard corpus frequently use bare verbs in contexts requiring the 3SgPres –s, and
nine of these children show use of the 3SgPres –s at chance level. Additionally, all
eleven children commit agreement marking errors on auxiliaries, and six of them
mark the [AGR] feature on auxiliaries at chance level. Furthermore, the mean
percentage group scores for specification of the [AGR] feature on main verbs and
auxiliaries are 30.76% (SD=21.35) and 38.15% (SD=27.79), both below chance level.
However, unlike the specification of the [TNS] feature, there is no significant
correlation between the MLU and the child’s performance with agreement marking
on either main verbs or auxiliaries. The non-existence of such a significant
correlation may suggest that the SLI children in the Leonard corpus have a deficit in
the acquisition of agreement marking.
Moreover, there is a statistically significant difference between these
children’s performance with the specification of [TNS] and [AGR] features on main
verbs (t(10)=3.26, p=.009, two-tailed5). The SLI children in the Leonard corpus
underspecified the [AGR] feature much more frequently than the [TNS] feature.
However, ATOM makes no specific claims about such a difference. There are two
suggested explanations for this finding. One is that the agreement between the
functional head T of TP and its subject in Spec-TP involves two features, namely the
person [PERS] and number [NUM] features. Yet these two features are treated as
one single composite [AGR] feature within the ATOM account (Radford 2005a).
Since agreement involves valuation of two features, it may be a more difficult process
for these SLI children, since tense marking is only controlled by one single [TNS]
feature. The other explanation as suggested by Clahsen, Bartke & Göllner’s (1997)
Agreement Deficit Model is that the [TNS] feature on T is an interpretable feature
while the [AGR] feature on T is an uninterpretable feature, which causes major
problems for SLI children.
Nevertheless, a further theoretical problem with ATOM is that it assumes that
the head T, which contains only the uninterpretable [AGR] feature and the
interpretable [TNS] feature, can be underspecified for the [TNS] feature. Such an
assumption violates Chomsky’s (1995) claim that syntactic structures must not
4
5
The data is analysed in the Pearson correlation using SPSS software.
The data is analysed in a paired-samples t test using SPSS software.
140
Lin
contain heads to which no semantic interpretation can be assigned. Since the
interpretable [TNS] feature on T provides information about the time at which an
event takes place, T cannot be assigned any semantic interpretation if the [TNS]
feature is not specified, which causes the derivation to crash at the semantic interface
(Radford 2005a). An alternative account, suggested by Radford and Ramos (2001)
for ATOM and the omission of tense marking found in SLI grammar, is that T carries
an interpretable feature other than the [TNS] feature, namely a mood [MOOD] feature.
This feature correlates with the Nom case marking, and this will be discussed in the
next section. That the [TNS] feature is underspecified but not the [MOOD] feature
will not result in a crash at the semantic interface.
Additionally, the notion of underspecification adopted in ATOM leads to a
developmental question. That is, which feature carried by T is mastered before the
other one if either [TNS] or [AGR] (or even both) features can be underspecified? A
suggested answer to such a question is that tense is acquired before agreement since
the [TNS] feature is interpretable and the [AGR] feature is uninterpretable. This
answer is supported by Ingham’s (1998) longitudinal study of a single child (Radford
2005a).
Tense Marking, Agreement Marking and Case Marking
Since ATOM asserts that the case-marking of subjects is determined by
whether clauses contain an auxiliary/main verb which agrees in person and number
with the subject or not, it predicts that (i) Nom subject pronouns always co-occur
with verbal forms specified for agreement and that (ii) the subject pronoun will be in
the (Acc) default form with verbal forms not specified for agreement.
However, it is shown that the SLI children in the Leonard corpus seldom use
non-Nom subjects in their spontaneous speech, and the mean percentage suppliance
of the correct Nom case is 97.26% (SD=2.67). These figures show that these
children’s performance on case marking is almost adult-like. This suggests that the
few case-marking errors found among the SLI children in the Leonard corpus may be
case spellout errors rather than case assignment errors.
If the Nom case of subjects is indeed assigned by the specification of [AGR]
feature as claimed by ATOM, the overall high percentage suppliance of the correct
Nom case found in the Leonard corpus indicates that a high percentage suppliance of
agreement marking should be found among these SLI children. Nevertheless, this
expectation is not borne out by the data presented. One way for ATOM to deal with
this is to say that these SLI children have problems with verb morphology – i.e.
with retrieving the s-inflections or irregular s-inflected forms rather than with
marking agreement in the syntax. This is the core assumption of Gopnik and
Crago’s (1991) Rule Deficit Model. However, an alternative Mood-and-Agreement
model which argues that Nom case is assigned by a [MOOD] feature on T can also
account for the data in this study. If the few case-marking errors are indeed case
assignment errors but not spellout errors, it could be due to the underspecification of
the [MOOD] and/or [AGR] feature(s) within the Mood-and-Agreement model. A
deficit in specification of [TNS] and/or [AGR] feature(s) in SLI grammar will not
cause any problem for this alternative account.
On the other hand, ATOM successfully predicts that SLI children will not have
141
Lin
any problem with the case marking on personal pronoun objects. A 100% of
adult-like case marking on personal pronoun objects is found for all eleven SLI
children in the Leonard corpus.
Agreement Marking and A-movement
Although ATOM does not make any specific prediction about SLI children’s
performance with A-movement, it claims that they frequently underspecify the [AGR]
feature in obligatory contexts, which should result in frequent A-movement errors,
since Chomsky (1998, 1999) proposes that agreement plays an important role in
A-movement. In more detail, an active Probe in T by virtue of its uninterpretable
and unvalued [PERS] and [NUM] features searches for an active nominal Goal (by
virtue of its uninterpretable and unvalued [CASE] feature) to match and delete its
[PERS] and [NUM] features and to delete its uninterpretable [EPP]6 feature by
movement. Additionally, Chomsky claims that the [EPP] feature on T can attract an
expletive such as there which carries only the [PERS] feature. Hence, he assumes
that T attracts a constituent agreeing with T in at least one of its φ-features. In other
words, the movement of the nominal from Spec-vP, Spec-VP or VP-complement to
Spec-TP requires T to attract the closest nominal with which it agrees in respect of all
φ-features carried by the nominal.
However, it is found that the SLI children in the Leonard corpus do not
commit any word order errors in obligatory contexts where A-movement is required.
These children correctly move subjects to the Spec-TP position even when they
commit other errors such as tense/agreement marking errors and auxiliary-omission
errors in the sentences below:
(1)
(2)
(3)
(4)
(5)
‘It not work’ (Child B)
‘He not clean’ (Child G)
‘It not working’ (Child H)
‘This be not the doctor’ (Child H)
‘It not this one’ (Child J)
According to Chomsky’s claim, the observation that these SLI children do not
have any problem with A-movement suggests that they specify person and/or number
agreement between T and the moved subject. Moreover, since these children
correctly assign Nom case to the subjects, it indicates that agreement is fully specified.
However, agreement marking errors and auxiliary-omission errors are found in the
sentences above. ATOM cannot provide an explanation for these findings. An
alternative account is that all these agreement marking errors and auxiliary-omission
errors are merely spellout errors. The agreement marking errors are not due to the
underspecification of the [AGR] feature (or the [PERS] and [NUM] features).
6
It is ‘originally an abbreviation for the Extended Projection Principle, which posited that every T
constituent must be extended into a TP projection which has a specifier. In more recent work, the
requirement for a T constituent like will to have a specifier is said to be a consequence of T carrying an
[EPP] feature requiring it to project a specifier. The EPP Generalisation specifies the conditions under
which the [EPP] feature carried by a head is deleted via use of an expletive or movement’ (Radford,
2004: 450).
142
Lin
Additionally, this account indicates that these SLI children’s low percentage
suppliance of the suffix –s in obligatory contexts may be the result of a problem
retrieving the regular s-inflection or irregular s-inflected forms.
Tense Marking, Agreement Marking and Auxiliary Inversion
Although ATOM does not make any specific prediction about SLI children’s
performance with auxiliary inversion, it maintains that SLI children sometimes
underspecify [TNS] and/or [AGR] feature(s) on T, which should lead to problems
with auxiliary inversion, since current thinking within Minimalism suggests that
auxiliary inversion is triggered by either an affix on C carrying a [TNS] feature
requiring it to attract the closest head marked for tense (Pesetsky & Torrego 2001,
Radford 2004) or an affix on C carrying a set of strong agreement features requiring it
to attract the closest head specified for agreement (Radford 2005b). More
specifically, if tense is the factor triggering auxiliary inversion, these children’s
performance on auxiliary inversion is expected to reflect their performance on tense
marking. If agreement is the factor triggering inversion, their performance on
auxiliary inversion should have the same frequency as agreement marking and Nom
case assignment, for Chomsky (2005) claims that C hands over its agreement
properties to T so that T must be underspecified for agreement and leads to a
non-nominative subject if the agreement features on C are not specified.
It is interesting to explore the extent to which the Leonard children’s
performance on auxiliary inversion can be explained in terms of ATOM and the
theories for the mechanism of auxiliary inversion. It is found that the overall
frequency with which these children supply auxiliary inversion (52.45%) is more
similar to their tense suppliance rate (53.87%) than to their agreement suppliance rate
(34.78%) and nominative case suppliance rate (97.26%). Additionally, according to
the finding in the previous section, the [AGR] feature seems to be unimpaired in
these SLI children’s grammars. Therefore, their problems with auxiliary inversion
may be due to the underspecification of the [TNS] feature.
4. Conclusion
The Agreement and Tense Omission Model for syntactic errors produced by
SLI children has been evaluated on the basis of the Leonard corpus in the CHILDES.
In the previous section, it was shown that the SLI children in the Leonard corpus did
experience difficulties in the use of the 3SgPres –s and the preterite verb forms as
predicted by ATOM. However, it was found that these SLI children mark tense
better than agreement.
This discrepancy cannot be explained by ATOM. In
addition, it was shown that case marking is unimpaired in these children’s grammars.
Therefore, an alternative account which claims that Nom case is assigned by an
interpretable [MOOD] feature on T was adopted in order to account for the findings.
Furthermore, it was found that these SLI children do not have any problem with
A-movement. This suggests that agreement marking errors are not due to the
underspecification of the [AGR] feature on T as proposed by ATOM. They may
instead be merely spellout errors. By contrast, it was shown that these SLI
children’s problems with auxiliary inversion may be due to the underspecification of
143
Lin
the [TNS] feature. Hence, this indicates that their tense marking errors are caused
by the underspecification of the [TNS] feature on T. However, the proposal that the
interpretable [TNS] feature may sometimes be underspecified requires the functional
head T to have another interpretable feature: [MOOD]. Such an amendment
prevents the derivation of sentences from crashing at semantic interface. In
summary, the primary deficit in SLI children is the specification of the [TNS] feature,
and this is arguably due to the fact that tense is a conceptually complex notion.
References
Chomsky, N. (1995). The minimalist program. Cambridge, MA: MIT Press.
Chomsky, N. (1998). Minimalist inquiries: the framework. In Martin, R., Michaels, D.
& Uriagereka, J. (eds.), Step by step: essays on minimalism in honor of Howard
Lasnik. Cambridge: MIT Press. 89 – 155.
Chomsky, N. (1999). Derivation by phase. In Kenstowicz (ed.), 1 – 52.
Chomsky, N. (2005). On phases. Ms., MIT.
Clahsen, H. (1989). The grammatical characterization of developmental dysphasia.
Linguistics 27. 897 – 920.
Clahsen, H., Bartke, S., & Göllner, S. (1997). Formal features in impaired grammars:
A comparison of English and German SLI children. Journal of Neurolinguistics
10. 151 –171.
Eisenbeiss, S., Bartke, S., & Clahsen, H. (2005). Structural and lexical case in child
German: evidence from language-impaired and typically-developing children.
Language Acquisition 13. 3 – 32.
Gopnik, M., & Crago, M. (1991). Familial aggregation of a developmental disorder.
Cognition 39. 1 – 50.
Ingham, R. (1998). Tense without agreement in early clause structure. Language
Acquisition 7. 51 – 81.
Kenstowicz, M. (ed.) (2001). Ken Hale: A life in language. Cambridge, MA: MIT
Press.
Leonard, L. (1998). Children with specific language disorders. Cambridge, MA: MIT
Press.
Leonard, L. (2004). Specific language impairment in children. In Kent, R. (ed.), The
MIT encyclopedia of communication disorders. Cambridge, MA: MIT Press.
402 – 404.
Pesetsky, D., & Torrego, E. (2001). T-to-C movement: Causes and consequences. In
Kenstowicz (ed.), 355 – 426.
Radford, A. (2004). Minimalist syntax: exploring the structure of English. Cambridge:
Cambridge University Press.
Radford, A. (2005a). Children’s English: principles-and-parameters perspectives. Ms.,
University of Essex, Colchester.
Radford, A. (2005b). English syntax revisited: an introduction. Ms., University of
Essex, Colchester.
Radford, A., & Ramos, E. (2001). Case, Agreement and EPP: evidence from an
English-speaking Child with SLI. Essex Research Reports in Linguistics 36.
42 – 81.
144
Lin
Rice, M., Wexler, K., & Cleave, P. (1995). Specific language impairment as a period
of extended optional infinitive. Journal of Speech and Hearing Research 38.
850 – 863.
Schütze, C. & Wexler, K. (1996). Subject case licensing and English root infinitives.
In Stringfellow, A., Cahana-Amitay, D., Hughes, E. & Zukowski, A. (eds.),
Proceedings of the 20th Annual Boston University Conference on Language
Development (Vol. 2). Somerville: Cascadilla Press. 670 – 681.
Wexler, K. (1994). Optional infinitives, head movement and the economy of
derivations in child grammar. In Lightfoot, D. & Hornstein, N. (eds.), Verb
movement. Cambridge: Cambridge University Press. 305 –350.
Wexler, K. (1996). The development of inflection in a biologically based theory of
language acquisition. In Rice, M. L. (ed.), Toward a genetics of language.
Mahwah: Erlbaum. 113 –144.
Wexler, K., Schütze, C. & Rice, M. (1998). Subject case in children with SLI and
unaffected controls: evidence for the Agr/Tns Omission Model. Language
Acquisition 7. 317 – 344.
Yi-An Lin
Department of Linguistics
Faculty of Modern and Medieval Languages
University of Cambridge
Sidgwick Avenue,
Cambridge, CB3 9DA
United Kingdom
yal23@cam.ac.uk
145
Download