Transcript FULL TEXT - Language and Cognitive Neuroscience Lab at UW

The Role of Morpho-Phonological Factors in Agreement Production:
When Singular and Plural are Both Grammatical
http://lcnl.wisc.edu
Jelena Mirković and Maryellen C. MacDonald
Language and Cognitive Neuroscience Lab, University of Wisconsin-Madison
Theories I: Grammatical Features
A central debate in language production research is the extent to which the computation of noun-verb
agreement is an autonomous syntactic process or whether it is constrained by non-syntactic factors (e.g.
Bock et al. 2001; Thornton & MacDonald, in press). The primary data for these alternatives have been the
rates of agreement errors in fragment completion task (a speaker completes a complex noun phrase like
"the key to the cabinets" with a verb that agrees with the local noun "cabinets" rather than the head noun
"key"). Evidence for non-syntactic influences on agreement is mixed in these studies. Recently several
researchers have identified constructions in which several grammatical options are available (Haskell &
MacDonald, submitted; Hemforth & Konieczny, 2002). These constructions are potentially quite
informative, because subtle non-syntactic effects may be more evident here than in cases where only one
agreement pattern is grammatical. We investigate another case of this sort, subject-verb number agreement
with certain quantifier phrases in Serbian such as "five cows", for which both singular and plural verbs are
grammatical.
Introduction
Number Marking and Number Morphing (Bock et al., 2001)
Message Formulation
Grammatical Encoding
NP
In the case of attraction errors, the
blocking mechanism failed.
VP
are
NP1
PP
Det
N
P
the
key
to
NP2
Det
the
N
Quantifier + Noun:
noun case depends on the quantifier
masculine:
1 konj-NOM.SG.
1 horse
trči-SG.
runs
2, 3, 4 krave-NOM.PL.
2, 3, 4 cows
2, 3, 4 konja-DUAL
2, 3, 4 horses
trče-PL.
run
5, 6, ... krava-GEN.PL.
5, 6, ... cows
5, 6, ... konja-GEN.PL.
5, 6, ... horses
trče-PL./trči-SG.
run/runs
cabinets
krava-GEN.PL.=NOM.SG.
Hierarchical Feature Passing (Vigliocco & Nicol, 1998)
number:
(3) The little girl-SG. is-SG. beautiful.
gender:
(4) La-FEM. niña-FEM. pequeña-FEM. es linda-FEM.
person:
(5) She-3rd.P.SG. sings-3rd.P.SG.
(6) Gledam malu-ACC. devojčicu-ACC.
I’m watching the little girl.
The result of processing is determined by the interaction of multiple
graded, probabilistic constraints (MacDonald et al., 1994; Trueswell &
Tanenhaus, 1994; Haskell & MacDonald, in press; Thornton &
MacDonald, in press)
syntax
semantics
The focus of this study is subject-verb number agreement production
(for example, (1)).
singular
plural
morphology
How to Study Subject-Verb
Agreement Production?
the key to the cabinets
•
•
the key to the cabinets is on the shelf
•
the key to the cabinets-PL. are-PL. on the shelf
head noun
local noun
 the subjects erroneously produce the verb that agrees in number
with the local noun, instead of the head noun
Experiment 1:
Two Forms Grammatical
pragmatics
singular
plural
morphology
5, 6, ... krava-GEN.PL=NOM.SG 5, 6, ... konja-GEN.PL. trče-PL/trči-SG
5, 6, ... horses
5, 6, ... cows
run/runs
- 40 Ss, native speakers of Serbian
- fragment completion task, with preambles with feminine or masculine
nouns (e.g. five cows)
Error-free Subject-Verb Agreement
and is supposed to repeat it and complete the sentence, e.g.
Sometimes the subjects produce attraction errors:
semantics
Competition produces variability in
responses!
Fragment completion task:
The subject is presented with a sentence preamble like
If case homophony and distributional information influence agreement
production, then the proportion of singular verbs with feminine nouns
should be higher than with masculine nouns.
syntax
pragmatics
•
Most of the studies of subject-verb agreement production were
focused on errors!
Recently, however, several constructions where more than one
option is grammatically correct have been studied (Haskell &
MacDonald, 2002; Hemforth & Konieczny, 2002):
 subtle non-syntactic effects may be more evident here than in
constructions where only one agreement pattern is grammatical.
In Serbian (and possibly other Slavic languages) there is a
construction where both singular and plural verb forms are allowed:
Quantifier + NP:
5 krava-GEN.PL.
trče-PL./trči-SG.
5 cows
run/runs
Experiment 1 investigates the role of morphophonological factors
in agreement production in this case.
Both singular and plural
verbs are grammatical!
Genitive plural form of feminine nouns is homophonous with nominative
singular form. Nominative form is the agreement controller in subjectverb agreement.
100
Percentage of Responses
Across different languages, it includes properties such as:
Theories II: Agreement Production as
a Constraint Satisfaction Process
singular verb
plural verb
misc.
80
60
40
20
Results:
• significantly more singular verb
responses in preambles with
feminine nouns
• miscellaneous responses are
equally distributed across
genders (i.e. the effect is not
coming from comprehension
difficulties)
0
Feminine
Experiment 2:
One Form Grammatical
Verb:
feminine:
1 krava-NOM.SG.
1 cow
(1) The bird-SG. sings-SG.
(2) The birds-PL. sing-PL.
case:
S
Phonological Encoding
Agreement is a systematic covariation of the properties of linguistic
elements which indicates a relationship between them.
e.g.
= Number marking Number morphing
Quantifiers in Serbian
Masculine
Normal, error-free agreement production is influenced by
distributional morpho-phonological factors: homophony with
nominative singular form in feminine nouns promotes the use of
singular verbs.
2, 3, 4 krave-NOM.PL.
2, 3, 4 cows
2, 3, 4 konja-DUAL
2, 3, 4 horses
trče-PL.
run
Only plural form
is grammatical!
konja-DUAL .=GEN.SG.
In masculine nouns, the archaic dual form is homophonous with
genitive singular form. Even though genitive form is not the agreement
controller in subject-verb agreement, it serves that function in adjectivenoun number agreement.
If case homophony and distributional information influence agreement
production then the number of singular verb forms (agreement errors)
in masculine nouns should be higher than in feminine nouns.
- 40 Ss, native speakers of Serbian
- same methodology as for Exp. 1 (fragment completion task: three cows)
singular verb
plural verb
misc.
100
Percentage of Responses
Abstract
80
60
40
20
0
Feminine
Results:
• significantly more singular verb
responses in preambles with
masculine nouns
• more plural verb responses in
preambles with feminine nouns
(p=0.088)
• equal distribution of
miscellaneous responses in two
genders
Masculine
Agreement production is influenced by distributional morphophonological factors, which is in this case indicated by the increase
of agreement errors in masculine nouns.
Conclusions
• Subject-verb agreement production can be influenced by morphophonological distributional information, when both one and more than
one grammatical alternatives exist.
• Together with other results showing semantic (Thornton &
MacDonald, in press; Vigliocco et al., 1996) and phonological
(Haskell & MacDonald, in press; Hartsuiker et al, submitted) effects,
our studies show that agreement production is well described in terms
of a constraint satisfaction process.
Please see handout for references.
Thanks to: Laboratory for Experimental Psychology, University of Belgrade, Belgrade, Serbia & Montenegro
This work is supported by NIH Grant P50 MH 64445 and the Vilas Trust, UW-Madison.