Transcript Document
Rules vs. Constructions A debate on question-acquisition
Lucia Pozzan, Lidiya Tornyova & Virginia Valian
IASCL 2011
Special thanks to L anguage A cquisition R esearch C enter Team Margarita Zeitlin Syelle Graves Erin Quirk Paul Feitzinger Nathan LaFave 2
English Main Questions
• Subject-Auxiliary Inversion Declarative: John is eating pizza Yes/no question: Is Wh-question: What John is eating pizza?
John eating? • Children’s questions: lack of inversion
Why my dog is digging a hole?
Katie’s brother is feeding the doll?
3
Our View
• • • Input is important but it is not the only factor Differences in syntactic properties are reflected in the input and, therefore, in inversion patterns across languages Children analyze input in terms of syntactic features, categories, and operations 4
Research Questions
• • Study 1 (Tornyova & Valian): Are inversion patterns in acquisition determined by syntactic properties of the adult language (reflected in the target input)?
Study 2 (Pozzan & Valian): Can input frequency alone account for inversion patterns in English-learning children? 5
Study 1: Tornyova & Valian
• • Both Bulgarian and English display inversion in main wh- and yes/no questions Different properties of question formation
Bulgarian English
6
Elicited Imitation
• • • 4 groups: 2 Bulgarian (n=27, ages 2;2 - 3;3; Mean 2;9 ) 2 English (n=20, ages 2;4 - 3;2; Mean 2;9) Imitated 24 wh- or yes/no questions 7
Procedure
Bulgarian wh-question
Kude e igral Ivan s tebe?
Where aux-sum played Ivan with you
Bulgarian yes/no question
S tebe li e igral Ivan?
With you li aux-sum played Ivan
English wh-question
Where did John play with you?
English yes/no question
Was John playing ball with you?
8
100 90 80 70 60 50 40 30 20 10 0
Inversion by Question Type
No aux Non-inversion Correct inversion Wh-Q Y/N-Q Bulgarian Wh-Q English Y/N-Q 9
• •
Summary
Children are sensitive to the syntactic regularities that underlie input differences Level of syntactic consistency predicts differences in performance
Bulgarian English
10
Study 2 Elicited Production and Input
• • Do frequencies of questions in the adult input account for children’s production patterns? • How should frequency of inversion be measured?
Are production patterns better accounted for in terms of abstract categories (e.g., arguments vs. adjuncts)? 11
L1 Production Participants & Materials
• • • • N = 38 monolingual children Age: 4;3 (Median: 4;2 Range: 3;2-5;8) SPELT: 33/40 Materials: 16 main questions
auxiliary
is are
Total wh-
4 (what, which, why, when) 4 (what, which, why, when)
8
4
8
yes/no
4 12
Protocol
This is an asking game. This is Katie and this is her mom. Katie wants to know some things. We are going to help her ask her mom questions.
“Why my dog is digging a hole?”
13
1,00 0,80 0,60 0,40 0,20 0,00 correct
Question-type
non-inversion no aux other double aux wh yes/no Wilcoxon Signed Ranks: Z = 2.5, p=.012
14
correct
Wh-type
non-inversion no aux other double aux Wilcoxon Signed Ranks: Z = 2.5, p=.011
15
correct Wh- by auxiliary non-inversion no aux other double aux 16
• Can (token) frequency in adult input account for the observed pattern?
• No input data on these particular children. Assumption: adult input to children is fairly homogeneous 17
How to measure frequency?
• Absolute Frequency (inverted main questions ): • Inverted wh-: Why are you laughing?
• Inverted yes/no: Are you laughing?
• Relative Frequency (inverted main / all questions): • Non inverted wh-: I don’t know why you are laughing.
• Non inverted yes/no: You are laughing?
18
CHILDES Corpora
SEARCH: Wh-elements:
what , which , when , why
Auxiliary and copula:
is
,
are
Corpus
Bates Bloom 70 Clark Gleason Snow Valian
Total
# Children
27 3 1 24 1 21
77
Age Range
1;8-2;4 1;4-2;10 2;3-3;2 2;1-5;2 2;3-3;9 1;9-2;8
1;8-5;2
Adult Input Utterances
11,274 40,385 32,349 37,698 19,801 26,250
167,757
19
Corpus
Bates Bloom 70 Clark Gleason Snow Valian
Total
What 735 1339 1092 1008 800 1423
6,397
Absolute Frequency (inverted questions)
Question-Type
Which 5 16 16 26 23 35
121
When 3 12 2 4 5 6
32
Why 19 33 125 27 21 14
239
Yes/No 258 1056 299 487 59 581
2,740
20
Relative Frequency
(inverted/all questions)
Question-type Corpus
What Which When Why Yes/No Bates 735/760 (97%) 5/5 (100%) Bloom 70 1339/1428 (94%) 19/22 (86%) Clark 1092/1190 (92%) Gleason 1008/1205 (84%) 17/20 (85%) 28/29 (96%) Snow 800/283 (97%) 24/27 (89%) Valian 1423/1599 (89%) 37/38 (97%)
Total 6397/7005 (91%) 130/141 (92%)
3/3 12/13 2/5 4/4 6/10
32/40
(100%) (92%) (40%) (100%) 5/5 (100%) (60%)
(80%)
19/21 (90%) 258/336 (77%) 33/35 125/132 27/33 21/25 14/17
239/263
(94%) (95%) (82%) (84%) (83%)
(91%)
1056/1447 299/766 487/772 59/70 581/919 (73%) (45%) (63%) (84%) (63%)
2740/2210 (65%)
21
Results
Absolute Frequency: errors should occur in which,
when
and
why
Relative Frequency: errors should occur in yes/no and
when
-questions Results: inversion errors only occur in why questions when and 22
Take-home Message
• • •
Study 1:
input does matter! A grammar in which operations are implemented uniformly is a ‘simpler’ grammar
Study 2:
elements pattern together according to syntactic category, not just (token) frequency In progress: • • Token vs. Type Frequency Wh- + is/are + NP combinations • Relative frequency (counting all inverted and non-inverted strings) 23
Bonus Slide Wh- + is/are +
NP
combinations
Auxiliary NP-Subject
is
What
brother/dog/he other 385 4670
When
2 16
Which
3 90
Why
Yes/no 57 98 124 1691
are
Total (is)
you other
Total (are)
5056
786 552
1338 18
11 3
14 93
10 18
28 155 1815
50 34
84
716 212
928
24
Bonus Slide 2 Overall Correct Imitation
All questions
Bulgarian- and English-speaking children show similar overall correct imitation rates 100 90 80 30 20 10 0 70 60 50 40 73
Bulgarian English
74 25