Issue

04/22/23 1

Issue

• Claim: acquisition of discourse integration is an extended process

• Even up to 6 yrs of age children may fail in correct discourse-linking:– Pronouns– Definite articles– Tense

(Karmiloff-Smith 1981; Avrutin 1999; & others)

04/22/23 2

But…

• In ordinary everyday language use, children do no seem to have much trouble with phrases that rely on discourse integration, notably ellipses, neither in production nor comprehension.

04/22/23 3

Examples (Sarah corpus)

*CHI: I drink it all up .*CHI: give me some more .*CHI: a lot .*MOT: I don't see any more .*CHI: yes you do .*MOT: want a little milk ?*MOT: want some ?*CHI: (a)n(d) shake it all up .*CHI: a bigger one ?*MOT: mmhm .

04/22/23 4

Question

• How do children understand such expressions?

04/22/23 5

Proposals

1. Heavy reliance on discourse context (nonverbal and visual information)• discourse context dominates syntax

2. Children are in fact capable of reconstructing ellipsis• syntax dominates discourse context

04/22/23 6

Our hypothesis

• Ellipsis reconstruction, and, hence, discourse linking through ellipsis, is a very early attainment in language acquisition.

• Reason: ellipsis reconstruction depends on syntactic configuration (in contrast to discourse anaphors such as pronominals and determiners).

• Syntax is (very) early and can support the interpretation of ellipsis.

04/22/23 7

Hypothesis:

The acquisition of ellipsis should simply follow in the footsteps of the initial acquisition of grammar. As each new level is constructed, the child can construct a parallel level in silence.

04/22/23 8

Aims

• To show that children understand ellipsis at an early age

• To show that they do so by linguistic reconstruction, not deixis or, more generally, reliance on (nonverbal) context.

04/22/23 9

Means

• Three experiments probing children’s understanding of nominal ellipsis1. preferential looking (English)

[NP some ___ ]2. Sentence-picture matching task (English and

Dutch) [NP two ___ ]

3. Truth-value judgment task (Dutch) [NP two ___ ] & effect of there-insertion

04/22/23 10

Experiment 1(Jones, Hirsh-Pasek & Roeper, in preparation)

• Can young children infer the object of noun phrase ellipsis?

• Subjects: 18 3-year-olds, range = 36.00 - 46.99 months, M = 40.78

• Procedure:– Two labeled transitive action sequences, each

followed by one test trial– Participants asked to point to the video clip best

representing the noun phrase elliptical sentence.• Conditions counterbalanced for target order

04/22/23 11

“John has socks”

“Kate is cooking pancakes.”

Can you find: “John wants to eat some.”

Can you find: “John wants some.”

04/22/23 12

Experiment 1: Results

• Initial analyses show that 3-year-olds pointed to the target action 77.78% of the time, a result significantly different from chance, t(17) = 3.82, p = .001.

• No effects of gender or target order.• Tentative conclusion:

3-year-olds reconstruct the missing element in the elliptic expression ‘some __’.

04/22/23 13

Experiment 2(Wijnen, Roeper & Van der Meulen, 2004)

• Participants:a. 28 American English-speaking children; mean

age 53.6 months (4:6, range 40-69)b. 47 Dutch-speaking children, mean age 41.5

months (3;6, range 28-57)• Task: sentence-picture verification• Materials

– 15 short stories, ending in pertinent questions, combined with different pictures

04/22/23 14

(introductory phrases)Some kids are playing in the sandbox.Are two upside down?

kids = argumentIn the sandbox = adjunct

Control Adjunct mism.

Arg+Adj mism.

04/22/23 15

Reconstruction

Some kids are playing in the sandbox.Are two upside down?

Two = two [kids]ARG [in the sandbox]ADJ

04/22/23 16

Experiment 2 - Results

0102030405060708090

100

Control Adj mismatch A+A mismatch

EnglishDutch

04/22/23 17

Experiment 2 - conclusion

• Results indicate adequate discourse integration.

• Reconstruction appears to take place (cf. difference control/mismatch conditions)– This is syntactic integration

• Question:Difference Eng-Du ~ related to “er/there”?

04/22/23 18

There ~ Er

• Eng: Some kids are in the sandbox.Are two upside down?

• Du: Er spelen kinderen in de zandbak.Staan er twee op hun kop?

• function of er~there:– expletive/existential, or contrastive– contrastive ‘there’ the bare cardinal will be

taken to denote a parallel (contrast) set, I.e., “two girls somewhere else”

– possibly: contrastive function is acquired earlier than expletive/existential function

04/22/23 19

Experiment 3

• Aims– Replicate previous results with a better design and

slightly different procedure.– Test the er/there-effect without the language

confound.

04/22/23 20

Experiment 3 - method

• Participants: 26 Dutch-speaking children, mean age 4;6 (range 3-6)

• Procedure:– Truth-value judgment (sentence-picture

verification); children were instructed to help a girl robot ‘learn to speak’

• Materials:– 12 stories, paired with different pictures to

instantiate 4 conditions.

04/22/23 21

Control Arg mismatch

Adj mismatch A+A mismatch

Drie jongens spelen in de zandbak, en twee zitten [ER] op een emmertje.

Control Arg. mismatch

Adj. mismatch Arg. + Adj. Mm

04/22/23 22

Experiment 3: expectations

• Control– –ER : yes

+ER: mixed (both subset [full reconstruction] and parallel set [no adj reconstr] reading are o.k.)

• Argument mismatch– no, both in +ER and –ER

• Adjunct mismatch– –ER no

+ER mixed (yes certainly possible)• A+A mismatch

– no, both in +ER and –ER‘yes’ only possible through deixis: ‘two [anything] [anywhere]’

04/22/23 23

Experiment 3: results

• 5+ year-olds have a very strong preference for ‘yes’ answers across the board

• we present results of 3-4 yr. olds only

04/22/23 24

Experiment 3: Results –ER

0102030405060708090

100

Control Argmismatch

Adjmismatch

A+Amismatch

% yes

04/22/23 25

Experiment 3 Summary of –ER results

• ‘yes’ preference is quite strong (task effect?)• Control vs. Arg+Adj-mismatch

– Clear difference as expected• Control vs. Adjunct mismatch

– smaller difference• Control vs. Argument mismatch

– Hardly any difference possibly a materials artefact

04/22/23 26

Experiment 3: Results +ER

0102030405060708090

100

Control Argmismatch

Adjmismatch

A+Amismatch

% yes

04/22/23 27

Experiment 3Summary of +ER results

• Control vs. Adjunct mismatch– in line with expectation; ‘er’ makes ‘parallel’

interpration (i.e., contrast) more acceptable, and suppresses the acceptability of full reconstruction interpretation

• Control vs. A+A mismatch– as expected

• Control vs. Argument mismatch– unexpected; possibly a materials artefact

04/22/23 28

Experiment 3: –ER vs +ER

0102030405060708090

100

control Argmismatch

Adjmismatch

A+Amismatch

-ER+ER

04/22/23 29

Experiment 3 - Conclusion

• Control–A+A-mismatch difference supports the idea that children reconstruct.

• But the difference between Control and partial mismatch conditions is less clearcut.– arg.mismatch results may be artefactual.

• The effect of +/-ER is marked – including ‘er’ renders the full reconstruction interpretation (control cond.) less acceptable, and favors contrastive interpretations.

04/22/23 30

General conclusions

• Results suggest that children are capable of reconstruction.

• However, the percentage ‘yes’ responses does not drop to zero in the (partial) mismatch conditions ‘deictic leakage’

• Young children are sensitive to the interpretive effects of er/there.– There is (predominantly) contrastive (cf. differences between

younger and older children). It blocks syntactic reconstruction of the original locative. This corroborates our syntactic reconstruction account.