Language Acquisition

1
Language Acquisition

• 3.

Elena Lieven, MPI-EVA, Leipzig School of
Psychological Sciences, University of Manchester
2
Outline for Session 3

• MAIN TOPICS
• Errors and the ATOM
• Errors in syntactic questions
• Learning more complex constructions
• Constituency, embedding and island constraints

ENDNOTE Complex syntax and input Metalinguistic
awareness
3
Errors in the use of non-nominative subjects
4
The Agreement/Tense Omission Model (ATOM)
Schütze Wexler, 1998

• Predicted to occur
• AGR/TNS NOM Im going He goes
• AGR/-TNS NOM I going He go
• -AGR/TNS -NOM Me going Him go
• Her gone Her went
• Predicted not to occur
• (or to occur at levels compatible with noise)
• AGR/TNS -NOM Me am going Him goes

5
Errors that shouldnt happen!
Anne Manchester Him doesnt And her
has Probably hers a baby A big girl now, her
is I think her was crying for me
Becky Manchester Where does him go? Her is
gonna make a dinner Hims eating you, crocodile
Abe Kuczaj Sometimes her barks nice, sometime
her donts Her has a tummy ache
6
Testing the ATOM Pine et al., 2006

• Expected and observed rates of agreement with
  non-nominative subjects
• Nina 3psg
• he/she him/her Rate
• Agreeing
• Actual 213 10 4.5
• Expected (160.8) (62.2) (27.9)
• Non-agreeing
• Actual 185 144
• Expected (237.2) (91.8)

7
Ninas 3psg by gender rates
Expected rate of non-nominative subjects with
agreeing verbs (3.3) Actual Rate 1.4
Expected rate of non-nominative subjects with
agreeing verbs (90.4) Actual Rate 53.8
8
Errors in syntactic questions
9
Errors in questions

• Omission Where he go
• Double marking Where does he does go?
• Non-inversion Where he does go?
• Agreement errors Where does you go?
• Case errors Where does her go?

10
Ambridge, Rowland, Theakston, Tomasello (in
press)
Adult Ask her why the dog is sleeping. Child
Why is the dog sleeping? Adult Ask her where
the pig can swim. Child Where can the pig swim?
4 year olds

• MAIN RESULT different number errors for
• different wh- words
• different auxiliaries
• same auxiliary w/ diff number (e.g., do
  does)

11
Accounting for patterns of error

• The error rate is low because children are
  learning constructions with slots
• High frequency frames should be protected from
  error
• Errors will occur when there isnt a frame

12

• Data
• Adam 23 410 (Brown 1973)
• All wh-questions requiring inversion
• Patterns of wh auxiliary (e.g. where does, what
  has)

• Results
• Almost all wh-word auxiliary combinations are
  produced either inverted or non-inverted - only
  3 out of 49 showed optional inversion
• The combinations that the child inverts are more
  frequent in the input than those that he fails
  to invert

Rowland Pine, 2000
13
Errors based on frames?
Non-inversion M. You dont throw things C. Why
you dont throw things?
Omission Why you like cakes?
? Why X You like cakes
Double marking Why dont you dont like cakes?
? Why dont X You dont like cakes
Agreement errors Where does you go?
? Where does X go? You
14
Error rates in syntactic questions
Rowland, in submission
15
Summary

• Errors should not be summed across forms
• Errors can be analysed in terms of patterns in
  the input
• Errors, correctly analysed, present important
  challenges to all theoretical positions that have
  to answered in terms of the theory

16
Relating partially incompatible constructions
17
Gaps with precedents at 30

• Brian 30
• What you doing GAP? 0
• What you doing GAP there? ADD there
• What I owe GAP you? 0
• What say GAP? 0
• What say my computer GAP? ADD NP
• Where that come from GAP? Where did that come
  from? (M)
• Wheres it gone GAP ? 0
• Wheres it gone GAP now ADD now
• Annie 30
• Where can he park GAP? Where can N park?
• What did you hurt GAP? 0
• And what did she do GAP? And what did NP do? (M)

18
Gaps without precedents

• Brian 30
• What your found GAP? You found X
• What I love GAP then? I love X
• Where you been to GAP? I been to X
• What I bought it in GAP? I bought it in X
• Annie 30
• What can we do GAP with that? We can do X with
  that
• And what that done GAP? That done X

19

• Wh construction Object of verb/preposition is
  utterance initial
• Whato aux V?
• Verb frame Object comes after the verb
• Ns V No

Profile determinance (Langacker) Since a
question is intended, the Wh construction is
schematic for the composite
utterance Note that children make mistakes over
this What I found it? What he done it?
20
Learning more abstract constructions
21
Building up constructions from prior constructions
Schematisation and analogy
Constituency, embedding and island constraints
22
Building new constructions from old parts
23
Diessel Tomasello
24
S-COMPLEMENTSDiessel Tomasello, Cognitive
Linguistics (2001)

• Subjects Adam, Eve, Sarah, Naomi, Peter, Nina -
  1 to 5 years
• Complex Ss 2807 tokens
• Examples from Sarah Examples from Nina
• I think hes gone See that monkey crying
• I think its in here See Becca sleeping
• I think my daddy took it See that go
• I think I saw one See my hands are washed
• its a crazy bone, I think See he bites me
• I think dis is de bowl See him lie down

25
Subjects in Complex Ss
1-P 2-P 3-P Lex Imp Guess 100 -- -- -- -- Bet
100 -- -- -- -- Mean 52 48 -- -- -- Know 36 55 0
5 04 -- Think 85 13 02 -- -- Wish 97 -- -- 03 --
Hope 88 12 -- -- -- See 07 01 01 -- 91 Look
-- -- -- -- 100 Watch -- -- 11 -- 89 Remember 6 6
-- -- 88
- Virtually no complementizers - Virtually no
non-present tenses - Virtually no modals or
negations
26
RELATIVE CLAUSESDiessel Tomasello, Cognitive
Linguistics (2000)

• - Subjects 4 CHILDES children from 19 to 51
• - Total of 324 relative clauses
• Heres the toy that goes around.
• Thats the sugar that fell out.
• Theres the ball I bought
• Thiss the bird that sings.
• Thats the one that goes moo.
• Heres the boy that ran into the water.

27

• Earliest All
• NP ONLY
• The girl that came with us .05 .19
• PRESENTATIONALS
• This is the car that turns around .75 .47
• OBLIQUES
• Im going to the zoo that has snakes 0 .06
• OBJECT
• She has a bathtub that goes with it .20 .26
• SUBJECT
• The one that not finished is up there 0 .01
• 50 of these Look at all the chairs Peters
  got

28
Brandt, Diessel, Tomasello (in progress) on
German
One German child, dense corpus, over 900 relative
clauses, age 2 - 5.
V2 Relative Clauses Grosser Wal, der hat
Zahnschmerzen Muscheln, die kann man essen
MOTHER (20 V2) Ich habe einen
Bauernhofkaese, der kommt von Frankreicher
Bauernhof .
Real Relative Clauses Wo ist ein Wal, der
Zahnschmerzen hat
29
3 years old matches adult
30
Initial Complex Constructions

• Early S-comps and relative clauses have
  restricted range of forms gt esp. in matrix
  clause
• No general rules for Wh-Questions (tough
  movement, binding) gt some items easier than
  others, even when all the words are well
  established

31
The construction conspiracy hypothesisAbbot
Smith Behrens, submitted

• Using German dense corpus Leo
• Development of sein- and werden-passive
  constructions

32
Related constructions for the SEIN- and
WERDEN-passives
33
Leo versus mother(werden- versus sein-passive)
34
Leos sein-passive and related constructions
(3rd ps. sg. only)
35
Leos werden-passives and related constructions
(3rd ps. sg. only)
36
Schematisation and analogy
37

• Schematisation overlap in lexical material,
• constituents and meaning
• Analogy no shared material, overlap in
• constituents and meaning
• car pulling boat
• truck pulling car

Gentner et al.
38
Analogy

• Already have a number of schemas
• Already know a lot about variations in NPs
• Create a construction with no lexical material
  that analogises across the schemas

39
The boys chopping the tree Hes chopping it
The dogs eating the toy Hes eating it
ANALOGY
cline
NPs VERBing NP

• role of type frequency in VERB and NPs
• role of varied nouns and single pronoun in NPs
• role of verb semantics (Goldberg Casenheiser)

40
Constraining argument overgeneralisations
41
Transitivity Overgeneralizations

• Mommy, can you stay this open?
• I come closer so it wont fall.
• Dont giggle me.
• She came it over there.
• I want to stay this rubber band on.
• Eva wont stay things where I want them to be.
• You cried her.
• Will you climb me up there?
• Kannst Du mich hochklettern?

42
Mummy, hes giggling me
? Hes making me giggle

• Laugh learned early
• entrenched
• no errors
• Chortle learned late
• abstraction achieved
• no errors
• Giggle - not entrenched
• errors

43
Three constraining factors working over
developmental time.
Growing abstractness of the transitive
construction
Many overgeneralizations b/c not entrenched
Preemption
Verb Subclasses
No overgeneralizations b/c Verb Islands
Giggle Chortle Laugh
Entrenchment
Low overgeneralzations b/c preemption and verb
subclasses in addition to entrenchment
44
What determines overgeneralisations?
Brooks et al, 1999

• Fixed Transitivity Verbs

More Entrenched Less Entrenched Hit Strike Ta
ke Remove Come Arrive Disappear Vanish
Method
Children (3-8 yrs.) see transitive event and then
are asked mismatching question - What
happened with PATIENT? - What did AGENT
do?
Results
More entrenched verbs overgeneralised less often
45
Grammaticality judgements relate to entrenchment

• 5-year-olds, 8-year-olds and adults heard
  sentences modelling argument structure errors
  with high and low frequency verbs
• She disappeared the rabbit
• She vanished the rabbit
• He came her to school
• He arrived her at school
• Children asked to indicate whether sentences
  sounded okay or silly
• Adults asked to rate sentences on a scale

Theakston, 2004
46
Adults judgments of ungrammatical sentences
0 ungrammatical, 7 grammatical
47
Constituency, embedding and island constraints
48
Children cant learn grammar from what they hear
Johnnie should clean his teeth
INVERSION
GAPS
What
should Johnnie
49
Moving constituents not words
happy?
is
Mary
happy
RULE? change the order of the first noun and
the first verb
The boy
who
is
smoking
is crazy
The boy
is
The boy who is smoking
is
is crazy
__crazy?
The boy who is smoking
RULE? change the order of the SUBJECT and the
MAIN VERB
50
Constituency

• Of course constituency cannot be learned from
  linear strings
• But this is a straw person argument and a very
  naïve view of input
• If learning form X, the relevant input consists
  of form X
• Children build up knowledge of constituents over
  time and use this to parse more complex sentences
• The input can give indirect evidence

51
It belongs to me The bike with wheels belongs to
me The cats my dog chases belong to our
neighbour
52
The boy wearing a green hat who I saw yesterday
in the park is very nice
The boys nice The bus is red They are really
really good Theyre the ones I want The ones
over there are the ones I want The boy wearing a
green hat is nice The girl who dances The one I
saw in the playground The one that turns round is
mine
Constructions gain more constituents Constituents
become more complex Form-meaning mappings become
more abstract
Over time
53
Modeling inversion with embedding

• Network trained on well-formed sentences
• Types and frequencies mimiced the Manchester
  corpus input
• Many sentences as follows
• Mary is happy Timmy can swim awfully fast
• Is Mary happy? Can Timmy swim fast?
• No sentences
• The boy who is smoking is crazy
• Tested on
• Is the boy who smoking is crazy?
• Is the boy who is smoking crazy?
  Clear preference for this

Lewis Elman, 2001
54
Unbounded dependencies and island constraints
Unbounded dependencies cannot be created for
every part of a sentence
Who did she see the report that was about She
saw the report that was about it Who did that
she knew bother him That she knew X bothered him
55
Information structure and unbounded dependencies

• Backgrounded constructions are islands
• e.g. restrictive relative clauses and noun
  complements
• The potential extraction site must be at least a
  potential focus domain
• this introduces a cline of acceptability into
  grammaticality judgements depending on the degree
  of backgrounding
• e.g. some presentational relatives can serve to
  convey the main assertion of the clause
• John is the sort of guy that I dont know a lot
  of people who think well of Cullicover

Lambrecht, Van Valin, Goldberg
56
Where does complex syntax come from?
57
Multi-clausal wh-questionsWho did the lion know
that swam in the pond?

• Very rare in childrens naturalistic speech
• 12 children aged 12 60 Stromswold, 1995
• 13,000 utterances with who, what, which
• 200 with gap in embedded clause
• (e.g. What wouldya like to have? What he
  want to play with?) and these are relatively
  simple infinitival complements
• Experiments are often with older children and
  sometimes with very unnatural sentences
• Some 3-year-olds and more older children show
  evidence of observing various hypothesised
  syntactic constraints
• This has been taken as evidence of innate
  knowledge of these unlearnable constraints
• However there is here also a cline in adult
  judgements of acceptability for these structures
  (Goldberg)

58
Complex syntax and input parentsHuttenlocher
et al. , 2002

• Measures
• Proportion of complex sentences (adults)
• Number of noun phrases per utterance (children)
• Comprehension task multi-clausal sentences
  (children)
• Results
• Proportion of complex sentences by parents the
  only significant predictor (r .41, plt.01)

59
Complex syntax and input teachers

• Participants
• 40 nursery classrooms high SES low SES, mixed
  SES
• 1 teacher per class
• Measures
• Teacher
• Proportion of complex sentences
• Number of noun phrases per utterance
• Observation of Quality of teaching
• Children
• Comprehension task multi-clausal sentences and
  sentences with varying numbers of noun phrases at
  time 1 and time 2
• Maths task

60
Results

• Comprehension was related to SES at the beginning
  of the year
• The childrens growth scores over the year were
  not related to SES
• They were significantly and strongly related to
  the teachers proportional use of complex
  constructions

61
End!