Can we devise a new measure of syntactic complexity for language acquisition

1
Can we devise a new measure of syntactic
complexity for language acquisition? Dunaújváros,
March 18, 2004 Marijan Palmovic Laboratory for
Psycholinguistic Research University of
Zagreb Marijan.Palmovic_at_public.srce.hr http//publ
ic.srce.hr/labpolin/
2

• Mean Lenght of Utterance MLU a traditional
  measure
• Simple and good measure for assessing syntactic
  development
• but
• there are problems in morphologically rich
  languages
• -not valid crosslinguistically

In Croatian - Does not discriminate between
children of various developmental rate?
3
Croatian corpus (CHILDES) MLU in morphemes
4
Croatian corpus (CHILDES) MLU in words
5
Various measures for assessing lexical
development 1
Type/token ratio (TTR)

• Generally
• Sensible to the size of the file
• - larger the file, smaller the TTR
• -- questionable application in morphologically
  rich languages

• Specifically, for Croatian
• Not an automatized measure
• - does not reflect lexical complexity
• -- ratio always around 1

6
Various measures for assessing lexical
development 2

• Number of different words in a standard number of
  utterances
• - longer the sentences (and therefore MLU),
  higher the NDW.

NDW Number of Different Words
Fixing the number of tokens, not utterances
VOCD a smarter TTR
- Accounts for the file size -- takes randomly
chosen samples of text
7
RRG

• Role and Reference Grammar
• Van Valin, LaPolla, 1997, Syntax

8
The overall organization of the RRG (Van Valin,
LaPolla)
9

• Aktionsart
• State
• Achievement
• Accomplishment
• Activity
• Active accomplishment

Causative counterparts -causative state,
causative achievement
The ice melted. Hot water melted the
ice. Accomplishment causative accomplishment
10
Lexical representation for the basic Aktionsart
classes (Van Valin, LaPolla, 1997)
John ate the fish. do (John, eat (John,
fish))
Macroroles actor undergoer
11
U
A
do (John, eat (John, fish))
12
Antonija (CHILDES) Table representing the
number of different verb classes - the first
step
13
Antonija, 19 ANT Ja idem tebi. Eng
I am going to you. ANT Ja cu ovo
drati. Eng I will hold this. ANT Necu
tebi dati olovku. Eng I will not give you
the pencil.
14
(No Transcript)
15
Complex sentences as represented in RRG
16
Interclausal Relations Hierarchy - IRH
17
9
CL COO
Attach points to different juncture and nexus
types to modify the scale obtained from
points given for logical form variety
8
CL SUB
7
CL COSUB
6
CORE COO
5
CORE SUB
4
CORE COSUB
3
NUC COO
2
NUC SUB
1
NUC COSUB
18
An example Strongest connection nuclear
cosubordination English examples John pushed
open the door. Vince wiped the table clean.
Prediction -should be acquired first!
19
But in Croatian, these sentences are translated
differently John je gurnuo vrata i otvorio
ih. Vinko je ocistio stol.
20

• The question about the acquisition of complex
  sentences is reduced to the question about the
  childs ability to learn IRH with the stronger
  connection ocurring first.
• A child can deduce other grammatical properties
  of complex sentences, for example, the ones
  regarding the operators of tense and aspect.
• Child starts to use clause linkage constructions
  before mastering the total range of constructions
  and operators possible in simple sentences.
• e.g. Hocu piti cajeka. I want to drink tea.
• Modify our scale with IRH (add 1 - 9 points)

21
Examples
Antonija, 17 two word sentences, e.g. Hocem
ovo. Want this. 50 Antonija, 19 core
cosubordination Idemo pjevati! Lets sing! 45
Antonija, 110 clausal coordination Nis-am
pjev-a-la, nego ja
s-am pak-a-la. Be_not-1SG
PRES sing-5-PARTF but I
be-1SGPRES cry-5-PARTF 'I was not singing,
but I was crying.'
49
22
Comparison with the MLU
Antonija, 17 two word sentences 5 MLU
(words) 1,57 MLU (morphemes) 2,29 Antonija,
19 core cosubordination 9 MLU (words)
2,43 MLU (morphemes) 3,53 Antonija, 110
clausal coordination 13 MLU (words)
2,19 MLU (morphemes) 3,43
23
Problems

• It is not easy to apply it is not an automatic
  measure
• Selection of verb classes can be problematic due
  to the imperfective perfective oppositions of
  verbs
• Questionable predictivity regarding the complex
  sentences children like to add sentences with
  conjection i and.