AAAL 2013, Dallas, TX, 16-19 March
The development of English L2 writing complexity A longitudinal and multidimensional analysis
Bram Bulté (
[email protected])
Alex Housen (
[email protected]) MULTI-L (Centre for Studies on Multilingualism, Second Language Learning & Teaching) 1
Rationale • Complexity as a basic and valid § descriptor of language performance § indicator of language proficiency § index of language progress and development (eg. Wolfe-Quintero et al 1998; Polio 2001; Ortega 2003, 2012; Unsworth 2005, 2008)
=> Claim: L2 systems & L2 production become more ‘complex’ over time … •
Increased range, breadth, depth, sophistication, compositionality, etc. of lexical and morpho-syntactic elements
⇒ BUT: is ‘more complex’ = ‘more developed’ = ‘more complex’? -> empirical question! Pallotti (2009), Bulté & Housen (2012)
. 2.
Risk of circular reasoning Later developed/ acquired More advanced
Better
More complex More proficient
More difficult More developed
3.
Approaches to L2 complexity Quantitative property: number (& nature) of (a) constituent components and (b) relationships between components (Rescher 1998; Dahl 2004)
Bulté & Housen (2012)
. 4.
Studies on complexity development • Few true longitudinal studies of L2 (writing) complexity development (see e.g. Polio 2012) • Focus on short period only • Few data collections
• Recently: studies from DST perspective •
Verspoor et al (2008); Spoelman & Verspoor (2010), Vyatkina (2012, 2013)
=> L2 (complexity) development: • Irregular, non-linear development • Individual variation • Complex interplay between different complexity components
. 5.
Empirical Study The development of L2 Writing Complexity in ESL by Dutch-speaking young learners
. 6.
Research questions (1)
. 7.
•
How can L2 complexity be conceptualised for understanding L2 performance, L2 proficiency and L2 development?
•
How can L2 complexity (as a multicomponential construct) be reliably, validly and feasibly measured?
•
Which aspects of L2 complexity show developmental trends/patterns?
•
How do the different complexity dimensions interact, synchronically and diachronically?
Research questions (2)
. 8.
•
How representative are average group complexity scores for individual learners?
•
What is the extent of inter- and intra-individual variation, and how does this variation develop over time?
•
(How) can we combine different complexity measures to provide a more comprehensive picture of complexity?
•
How can multidimensional complexity development be visualized?
Participants & Data collection 10 learners of L2 English
11 writing tasks
• L1 Dutch
• Collected by Verspoor, Smiskova & colleagues
• +/- 12 years old
10 learners
Context: Secondary school • 5 mainstream EFL teaching • 2-3 hrs/week • 5 bilingual education (CLIL)
• Oct 2007 – May 2009 (19 months) • Different time intervals (from 4 – 16 weeks) • Limit of 1000 characters • No time limit (in practice 10 mins)
• 50% curriculum in English • Informal topics (+/- 14 hrs/week • EFL 2-3 hrs/week
11 data collection points
. 9.
Writing prompts/topics
#
Prompt
#
Prompt
1
Your new school, friends and teachers
7
The most awful (or best) thing that happened to you during summer vacation
2
Your favourite pet animal
8
Rules at home. Do you think they are fair or not?
3
Saint Nicholas
9
Pretend you have just won 1000 euro
4
What will you do for Christmas?
10
Write about someone you really admire. Why?
5
Your favourite holiday
11
What would you like to be when you grow up / what kind of job would you like to have?
6
The most awful (or best) thing that happened to you at school so far
. 10.
Quantitative complexity metrics (1) 10 Syntactic complexity measures: • Overall Cx: mean length of sentence (MLS); mean length of T-unit (MLTU). • Sentential Cx (clause linking): • Sentence type measures tapping different clause combining strategies: • • • •
Simple sentence ratio (SS) Compound sentence ratio (CdS) Complex sentence ratio (CxS) Compound complex sentence ratio (CdCxS)
• Coordinate clause ratio (coordinated clauses / sentence) (CoordCl/S) • Subclause ratio (subclauses / clause) (SubCl/Cl - SCR) • Clausal Cx: mean length of finite clause (MLCfin). • Phrasal Cx: mean length of noun phrase (MLNP).
. 11.
Quantitative complexity measures (2) 3 Lexical complexity measures: • Word length: letters / content word (MLW) • Lexical Richness: variation in and number of word types used
-> Guiraud index for content words (G) • Lexical Sophistication: variation in and number of 'basic' (frequent) vs. 'advanced' (less frequent) word types used
-> Advanced G (Adv G) (based on ‘basic’ vs. ‘advanced’ word lists compiled by P. Nation)
. 12.
Today’s focus 1. Significant linear trends for different complexity dimensions. 2. Individual developmental trajectories vs. group trends. 3. Inter-individual variation. 4. Interactions between different complexity dimensions. 5. Combining scores on different complexity measures (composite complexity measure).
. 13.
Results – Group development Syntactic complexity F
Lexical complexity ηp²
p
F
ηp²
p
MLS
3.545
.001
.307
MLW
1.906
.056
.192
MLTU
7.499
.000
.484
G
7.245
.000
.475
SS
3.996
.000
.333
Adv G
5.424
.000
.404
CdS
1.218
.292
.132
CxS
4.291
.000
.349
CdCxS
2.016
.042
.201
CxS+CdCxS
6.244
.000
.438
.692
.729
.080
10.169
.000
.560
MLCfin
1.093
.377
.120
MLNP
5.453
.000
.405
CoordCl/S SubCl/Cl
. 14.
Main Findings • Significant progress on all scores, except for CdS, CoordCl/S, MLCfin & Word length. • Strongest effect size for Subclause ratio (.560), followed by MLTU (.484), Guiraud (.475) & % Cx+CdCx S (.438). • Syntactic: subordination shows more linear development than coordination. Phrasal complexity increases, clausal complexity does not. • Lexical: increase in richness and sophistication.
Results – Individual and group development Subclause Ratio (SCR) 0.7
0.6
101 105
0.5
109 122
0.4
126 103
0.3
108 112
0.2
119 120
0.1
AVG
0
. 15.
Results – Individual linear trends SCR 0,6 101 0,5 0,4
0,3 0,2 0,1
105 109 122 126 103 108 112
119 120
0
. 16.
Results – Individual trajectory (L109) SCR (L109) 0.7
0.6
0.5 109
0.4
AVG GROUP Avg (t-1; t; t+1)
0.3
Progmin Progmax
0.2
0.1
0
. 17.
Results – Individual trajectory (L120) SCR (L120) 0.6
0.5
0.4 109 AVG GROUP
0.3
Avg (t-1; t; t+1) Progmin
0.2
Progmax
0.1
0
. 18.
Results – Interindividual variation CoV - SCR 1,2
1 0,8 0,6 0,4 0,2
0
. 19.
Results – Interactions between dimensions SCR + MLNP (Group) 1.5
1
0.5 Z SCR 0
Z MLNP SCR + MLNP
-0.5
-1
-1.5
. 20.
Results – Moving correlations SCR & MLNP 0 -0,1 -0,2 -0,3 -0,4 -0,5 -0,6 -0,7
-0,8 -0,9
-1
. 21.
Results – Visualizing concurrent development SCR & MLNP (Group) 1.5
10 7
1
9
0.5
-1.5
5
-1
6
-0.5
1
0
3 4
-0.5
-1 2
-1.5
. 22.
11
0 0.5
1
8
1.5
Results – Combining complexity dimensions 3
Synt Cx 3 & Lex Cx 3 (Group)
2
1
0
SyntCx3 LexCx3
-1
-2
-3
. 23.
Results: multiple linear regression (time) • Model including SCR, MLNP, G & MLW (R2 = .533) •
(Formula: y = -35.648 + 21.246*SCR + 4.349*MLNP + 1.934*G + 3.755*MLW)
• Correlation between individual predicted scores and time • Average: R = .807; R2 = .652 • St Dev = .076; Min = .662; Max = .890
. 24.
Results: multiple linear regression (time) • Model including SCR, MLNP, G & MLW (R2 = .533) 20
15
101
105 109 122
10
126 103 108 112 119
5
120 Average
101
105 109 122
126 103 108 112 119
120 Avg - STDEV
Avg + STDEV
0 0
5
-5
. 25.
10
15
20
Summary Group development: •
Selective components of the complexity of texts written by L2 learners increase over time (lexical richness & sophistication, length of syntactic units & subordination).
•
Highest effect sizes for SCR, MLTU, G.
Individual development: •
High degree of variability.
•
Not all learners follow observed (average) group trends.
•
Inter-individual variation decreases over time for certain measures.
Interaction between complexity dimensions in development: •
Subordination & phrasal elaboration seem to be both connected as well as competing growers.
Combining complexity dimensions: •
Standardised scores can be combined to obtain a more global picture.
•
Multiple linear regression to combine measures in model that best fits development. . 26.
Conclusions • Need for conceptual clarity: complexity / difficulty / developmental timing / quality (minimise risk of circular reasoning) • L2 complexity is complex (multiple dimensions, layers, components) • Necessary to look at each dimension/layer/component individually • BUT also look at (a) the (complex) relationships between them and (b) their combined effect in L2 development • (Validity of) complexity measures as measures of development •
L2 (writing) complexity, as defined and operationalized in this study, clearly increases over time.
•
At group level: ± smooth (linear, consistent) development.
•
BUT, at the individual level: high degree of both intra- and interindividual variability (non-linear, irregular)
=> General complexity measures can capture development in broad strokes, but … . 27.
Conclusions (2) • ….a good index of development should: § Increase over time § Increase linearly with development § Indicate if learner is beginner, intermediate, advanced § Cover the full developmental trajectory § Capture both short-term and long-term changes § Be insensitive to variations in terms of L1 background, task
type, … ⇒ is (absolute, linguistic) complexity the way to go? ⇒ perhaps difficulty (defined as ‘developmental timing’) is a more useful construct to help understand the dynamics of L2 development? 28. .
Thank you!
[email protected] [email protected]
. 29.