1. Anik Wulyani, PhD candidate
Automatic measures of cohesion and lexical proficiency in L2 writing: A case study
Anik Wulyani, PhD candidate
School of Linguistics and Applied Language Studies, Victoria University of Wellington
Introduction
Can you see the difference?
Table 2b Expected Results, Additional Indices
Why are the cohesion indices of the present study different from Crossley & McNamara (2009)?
In terms of cohesion and vocabulary use which one do you think is the blog post of an L1 writer?
How do you know?
This study compared the use of cohesion devices and lexical knowledge in L1 and L2 blog entries, using the Coh-Metrix tool ( The present study is based on the research described in Crossley and McNamara (2009) who studied the lexical differences in L1 and L2 writing by using Coh-Metrix.
The topic of automated measures of L2 writing has attracted considerable interests recently, due to potentially better efficiencies and higher effectiveness of computational measures, compared to human raters.
In the present study, the Coh-Metrix tool ( is used to automatically evaluate lexical knowledge and the use of cohesive devices in the L2 (English) writing of an Indonesian writer, and to compare these measures with those generated for L1 writers.
In my study, written entries from an L2 writer’s (
and an L1 writer’s ( blogs
are analysed for indices of lexical knowledge and cohesion, and the results are compared.
Indices
Expected
Indices # 48: Lexical diversity, Type-Token Ratio, content word lemmas
L1 > L2
Indices # 49: Lexical Diversity, Type-Token Ratio, all words
Indices # 50: Lexical Diversity, MTLD, all words
Indices # 51: Lexical Diversity, VOCD, all words
Indices # 95: CELEX Log frequency for all words, mean
L1 < L2
Indices # 99: Concreteness for content words
L1<L2
Indices # 104: Hypernymy for verbs
Indices # 105: Hypernymy for noun and verbs
Because:
The L1 writer appeared to be more casual in organizing his texts (few argument overlap) and to expect his readers to have high knowledge to understand his texts (few verb incidences)
The L2 writer seemed to have a formal writing or text organization (greater argument overlap) and to use more cohesion devices (more verb incidences) to assist her readers to better understand her texts
 Graesser and McNamara, 2011; McNamara, Louwerse, McCarthy, & Graesser, 2010; & O’Reilly & McNamara, 2007
Those 15 indices from Coh-Metrix were chosen because previous related studies showed that they were able to differentiate L1 writing to
L2 writing (Crossley & McNamara, 2009; Crossley & McNamara, 2010; Kyle, 2011; McNamara, Crossley & McCarthy, 2010; & Crossley, Salsbury, McNamara, & Jarvis, 2010).
Sample 1 Blog post
Today I guide my students to continue design their summative task and formative tasks. I ask my students to list down possible things they can do for their summative and formative tasks. I asked them to list down as many action as possible (actions which they have done in their previous summative or their lies). Then they need to classify the actions which can be feasibly done for their summative tasks.
Implications
Sample 2 Blog post
1. Practicality  Coh-Metrix is able to distinguish L1 and L2 texts  using linguistics features (cohesion and lexical indices)
2. Assessment  Coh-Metrix is a powerful tool to measure L2 writing
In my blog I have tried, so far, to address general issues to do with presenting, conferences, writing abstracts etc etc. As with every other blogger, my ‘thoughts’ have been personal, of course, but I have tried to exercise some dispassion.
But not this time. Just for once I want to tell you how I feel – or rather what it felt like (and then see if there is anything to learn from that). What I am trying to say is that this post is going to be incredibly personal, and I hope you will forgive me for that.
Research Question:
Can indices of cohesion and lexical use generated by Coh-Metrix be used to distinguish between L1 and L2 blog writers?
Limitations
A case study subjects of the study  one L1 writer and one L2 writer
Sample texts: relatively low in number, raising the question of generalizability
Future studies using Coh-Metrix may benefit from using:
large corpus
not only cohesion and lexical indices
Design & Methodology
Blogging among in-service teachers
Main research
EFL teachers in Indonesia
What does Coh-Metrix 3.0 and MANOVA with alpha level .05 tell us about the cohesion and lexical use?
Pilot study
Coh-Metrix Tool 3.0
One L1 blog writer, native speaker of English, the UK
One L2 blog writer, EFL teacher, Indonesia
Automatic measures of cohesion and lexical proficiency in L2 writing: A case study
Three cohesion indices showed significant differences: argument overlap (#29), LSA given/new (#46), and verb incidence (#85)
Three lexical indices reached statistical differences: lexical diversity #50, #51, and hypernymy for nouns and verbs (#105)
Cohesion:
The L1 writer appeared to produce more cohesive texts than the L2 writer
The L2 writer was more likely to produce texts with more new information than the L1 writer.
The L2 writer seemed to use more verbs or more spatial cohesion than the L1 writer
4. Lexical Proficiency:
The L1 writer seemed to have more diverse lexicon than the L2 writer
The L2 writer appeared to use less diverse vocabulary than the L1 writer
The L1 writer used significantly more conceptually abstract and hierarchically connected words than the L2 writer
Future Research
Future research  English Language Teaching in second language context  L2 writing proficiency  L2 blog writers  cohesion and lexical indices
L 1 writer:
13 blog posts
L2 writer:
13 blog posts
References
Coh-Metrix Tool 3.0
Cohesion indices:
5 indices
Lexical indices:
10 indices
Cohesion indices:
5 indices
Lexical indices:
10 indices
Crossley, S.A. & McNamara, D. S. (2009). Computational assessment of lexical differences in L1 and L2 writing. Journal of Second Language Writing, 18, 119–135
Crossley, S. A., & McNamara, D. S., (2010). Predicting second language writing proficiency: the roles of cohesion and linguistic sophistication. Journal of Research in Reading, 35 (2), 1–21. DOI: /j x
Crossley, S. A., Salsbury, T., McNamara, D. S., & Jarvis, S. (2010). Predicting lexical proficiency in language learner texts using computational indices. Language Testing, 20 (10), 1–20. doi: /
Graesser, A. C. & McNamara, D. S. (2011). Computational analyses of multilevel discourse comprehension. Topics in Cognitive Science, 3,
Kyle, K. (2011). Objective measures of writing quality. Master Thesis. Colorado State University, Fort Collins: USA
McNamara, D. S., Crossley, S.A., & McCarthy, (2010). Linguistic features of writing quality. Written Communication, 27 (1), doi: /
McNamara, D. S., Louwerse, M. M., McCarthy, P.M., & Graesser, A. C. (2010). Coh-Metrix: Capturing linguistic features of cohesion. Discourse Processes, 47 (4),
O’Reilly, T., & McNamara, D. S. (2007). Reversing the reverse cohesion effect: Good texts can be better for strategic, high-knowledge readers. Discourse Processes, 43 (2),
Table 1 Differences between Crossley & McNamara’s study & present study
Crossley & McNamara (2009)
Present study
Seven variables (indices)
Fifteen variables (indices)
Corpus of L1 (undergraduates in USA) & of L2 (the International Corpus of Learner English)
Blog posts of L1 (a teacher and trainer in English Language Teaching) & of L2 (an Indonesian English teacher)
Academic writing/argumentative essays
Blog writing
compared
MANOVA
Current study versus Crossley & McNamara (2009)
Are they significantly different?
Indices
Current study
Crossley & McNamara (2009)
Argument overlap, #29
L1 < L2
L1 > L2
LSA given/new, #46
Verb Incidence, #85
Lexical diversity, #50
Lexical diversity, #51
Hypernymy for nouns & verbs, #105
L1> L2
Results
Table 2a Expected Results (Crossley & McNamara, 2009)
Indices
Expected
Indices # 29: Argument overlap
L1 > L2
Indices # 46: Latent Semantic Analysis (LSA) give/new
L1 < L2
Indices # 85: Verb incidence
Indices # 94: CELEX word frequency for content words, mean
Indices # 97: Age of acquisition
Indices # 102: Polysemy
Indices # 103: Hypernymy for nouns
Before discussing the findings of the study, please check the samples of the blog posts I have put in the next section!
address: