O n f e r e n c e

Elena Cotos, Stephen Gilbert, Jivko Sinapov 
Iowa State University, Iowa, United States of America 
[e-mail], [e-mail], [e-mail] 
NLP-Based Analysis of Rhetorical Functions for AWE Feedback 
Bio data 
Dr. 
Elena Cotos
is the Director of the Academic Communication Program in the 
Graduate College at Iowa State University. She is also Adjunct Assistant Professor in the 
TESL/Applied Linguistics and Technology Program, English Department. Her research 
interests include AWE, I/CALL, language assessment, genre analysis, and learner 
corpora. Dr. Cotos’ work has been published in 
professional journals, edited volumes, 
and her new book “Genre
-
based Automated Writing Evaluation for L2 Research Writing” 
will be published in the forthcoming year by Palgrave McMillan. 
Dr. 
Stephen Gilbert
is the Associate Director of the Virtual Reality Application Center 
and Assistant Professor of Industrial and Manufacturing Systems Engineering at Iowa 
State University. His research focuses on human-computer interaction, E-learning design 
and development, software usability, user-centered design, instructional technologies, 
and distance education. Dr. Gilbert’s work appeared in The International Journal for 
Artificial Intelligence in Education and in numerous peer-reviewed conference 
proceedings. 
Dr. 
Jivko Sinapov
specializes in computer science and human-computer interaction. He 
is currently a Postdoctoral Researcher and Instructor affiliated with ISU's Human-
Computer Interaction program. His current research interests include robotics, 
computational perception, machine learning, and natural language 
processing. 
Abstract 
Automated Writing Evaluation (AWE) platforms (e.g. Criterion, MyAccess, WriteToLearn, 
Folio, etc.) have employed natural language processing (NLP) in a variety of ways to 
provide language learners with formative feedback on essay writing. NLP approaches 
have been successfully exploited for feedback on such aspects of the writing construct as 
grammar, usage, mechanics, style, organization, coherence, and content. However, 
automated analysis of rhetorical intent, especially in genres other than the essay, 
remains a challenging methodological problem. Finding solutions to this problem requires 
an interdisciplinary research agenda involving applied linguists, computational linguists, 
and computer science experts, whose combined efforts would result in the creation of 
more advanced intelligent feedback systems. 
Our interdisciplinary team explored the challenge of automated genre analysis with the 
applied purpose of developing a new genre-based AWE, the Research Writing Tutor 
(RWT). RWT targets the research article genre and generates feedback on the 
communicative effectiveness of learner discourse. In this paper, we first review the 
approaches to and implementations of automated discourse categorization in order to 
provide a background for our automated genre analysis methodology.

---

-118- 
2014 CALL Conference 
LINGUAPOLIS
www.antwerpcall.be 
Then, we describe the development of an automated discourse classification system 
which employs a 3-
step process of discourse structure identification: feature selection 
from annotated corpus data, sentence representation, and training leading to sentence-
level classification into rhetorical moves and steps (Swales, 1981, 1990, 2004). Lastly, 
we demonstrate how our categorization approach was applied to the generation of RWT 
feedback on the rhetorical functions of individual sections of the research article. With 
this work, we not only address a challenging NLP task, but also pave the road for future 
AWE endeavours that will facilitate construction of meaning in academic genres.