Home Articles Reader Opinion Editorial Book Reviews Discussion Writers Guide About TCRecord
transparent 13

Artificial Intelligence in Second Language Learning. Raising Error Awareness

reviewed by Mathias Schulze - March 21, 2006

coverTitle: Artificial Intelligence in Second Language Learning. Raising Error Awareness
Author(s): Marina Dodigovic
Publisher: Multilingual Matters, Clevedon
ISBN: 1853598291, Pages: 303, Year: 2005
Search for book at Amazon.com

The book, which appeared in the Second Language Acquisition series, consists of six chapters, acknowledgements section, introduction, conclusion, three appendices (test case sentences and two informal software reviews), the bibliography, and an index. Chapters 1 to 5 provide detailed insights into the pedagogic, linguistic and design considerations, and decisions for software for advanced learners of English for Academic Purposes (EAP) which contains an error detection module that relies on natural language processing (NLP)—a branch of artificial intelligence research. It is mainly the second part of chapter 5 which describes selected aspects of software design and implementation. Chapter 6 discusses issues of software evaluation in CALL (computer-assisted language learning). “The author of this book has made an effort to try the computer out in its capacity of NNS [non-native speaker] error correction” (p. 139). Dodigovic describes her way to a web-based, language-learning tutor as a vehicle for the implementation of remediation strategies for persistent errors. “As the latter required nothing short of a miracle, artificial intelligence was isolated as the only concept that might work” (p. 10).

Chapter 1: “Can Another Language Be Learnt?” discusses in some detail second language acquisition theories as they pertain to the project at hand, identifies characteristics of the EAP students at this Australian university (characteristics which they possibly share with many others), and describes a first needs analysis. NLP is “deemed capable of providing this group of students with automatic analysis of their typed English sentences supplying them also with meaningful feedback concerning the grammatical correctness of their output” (p. 38). With its general scope, this is a very ambitious project aim.

Research and development in computer-assisted language learning (CALL) are discussed in chapter 2. Here Dodigovic argues for a development process which is research-based. In her own project, she identifies four subprojects: one development project (creation of course material for EAP students) and three research projects (contrastive analysis of student and professional academic writing, error analysis of students’ writing, needs analysis to obtain information on preferred learning styles and attitudes to CALL). The next chapter focuses on one particular design decision: the reasons for Dodigovic’s choice of the web as the delivery medium for the course materials.

Chapter 4 starts off with a good overview of error analysis and correction. It continues with a brief sketch of some developments and discussions in NLP-based CALL. This discussion is interrupted by an excursion into speech processing which does not appear to play a role in either the further discussion or the design of the language learning software. In her concluding remarks for this chapter, Dodigovic answers the question she posed in the chapter title—“Can Computers Correct Language Errors?”—by stating “that the computer has the ability to capture and correct errors in speech, grammar and writing” (p. 137), but concedes that these “abilities may be limited at the moment” (p. 137).

The next chapter—“How to Develop an Artificially Intelligent Language Tutor?”—talks more about design decisions and a needs analysis carried out by Suphawat, a postgraduate student, as well as the challenges posed by academic writing that are particularly salient for non-native writers. Dodigovic argues that it is more important for her system to “to identify errors . . . and suggest some ways of making the underlying grammar transparent to the student” (p. 188). In the second part of chapter 5, she takes the reader on a walkthrough of the sentence correction element of the program. Through a variety of screenshots (pp. 195–231) illustrating the analysis of a number of sample sentences, she shows what students might see when using the system via a web interface.

Chapter 6 starts off by discussing the general concept of evaluation in CALL. Dodigovic evaluated her program through a comparative error analysis (pre- and post-test texts by the same group of students). She asserts that there is a significant positive difference between the two sets of test results. The questions on the pre-test, pilot lesson on malaria from the system, and of the post-test were very similar (p. 256); the method of answering the questions was different in the pre-test (multiple-choice grammaticality judgment) as compared to the lesson and the post-test (one-sentence answers). The questions in the latter two sets are surprisingly similar (e.g., p. 202 and p. 277). This would be a good reason for the students’ ability to answer these questions with fewer errors after having used the system. It does not prove, however, that the system facilitates language learning, language acquisition, or language awareness.

The book provides detailed information about the preparation of an ICALL project. It makes transparent the reasons for design decisions, which are based on second-language acquisition theories, error analysis, and second language writing. With their interesting insights, these are the strong sections of the book. On the other hand, Dodigovic describes the implementation of the NLP components of her system with much less detail and elaborates less on her design decisions. For example, the decision not to follow an established grammatical formalism for the implementation of the parser grammar is neither well-motivated nor does the project profit from it. Equally problematic is the decision to rely on a small set of buggy rules. Particularly since some of them have the student feedback “hard-wired” in them (p. 192, where one rule seems to cover the sentence “There is a problem occur.” only). Dodigovic concedes in her conclusion that “our project was notably far from completed” (p. 264). It is difficult to estimate the coverage and sophistication of the grammar and error diagnosis algorithms because hardly any information is provided. The lack of discussion of hard problems in natural language processing of learner language in CALL is a weakness of the book.

Overall, Dodigovic succeeds in showing the complex network of variables which need to be considered when designing language learning software with error-diagnostic components.

Cite This Article as: Teachers College Record, Date Published: March 21, 2006
https://www.tcrecord.org ID Number: 12348, Date Accessed: 12/6/2021 3:47:14 PM

Purchase Reprint Rights for this article or review
Article Tools
Related Articles

Related Discussion
Post a Comment | Read All

About the Author
  • Mathias Schulze
    University of Waterloo
    E-mail Author
    MATHIAS SCHULZE is an Assistant Professor of German at the University of Waterloo. Mat's research focus is the application of linguistic theory to computer-assisted language learning (CALL). He has published on language technology in CALL and the acquisition of grammar through CALL. Currently, he is working on the computational implementation of German grammar, aspects of online teaching and learning, and structural aspects of the German used in the Waterloo region.
Member Center
In Print
This Month's Issue