The ICICLE Project

ICICLE (Interactive Computer Identification and Correction of Language Errors) is a project designed to provide writing assistance for second language learners of English, specifically American Sign Language natives. The system will analyze written English texts (usually short scholastic essays) from deaf individuals, identifying possible errors and generating tutorial text (identifying errors, suggesting corrections, discussing grammatical rules) tailored to each writer's level of language competence and particular learning strengths.

Table of Contents Background Purpose Method Design Progress Publications Acknowledgements

Contributors Diego Matias Dominguez, Lisa Michaud, Kathy McCoy, R. Jeffrey Morriss, Chris Pennington, David Schneider, Linda Suri

The problem of deaf literacy has been well-documented and has far reaching effects on every aspect of deaf students' education. Though data on writing skills is difficult to obtain, we note that the reading comprehension level of deaf students is considerably lower than that of their hearing counterparts.

Some deaf people use American Sign Language (ASL). ASL is a visual-gestural language whose grammar is distinct and independent of the grammar of English or any other spoken language, with a structure that is radically different from that of English, being much more similar to that of Chinese. The order of signs in ASL does not correspond to the word order of English, and ASL includes non-manual behavior such as posture and facial expression for morphological and grammatical purposes. This contributes further to the distance between ASL, which encourages simultaneous communication of information, and written English, which is completely sequential. Because of these differences, we hold that ASL natives acquiring English are essentially taking on a task of second language acquisition, learning a new and distinct language that does not share many features with their own.

Adding to the difficulties that rise from the grammatical differences between the languages is the fact that ASL has no accepted written form, eliminating the opportunity other second language learners have to establish literacy skills in a fluent native language and then transfer those skills to the new language being learned. Deaf learners of English also have little to no exposure to English input keyed down to their language level, in an easily-accessed form such as the input a hearing learner can have just from listening to the language. This underscores the need for a way to tutor deaf learners so that their unique needs as language learners are met.

The goal of this project is to develop a computer tool to provide tutoring to improve the written English of deaf writers. The envisioned program will accept a writing sample, usually several sentences in length, and analyze the document for errors. It then will select which errors are within the writer's grasp of understanding, and construct a tutorial response appropriate to both the learner's level of English and his learning style and strengths. Essentially, the system will work toward a higher level of proficiency in written English for the writer using selective correction and individualized tutoring.

The design of this program is based on the belief that English should be viewed as a second language for many deaf people, as explained above. Due to the individual's relative lack of exposure to English, his initial concept of English grammar rules should be largely based on what he understands of ASL, his native language. As his learning progresses, more and more of his English usage settles into correct English patterns, but there is still a (decreasing) realm of language features the learner has yet to acquire. Between the acquired English proficiency and the future knowledge lies a realm of experimentation, where the learner is varying in his usage as he tries out new knowledge and hypotheses about how the English rules should work. It has been shown in second language acquisition research that this realm is where most of the errors in a learner's usage are made, and that here is where correction and tutoring can be most beneficial. Therefore, our system will model a learner's level of acquisition and determine this realm of experimentation for use in identifying errors and tailoring the corrections.

To work toward a model of English acquisition for a deaf learner, we have analyzed writing samples and we have a taxonomy of errors typical of the deaf individual. We are now working on developing a concrete order of acquisition to show which features are mastered in what order so that a user model can be constructed on which the learner's proficiency is shown.

The system we are designing will consist of two phases. In the first phase, the system will identify the errors present in a learner's text. To do this, it relies on a grammar of English which has been augmented with a set of error rules which capture the errors in our taxonomy. When errors are ambiguous, the user model may be used to determine what is the most likely error given the learner's level of proficiency. The identified errors are then filtered, and the ones most useful for tutoring are passed to the second phase, where the tutorial response is generated.

Now that the content is chosen, the manner of the response must be selected. The system will have at its disposal several possible tutorial techniques, from a simple identification of the grammar rule in violation to a production of examples of similar sentences using the grammar rule correctly. Which technique is chosen will depend on the individual learner; the system will track the success of the various techniques and tailor its responses to use those which produce the most success with the student over multiple sessions. The system will also take into account what the student knows already, and will avoid giving lengthy explanations when unnecessary. The student will always have the option to request more information, to ask for terms to be defined or explained, or to explore other methods of tutoring according to his tastes. These preferences will be recorded and considered in subsequent sessions.

In essence, the system is designed to work with a learner over multiple sessions, to test both its own success and the learner's progress over time.

The initial analysis of writing samples has resulted in a taxonomy which is already implemented in the rules used to identify errors in learner text. A prototype system now exists with these rules, and in a window-based environment can load in a user's text, analyze it, and identify and highlight all errors, giving a simple one- or two-sentence explanation of the error. Errors of different types are highlighted in different colors, and the user can elect to view all errors of only a certain type. He can also pop up a sub-window to edit any sentence, re-entering it into the text and causing an immediate re-analysis.

Future plans include extensive work on the response generation process. Research is currently being started on learning strategies in the second language acquisition process and how to generate text intelligently (as opposed to giving pre-prepared reponses).

Schneider, David and Kathleen F. McCoy. Recognizing Syntactic Errors in the Writing of Second Language Learners. In Proceedings of the Thirty-Sixth Annual Meeting of the Association for Computational Linguistics and the Seventeenth International Conference on Computational Linguistics (COLING-ACL), Volume 2, Montreal, Quebec, Canada, August 10-14, 1998.
[abstract] [postscript (328K)]

Michaud, Lisa N., and Kathleen F. McCoy. Planning Text in a System for Teaching English as a Second Language to Deaf Learners. In Proceedings of Integrating Artificial Intelligence and Assistive Technology, an AAAI '98 Workshop, Madison, Wisconsin, July 26, 1998.
[abstract] [postscript (317K)]

Michaud, Lisa N.. Tutorial Response Generation in a Writing Tool for Deaf Learners of English (an abstract and poster presentation). In Proceedings of the Fifteenth National Conference on Artificial Intelligence (AAAI '98), Madison, Wisconsin, July 26-30, 1998.
[postscript (145K)]

McCoy, K. F., and Masterman, L. N. (1997), A Tutor for Teaching English as a Second language for Deaf Users of American Sign Language, in Proceedings of Natural Language Processing for Communication Aids, an ACL/EACL'97 Workshop, Madrid, Spain, July, 1997.
[abstract] [postscript (391K)]

McCoy, K. F., Pennington, C. A., & Suri, L. Z. (1996) English error correction: A syntactic user model based on principled "mal-rule" scoring. In Proceedings of UM-96, the Fifth International Conference on User Modeling. Kailua-Kona, Hawaii, January 1996, pp. 59-66.
[abstract], [text (45K)], [postscript (388K)]

Suri, L. Z. (1993). Extending focusing frameworks to process complex sentences and correct the written English of proficient signers of American Sign Language. Technical Report 94-21, Department of Computer and Information Sciences, University of Delaware, Newark, DE.
[abstract], [postscript (1.07M)]

Suri, L. Z., & McCoy, K. F. (1993) Correcting discourse-level errors in a CALL system for second language learners. Technical Report 94-02, Department of Computer and Information Sciences, University of Delaware, Newark, DE.
[abstract], [postscript (154K)]

Suri, L. Z., & McCoy, K. F. (1993). A methodology for developing an error taxonomy for a computer assisted language learning tool for second language learners. Technical Report 93-16, Department of Computer and Information Sciences, University of Delaware, Newark, DE.
[abstract], [postscript (292K)]

Suri, L. Z. (1992). Correcting illegal NP omissions using local focus. In Proceedings of the 30th Annual Meeting of the Association of Computational Linguistics (pp. 273-275). University of Delaware, Newark, DE.
[abstract]

Suri, L. Z. (1992). Using local focus to correct illegal NP omissions (a Ph.D. proposal). Technical Report 93-07, Department of Computer and Information Sciences, University of Delaware, Newark, DE.
[abstract], [text (94K)], [postscript (301K)]

McCoy, K. F., & Suri, L. Z. (1991). Natural language processing principles for improving deaf writing. Rehabilitation R & D Progress Reports / Journal of Rehabilitation Research and Development, 29, 331.
[abstract], [text 4K], [postscript (50K)]

Suri, L. Z. (1991). Language transfer: A foundation for correcting the written English of ASL signers. Technical Report 91-19. Department of Computer and Information Sciences, University of Delaware.
[abstract]

Suri, L. Z., & McCoy, K. F. (1991). Language transfer in deaf writing: A correction methodology for an instructional system. Technical Report 91-20. Department of Computer and Information Sciences, University of Delaware.
[abstract]

This work has been supported in part by a NSF grant #IRI-9010112. Additional support has been provided by the Nemours Foundation.