Full Program »

Feedback to Grammatical Errors in German as Second Language Focused on the Learner’s Personal Acquisition Level

In the research area of Intelligent Computer-Assisted Language Learning (ICALL), grammar/linguistic-awareness teaching figures prominently (cf. Roehr, 2007; Amaral&Meurers, 2011; Meurers, 2013). As result of a corpus study evaluating written essays of 220 French-speaking learners of German as second language (GSL) in primary and secondary schools in Geneva/Switzerland, Erika Diehl and colleagues (2000) report that even under classroom conditions, pupils use similar acquisition strategies to those observed in first language acquisition. For the three strands in grammar acquisition: (1) verbal morphology, (2) verb placement and (3) case marking, they identified dedicated acquisition stages—which do not reflect the grammar instruction lessons. In general, none of the consecutive steps in each individual strand can be left out, i.e. vertically no deviation is possible, whereas horizontal alignment within each strand is a matter of personal individualization. According to Diehl et al. (2000), these results strongly support Pienemann’s so-called “teachability hypothesis” (1989), i.e. grammar instruction has only a chance to be effective if it takes natural acquisition orders and strategies into account. Similarly, for learners of English, Varnosfadrani & Ansari (2011) claim that “[their] finding lends support to suggestions that corrective feedback (like other types of form-focused instruction) needs to take into account learners’ cognitive readiness to acquire features.”

Based on these empirical findings, we present a GSL-e-learning system where the learner can freely construct phrases/sentences. The system notifies only syntactic errors up to a personal level of proficiency with German. In contrast, the system ignores/overlooks errors on later stages the user has not yet mastered. In order to identify the learner’s current problem, the student’s and the system’s generation process become aligned. For any action by the learner, e.g. adding/inflecting word forms or (re-)arranging word order, a natural-language paraphrase-generator calculates whether the linguistic construction is correct—scaffolded writing (Harbusch&Kempen, 2011). For a recent reimplementation, see COMPASSIII (Harbusch et al., 2013). In our presentation, we illustrate the necessary modifications of the COMPASSIII system allowing the generation of licensed errors. For that purpose, we define underspecification rules generating wrong constructions on any level in each strand by modifying the underlying syntactic grammar formalism Performance Grammar (PG) (Harbusch&Kempen, 2002; Kempen&Harbusch, 2002). Strands (1) and (3) according to Diehl et al. (2000) coincide with independently applicable dominance-structure construction-rules. In strand (1), Conjugation features of the Verb(s) become underspecified/weakened; in strand (3), Noun-Phrase Declination-features for a chosen grammatical function. Word-ordering rules in PG are related to strand (2) where underspecification rules vary the Verb’s position.
In the beginning of a session with our system, the learner’s proficiency level within each strand has to be determined. Accordingly, the underspecification rules get divided into two sets. For any construction the learner enters, the generator calculates—based on the two underspecification rule sets: (1) all paraphrases with errors the learner can already master (notified as feedback) and (2) those beyond that scope (in order to avert arbitrary lists of words as input). The strategy how to present feedback for type-(1) errors in our system follows empirical findings (see, e.g., Kartchava, 2012).
Amaral, L. & Meurers, D. (2011). On Using Intelligent Computer-Assisted Language Learning in Real-Life Foreign Language Teaching and Learning. ReCALL, 23 (1):4–24.
Diehl, E., Christen, H., Leuenberger, S., Pelvat, I. & Studer, T. (2000). Grammatikunterricht: Alles für der Katz? Tübingen: Niemeyer.
Harbusch, K., Härtel, J. & Cameran, C.-J. (2013). COMPASSIII: Teaching L2 grammar graphically on a tablet computer. Proc. of SLaTE 2013, Grenoble, France.
Harbusch, K. & Kempen, G. (2002). A quantitative model of word order and movement in English, Dutch and German complement constructions. Proc. of the 19th COLING, Taipei.
Harbusch, K. & Kempen, G. (2011). Automatic online writing support for L2 learners of German through output monitoring by a natural-language paraphrase generator. In: M. Levy, F. Blin, C. Bardin Siskin & O. Takeuchi (eds.), WORLDCALL, New York: Routledge, pp. 128–143.
Kartchava, E. (2012). Noticeability of corrective feedback, L2 development and learner beliefs. PhD Thesis, Université de Montréal, Canada.
Kempen, G. & Harbusch, K. (2002). Performance Grammar: A declarative definition.
In: Theune, M., Nijholt, A. & Hondorp, H. (eds.), Computational Linguistics in the Netherlands 2001. Amsterdam: Rodopi, pp. 148–162.
Meurers, D. (2013). Natural Language Processing and Language Learning. In: Chapelle, C.A. (ed). Encyclope-dia of Applied Linguistics, Oxford, UK: Blackwell, pp. 1–13.
Pienemann, M. (1989) Is language teachable? Psycholinguistic experiments and hypothesis. Applied Linguis-tics, 10 (1):52–79.
Roehr, K. (2007). Metalinguistic knowledge and language ability in university-level L2 learners. Applied Linguistics, 29 (2):173–199.
Varnosfadrani, A.D. & Ansari, D.N. (2011). The Effectiveness of Error Correction on the Learning of Mor-phological and Syntactic Features. World Journal of English Language, 1 (1):29–40.


Karin Harbusch    
Computer Science Department
Universität Koblenz-Landau

Karin Harbusch is professor of Computational Linguistics and Artificial Intelligence at the Computer Science Department of the University of Koblenz-Landau since 1995. Her primary interest is in natural-language generation and understanding by computer, in particular based on the formalisms of Tree Adjoining Grammar and Performance Grammar. Applied projects deal with tutorial e-learning systems for first- and second-language teaching, and with computational writing support for severely motor-impaired users. From 1989 till 1995 she held a position as senior researcher at the DFKI (German Research Center for Artificial Intelligence) in Saarbrücken, where she had received her PhD at Saarland University in 1989. At the DFKI, she participated in several language generation and knowledge representation projects, including the VERBMOBIL research program on spoken language translation.

Christel-Joy Cameran    
Computer Science Department
Universität Koblenz-Landau

Christel-Joy Cameran is a PhD student at the Computer Science Department of the University of Koblenz-Landau since 2013. Her primary interest is in tutorial e-learning systems for first- and second-language teaching.

Johannes Härtel    
Computer Science Department
Universität Koblenz-Landau

Johannes Härtel a Master student at the Computer Science Department of the University of Koblenz-Landau since 2013. His primary interest is in tutorial e-learning systems for first- and second-language teaching.


Powered by OpenConf®
Copyright ©2002-2013 Zakon Group LLC