Dictation is considered an efficient exercise for testing the language proficiency of learners of French as a Foreign Language (FFL). However, the traditional teaching approach to dictation reduces the instructional feedback efficiency. To remedy this, this study adopts a design-based research approach and builds an automatic error type annotation platform for dictation practice named FRETA-D (French error type annotation for dictation) to pursue intelligent pedagogical feedback for both FFL teachers and students. FRETA-D can automatically identify error boundaries as well as classify the errors into fine-grained error types in learners' dictation texts. FRETA-D features a dataset-independent classifier based on a framework with 25 main error types, which is generalized from French grammar rules and characteristics of frequent learner dictation errors. Five French teachers are invited to evaluate the appropriateness of automatically predicted error types of 147 randomly selected samples, and the acceptance rate reaches more than 85%. Automatic evaluation on 1,009 sentences by comparing with manually labeled references also shows promising results, reaching more than 85% consistency with human judgments. The accessibility of FRETA-D has also been confirmed by 50 Chinese undergraduate FFL learners with different professional backgrounds. FRETA-D facilitates conducting dynamic statistical analysis of learners'error types. And we share the same findings with previous studies that there exist causal links between the dictation errors and learners' mastery of French phoneme and grapheme.
Keywords: AI program; artificial intelligence; independent study; instant feedback; language education; language learning.
Copyright © 2023 Qin, Luo and Zhai.