This paper presents the set-up and results of the rule-based Cooperative State University Karlsruhe (CSU-K) system for the 2nd edition of the shared spoken CALL ESL task. The data was collected from Swiss teenage students using a speech-enabled online tool for English conversation practice. The tool should eventually be able to judge student input with respect to syntactic and semantic correctness. The tasks consisted of training data of a German text prompt with the associated audio file containing an English language response by the students. In the second edition of the task, 6.698 utterances were provided in addition to the 2017 task. The contribution of this paper is a further look at how rule-based systems can be employed for these sorts of tasks. Meaning and grammar are treated separately in order to classify the language as correct. A number of experts were constructed to deal separately with different POS such as nouns, adjectives, verb usage and pronouns or determiners. Distance measurements derived from Doc2Vec where then employed between utterance and prompt responses. A D-value of 10.08 is reported on the final 2nd Edition evaluation test files.
Cite as: Jülg, D., Kunstek, M., Freimoser, C.P., Berkling, K., Qian, M. (2018) The CSU-K Rule-Based System for the 2nd Edition Spoken CALL Shared Task. Proc. Interspeech 2018, 2359-2363, doi: 10.21437/Interspeech.2018-1000
@inproceedings{julg18_interspeech,
author={Dominik Jülg and Mario Kunstek and Cem Philipp Freimoser and Kay Berkling and Mengjie Qian},
title={{The CSU-K Rule-Based System for the 2nd Edition Spoken CALL Shared Task}},
year=2018,
booktitle={Proc. Interspeech 2018},
pages={2359--2363},
doi={10.21437/Interspeech.2018-1000}
}