Automatic detection of grammatical structures from non-native speech

This study focuses on the identification of grammatical structures that could serve as indices of the grammatical ability of non-native speakers of English. We obtain parse trees of manually transcribed non-native spoken responses using a statistical constituency parser and evaluate its performance on noisy sentences. We then use the parse trees to identify the grammatical structures of the Index of Productive Syntax (IPSyn), previously found useful in evaluating grammatical development in the context of native language acquisition. Empirical results of this study show: a) parsing ungrammatical sentences using a probabilistic parser suffers some degradation but is still useful for further processing; and b) automatic detection of the majority of the grammatical structures measured by IPSyn can be performed on non-native adult spoken responses with recall values more than 90%. To the best of our knowledge, this is the first study which explores the relationship between parser performance and the automatic generation of grammatical structures in the context of second language acquisition.