Rule-based grammatical error detection on spontaneous children’s speech

Authors: Christopher Gebauer, Lars Rumberg, Fabian Witt, Edith Beaulac, Hanna Ehlert, Jörn Ostermann

Abstract:

Successful language acquisition is fundamental for the participation in all relevant areas of life. Unfortunately, developmental language disorders (DLD) are the most common developmental disorders during childhood. An early identification of children with DLD is necessary to ensure educational success and to significantly reduce the risk for issues in mental health, social behavior, and skill development in various areas. In this work we explore the suitability of a rulebased grammatical error correction system for written text, to assess grammatical correctness of spontaneous children’s speech. We evaluate the tool’s capabilities on kidsTALC, a corpus containing dialogues of children with speech language therapists in different elicitation contexts, using the manual and automatically generated transcriptions. A qualitative analysis reveals that disfluencies and a lack of context are the leading causes for misclassifications. Nevertheless, the system achieves a FG1-score of 0.92 when applied to manual transcriptions.