An NLP pipeline as assisted transcription tool for speech therapists

This work presents the design of a computer assisted transcription system for speech-language therapists and an evaluation of its core-module: the NLP pipeline. This pipeline combines a tokenizer, a lemmatizer, a part-of-speech tagger and a spellchecker in order to perform a semi-automatic annotation of speech transcriptions. The implemented module has been evaluated on a corpus of spoken interaction of children with Developmental Language Disorder (DLD) with the caregiver. Results are promising in automatic error detection (F-measure of 0.547 against a Ground Truth of 0.616) but low in automatic error correction, and confirm the effectiveness within an assisted transcription tool.