Detecting Problems in Spoken Child-Computer Interaction
Dino Seppi, Matteo Gerosa, Björn Wolfgang Schuller, Anton M. Batliner, Stefan Steidl · OPUS (Augsburg University) · 2008
In this paper we describe the effectiveness of some linguistic features for detecting problems in spoken child-computer interactions.To this aim, we use an Automatic Speech Recognizer for generating the spoken word chain, and a word tokenizer for obtaining the lexical and stemming information.Automatic classification of each turn is eventually achieved by exploiting the frequencies of tokens' classes.The impact of ASR and tagger accuracy on automatic detection are discussed by comparing fully automatic with manually corrected approaches.