Automated grammatical tagging of child language samples.

ABSTRACT

Recent studies of the automated grammatical categorization ("tagging") of words using probabilistic methods have reported substantial levels of accuracy-over 95% agreement with manual tagging for words from a variety of texts. However, the texts with which this method has been tested were written by adults and edited by publishers. The present study examined the accuracy with which such methods could tag transcribed conversational language samples from 30 normally developing children. On a word-by-word basis, automated accuracy levels ranged from 92.9% to 97.4%, averaging 95.1%. Accuracy at correctly tagging whole utterances was lower, ranging from 60.5% to 90.3%, with an average of 77.7%. Probabilistic methods of coding language samples hold potential as a viable tool for child language research. Further study and improvement of automated grammatical tagging is warranted and necessary before widespread use can be made of this technology.