Abstrakt
The paper presents a longitudinal corpus of transcribed spontaneous child-adult interactions in Czech. It consists of 99,388 tokens in 42,103 utterances produced by seven children between ca 1.5 to 3.5 years of age, and 238,211 tokens in 61,252 utterances produced by their close caregivers in everyday situations at home.
The corpus covers language production of the children from the mean length of 1.01 word per utterance up to 5.33 words per utterance. The length of the recorded period ranges for individual children from 11 to 27 months.
The transcripts of both child and adult utterances were lemmatized and tagged using MorphoDiTa, a tool for automatic morphological analysis of Czech. The annotation was transformed into the MOR format used within CHILDES, a database dedicated to corpora of first language acquisition.
Detailed manual checking was performed on the annotation of all children's utterances. Data from three children were used for a comparison of part-of-speech classification before and after manual checking, data from one child was additionally analyzed for differences in morphological tagging proper.
The number of differences was rather low, with (expected) limitations in the areas of part-of-speech classification for uninflected words, annotation of homonymous forms, and annotation of child-specific words. The corpus represents an important contribution to the research of child language with special significance for Slavic languages and other morphologically rich inflecting languages, which are still underrepresented in the study of first language acquisition.