Abstract
Background
Mean length of the utterance (MLU) and Index of productive syntax (IPSyn) are well-established measures of language acquisition. Previous research revealed that different types of MLU are suitable for different languages (Parker & Brorson, 2005; Tomas & Dorofeeva, 2019;Zajacová et al., 2011). So far no one has compared which one is suitable for Czech. Likewise, IPSyn (Altenberg, Roberts, & Scarborough, 2018) is a widely used measure (Washington et al., 2019 for English and Jamaican Creole; Nieminen, 2009 for Finish) that has not yet been adapted to many Slavic languages (the only exception is Russian to our knowledge;Chernobilsky, 2009). Our research aims to compare MLU measured in various units and on different sample lengths with each other as well as with IPSyn and test-based measures at two points in time with the same children.
Methods
Recordings of 110 children were made during free play at two time points: 2;6 years and then at 3;11.On both occasions, their receptive vocabulary (VOC) and grammar (GRAM) were tested and then used as test-based measures of their language. For transcript-based measures, we compared MLU in syllables, morphemes, and words in transcripts of different lengths (50, 75, 100, and all available utterances) and we adapted the IPSyn measure to the needs of the Czech language and the constraint of working only with the morphological annotation so far.
Results
The different MLU measures correlated closely in all lengths of transcripts (all r's > 0,89) with the highest values for transcripts of all utterances (see Table 1). For further analyses, we used MLU measured in words in all utterances. We found strong correlations between transcript-based measures of MLU and IPSyn in both time points and no or moderate correlations between test-based measures of grammar and vocabulary. At the first point in time, the grammar test showed a higher correlation with IPSyn than with MLU which leveled out at the second point in time (see Table 2). Regression models showed that both MLU and IPSyn in 2;6 years predict themselves in 3;11(β = 0.35 and 0.37, respectively). In the case of IPSyn, vocabulary showed a unique effect above and beyond the factors of other predictors (β = 0.26), in the case of MLU, vocabulary was at the limit of significance.
Discussion
Our results confirm that MLU in words is an adequate measure. This is an important finding for Czech and potentially other Slavic languages since their highly inflectional character complicates the automatic calculation of MLU in morphemes or syllables.IPSyn is a valid measure and it shows reasonable predictive validity, with later IPSyn scores being predicted by earlier IPSyn scores and by a test-based vocabulary measure. IPsyn also correlates closely with MLU. It also appears to be a better measure of grammar than MLU since at the first point of time it correlates with grammar test more than does MLU.