Poster #120 - Language Sample Analysis of All-Day Language Recordings: Are there differences in existing sample selection methods?

• In Event: 1-115 - Poster Session 03
  In Poster Session: PS 03 Section - Methods, History, Theory

Integrative Statement

All-day recordings of children’s home language environment using the Language Environment Analysis (LENA) have become increasingly common in developmental research. Approximately 6 studies using the LENA have been published annually since 2012; most employed LENA’s automated data to describe children’s language input, specifically frequency counts of adult words and child-adult conversational turns (Greenwood et al., 2018). However, LENA offers great potential to explore additional input characteristics relevant to language development. Language sample analysis permits deeper exploration of LENA data through transcription and coding of discrete audio segments for more precise linguistic variables. To date, 8 studies using the LENA have undertaken language sample analysis. Yet, consensus on how to select audio samples has not been reached. Researchers have selected audio segments by the highest number of adult words (Burgess et al., 2013; Weisleder & Fernald, 2013), conversational turns (Gilkerson et al., 2015; Kashinath et al., 2015), child vocalizations (Canault et al., 2015; Caskey et al., 2013; Oller, 2010; Wood et al., 2014), and percentage of speech within 6 feet of the child (Cycyk & Hammer, in preparation). Because sampling techniques likely differ in the information provided (e.g., Tamis-LeMonda et al., 2017), the purpose of this study is to evaluate these sample selection methods to guide researchers using the LENA. Specifically, we asked if the language input data obtained from each of the aforementioned selection methods differed. Methods. Thirty-one Spanish-speaking families participated. Their children averaged 19.73 months of age (SD = 3.19). Families recorded a mean total of 14 hours, 44 minutes using the LENA. Four audio segments, each 5 minutes long, were selected from every family’s complete audio data using a specific selection method: (1) highest number of adult words (AWC); (2) highest number of conversational turns (CTC); (3) highest number of child vocalizations (CVC); (4) highest percentage of speech within 6 feet of the child. Transcription and coding of language samples is underway. Along with the LENA automated values for AWC and CTC, these input variables will be explored: total number of words, total number of different words, mean length of utterance, and proportion of child-directed utterances. Differences between sample selection methods will be determined via repeated measures ANOVAs, comparing the means of each input variable by selection method. Results. At submission, language sample transcription was 25% complete. Preliminary comparisons of mean AWC and CTC by samples selected for highest AWC, CTC, and CVC are complete. ANOVAs revealed significant effects of sample selection method on AWC [F(2, 30 = 92.89, p < 0.01] and CTC [F(2, 90) = 14.94, p < 0.01]. The highest mean values for AWC and CTC corresponded to samples selected for highest AWC and CTC, respectively. Significant differences in mean values for AWC and CTC were observed across sampling methods (see figures 1 and 2). For example, AWC was lowest for samples selected by highest number of child vocalizations. This suggests that specific sampling methods are suited to assessing particular input features and that methodological guidance is needed on sample selection for language sample analysis.

Authors

• Lauren M Cycyk, University of Oregon
  Non-Presenting Author

• Stephanie De Anda, University of Oregon
  Presenting Author

• Maria Kapantzoglou, Portland State University
  Non-Presenting Author

• Aquiles Iglesias, University of Delaware
  Non-Presenting Author