Using LENA to evaluate child language input and language learning in audio environment featuring three languages
For nearly all, language is a birthright. The capacity for language acquisition runs deep in human biology. Unless human communicative interaction is essentially absent, each child will learn at least one language by the time he or she is four years old.
Two conflicting proposals are on the table regarding what language experiences propel this development. One, pertaining to a nativist approach, holds that input in the general audio environment drives language development regardless of how it is presented. The other, based on a functionalist approach, posits that parents and other caregivers adjust their talk to the child to make it more comprehensible; this approach implies that input directed toward the child is particularly significant in determining the emergence of language structure and language choice.
An interesting test of these conflicting proposals is possible in the context of multilingual learning. With all-day home recordings, it is now possible to assess language input in a way that is representative of the total input pattern and to determine, for example, the relation between the amount of input vocabulary and child's vocabulary learning in each language. Perhaps most important is the evaluation of the amount of input in each language that is directed toward the child. Some parents may use one or more languages when they talk with a child and an entirely different language when they talk with one another.
The following is a brief report of a study presented at the ASHA convention in November 2008. Using the LENA System to capture data from the natural language environment of a multilingual household, I set out to contribute a perspective on the role of input in each language on multilingual learning and to test the ability of LENA to provide a basis for representative sampling of vocal interaction at home. In the study, the key child was my own daughter. Her Austrian mother's native language is German and second language is English; my native language is English, but I spoke German to the child during the period of the study; and the child's governess spoke almost exclusively in Spanish. Although the parents spoke only in German and the governess spoke only in Spanish to the child, the parents spoke English to one another, and the broader audio environment (i.e., "out on the street") was overwhelmingly English.
Over a 13-month period—when the child was 11 months to 24 months of age—I used LENA to make 11 recordings, each approximately 10 hours in duration and representing all languages of input. Using the LENA reports from the 11 sessions, I isolated 39 five-minute samples that had captured high language volubility for both adults and the child.
Mostly using records that had been kept by paper and pencil, but partly based on the LENA recordings, I estimated the proportions of time spent by the various caregivers with the child. More than a quarter of the time consisted of the mother speaking German with the child. Another quarter was the two parents speaking German to the child and English to each other. Roughly 45 percent of the time was distributed across three other circumstances fairly equivalently: The governess speaking Spanish with the child; the father speaking German with the child; and other circumstances "out on the street," which consisted of a mix of the three languages, depending on who was with the child and whom they encountered.
I then used the LENA reports to quickly locate high volubility periods and count the words spoken by each adult and the child in each of the three languages. A fluent speaker of all three languages, I did these counts myself. The counts also included an indication regarding whether each adult word was spoken to the child or to some other individual. Finally the counts were used to calculate and graph raw word counts, proportional word counts (adjusted for input time distribution by language), and lexical types from the child compared with proportional input amounts.
Here are the results in a nutshell. The child spoke about as many words in German as one would have predicted based on how much German she heard. However, almost all the German she heard was directed to her. Interestingly, she spoke many times less English than would have been predicted by the amount of English she heard. The key here is that English was almost never directed to her. Likewise, she spoke Spanish words many times more than English words even though the estimated total input in Spanish and English was about equal. The difference? The Spanish in her environment was usually directed to her.
These findings offer support for a functionalist viewpoint. The type of input appeared to matter in a big way—whether speech was directed toward the child played a huge role in the language she learned to use.
The LENA System proved essential to this study. It made it possible to acquire daylong recordings and to locate representative periods of vocal activity so that a reasonable estimate could be made of the child's total language input and output. Consequently, it was possible to address the role of input types in the child's learning at relatively low cost, especially in the time needed to make the word counts. The most important aspect of this study, for me, is that it supplies additional confidence in the potential of all-day recording and automated word and utterance counts as a basis for representative sampling by lexical or phonetic transcription of the home language environment. The LENA System is, in my opinion, a very important new tool for the study of child language.
