The validity of LENA technology for assessing the linguistic environment and interactions of infants learning Hebrew and Arabic

From data to discovery: Technology propels speech-language research and theory-building in developmental science

Research on speech and language development has a long history, but in the past decade, it has been transformed by advances in recording technologies, analysis and classification tools, and AI-based language models. We conducted a systematic literature review to identify recently developed (semi-)automatic tools for studying speech-language development and learners' environments in infants and children under the age of 5 years. The Language ENvironment Analysis (LENA) system has been the most widely used tool, with more and more alternative free- and/or open-source tools emerging more recently. Most studies were conducted in naturalistic settings, mostly recording longer time periods (daylong recordings). In the context of vulnerable and clinical populations, most research so far has focused on children with hearing loss or autism. Our review revealed notable gaps in the literature regarding cultural, linguistic, geographic, clinical, and social diversity. Additionally, we identified limitations in current technology-particularly on the software side-that restrict researchers from fully leveraging real-world audio data. Achieving global applicability and accessibility in daylong recordings will require a comprehensive approach that combines technological innovation, methodological rigour, and ethical responsibility. Enhancing inclusivity in participant samples, simplifying tool access, addressing data privacy, and broadening clinical applications can pave the way for a more complete and equitable understanding of early speech and language development. Automatic tools that offer greater efficiency and lower cost have the potential to make science in this research area more geographically and culturally diverse, leading to more representative theories about language development.

The use of Language ENvironment Analysis in autism research: A systematic review

The Language ENvironment Analysis (LENA) system has been used increasingly in research to record and evaluate the everyday speech of autistic children and their families. However, it is unclear how researchers are using LENA and whether the system is well-suited for work with autistic individuals. The purpose of this systematic review is to summarize the use of LENA in autism research, to highlight the strengths and limitations of the system as identified by researchers, and to provide recommendations for future research and clinical use. Forty-two studies that used LENA with samples of autistic children were identified through a systematic database search. Researchers using LENA in autism did so across a variety of ages, settings, and analytical approaches. Most studies used LENA within recommended guidelines. The most common purpose of using LENA was for exploratory research. Noted strengths of the LENA system included ecological validity, cost-effectiveness, and timely clinical feedback. Limitations included lower rates of speaker identification compared to human coders and limited information regarding speech context and language development. This systematic review provides key insights into the methods surrounding LENA use in autism research and serves to inform researchers and clinicians on best practices for future use with this technology.Lay abstractIn research, language ability has historically been measured using structured tasks in laboratory settings. In recent years, there has been a growing emphasis on the need to instead capture language ability in an individual's natural setting (i.e. through social interaction or in their home). Considering natural language may be particularly important for the autistic population, as an autistic child's language ability can be very different depending on the setting. One common tool for capturing natural language is the LENA recording system, which takes audio recordings over long periods of time and provides estimates of children's and caregivers' speech. The purpose of this systematic review is to summarize the use of LENA in autism research, to highlight the strengths and limitations of the system as identified by researchers, and to provide recommendations for future research and clinical use. We identified 42 autism studies that used LENA in a variety of ways and settings. Most studies used LENA within the guidelines put forth by its creators, and it was most commonly used to understand speech or speech development for autistic children. LENA is a useful tool for clinicians and caregivers to gain some insights into child speech, but those considering using it should be aware of concerns about its accuracy and limitations about the information it provides. In this review, we supplement the official LENA guidelines with specific suggestions for use with the autistic population.

Validation of the Language ENvironment Analysis in Swedish Children

PURPOSE

The Language ENvironment Analysis (LENA) is a technological tool designed for comprehensive recordings and automated analysis of young children's daily language and auditory environments. LENA recordings play a crucial role in both clinical interventions and research, offering insights into the amount of spoken language children are exposed to in their homes, including adult word count (AWC) and child vocalization count (CVC). Although LENA was initially developed for American English, it has been validated in various other languages. This study aimed to evaluate the reliability of LENA in a Swedish context by comparing its automatic estimates with human transcriptions.

METHOD

Thirty-six children aged 11-29 months participated in the validation study. A total of 540 min (9 hr) of recordings were transcribed by one experienced speech-language pathologist and two special educators. Interrater agreement over 150 min was notably higher for CVC (intraclass correlation coefficient [ICC] = .97, 95% confidence interval [CI] [.91, .99], r = .96, p < .001) compared to AWC, which yielded an ICC of .82 (95% CI [.15, .96], r = .80, p = .006).

RESULTS

Both estimated CVC and AWC (n = 36) were significantly correlated with human transcriptions (r = .79 and .82, respectively; p < .001).

CONCLUSIONS

The findings align with prior validation studies, indicating that LENA is suitable for use in a Swedish context, particularly for families with children aged 11-29 months. Nonetheless, additional validation studies are necessary, particularly focusing on younger infants, to enhance our understanding of the tool's reliability in preverbal children and their caregivers.

Auditory sequence learning with degraded input: children with cochlear implants ('nature effect') compared to children from low and high socio-economic backgrounds ('nurture effect')

Implicit sequence learning (SL) is crucial for language acquisition and has been studied in children with organic language deficits (e.g., specific language impairment). However, language delays are also seen in children with non-organic deficits, such as those with hearing loss or from low socioeconomic status (SES). While some children with cochlear implants (CI) develop strong language skills, variability in performance suggests that degraded auditory input (nature) may affect SL. Low SES children typically experience language delays due to environmental deprivation (nurture). The purpose of this study was to investigate nature versus nurture effects on auditory SL. A total of 100 participants were divided into normal hearing (NH) children, young adults, CI children from high-moderate SES, and NH children from low SES who were tested with two Serial Reaction Time (SRT) tasks with speech and environmental sounds, and with cognitive tests. Results showed SL for speech and nonspeech stimuli for all participants, suggesting that SL is resilient to degradation of auditory and language input and that SL is not specific to speech. Absolute reaction time (RT) (reflecting a combination of complex processes including SL) was found to be a sensitive measure for differentiating between groups and between types of stimuli. Specifically, normal hearing groups showed longer RT for speech compared to environmental stimuli, a prolongation that was not evident for the CI group, suggesting similar perceptual strategies applying for both sound types; and RT of Low SES children was the longest for speech stimuli compared to other groups of children, evidence of the negative impact of language deprivation on speech processing. Age was the largest contributing factor to the results (~ 50%) followed by cognitive abilities (~ 10%). Implications for intervention include speech-processing targeted programs, provided early in the critical periods of development for low SES children.

Validation of the Language ENvironment Analysis (LENA) Automated Speech Processing Algorithm Labels for Adult and Child Segments in a Sample of Families From India

PURPOSE

The Language ENvironment Analysis (LENA) technology uses automated speech processing (ASP) algorithms to estimate counts such as total adult words and child vocalizations, which helps understand children's early language environment. This ASP has been validated in North American English and other languages in predominantly monolingual contexts but not in a multilingual context like India. Thus, the current study aims to validate the classification accuracy of the LENA algorithm specifically focusing on speaker recognition of adult segments (AdS) and child segments (ChS) in a sample of bi/multilingual families from India.

METHOD

Thirty neurotypical children between 6 and 24 months (M = 12.89, SD = 4.95) were recruited. Participants were growing up in bi/multilingual environment hearing a combination of Kannada, Tamil, Malayalam, Telugu, Hindi, and/or English. Daylong audio recordings were collected using LENA and processed using the ASP to automatically detect segments across speaker categories. Two human annotators manually annotated ~900 min (37,431 segments across speaker categories). Performance accuracy (recall and precision) was calculated for AdS and ChS.

RESULTS

The recall and precision for AdS were 0.62 (95% confidence interval [CI] [0.61, 0.63]) and 0.83 (95% CI [0.8, 0.83]), respectively. This indicated that 62% of the segments identified as AdS by the human annotator were also identified as AdS by the LENA ASP algorithm and 83% of the segments labeled by the LENA ASP as AdS were also labeled by the human annotator as AdS. Similarly, the recall and precision for ChS were 0.65 (95% CI [0.64, 0.66]) and 0.55 (95% CI [0.54, 0.56]), respectively.

CONCLUSIONS

This study documents the performance of the ASP in correctly classifying speakers as adult or child in a sample of families from India, indicating recall and precision that is relatively low. This study lays the groundwork for future investigations aiming to refine the algorithm models, potentially facilitating more accurate performance in bi/multilingual societies like India.

SUPPLEMENTAL MATERIAL

https://doi.org/10.23641/asha.27910710.

Establishing the reliability of metrics extracted from long-form recordings using LENA and the ACLEW pipeline

Long-form audio recordings are increasingly used to study individual variation, group differences, and many other topics in theoretical and applied fields of developmental science, particularly for the description of children's language input (typically speech from adults) and children's language output (ranging from babble to sentences). The proprietary LENA software has been available for over a decade, and with it, users have come to rely on derived metrics like adult word count (AWC) and child vocalization counts (CVC), which have also more recently been derived using an open-source alternative, the ACLEW pipeline. Yet, there is relatively little work assessing the reliability of long-form metrics in terms of the stability of individual differences across time. Filling this gap, we analyzed eight spoken-language datasets: four from North American English-learning infants, and one each from British English-, French-, American English-/Spanish-, and Quechua-/Spanish-learning infants. The audio data were analyzed using two types of processing software: LENA and the ACLEW open-source pipeline. When all corpora were included, we found relatively low to moderate reliability (across multiple recordings, intraclass correlation coefficient attributed to the child identity [Child ICC], was < 50% for most metrics). There were few differences between the two pipelines. Exploratory analyses suggested some differences as a function of child age and corpora. These findings suggest that, while reliability is likely sufficient for various group-level analyses, caution is needed when using either LENA or ACLEW tools to study individual variation. We also encourage improvement of extant tools, specifically targeting accurate measurement of individual variation.

The Language ENvironment Analysis system (LENA): A validation study with Italian-learning children

This study is a validation of the LENA system for the Italian language. In Study 1, to test LENA's accuracy, seventy-two 10-minute samples extracted from daylong LENA recordings were manually transcribed for 12 children longitudinally observed at 1;0 and 2;0. We found strong correlations between LENA and human estimates in the number of Adult Word Count (AWC) and Child Vocalisations Count (CVC) and a weak correlation between LENA and human estimates in Conversational Turns Count (CTC). In Study 2, to test the concurrent validity, direct and indirect language measures were considered on a sample of 54 recordings (19 children). Correlational analyses showed that LENA's CVC and CTC were significantly related to the children's vocal production, a parent report measure of prelexical vocalizations and the vocal reactivity scores. These results confirm that the automatic analyses performed by the LENA device are reliable and powerful for studying language development in Italian-speaking infants.

The Myth of Women's Advantage in Using Child-Directed Speech: Evidence of Women Versus Men in Single-Sex-Parent Families

PURPOSE

Differences between child-directed speech (CDS) by women and men are generally explained by either biological-evolutionary or gender-social theories. It is difficult to tease these two explanations apart for different-sex-parent families because women are usually also the main caregivers. Thus, this study aims to examine the influence of parental sex on CDS by investigating men and women who are in same-sex-parent families.

METHOD

Twenty same-sex-parent families participated in the study-10 families in which the parents were two men and 10 families in which the parents were two women. The families were matched for toddler age (range: 9-24 months) and sex. CDS was recorded using the Language Environment Analysis (LENA) device for 16 hr during a day. Each parent was also audio-recorded during a 30-min play session with his or her child.

RESULTS

No difference was found between men and women across all the LENA measures, namely, adult word count, conversational turns count, and child vocalization count. The analysis of speech samples during parent-child play showed no difference between men and women in mean length of utterance and number of nouns, verbs, and adjectives. Pragmatic speech acts of initiations, responses to infants' actions, or vocalizations were similar in both sexes. Women used more "teaching" utterances than men, and men who were main caregivers used more "teaching" utterances than men who were secondary caregivers. Across both sexes, secondary caregivers used more "requests for actions" compared to main caregivers.

CONCLUSIONS

The present findings support a functional-social approach and not a biological approach for explaining the use of CDS by men and women. These findings have clinical implications on the involvement of men in early intervention programs.