A Sound Vocabulary
The size of a child's vocabulary as they enter kindergarten is a pivotal measure. The fewer words they know, the more likely they are to have poor performance in school, engage in antisocial behaviors, end up in the correctional system, and ultimately take lower-paying, less-secure jobs.
But predicting the size of a child’s vocabulary and finding ways to efficiently treat early childhood speech delays, deficits, and disorders is no easy feat. That is why Department of Speech-Language-Hearing Sciences Professor Benjamin Munson and his intercollegiate team of colleagues have dedicated more than 20 years to understanding the patterns of early childhood speech and the impact they may have on a child’s future speech and vocabulary.
Together with Dr. Jan Edwards of the University of Maryland (previously of UW-Madison) and Mary Beckman of Ohio State University, Munson has recently completed a five-year study that is yielding surprising results and overturning long-standing assumptions in their field.
New Methods, New Findings
The question that Munson, Edwards, and Beckman set out to explore was whether the sounds made in early childhood were in any way correlated to the eventual size of a child's vocabulary.
Up until that point, the conventional wisdom was that there was only a weak relationship between speech production ability (a child's ability to produce certain sounds) and vocabulary size, but the team was not convinced. As they began to collect data, they came upon a glaring problem: the tools used to measure and document the sounds children were making were imprecise. They found that using phonetic transcription, the standard method for documenting speech learning, missed a great deal of the subtleties and complexities of young children's speech production.
"You can have 100 kids making the same sound and it would be transcribed phonetically the same way," Munson explains. "So up till now, people have said that speech production isn't related to vocabulary size because we have these hundred kids who are making the sound correctly and yet they still have differences in vocabulary size. But there could be 100 different levels of mastery of those sounds if you use a finer-grained measure."
The team developed three powerful new models for more accurately recording, measuring, and assessing their data. First, they turned to computing power to analyze the spectral detail in speech, using it to predict the motions made by the mouth and tongue to produce certain sounds.
Second, they used crowd-sourcing to help analyze their data. Asking groups of people to rate the pronunciation accuracy of the same recording eliminated the problem of one researcher assigning definitive values to a subjective piece of data based on his or her own personal perceptions of pronunciation.
Lastly, they have begun to use ultrasound imaging to record precisely how certain children made certain sounds. It is a completely non-invasive way to accurately measure subtle differences in speech production in children as young as two and three.
Sound Learning Leads to Word Learning
The results are allowing Munson and the team to better understand the complex relationship between early sounds and vocabulary. And because the scope of the project was so large—it was conducted across two testing sites with well over 200 preschool children, among them children with hearing impairment who use cochlear implants, late-talking children, and children from a variety of socioeconomic backgrounds—it has given them greater depth of insight into the connection.
"Sound structure and vocabulary knowledge are related in really important ways," he explains. "At the early stage of development, speech perception and production and sound acquisition are the foundation upon which word-learning is built. And later in development, your vocabulary is the filter through which you learn new words."
In other words, the better a child correctly perceives a sound and is able to replicate it, the more likely it is that he or she will become what Munson calls a "good word learner." He explains that those children who have a good basis of understanding of word sounds can "scaffold" those skills and word learning takes off exponentially.
Because each child in the study was assessed three times, the team is able to look at which children were the fastest word-learners and why. Was it those with specific speech production or speech perception? Do any specific groups (for example, the hearing impaired children) follow any unique pattern? Are there particular sounds that have more or less impact on future vocabulary size?
Scaffolding for Success
It is those questions that will propel the team’s future research. With the data-collection portion of the study complete, Munson, Edwards, and Beckman will continue to mine the data and publish their results—even as they prepare grant proposals for their next project.
"It never ends," laughs Munson. "I'm interested in taking these new technologies and new methods to better automate our assessment of speech production accuracy, creating speech algorithms that automatically assess children’s speech production accuracy, and making it more accessible to more people."
He hopes that the data will help those in the field think about more accurate measures of articulation and begin to dispel the notion that speech production and vocabulary size should not be correlated
Ultimately, the goal is to give those who work in clinical settings more resources to help children overcome vocabulary deficiencies more quickly and efficiently.
"It's much harder to teach a child with a 10,000 word deficit 10,000 new words. Each one of those words takes a long time," Munson says, "But if you set a child up to be a good word-learner by teaching them how to build a good scaffold with the sound structure of language, how to build a good foundation of how to learn a new word, hopefully we’re setting children up for success in language learning later in life."