Summary: Researchers examine how infants discover which acoustic distinctions in speech are meaningful in their native language, showing that natural conversational speech provides the contextual signals needed for this learning.
Source: University of Maryland
Babies can tell many speech sounds apart shortly after birth, and by around their first birthday they tune into the sound contrasts that matter for their native language. Despite this well-established developmental pattern, researchers are still working to understand how infants determine which acoustic differences are contrastive—that is, which differences can change the meaning of words.
For example, in English the sounds represented by the letters b and d are contrastive: swapping the initial consonant in “ball” for a d produces a different word, “doll.”
A new study published in Proceedings of the National Academy of Sciences (PNAS) by two computational linguists with the University of Maryland sheds light on how infants might learn these contrasts. The paper argues that infants can exploit the contexts in which sounds appear to decide which acoustic dimensions are functionally contrastive in their language.
Historically, researchers expected clear, consistent acoustic differences between contrastive sounds—for instance, short versus long vowels in Japanese. While such differences are apparent in carefully produced speech, natural, spontaneous speech is much more variable and often blurs these distinctions.
“This is one of the first phonetic learning accounts shown to work on spontaneous speech data, suggesting that infants could indeed learn which acoustic dimensions are contrastive from everyday language,” says Kasia Hitczenko, lead author of the study.
Hitczenko completed her Ph.D. in linguistics at the University of Maryland in 2019 and is now a postdoctoral scholar at the Cognitive Sciences and Psycholinguistics Laboratory at the École Normale Supérieure in Paris.
The study proposes that infants make use of contextual clues—such as neighboring sounds or broader linguistic surroundings—to interpret acoustic variability. The researchers evaluated this idea using two case studies that defined “context” in different ways and compared data from Japanese, Dutch, and French.
The team analyzed speech samples divided by context and plotted vowel durations within each context. In Japanese, those distributions differed markedly across contexts: some contexts were dominated by shorter vowel durations while others contained more long vowels. In contrast, French showed similar vowel-duration distributions across contexts, reflecting the language’s lack of a short/long vowel contrast.
These results indicate that contrastive dimensions tend to vary more by context than noncontrastive ones, creating a signal infants could exploit. In other words, even when raw acoustic measures overlap across sounds, the contexts in which sounds occur can reveal which acoustic cues are linguistically relevant.
“This work offers a compelling explanation for how infants learn sound contrasts from natural speech,” says co-author Naomi Feldman, an associate professor of linguistics with an appointment in the University of Maryland Institute for Advanced Computer Studies (UMIACS). She notes that the contextual signal studied appears broadly consistent across languages and may generalize to other kinds of contrasts beyond vowel length.
The published paper builds on Hitczenko’s doctoral dissertation, which explored how context can be used for phonetic learning and perception from everyday speech.
About this linguistics research news
Author: Press Office
Source: University of Maryland
Contact: Press Office – University of Maryland
Image: The image is in the public domain
Original Research: Closed access.
“Naturalistic speech supports distributional learning across contexts” by Kasia Hitczenko et al., PNAS
Abstract
Naturalistic speech supports distributional learning across contexts
Infants are born able to discriminate most speech sounds, and by around one year of age they become specialized listeners tuned to the contrasts of their native language. This developmental shift implies that infants learn which acoustic dimensions are contrastive and begin to prioritize those dimensions when perceiving speech.
Yet speech is highly variable and many sounds overlap acoustically, leaving open the question of how infants distinguish contrastive from noncontrastive dimensions in realistic learning environments.
The authors demonstrate that infants could learn contrastive acoustic dimensions despite acoustic overlap because sounds that share similar acoustics often occur in different contexts. That cross-linguistic contextual variation produces a detectable signal: acoustic distributions vary more by context along contrastive dimensions than along noncontrastive ones.
By documenting this contextual pattern in natural speech, the study offers a plausible mechanism for how infants learn about sound contrasts in everyday language exposure, addressing a long-standing puzzle in early language acquisition.