The learnability consequences of Zipfian distributions in language

Immature vocalizations elicit simplified adult speech across multiple languages

Learning to speak takes place during a prolonged period of immaturity, which confers advantages for communicative development. Social partners, required for survival in early development, afford feedback for immature vocalizations like babbling and early speech. Feedback, in the form of changes to the linguistic structure of adult speech in response to infant vocalizations, may guide the earliest stages of language acquisition. In a cross-linguistic study of 1,586 transcripts, spanning 13 languages from 5 language families, we investigated whether caregiver talk was consistently influenced by children's (aged 5-30 months) immature speech. Across languages, we found that most caregivers significantly simplified their linguistic structure in response to children's immature speech, resulting in reduced lexical diversity, shorter utterance lengths, and higher likelihoods of single-word utterances. Children's vocalizations elicited learnable language from caregivers, highlighting a potentially widespread feature of language use that is catalyzed by immature behavior. Thus, altriciality allows for immature speech to be a social tool, creating opportunities for learning during social interaction.

Speech Stream Composition Affects Statistical Learning: Behavioral and Neural Evidence

Statistical learning (SL), the ability to extract patterns from the environment, has been assumed to play a central role in whole cognition, particularly in language acquisition. Evidence has been gathered, however, from behavioral experiments relying on simplified artificial languages, raising doubts on the generalizability of these results to natural contexts. Here, we tested if SL is affected by the composition of the speech streams by expositing participants to auditory streams containing either four nonsense words presenting a transitional probability (TP) of 1 (unmixed high-TP condition), four nonsense words presenting TPs of 0.33 (unmixed low-TP condition) or two nonsense words presenting a TP of 1, and two of a TP of 0.33 (mixed condition); first under incidental (implicit), and, subsequently, under intentional (explicit) conditions to further ascertain how prior knowledge modulates the results. Electrophysiological and behavioral data were collected from the familiarization and test phases of each of the SL tasks. Behavior results revealed reliable signs of SL for all the streams, even though differences across stream conditions failed to reach significance. The neural results revealed, however, facilitative processing of the mixed over the unmixed low-TP and the unmixed high-TP conditions in the N400 and P200 components, suggesting that moderate levels of entropy boost SL.

Convergent evolution in whale and human vocal cultures

The complex songs of humpback whales conform to fundamental laws of language.

Whale song shows language-like statistical structure

Humpback whale song is a culturally transmitted behavior. Human language, which is also culturally transmitted, has statistically coherent parts whose frequency distribution follows a power law. These properties facilitate learning and may therefore arise because of their contribution to the faithful transmission of language over multiple cultural generations. If so, we would expect to find them in other culturally transmitted systems. In this study, we applied methods based on infant speech segmentation to 8 years of humpback recordings, uncovering in whale song the same statistical structure that is a hallmark of human language. This commonality, in two evolutionarily distant species, points to the role of learning and cultural transmission in the emergence of properties thought to be unique to human language.

Zipfian distributions facilitate children's learning of novel word-referent mappings

The word-frequency distributions children hear during language learning are highly skewed (Zipfian). Previous studies suggest that such skewed environments confer a learnability advantage in tasks that require the learner to discover the units that have to be learned, as in word-segmentation or cross-situational learning. This facilitative effect has been attributed to contextual facilitation from high frequency items in learning lower frequency items, and to better learning under the increased predictability (lower entropy) of skewed distributions. Here, we ask whether Zipfian distributions facilitate learning beyond the discovery of units, as expected under the predictability account. We tested children's learning of novel word-referent mappings in a learning task where each mapping was presented in isolation during training, and did not need to be dicovered. We compared learning in a uniform environment to two skewed environments with different entropy levels. Children's learning was overall better in the two skewed environments, even for low frequency items. These results extend the facilitative effect of Zipfian distributions to additional learning tasks and show they can facilitate language learning beyond the discovery of units.

When is a word in good company for learning?

Although identifying the referents of single words is often cited as a key challenge for getting word learning off the ground, it overlooks the fact that young learners consistently encounter words in the context of other words. How does this company help or hinder word learning? Prior investigations into early word learning from children's real-world language input have yielded conflicting results, with some influential findings suggesting an advantage for words that keep a diverse company of other words, and others suggesting the opposite. Here, we sought to triangulate the source of this conflict, comparing different measures of diversity and approaches to controlling for correlated effects of word frequency across multiple languages. The results were striking: while different diversity measures on their own yielded conflicting results, once nonlinear relationships with word frequency were controlled, we found convergent evidence that contextual consistency supports early word learning. RESEARCH HIGHLIGHTS: The words children learn occur in a sea of other words. The company words keep ranges from highly variable to highly consistent and circumscribed. Prior findings conflict over whether variability versus consistency helps early word learning. Accounting for correlated effects of word frequency resolved the conflict across multiple languages. Results reveal convergent evidence that consistency helps early word learning.

Cultural evolution creates the statistical structure of language

Human language is unique in its structure: language is made up of parts that can be recombined in a productive way. The parts are not given but have to be discovered by learners exposed to unsegmented wholes. Across languages, the frequency distribution of those parts follows a power law. Both statistical properties-having parts and having them follow a particular distribution-facilitate learning, yet their origin is still poorly understood. Where do the parts come from and why do they follow a particular frequency distribution? Here, we show how these two core properties emerge from the process of cultural evolution with whole-to-part learning. We use an experimental analog of cultural transmission in which participants copy sets of non-linguistic sequences produced by a previous participant: This design allows us to ask if parts will emerge purely under pressure for the system to be learnable, even without meanings to convey. We show that parts emerge from initially unsegmented sequences, that their distribution becomes closer to a power law over generations, and, importantly, that these properties make the sets of sequences more learnable. We argue that these two core statistical properties of language emerge culturally both as a cause and effect of greater learnability.

Zipfian Distributions in Child-Directed Speech

Across languages, word frequency and rank follow a power law relation, forming a distribution known as the Zipfian distribution. There is growing experimental evidence that this well-studied phenomenon may be beneficial for language learning. However, most investigations of word distributions in natural language have focused on adult-to-adult speech: Zipf's law has not been thoroughly evaluated in child-directed speech (CDS) across languages. If Zipfian distributions facilitate learning, they should also be found in CDS. At the same time, several unique properties of CDS may result in a less skewed distribution. Here, we examine the frequency distribution of words in CDS in three studies. We first show that CDS is Zipfian across 15 languages from seven language families. We then show that CDS is Zipfian from early on (six-months) and across development for five languages with sufficient longitudinal data. Finally, we show that the distribution holds across different parts of speech: Nouns, verbs, adjectives and prepositions follow a Zipfian distribution. Together, the results show that the input children hear is skewed in a particular way from early on, providing necessary (but not sufficient) support for the postulated learning advantage of such skew. They highlight the need to study skewed learning environments experimentally.