One of the many tasks facing the typically-developing child language learner is learning to discriminate between distinctive sounds that make up words in their native language. We investigate whether multimodal information---specifically adult speech coupled with video frames of speakers' faces---benefits a computational model of phonetic learning. We introduce a method for creating high-quality synthetic videos of speakers' faces for an existing audio corpus. Our learning model, when trained and tested on audiovisual inputs, achieves 8.1% relative improvement on a phoneme discrimination battery compared to a model trained and tested on audio-only input. It outperforms the audio model by 3.9% when tested on audio-only data, suggesting that visual information facilitates the acquisition of acoustic distinctions. In noisy audio environments, our audiovisual model recovers 67% of the loss in performance of the audio model relative to non-noisy environments. These results demonstrate that visual information benefits an ideal learner and illustrate multiple ways that children might leverage visual cues when learning to discriminate speech sounds.

Current computational models of word learning make use of correspondences between words and observed referents, but as of yet cannot—as human learners do—leverage information regarding the meaning of other words in the lexicon. Here we develop a Bayesian framework for word learning that learns a lexicon from multiword utterances. In a set of three simulations we demonstrate this framework’s functionality, consistency with experimental work, and superior performance in certain learning tasks with respect to a Bayesian word leaning model that treats word learning as inferring the meaning of each word independently. This framework represents the first step in modeling the potential synergies between referential and distributional cues in word learning

How do adults understand children's speech? Children's productions over the course of language development often bear little resemblance to typical adult pronunciations, yet caregivers nonetheless reliably recover meaning from them. Here, we employ a suite of Bayesian models of spoken word recognition to understand how adults overcome the noisiness of child language, showing that communicative success between children and adults relies heavily on adult inferential processes. By evaluating competing models on phonetically-annotated child language from the Providence corpus, we show that adults' recovered meanings are best predicted by prior expectations fitted specifically to the child language environment, rather than to typical adult-adult language. After quantifying the contribution of this "child-directed listening" over developmental time, we discuss the consequences for theories of language acquisition, as well as the implications for commonly-used methods for assessing children's linguistic proficiency.

We investigate the development of children’s early grammat-ical knowledge using the test case of the English regular plu-ral. Previous research points to early generalization, with chil-dren applying an abstract morphological rule to produce novelplurals well before 24 months. At the same time, childrenuse the plural inconsistently with familiar object words, anddemonstrate limited receptive knowledge of the plural in theabsence of supporting linguistic features. In the first studyto test knowledge of the plural within participants using aparadigm matched across comprehension and production, weconduct two experiments with n = 52 24-36-month-olds: aneyetracking task to evaluate what they understand, and a sto-rybook task to test how they use the plural. We manipulateboth novelty (novel vs. familiar object words) and phonolog-ical form (/s/ vs. /z/ plurals). We find strong, age-related ev-idence of productive knowledge of the plural in an expressivetask, but do not find evidence of receptive knowledge in thesesame children.

Vector-space models of semantics represent words ascontinuously-valued vectors and measure similarity based onthe distance or angle between those vectors. Such representa-tions have become increasingly popular due to the recent de-velopment of methods that allow them to be efficiently esti-mated from very large amounts of data. However, the ideaof relating similarity to distance in a spatial representationhas been criticized by cognitive scientists, as human similar-ity judgments have many properties that are inconsistent withthe geometric constraints that a distance metric must obey. Weshow that two popular vector-space models, Word2Vec andGloVe, are unable to capture certain critical aspects of humanword association data as a consequence of these constraints.However, a probabilistic topic model estimated from a rela-tively small curated corpus qualitatively reproduces the asym-metric patterns seen in the human data. We also demonstratethat a simple co-occurrence frequency performs similarly toreduced-dimensionality vector-space models on medium-sizecorpora, at least for relatively frequent words.

Because words have multiple meanings, language users must often choose appropriate meanings according to the context of use. How this potential ambiguity affects first language learning, especially word learning, is unknown. Here, we present the first large-scale study of how children are exposed to, and themselves use, ambiguous words in their language learning environments. We tag 180,000 words in two longitudinal child language corpora with word senses from WordNet, focusing between 9 and 51 months and limiting to words from a popular parental vocabulary report. We then compare the diversity of sense usage in adult speech around children to that observed in a sample of adult language, as well as the diversity of sense usage in children's own productions. To accomplish this we use a Bayesian model-based estimate of sense entropy, a measure of diversity that takes into account uncertainty inherent in small sample sizes. This reveals that sense diversity in caregivers' speech to children is similar to that observed in a sample of adult-directed written material, and that children's use of nouns---but not verbs---is similarly diverse to that of adults. Finally, we show that sense entropy is a significant predictor of vocabulary development: children begin to produce words with a higher diversity of adult sense usage at later ages. We discuss the implications of our findings for theories of word learning.

The ability to rapidly recognize words and link them to referents in context is central to children's early language development. This ability, often called word recognition in the developmental literature, is typically studied in the looking-while-listening paradigm, which measures infants' fixation on a target object (vs. a distractor) after hearing a target label. We present a large-scale, open database of infant and toddler eye-tracking data from looking-while-listening tasks. The goal of this effort is to address theoretical and methodological challenges in measuring vocabulary development. We present two analyses of the current database (N=1,320): (1) capturing age-related changes in infants' word recognition while generalizing across item-level variability and (2) assessing how a central methodological decision -- selecting the time window of analysis -- impacts the reliability of measurement. Future efforts will expand the scope of the current database to advance our understanding of participant-level and item-level variation in children's vocabulary development.