About

Professor Michael C. Frank did his undergraduate degree at Stanford in Symbolic Systems and his PhD work at MIT. He is broadly interested in the relationship between language and cognition, especially as it relates to children's early language development.

Research topics

• Computer Science
• Psychology
• Artificial Intelligence
• Developmental psychology
• Cognitive psychology
• Mathematics
• Social psychology
• Linguistics
• Sociology
• Statistics
• Machine Learning
• Cognitive science
• Data science
• Neuroscience
• Communication
• World Wide Web
• Natural Language Processing
• Pathology
• Applied psychology
• Epistemology
• Medicine
• Philosophy
• Geography

Selected publications

• ManyBabies 5: A large-scale investigation of the proposed shift from familiarity preference to novelty preference in infant looking time

  2023 · 28 citations

  • Psychology
  • Cognitive psychology
  • Developmental psychology

  Much of our basic understanding of cognitive and social processes in infancy relies on measures of looking time, and specifically on infants’ visual preference for a novel or familiar stimulus. However, despite being the foundation of many behavioral tasks in infant research, the determinants of infants’ visual preferences are poorly understood, and differences in the expression of preferences can be difficult to interpret. In this large-scale study, we test predictions from the Hunter and Ames model of infants' visual preferences. We investigate the effects of three factors predicted by this model to determine infants’ preference for novel versus familiar stimuli: age, stimulus familiarity, and stimulus complexity. Drawing from a large and diverse sample of infant participants (minimum expected sample size N = 1,280), this study aims to provide empirical evidence for a robust and generalizable model of infant visual preferences, leading to a more solid theoretical foundation for understanding the mechanisms that underlie infants’ responses in common behavioral paradigms. Moreover, we hope that our findings will guide future studies that rely on infants' visual preferences to measure cognitive and social processes.

  PublisherOA PDFDOI

• From partners to populations: A hierarchical Bayesian account of coordination and convention.

  Psychological Review · 2022 · 65 citations

  • Computer Science
  • Artificial Intelligence
  • Computer Science

  over multiple timescales. Partner-specific common ground quickly emerges from social inferences within dyadic interactions, while community-wide social conventions are stable priors that have been abstracted away from interactions with multiple partners. We present new empirical data alongside simulations showing how our model provides a computational foundation for several phenomena that have posed a challenge for previous accounts: (a) the convergence to more efficient referring expressions across repeated interaction with the same partner, (b) the gradual transfer of partner-specific common ground to strangers, and (c) the influence of communicative context on which conventions eventually form. (PsycInfo Database Record (c) 2023 APA, all rights reserved).

  PublisherDOI

• A Multilab Study of Bilingual Infants: Exploring the Preference for Infant-Directed Speech

  Advances in Methods and Practices in Psychological Science · 2021 · 77 citations

  • Sociology
  • Psychology
  • Developmental psychology

  From the earliest months of life, infants prefer listening to and learn better from infant-directed speech (IDS) than adult-directed speech (ADS). Yet, IDS differs within communities, across languages, and across cultures, both in form and in prevalence. This large-scale, multi-site study used the diversity of bilingual infant experiences to explore the impact of different types of linguistic experience on infants' IDS preference. As part of the multi-lab ManyBabies 1 project, we compared lab-matched samples of 333 bilingual and 385 monolingual infants' preference for North-American English IDS (cf. ManyBabies Consortium, 2020: ManyBabies 1), tested in 17 labs in 7 countries. Those infants were tested in two age groups: 6-9 months (the younger sample) and 12-15 months (the older sample). We found that bilingual and monolingual infants both preferred IDS to ADS, and did not differ in terms of the overall magnitude of this preference. However, amongst bilingual infants who were acquiring North-American English (NAE) as a native language, greater exposure to NAE was associated with a stronger IDS preference, extending the previous finding from ManyBabies 1 that monolinguals learning NAE as a native language showed a stronger preference than infants unexposed to NAE. Together, our findings indicate that IDS preference likely makes a similar contribution to monolingual and bilingual development, and that infants are exquisitely sensitive to the nature and frequency of different types of language input in their early environments.

  DOI

• Unsupervised neural network models of the ventral visual stream

  Proceedings of the National Academy of Sciences · 2021 · 328 citations

  • Artificial Intelligence
  • Artificial Intelligence
  • Computer Science

  Deep neural networks currently provide the best quantitative models of the response patterns of neurons throughout the primate ventral visual stream. However, such networks have remained implausible as a model of the development of the ventral stream, in part because they are trained with supervised methods requiring many more labels than are accessible to infants during development. Here, we report that recent rapid progress in unsupervised learning has largely closed this gap. We find that neural network models learned with deep unsupervised contrastive embedding methods achieve neural prediction accuracy in multiple ventral visual cortical areas that equals or exceeds that of models derived using today's best supervised methods and that the mapping of these neural network models' hidden layers is neuroanatomically consistent across the ventral stream. Strikingly, we find that these methods produce brain-like representations even when trained solely with real human child developmental data collected from head-mounted cameras, despite the fact that these datasets are noisy and limited. We also find that semisupervised deep contrastive embeddings can leverage small numbers of labeled examples to produce representations with substantially improved error-pattern consistency to human behavior. Taken together, these results illustrate a use of unsupervised learning to provide a quantitative model of a multiarea cortical brain system and present a strong candidate for a biologically plausible computational theory of primate sensory learning.

  DOI

• SAYCam: A Large, Longitudinal Audiovisual Dataset Recorded From the Infant’s Perspective

  Open Mind · 2021 · 123 citations

  Senior authorCorresponding
  • Computer Science
  • Artificial Intelligence
  • Psychology

  We introduce a new resource: the SAYCam corpus. Infants aged 6-32 months wore a head-mounted camera for approximately 2 hr per week, over the course of approximately two-and-a-half years. The result is a large, naturalistic, longitudinal dataset of infant- and child-perspective videos. Over 200,000 words of naturalistic speech have already been transcribed. Similarly, the dataset is searchable using a number of criteria (e.g., age of participant, location, setting, objects present). The resulting dataset will be of broad use to psychologists, linguists, and computer scientists.

  PublisherOA PDFDOI

• Adults’ and Children’s Comprehension of Linguistic Disjunction

  Collabra Psychology · 2021 · 8 citations

  Senior authorCorresponding
  • Computer Science
  • Psychology
  • Linguistics

  Disjunction has played a major role in advancing theories of logic, language, and cognition, featuring as the centerpiece of debates on the origins and development of logical thought. Recent studies have argued that due to non-adult-like pragmatic reasoning, preschool children’s comprehension of linguistic disjunction differs from adults in two ways. First, children are more likely to interpret “or” as “and” (conjunctive interpretations); Second, children are more likely to consider a disjunction as inclusive (lack of exclusivity implicatures). We tested adults and children’s comprehension of disjunction in existential sentences using two and three-alternative forced choice tasks, and analyzed children’s spontaneous verbal reactions prior to their forced-choice judgments. Overall our results are compatible with studies that suggest children understand the basic truth-conditional semantics of disjunction. Children did not interpret “or” as “and”, supporting studies that argue conjunctive interpretations are due to task demands. In addition, even though our forced-choice tasks suggest children interpreted disjunction as inclusive, spontaneous verbal reactions showed that children were sensitive to the adult-like pragmatics of disjunction. Theoretically, these studies provide evidence against previous developmental accounts, and lend themselves to two alternative hypotheses. First, that preschool children’s pragmatic knowledge is more adult-like than previously assumed, but forced-choice judgments are not sensitive enough to capture this knowledge. Second, children may have the knowledge of the relevant lexical scale themselves, but be uncertain whether a new speaker also has this knowledge (mutual knowledge of the scale).

  PublisherOA PDFDOI

• Action anticipation based on an agent's epistemic state in toddlers and adults

  2021 · 31 citations

  • Computer Science
  • Psychology
  • Epistemology

  Do toddlers and adults engage in spontaneous Theory of Mind (ToM)? Evidence from anticipatory looking (AL) studies suggests that they do. But a growing body of failed replication studies raised questions about the paradigm’s suitability. In this multi-lab collaboration, we test the robustness of spontaneous ToM measures. We examine whether 18- to 27-month-olds’ and adults’ anticipatory looks distinguish between two basic forms of an agent’s epistemic states: knowledge and ignorance. In toddlers [ANTICIPATED n = 520 50% FEMALE] and adults [ANTICIPATED n = 408, 50% FEMALE] from diverse ethnic backgrounds, we found [SUPPORT/NO SUPPORT] for epistemic state-based action anticipation. Future research can probe whether this conclusion extends to more complex kinds of epistemic states, such as true and false beliefs.

  PublisherOA PDFDOI

• Variability and Consistency in Early Language Learning

  The MIT Press eBooks · 2021 · 190 citations

  1st authorCorresponding
  • Computer Science
  • Computer Science
  • Natural Language Processing

  A data-driven exploration of children's early language learning across different languages, providing an empirical reference and a new theoretical framework. This book examines variability and consistency in children's language learning across different languages and cultures, drawing on Wordbank, an open database with data from more than 75,000 children and twenty-nine languages or dialects. This big data approach makes the book the most comprehensive cross-linguistic analysis to date of early language learning. Moreover, its data-driven picture of which aspects of language learning are consistent across languages suggests constraints on the nature of children's language learning mechanisms. The book provides both a theoretical framework for scholars of language learning, language, and human cognition, and a resource for future research. Wordbank archives data from parents' reports about their children's language learning using instruments in the MacArthur-Bates Communicative Development Inventory (CDI); its goal is to make CDI data available for study and analysis. After an overview of practical and theoretical issues, each of the book's empirical chapters applies a particular analysis to the Wordbank dataset, considering such topics as vocabulary size, demographic variation, syntactic and semantic categories, and the relationship between vocabulary growth and grammar. The final three chapters draw on the preceding chapters to quantify variability and consistency, consider the bird's eye view of language acquisition afforded by the data, and reflect on methodology.

  DOI

• Moderated Online Data-Collection for Developmental Research: Methods and Replications

  Frontiers in Psychology · 2021 · 68 citations

  • Computer Science
  • Psychology
  • Data science

  Online data collection methods are expanding the ease and access of developmental research for researchers and participants alike. While its popularity among developmental scientists has soared during the COVID-19 pandemic, its potential goes beyond just a means for safe, socially distanced data collection. In particular, advances in video conferencing software has enabled researchers to engage in face-to-face interactions with participants from nearly any location at any time. Due to the novelty of these methods, however, many researchers still remain uncertain about the differences in available approaches as well as the validity of online methods more broadly. In this article, we aim to address both issues with a focus on moderated (synchronous) data collected using video-conferencing software (e.g., Zoom). First, we review existing approaches for designing and executing moderated online studies with young children. We also present concrete examples of studies that implemented choice and verbal measures (Studies 1 and 2) and looking time (Studies 3 and 4) across both in-person and online moderated data collection methods. Direct comparison of the two methods within each study as well as a meta-analysis of all studies suggest that the results from the two methods are comparable, providing empirical support for the validity of moderated online data collection. Finally, we discuss current limitations of online data collection and possible solutions, as well as its potential to increase the accessibility, diversity, and replicability of developmental science.

  DOI

• Many Labs 5: Testing Pre-Data-Collection Peer Review as an Intervention to Increase Replicability

  Advances in Methods and Practices in Psychological Science · 2020 · 102 citations

  • Computer Science
  • Computer Science
  • Psychology

  Replication studies in psychological science sometimes fail to reproduce prior findings. If these studies use methods that are unfaithful to the original study or ineffective in eliciting the phenomenon of interest, then a failure to replicate may be a failure of the protocol rather than a challenge to the original finding. Formal pre-data-collection peer review by experts may address shortcomings and increase replicability rates. We selected 10 replication studies from the Reproducibility Project: Psychology (RP:P; Open Science Collaboration, 2015) for which the original authors had expressed concerns about the replication designs before data collection; only one of these studies had yielded a statistically significant effect ( p < .05). Commenters suggested that lack of adherence to expert review and low-powered tests were the reasons that most of these RP:P studies failed to replicate the original effects. We revised the replication protocols and received formal peer review prior to conducting new replication studies. We administered the RP:P and revised protocols in multiple laboratories (median number of laboratories per original study = 6.5, range = 3–9; median total sample = 1,279.5, range = 276–3,512) for high-powered tests of each original finding with both protocols. Overall, following the preregistered analysis plan, we found that the revised protocols produced effect sizes similar to those of the RP:P protocols (Δ r = .002 or .014, depending on analytic approach). The median effect size for the revised protocols ( r = .05) was similar to that of the RP:P protocols ( r = .04) and the original RP:P replications ( r = .11), and smaller than that of the original studies ( r = .37). Analysis of the cumulative evidence across the original studies and the corresponding three replication attempts provided very precise estimates of the 10 tested effects and indicated that their effect sizes (median r = .07, range = .00–.15) were 78% smaller, on average, than the original effect sizes (median r = .37, range = .19–.50).

  DOI

Recent grants

• Wordbank: An Open Repository for Developmental Vocabulary Data

  NSF · $502k · 2015–2018

• Collaborative Research: CompCog: Broad-coverage probabilistic models of communication in context

  NSF · $428k · 2015–2019

• Interactions between word learning and visual category development in infancy

  NSF · $138k · 2017–2019

Frequent coauthors

• Mélanie Söderström

  Stanford University

  73 shared

• Krista Byers‐Heinlein

  Concordia University

  63 shared

• J. Kiley Hamlin

  University of British Columbia

  58 shared

• Christina Bergmann

  53 shared

• Heidi A. Baumgartner

  University of Manitoba

  53 shared

• Bria Long

  48 shared

• Manuel Bohn

  Leuphana University of Lüneburg

  46 shared

• Kyle MacDonald

  41 shared

Labs

• LangCog LabPI

Education

• PhD, Brain and Cognitive Sciences

  Massachusetts Institute of Technology

  2010

Similar researchers at Stanford University

• Yi Cuih-287
• Zhenan Baoh-224
• Karl Deisserothh-218
• Michael Snyderh-215
• Roger Blandfordh-205