About

Gareth Roberts is an associate professor in Linguistics at the University of Pennsylvania, where he currently serves as Graduate Chair. He is also a member of the Psychology Graduate Group and the founder and director of the Cultural Evolution of Language Lab. Additionally, he is the founder and co-director of the Social and Cultural Evolution Working Group at Penn. His research primarily involves conducting experiments in which participants communicate using artificial languages or construct new languages from scratch. These experiments aim to shed light on the role of social and communicative pressures in shaping the emergence and evolution of language and other communication systems. Broadly, his interests encompass issues related to language evolution and change, as well as cultural evolution in general.

Research topics

• Computer Science
• Artificial Intelligence
• Sociology
• Linguistics
• Psychology
• Anthropology
• Epistemology
• Mathematics education
• Cognitive psychology
• Philosophy

Selected publications

• Iconicity Gives Communicators a Head-Start, Even if Only the Producer Experiences It

  Open Mind · 2026-01-01

  articleOpen accessSenior author

  Iconicity has increasingly come to be recognized as widespread in language. It plays a particularly important role in bootstrapping new referring expressions in existing languages and in the emergence of new languages and other communication systems. The basis of this role has long been assumed to depend primarily on transparency of the iconic signal for the receiver, who benefits from being better able to identify its meaning. But might there also be producer-side advantages, distinct from this transparency-based comprehension benefit, that support communication? We investigated this using an experimental referential communication game in which dyads used a novel signaling medium to communicate fruits and vegetables. We manipulated both whether the producer could generate iconic signals and whether the receiver saw iconic or arbitrary signals. Results suggested that there was an iconicity benefit via stability in production even if the receiver was unable to perceive the iconicity (and therefore unable to benefit from transparency). However, while this gave dyads a substantial head-start, the lack of a benefit from iconicity for the receiver meant dyads in this condition still performed significantly less well overall than dyads with full iconicity.

  PublisherOA PDFDOI

• The benefits of one-sided iconicity

  Underline Science Inc. · 2025-06-18

  otherOpen access

  Iconicity has recently been shown to be widespread in language and to play a particularly important role in bootstrapping new referring expressions or even getting whole new languages off the ground. The basis of this role has long been assumed to depend primarily on transparency for the receiver of the iconic signal, but might there also be producer-side advantages that play a significant role? We investigated this using an experimental referential communication game in which dyads communicated fruit and vegetables. We manipulated whether the sender could generate iconic signals and whether the receiver saw them. Results suggested that iconicity gave dyads a head-start, via stability in production, even if the receiver did not perceive the iconicity. However, this benefit declined over time, most likely due to memory constraints on the receiver.

  PublisherDOI

• Central Limit Theorem for ergodic averages of Markov chains \& the comparison of sampling algorithms for heavy-tailed distributions

  ArXiv.org · 2025-12-20

  articleOpen accessSenior author

  Establishing central limit theorems (CLTs) for ergodic averages of Markov chains is a fundamental problem in probability and its applications. Since the seminal work~\cite{MR834478}, a vast literature has emerged on the sufficient conditions for such CLTs. To counterbalance this, the present paper provides verifiable necessary conditions for CLTs of ergodic averages of Markov chains on general state spaces. Our theory is based on drift conditions, which also yield lower bounds on the rates of convergence to stationarity in various metrics. The validity of the ergodic CLT is of particular importance for sampling algorithms, where it underpins the error analysis of estimators in Bayesian statistics and machine learning. Although heavy-tailed sampling is of central importance in applications, the characterisation of the CLT and the convergence rates are theoretically poorly understood for almost all practically-used Markov chain Monte Carlo (MCMC) algorithms. In this setting our results provide sharp conditions on the validity of the ergodic CLT and establish convergence rates for large families of MCMC sampling algorithms for heavy-tailed targets. Our study includes a rather complete analyses for random walk Metropolis samplers (with finite- and infinite-variance proposals), Metropolis-adjusted and unadjusted Langevin algorithms and the stereographic projection sampler (as well as the independence sampler). By providing these sharp results via our practical drift conditions, our theory offers significant insights into the problems of algorithm selection and comparison for sampling heavy-tailed distributions (see short YouTube presentations~\cite{YouTube_talk} describing our \href{https://youtu.be/m2y7U4cEqy4}{\underline{theory}} and \href{https://youtu.be/w8I_oOweuko}{\underline{applications}}).

  PublisherOA PDF

• Noise, time, and signal reduction in a laboratory communication game

  Journal of Language Evolution · 2025-12-10

  articleSenior author

  Abstract We present an experimental communication game study designed to investigate the effects of noise, context, and time pressures on communication systems. This study is inspired by and follows on from an earlier study, designed to model the emergence of grammatical focus using a simple nonlinguistic communication game paradigm in which participants communicated line figures by selecting cells in a grid to send to each other. In our study, the communication task was more challenging, and a confound between noise and effort pressures was removed. We found that signals became reduced over time, though to a lesser extent given noise, which made communication significantly harder. Time pressures reduced the effect of noise. In general, signals became more stable over time, but noise interfered with the stabilization process. Overall, this study supports the findings of earlier work that employed a much simpler communication task and contributes to our understanding of how information structure gets into language.

  PublisherDOI

• Exact Bayesian inference for Markov switching diffusions

  ArXiv.org · 2025-02-13

  preprintOpen accessSenior author

  We develop the first exact Bayesian methodology for the problem of inference in discretely observed regime switching diffusions. Switching diffusion models extend ordinary diffusions by allowing for jumps in instantaneous drift and volatility. The jumps are driven by a latent, continuous time Markov switching process. We address the problem through an MCMC and an MCEM algorithm that target the exact posterior of diffusion parameters and the latent regime process. The algorithms are exact in the sense that they target the correct posterior distribution of the continuous model, so that the errors are due to Monte Carlo only. We illustrate the method on numerical examples, including an empirical analysis of the method's scalability in the length of the time series, and find that it is comparable in computational cost with discrete approximations while avoiding their shortcomings.

  PublisherOA PDFDOI

• Central Limit Theorem for ergodic averages of Markov chains \& the comparison of sampling algorithms for heavy-tailed distributions

  arXiv (Cornell University) · 2025-12-20

  preprintOpen accessSenior author

  Establishing central limit theorems (CLTs) for ergodic averages of Markov chains is a fundamental problem in probability and its applications. Since the seminal work~\cite{MR834478}, a vast literature has emerged on the sufficient conditions for such CLTs. To counterbalance this, the present paper provides verifiable necessary conditions for CLTs of ergodic averages of Markov chains on general state spaces. Our theory is based on drift conditions, which also yield lower bounds on the rates of convergence to stationarity in various metrics. The validity of the ergodic CLT is of particular importance for sampling algorithms, where it underpins the error analysis of estimators in Bayesian statistics and machine learning. Although heavy-tailed sampling is of central importance in applications, the characterisation of the CLT and the convergence rates are theoretically poorly understood for almost all practically-used Markov chain Monte Carlo (MCMC) algorithms. In this setting our results provide sharp conditions on the validity of the ergodic CLT and establish convergence rates for large families of MCMC sampling algorithms for heavy-tailed targets. Our study includes a rather complete analyses for random walk Metropolis samplers (with finite- and infinite-variance proposals), Metropolis-adjusted and unadjusted Langevin algorithms and the stereographic projection sampler (as well as the independence sampler). By providing these sharp results via our practical drift conditions, our theory offers significant insights into the problems of algorithm selection and comparison for sampling heavy-tailed distributions (see short YouTube presentations~\cite{YouTube_talk} describing our \href{https://youtu.be/m2y7U4cEqy4}{\underline{theory}} and \href{https://youtu.be/w8I_oOweuko}{\underline{applications}}).

  PublisherDOI

• Simulation to optimize the laboratory diagnosis of bacteremia

  Microbiology Spectrum · 2024-09-24 · 1 citations

  articleOpen access

  Blood cultures are central to the management of patients with sepsis and bloodstream infection. Clinical decisions depend on the timely availability of laboratory information, which, in turn, depends on the optimal laboratory processing of specimens. Discrete event simulation (DES) offers insights into where optimization efforts can be targeted. Here, we generate a detailed process map of blood culture processing within a laboratory and use it to build a simulator. Direct observation of laboratory staff processing blood cultures was used to generate a flowchart of the blood culture laboratory pathway. Retrospective routinely collected data were combined with direct observations to generate probability distributions over the time taken for each event. These data were used to inform the DES model. A sensitivity analysis explored the impact of staff availability on turnaround times. A flowchart of the blood culture pathway was constructed, spanning labeling, incubation, organism identification, and antimicrobial susceptibility testing. Thirteen processes in earlier stages of the pathway, not otherwise captured by routinely collected data, were timed using direct observations. Observations revealed that specimen processing is predominantly batched. Another eight processes were timed using retrospective data. A simulator was built using DES. Sensitivity analysis revealed that specimen progression through the simulation was especially sensitive to laboratory technician availability. Gram stain reporting time was also sensitive to laboratory scientist availability. Our laboratory simulation model has wide-ranging applications for the optimization of laboratory processes and effective implementation of the changes required for faster and more accurate results. IMPORTANCE: Optimization of laboratory pathways and resource availability has a direct impact on the clinical management of patients with bloodstream infection. This research offers an insight into the laboratory processing of blood cultures at a system level and allows clinical microbiology laboratories to explore the impact of changes to processes and resources.

  PublisherDOI

• Communicative Pressures and the Emergence of Structure in Phonological Inventories

  SSRN Electronic Journal · 2024-01-01

  preprintOpen access1st authorCorresponding
  PublisherDOI

• Co‐Occurrence, Extension, and Social Salience: The Emergence of Indexicality in an Artificial Language

  Cognitive Science · 2023-05-01 · 6 citations

  articleOpen accessSenior authorCorresponding

  We investigated the emergence of sociolinguistic indexicality using an artificial-language-learning paradigm. Sociolinguistic indexicality involves the association of linguistic variants with nonlinguistic social or contextual features. Any linguistic variant can acquire "constellations" of such indexical meanings, though they also exhibit an ordering, with first-order indices associated with particular speaker groups and higher-order indices targeting stereotypical attributes of those speakers. Much natural-language research has been conducted on this phenomenon, but little experimental work has focused on how indexicality emerges. Here, we present three miniature artificial-language experiments designed to break ground on this question. Results show ready formation of first-order indexicality based on co-occurrence alone, with higher-order indexicality emerging as a result of extension to new speaker groups, modulated by the perceived practical importance of the indexed social feature.

  PublisherOA PDFDOI

• The emergence of phonological dispersion through interaction: an exploratory secondary analysis of a communicative game

  Frontiers in Psychology · 2023-05-24 · 2 citations

  articleOpen access1st authorCorresponding

  Introduction: Why is it that phonologies exhibit greater dispersion than we might expect by chance? In earlier work we investigated this using a non-linguistic communication game in which pairs of participants sent each other series of colors to communicate a set of animal silhouettes. They found that above-chance levels of dispersion, similar to that seen in vowel systems, emerged as a result of the production and perception demands acting on the participants. However, they did not investigate the process by which this dispersion came about. Method: To investigate this we conducted a secondary statistical analysis of the data, looking in particular at how participants approached the communication task, how dispersion emerged, and what convergence looked like. Results: We found that dispersion was not planned from the start but emerged as a large-scale consequence of smaller-scale choices and adjustments. In particular, participants learned to reproduce colors more reliably over time, paid attention to signaling success, and shifted towards more extreme areas of the space over time. Conclusion: This study sheds light on the role of interactive processes in mediating between human minds and the emergence or larger-scale structure, as well as the distribution of features across the world's languages.

  PublisherOA PDFDOI

Recent grants

• Key factors in the emergence of combinatorial structure: An experimental and computational approach

  NSF · $103k · 2020–2024

Frequent coauthors

• Bruno Galantucci

  23 shared

• Benjamin Langstein

  Yeshiva University

  7 shared

• Betsy Sneller

  7 shared

• Joshua B. Plotkin

  University of Pennsylvania

  5 shared

• Robin Clark

  5 shared

• Masha Fedzechkina

  Apple (United Kingdom)

  5 shared

• Steve Millington

  Manchester Metropolitan University

  4 shared

• Wei Lai

  H.B. Fuller (United States)

  3 shared

Labs

• Department of Linguistics, University of PennsylvaniaPI

Education

• Ph.D., Language evolution, language change, language variation, cultural evolution, experimental semiotics, social interaction

  Univ. of Edinburgh

  2010