[Home] [People] [What We Do]

What do we study in CONE LAB?

Scene from the film Arrival, depicting scientist holding a sign that says 'HUMAN' Our research in the CONE LAB is motivated by a love of languages, of learning languages, and of learning about how people learn languages! We work to understand people as biological and cultural participants in communication. And those people have brains and experiences that shape the languages they use, and are sometimes shaped by the languages, as well. Our past research has included speakers of English, Chinese, Spanish, French, and Attié, as well as bilingual speakers of these languages, adults, children, and babies! We’re especially interested in the kinds of solutions that learners come up with when provided conflicting information about world through their different languages. We also work with artificial languages that allow us to carefully control the learning challenges we create for our participants, and computational models that allow us to carefully control the abilities and experiences of simulated learners. We also combine these behavioral and computational approaches with brain imaging technologies like EEG, fNIRS, and fMRI to study the biological aspects of language learning and use. Many of the datasets we collect and tools we develop are shared openly. Please see recent publications for links to those resources.

Research Areas

Linguistic Categories & Bilingualism. Languages arise to support communities of language users, and therefore languages reflect the experiences and needs of people who use them…

To the extent that communities’ ways of organizing information differ, we might expect their languages to do the same. Sapir expressed this view in extreme, writing, “The worlds in which different societies live are distinct worlds, not merely the same world with different labels attached” (1929, in Language). Since then, decades of language and psychology research has shown that not only languages differ, but people who use languages differ too: Even highly skilled bilinguals perform simple language tasks like object naming differently than monolinguals of the same language. These differences may go undetected in fluency (speed and ease of speech), but subtle variations can result in misunderstandings, like ordering a “white wine” and instead receiving a bottle of vodka (a typical translation error between English and Chinese), or demand extra effort and resources, a familiar experience to anyone who felt exhausted after attending a lecture in their second language. Our research aims to discover and understand the differences between how languges and language users represent information and learn the cognitive consequences of these language disagreements.

Statistical Learning. Statistical learning is a mechanism for finding patterns in the environment, such as the order in which certain events occur (like syllables in speech) or the links between sounds and images (like the names of objects)…

The statistical learning mechanism is available from infancy, and probably supports some of babies’ first discoveries about language. However, statistical learning is also active throughout childhood and adulthood. We are interested in how this ability to find patterns helps children learn how to read, how differences in this ability contribute to different outcomes in literacy, and whether statistical learning experiments do a good job of representing the different ways children learn to read in different educational contexts. One of those contexts is bilingulism, where children or adults know two different languages (and thus, two different patterns). We explore how statistical learning mechanisms can be applied when two artificial languages provide different, competing patterns. These studies investigate how a simple learning mechanism like SL might contribute to the very complicated inferences that learners (infants, children, and adults) make as they navigate between languages.

Machine Learning and Brain Imaging with fNIRS. Combining brain imaging technology with machine learning tools allows cognitive neuroscientists to begin asking new questions about what information is represented in the brain’s activity, and our work is extending these techniques to new populations…

Using multivariate statistical model to classify or “decode” information in the brain was first achieved with functional MRI, allowing researchers to guess which image participants were looking at or which word they were reading, but MRI is expensive, requires a large dedicated facility, and follows strict safety protocols. Although many more applications of machine learning to brain imaging have emerged in recent years, the cost, portability, and child-friendliness of this technology remain huge obstacles in using it. Further, these limitations affect what people have the opportunity to participate in brain imaging studies, and therefore who the field of cognitive neuroscience treats as interesting populations for study. Our work adapts the machine learning approaches developed in fMRI, EEG, and MEG communities for funcational near-infrared spectroscopy (fNIRS). As a low-cost and portable technology, fNIRS is a crucial tool for extending the reach of cognitive neuroscience to young children and to people in more difficult-to-access regions of the world.


Team MCPA - Multivariate pattern analysis of adult and infant fNIRS data

Allô Alphabet - Phone-based literacy intervention in rural Côte d’Ivoire

SPIN Scorcerer - Automated text processing tool for detailed scoring of speech-in-noise transcriptions


Benjamin Zinszer (bzinszer@swarthmore.edu)

Unless otherwise noted, software available on this page is provided under the GNU General Public License v2.0. The GNU GPL is the most widely used free software license and has a strong copyleft requirement. When distributing derived works, the source code of the work must be made available under the same license. Redistribution of this code must also include (1) a copy of the enclosed license and copyright notice, (2) state what changes, if any, have been made to the code, (3) provide attribution to the authors of the code, in this case Benjamin Zinszer and the CONE LAB, and (4) remain under the same GNU GPL v2.0 license.