Our Research Program

Language is one of the few uniquely human cognitive abilities and a foundation of human culture and civilization. What cognitive and neural mechanisms enable us to produce and understand language? Since Dax’s, Broca’s, and Wernicke's seminal discoveries in the 19th century, a broad array of brain regions have been implicated in linguistic processing spanning frontal, temporal, and parietal lobes of both hemispheres, as well as subcortical and cerebellar structures. However, characterizing the precise contributions of these different structures to linguistic processing has proven challenging. The goal of our research program is to understand the computations we perform and the representations we build during language processing, and to provide a detailed characterization of the brain regions underlying these computations and representations both in healthy individuals and individuals with brain disorders.

Our research centers around three key questions:

  1. What is the internal architecture of the language network?
  2. How does the language network interact with other large-scale networks in the human brain (e.g., the domain-general “multiple demand” network or the network supporting social cognition)?
  3. What is the nature of inter-individual differences in the neural instantiation of language processing, and how do these differences relate to differences in behavior and genetic make-up?

Central to our research was our development of new techniques – adopted from fMRI methods that have been successful in the field of vision research – to functionally “localize” brain regions sensitive to high-level linguistic processing (Fedorenko et al., 2010; read more about our approach here). These regions, each present in every individual subject, can be defined in just a few minutes of scanning and are highly replicable, both within and across scanning sessions. Although most of our current work relies on functional MRI, we actively pursue behavioral research aimed at understanding the mechanisms underlying language comprehension and production. In addition, across several collaborations we use an array of other methods which enable us to tackle a wider range of research questions. These include intracranial recordings from epilepsy patients (ECoG); behavioral and neuroimaging investigations of patients with neurodevelopmental and acquired brain disorders; genotyping methods; etc.

Key Discoveries

1. High-level language-processing brain regions are functionally specialized for language

Language has been argued to share cognitive and neural machinery with a number of cognitive processes, including arithmetic processing, general working memory, cognitive control, and musical processing. However, evidence from patients with acquired brain damage suggests that language can be selectively damaged or preserved. We addressed this centuries-old question by examining the responses of language-responsive regions - defined in each brain individually - to several non-linguistic tasks, each tapping a mental process that has been argued to rely on the same resources as language. We showed that high-level language regions, including regions in the left inferior frontal cortex (in Broca's area), show little or no response to any non-linguistic task, in spite of the fact that these tasks activate cortical regions in close proximity to the language regions.

Although we are still a long way away from understanding the precise computations performed by the language regions, their functionally specific responses to language already help rule out some hypotheses (e.g., that left frontal lobe structures support language only via domain-general processes like cognitive control or working memory, or that language regions represent or process any complex temporally-unfolding stimuli, like music). These findings support a clear distinction between language and other cognitive processes, resolving the prior conflict between the neuropsychological and neuroimaging literatures.

Relevant papers

2. The language network is ubiquitously sensitive to both word-level (lexical) and syntactic / compositional semantic processing, as revealed with fMRI

In the early days of language research (e.g., Chomsky, 1965), the lexicon and the grammar (syntactic rules) were conceived of as distinct components of the human cognitive architecture. And although over the years the distinction began to blur, with grammars becoming increasingly lexicalized (e.g., Joshi et al., 1975; Schabes et al., 1988; Bresnan, 1982; Pollard & Sag, 1994; Goldberg, 1995; Bybee, 1998; Jackendoff, 2002, 2007; Culicover & Jackendoff, 2005), most frameworks still draw a distinction between stored linguistic representations and computational machinery used to combine these stored units (complex as they may be) in novel ways. In the cognitive neuroscience of language, most existing proposals argue for a particular brain region within the language network that supports syntactic, or more generally combinatorial, processing. (The location of this alleged region differs across proposals.) However, no brain region has been convincingly shown to selectively respond to lexical meanings, or to selectively respond to combinatorial (syntactic or semantic) information. Consistent with this picture, we found that each region in the extended language network is sensitive to both lexical and combinatorial information. Furthermore, using multi-voxel pattern analyses (MVPAs) we asked whether perhaps some brain regions represent lexical information more robustly and others represent combinatorial information more robustly, in spite of the fact that every brain region in the language network is sensitive to both kinds of information. We found instead that lexical information is represented more robustly throughout the language network than combinatorial information, consistent with the fact that content words carry more information (e.g., Shannon, 1949) than function words and word order. This result thus suggests that lexical information may play a more critical role than syntax in the representation of linguistic meaning.

However, this fMRI data pattern is also consistent with distinct, and perhaps localized, implementation of different linguistic computations obscured by the facts that (i) the fMRI signal detects neural activity with a delay of several seconds, and (ii) different regions of the language network are anatomically and functionally inter-connected (e.g., Blank et al., 2014). In other words, neural responses that are localized in both space and time may appear distributed and ubiquitous across all language regions because of the information spreading throughout the functionally integrated language network within a few hundred milliseconds. In an ongoing collaboration with Ziv Williams, we are using a combination of intracranial stimulation and recordings to characterize the rapid temporal dynamics and to identify the time-causal basis of different linguistic operations, so stay tuned.

Relevant papers

3. Brain regions of the fronto-parietal “multiple-demand (MD)” network are domain-general

A number of brain regions in the frontal and parietal cortices, as well as some subcortical regions, are active during a broad range of cognitive tasks in both humans and non-human primates (e.g., Miller & Cohen, 2001; Duncan, 2001, 2010). This network of regions has become known as the “cognitive control”, “task-related” or “multiple-demand (MD)” network and has been implicated as the core of human fluid intelligence (e.g., Woolgar et al., 2010). However, the early evidence for the domain-generality of these regions in humans came from meta-analyses of activation peaks from across fMRI studies (e.g., Duncan & Owen, 2000), a method known to overestimate activation overlap. To test whether these brain regions are truly domain-general, we examined overlap among several cognitive tasks, including tasks tapping arithmetic processing, working memory and cognitive control, in individual subjects and found strong evidence of overlap across these diverse tasks. These results demonstrate that a number of regions in the human brain are truly domain-general and plausibly support flexible human behavior.

Relevant papers

4. A potential neural marker of the construction of complex meanings

How do circuits of neurons in our brains construct and hold the meaning of a sentence? To start to address this question, we measured neural activity from the surface of the human brain in patients being mapped out before neurosurgery, as they read or listened to sentences. In many electrodes, neural activity increased steadily over the course of the sentence, but the same was not found when participants processed lists of words or pronounceable nonwords, or grammatical nonword strings ("Jabberwocky"). This build-up of neural activity appears to reflect neither word meaning nor syntax alone, but the representation of complex meanings.

Relevant papers

Current Funding

  • NIH NICHD (Pathway to Independence award to Ev Fedorenko)
  • The Simons Foundation (through the Simons Center for the Social Brain at MIT)
  • The Office of the Director of National Intelligence (ODNI), Intelligence Advanced Research Projects Activity (IARPA), via Air Force Research Laboratory (AFRL)

Past Funding

  • The John Templeton Foundation