EvLab Research

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? 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 representations and computations that underlie our linguistic ability, and to provide a detailed characterization of the brain regions that support language processing in both typical and atypical brains.

Key research directions

Across numerous projects, we are tackling a wide range of research questions. Most of them fall into one of the following categories:

  1. Language comprehension: How do we infer others’ thoughts from their utterances?
  2. Language production: How do we convert our thoughts into utterances?
  3. Language and executive functions: What role do domain-general executive resources, like attention, working memory, and cognitive control play in language comprehension/production?
  4. Language and social cognition: Given that linguistic communication is a social behavior, what is the relationship between the language system and the mechanisms that support social perception and cognition?
  5. Meaning: What is the nature of our conceptual representations, and how do they map onto linguistic forms?
  6. Plasticity: How do our minds and brains reorganize following early or late damage to the language system?
  7. Individual differences: How do individuals vary in their neural architecture, and how does this variability relate to behavioral variability in linguistic and cognitive abilities, and to differences in our genetic make-up, including in individuals with neurodevelopmental disorders?
  8. Bi- and multilingualism: How do bi- and multilingual individuals process different languages? What aspects of our experience affect neural responses to native and non-native languages?

Methods

We use a number of approaches, including:

  1. Behavioral experiments and corpus analyses
  2. fMRI
  3. ERP
  4. MEG
  5. Intracranial recordings (ECoG, SEEG) and stimulation
  6. Natural language processing
  7. Neural network modeling

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 linguistic processing (Fedorenko et al., 2010; read more about our approach here). Although originally developed for fMRI, this approach can be naturally extended to other imaging modalities, like MEG or intracranial recordings.

We primarily work with neurotypical adults, but for some projects, we turn to other populations, like individuals with developmental or acquired brain disorders, individuals who have sustained early brain damage, children and older adults, and individuals with special aptitude for language or unusual linguistic experiences (e.g., hyperpolyglots).

Key discoveries

1. 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, music processing, general executive processes (like working memory and cognitive control), action understanding, and aspects of social perception and cognition. However, evidence from patients with acquired brain damage suggests that language can be selectively impaired or preserved. Across a series of studies, we examined the responses of language-responsive regions to diverse non-linguistic tasks, each tapping a mental process that has been argued to rely on the same resources as language. We showed that language regions, including the one residing 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 sometimes 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 help rule out some hypotheses (e.g., that left frontal lobe structures support language only via domain-general processes like working memory or cognitive control, or that language regions represent or process abstract, content-independent, syntactic information). 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 meanings and combinatorial (syntactic/semantic) processing

Linguistic theorizing, empirical evidence from language acquisition and processing, and computational modeling have jointly painted a picture whereby lexico-semantic and syntactic processing are deeply inter-connected and perhaps not separable. In contrast, many current proposals of the neural architecture of language continue to endorse a view whereby certain brain regions selectively support syntactic/combinatorial processing, although the locus of such “syntactic hub”, and its nature, vary across proposals. Using diverse linguistic manipulations, we have searched for a dissociation between lexico-semantic and syntactic/combinatorial processing. In both fMRI and ECoG, we found that both lexico-semantic and syntactic conditions elicit robust responses throughout the left fronto-temporal language network. Further, no regions are more strongly engaged by syntactic than lexico-semantic processing, although some regions show the opposite pattern. In ECoG, this result manifested as a steady increase in neural activity over the course of the sentence, but no such increase was found when participants processed grammatical nonword strings (“Jabberwocky" sentences). Thus, syntactic/combinatorial processing does not appear to be separable from lexico-semantic processing within the language network, in line with strong integration between these two processes that has been consistently observed in behavioral and computational language research. The results further suggest that the language network may be generally more strongly concerned with meaning than syntactic form, in line with the primary function of language—to share meanings across minds.

Relevant papers
3. Composition is the core driver of the language network.

The fronto-temporal language network responds robustly and selectively to sentences. But the features of linguistic input that drive this response and the computations these language areas support remain debated. Two key features of sentences are typically confounded in natural linguistic input: words in sentences a) are semantically and syntactically combinable into phrase- and clause-level meanings, and b) occur in an order licensed by the language’s grammar. Inspired by recent psycholinguistic work establishing that language processing is robust to word order violations, we hypothesized that the core linguistic computation is composition, and, thus, can take place even when the word order violates the grammatical constraints of the language. This hypothesis predicts that a linguistic string should elicit a sentence-level response in the language network as long as the words in that string can enter into dependency relationships as in typical sentences. Across two fMRI experiments, we tested this prediction by introducing a varying number of local word swaps into naturalistic sentences, leading to progressively less syntactically well-formed strings. Critically, local dependency relationships were preserved because combinable words remained close to each other. As predicted, word order degradation did not decrease the magnitude of the neural response in the language network, except when combinable words were so far apart that composition among nearby words was highly unlikely. This finding demonstrates that composition is robust to word order violations, and that the language regions respond as strongly as they do to naturalistic linguistic input as long as composition can take place.

Relevant papers
4. The multiple demand (MD) network is highly domain-general and supports fluid intelligence.

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. This network of regions has become known as the “cognitive/executive control” or “multiple-demand (MD)” network and has been implicated as the core of human fluid intelligence. However, the early evidence for the domain-generality of these regions in humans came from meta-analyses of activation peaks from across fMRI studies, 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 at the core of which is the ability to reason about diverse problems (fluid intelligence). Indeed, we found that damage to the MD regions, but not language regions, leads to loss in IQ, and variation in the MD regions’, but not language regions’, activity predicts inter-individual variation in fluid intelligence.

Relevant papers
5. The domain-general multiple demand (MD) network does not support language comprehension.

Aside from the language-selective left-lateralized fronto-temporal network, language comprehension sometimes additionally recruits a domain-general bilateral fronto-parietal network implicated in executive functions: the multiple demand (MD) network. However, the nature of the MD network’s contributions to language comprehension remains debated. Across several studies using naturalistic story materials, we found that—in contrast to the language-selective regions—the domain-general MD brain regions do not closely ‘track’ linguistic stimuli, are not sensitive to how predictable a word is in context, and do not show a correlation with incremental behavioral measures of comprehension difficulty. Further, in a large-scale investigation using data from 30 diverse word and sentence comprehension experiments (481 unique participants, 678 scanning sessions), we found that the MD network responded more strongly in paradigms with an explicit task compared to passive comprehension paradigms. In fact, many passive comprehension tasks failed to elicit a response above the fixation baseline in the MD network, in contrast to strong responses in the language-selective network. In tandem, these results argue against a role for the MD network in core aspects of sentence comprehension like inhibiting irrelevant meanings or parses, keeping intermediate representations active in working memory, or predicting upcoming words or structures. These results suggest that the MD network’s engagement during language processing likely reflects effort associated with extraneous task demands.