18. Language I

What this lecture covers

The lecture asks whether language is a distinct system in the mind and brain, or whether it is inseparable from thought in general. It sits in a course on the human brain after weeks of building up functional MRI methods, and it uses those methods, plus evidence from brain damage, to answer the question. After watching, you can explain why patients who lose almost all language can still reason, do arithmetic, and understand other people's intentions, and why decades of brain-imaging studies appeared to contradict that finding.

The lecture opens with a recap of representational similarity analysis (RSA) before turning to its main topic. RSA compares whole patterns of pairwise similarity across conditions, whether those patterns come from fMRI voxels, neurons, behavioral ratings, or a hypothesized model, which makes it possible to compare representations across brain regions, species, and methods.

Key ideas

• Representational similarity analysis (RSA): a method that compares matrices of pairwise similarity between conditions (from voxels, neurons, ratings, or models) rather than a single binary classification.
• Compositionality: human language lets speakers combine words to express sentences never said before, unlike animal communication, which is typically restricted to narrow survival-related signals.
• Components of language: phonology (sound structure), lexical semantics (word meaning), syntax (word order and grammar), and pragmatics (inferring intent beyond literal words).
• Global aphasia: near-total loss of language after brain damage, studied in patients who nonetheless pass tests of causal reasoning, arithmetic, and understanding who-did-what-to-whom scenarios.
• Group analysis pitfall: anatomically aligning many subjects' brains and averaging can create the appearance of overlapping activation for two tasks even when no individual subject shows any overlap.
• Functional localizer: identifying a functional region (such as a language region) separately in each subject, rather than relying on anatomical alignment across a group.
• Language specificity: when language regions are localized individually per subject, they respond strongly to sentence comprehension but barely at all to arithmetic, spatial tasks, music, or other cognitive control tasks.

Walkthrough

Representational similarity analysis recap (0:10)

The lecture reviews RSA as a generalization of multivoxel pattern analysis. Instead of comparing patterns for only two conditions, RSA builds a matrix of pairwise similarities across many stimuli, in a brain region, in behavioral ratings, or in neural recordings, and correlates matrices across methods and species, as long as the same set of stimuli defines the axes.

What language is and how it differs from animal communication (12:21)

The lecture frames language as universal to neurologically intact humans, equally expressive across roughly 7,000 languages, and unique among species. A video of Chaser the border collie, who can identify a thousand toy names, is used to show that impressive word association is not the same as language: Chaser cannot combine words to express relationships such as who did what to whom, which the lecture treats as central to human language. The lecture then surveys phonology, lexical semantics, syntax, and pragmatics as the main components of language, before narrowing the rest of the lecture to syntax and semantics in sentence understanding.

Can you think without language? (24:31)

Students discuss whether thought is possible without language. The lecture brings in evidence that pre-verbal infants and animals lacking language can still think in sophisticated ways, and describes a study of the Pirahã, an Amazonian group whose language has no number words, who nonetheless succeed at approximate-magnitude matching tasks. This supports the idea that at least some forms of thought, such as the approximate number system, do not depend on language.

Evidence from developmental disorders and brain damage (32:39)

The lecture describes Williams syndrome, where low IQ and poor spatial reasoning coexist with unusually rich language, as early evidence that language and general cognition can dissociate. It then covers Paul Broca's 1861 identification of the left frontal lobe (Broca's area) as important for speech, based on a patient who lost speech but retained other thinking abilities. Writer Tom Lubbock's account of losing language to a brain tumor, while insisting his thoughts remained "vast, limitless," is presented as a first-person account of preserved thought without language.

Global aphasia and preserved nonverbal thought (39:45)

Rosemary Varley's studies of patients with global aphasia, who perform at chance on language tasks, show these patients succeed at ordering picture sequences, distinguishing real from invented objects, judging plausible events, performing arithmetic and logic tasks, appreciating music, and reasoning about other people's mental states. The lecture concludes that language and thought are not the same system in adulthood, while noting that language remains essential earlier in development, citing deaf children who acquire language late and show later delays in understanding others' thoughts.

The neuroimaging contradiction and the group-analysis problem (52:52)

The lecture identifies a conflict: patient data show language is separate from other thought, but most fMRI studies report that language regions overlap with regions active during arithmetic, music, and other tasks. It attributes this to a methodological flaw in group analysis, where brains are aligned anatomically and averaged, which can create apparent overlap even when no individual subject shows any.

Individually localized language regions show specificity (57:55)

The lecture presents a solution: define language regions functionally in each subject using a localizer contrasting sentences against non-word strings, validated for reliability, and for generalizing across reading versus listening, across languages, and across degraded versus intact speech. When Nancy Kanwisher and Evelina Fedorenko tested seven other cognitive tasks against these individually defined regions, the language regions showed little to no response to arithmetic, spatial working memory, music, or other cognitive control tasks, resolving the earlier contradiction and matching the patient evidence.

Before you watch

• Review multivoxel pattern analysis and functional localizers, covered in earlier lectures on the fusiform face area, since this lecture builds directly on both.
• Recall the approximate number system from the numerical cognition lecture, since it is used here as an example of thought without language.

Check your understanding

1. Why does a group fMRI analysis risk showing overlap between two tasks even when no individual subject's brain shows any overlap?
2. What evidence from global aphasia patients supports the claim that thought can occur without language?
3. How does the lecture distinguish language from general communication, using Chaser the border collie as an example?
4. Why does the lecture separate the question of whether language is needed during development from whether it is needed for ongoing adult thought?
5. What makes representational similarity analysis richer than a simple two-condition classification?