Language in Our Brain: The Origins of a Uniquely Human Capacity, by Angela D. Friederici Free

Angela D. Friederici’s new book is the crowning achievement of the long and prolific career of one of the founding figures of the cognitive neuroscience of language. In Language in Our Brain (LiOB), Friederici provides a synthesis of decades of empirical research on the neural bases of human language—mainly from her own lab at the Max Planck Institute for Human Cognitive and Brain Sciences in Leipzig and affiliated labs. The book is richly illustrated and comprises eight chapters with summaries, a foreword by Noam Chomsky, and a glossary of technical terms covering language science as well as current neuroscience methods. LiOB is intended both as a compendium of Friederici’s own work for the specialist and as a guide to the neurobiology of language for students and researchers in neighboring areas of science.

In the Introduction, Friederici makes explicit the ‘cultural horizon’ within which the project of LiOB is situated. There are two key ideas here. First, the modern neurological notion that brain anatomy constrains and shapes cognitive processes—an application of the traditional principle of materialist biology that ‘structure determines function’. This is exemplified by a quote by Paul Flechsig, which Friederici adopted as a motto for LiOB: ‘It is rather unlikely that psychology, on its own, will arrive at the real, lawful characterization of the structure of the mind, so long as it neglects the anatomy of the organ of the mind.’ Second, the basic tenets of the Minimalist Program in generative syntax and their implications for the biology of language—from neurology to acquisition and evolution—in particular the ideas that (1) structure building in syntax boils down to a single operation, called ‘Merge’, and (2) human language emerged as the result of a single change to brain structure, or a ‘slight rewiring of the brain’, which indeed made Merge possible (Chomsky 1995; Berwick and Chomsky 2016). To fully appreciate LiOB one needs to accept, even just temporarily, these ideas.

In Chapter 1, Friederici outlines a model of auditory language processing based on a review and summary of experiments mapping specific subprocesses in brain space and time using current neuroscience techniques. The model, developed by Friederici over the past 20 years, includes four main components: (1) an acoustic-phonological processor, (2) a largely left-hemispheric system for computing ‘syntactic and semantic relations’ in phrases and sentences, (3) a largely right-hemispheric system for processing prosody, stress, and intonation, and (4) an interpretation system for integrating information from subsystems (2) and (3), which are seen as parallel. One of the strengths of Friederici’s proposal is indeed its scope, the fact that the model aims to capture ‘every significant aspect of the structure and use of language’ (Chomsky’s foreword, p. ix).

The most pressing and stimulating questions arise from close inspection of subsystem (2). LiOB builds on Friederici’s classic Syntax First (SF) hypothesis: the view that an early stage of ‘syntactic phrase structure building’ precedes and feeds into the derivation of syntactic and semantic relations (p. 16). This process amounts to the identification of syntactic word category and its use in computing an ‘initial phrase structure’ (p. 32). It is claimed to occur within 200 ms from word onset, and is reflected in early left anterior negativity (ELAN) effects in event-related potentials (ERPs). Friederici’s case is generally appealing, but does not fully engage with contrary evidence. First, the criticism of the validity of ELAN findings by Steinhauer and Drury (2012) is relegated to a mere footnote (pp. 34, 241). Second, the study by van den Brink and Hagoort (2004) is seen by Friederici as a failure to provide evidence against SF (p. 36), but their data do show that lexical semantic relations can be computed before word category can be identified. One may then coherently reformulate SF as the weaker hypothesis that word category information is processed very rapidly as soon as it is available; however, in general, the brain handles first the information that becomes available first—such is the ‘opportunism’ of the language system (Jackendoff 2007). Mitigating some of the implications of SF, LiOB also posits (1) parallel computation of syntactic (reflected in ERPs by the LAN component) and semantic relations (N400)—these processes interact in the assignment of thematic roles (pp. 16, 62–7)—and (2) a subsequent stage at which the outputs of these different processes are integrated (P600; Fig. 1.21; p. 83). This interactive-and-integrative account realigns Friederici’s proposal with most current models of language processing.

In Chapter 2, titled ‘Excursions’, Friederici briefly surveys the relations between the model presented in Chapter 1, language production, and communication. Chapters 3 and 4 are the true core of LiOB. Here, Friederici amplifies some of the ideas presented in the final section of Chapter 1 on the ‘Functional Neuroanatomy of Language Comprehension’. The result is a detailed model of the structural (Chapter 3) and functional language network (Chapter 4).

Chapter 3 shows Friederici at her best. She deconstructs the anatomy of language networks in the brain, in particular the organization of white matter connectivity, using data from magnetic resonance tractography. The aim is to identify the critical anatomical pathways enabling bi-directional information transfer between brain areas, which may also support specific language-relevant operations. The success of this task—mapping anatomical pathways to linguistic functions one-to-one (Fig. 3.5; p. 120)—is crucial in order for LiOB to live up to Flechsig’s maxim and to the ‘structure determines function’ doctrine. Friederici’s hypothesis is that there are two dorsal pathways and two ventral pathways underlying language in the brain. The first dorsal stream subserves auditory-to-motor mapping (e.g. during speech production) and consists of fibers of the superior longitudinal fasciculus (SLF), connecting premotor cortex (BA 6) to the posterior superior temporal gyrus and sulcus (pSTG/STS). The second dorsal pathway links the pSTG/STS to the inferior frontal gyrus (IFG; BA 44) through the arcuate fasciculus (AF), and is seen by Friederici as a key circuit for ‘complex syntax’. The first ventral stream connects regions of the posterior temporal cortex—middle and superior temporal gyri—to more anterior portions of the IFG (BA 45/47) via the extreme capsule bundle, and is associated with semantic processing. The second ventral route connects anterior regions of the temporal lobe with the frontal operculum (FOP) via the uncinate fasciculus (UF), and is claimed to subserve ‘elementary combinatorics’ (p. 120). This is a left-lateralized network; the right hemisphere contributes more to processing prosody, stress, and intonation.

Chapter 4 lays out an incisive account of the neuroscience governing the language network. LiOB explains that there are common receptor-architectonic features to core language areas (IFG, i.e. BA 44/45/47, pSTG/STS, and Te2-auditory cortex) of the left hemisphere, but not of their right-hemispheric homologues. These properties set these regions apart from other cortical structures, not only in terms of fine neuroanatomy, but also functionally. Friederici describes how the different nodes of the language network interact via bottom-up and top-down processes: input-driven activation of auditory cortices and anterior temporal regions is followed by recruitment of inferior frontal cortex, corresponding to information transfer via ventral pathways, and resulting in first-pass syntactic (BA 44) and semantic (BA 45/47) computation; the inferior frontal cortex then drives further top-down processing in parietal and in posterior temporal cortex, where syntactic and semantic information are integrated. Neural oscillations, Friederici argues, could provide information on the ‘temporal domains’ at which local (within-area) and long-distance (across-areas) cortical processes unfold. An important remark by Friederici is that we know more about how information is transmitted than about how it is coded in the brain (pp. 133–4). Addressing this problem is one of the great challenges of neurobiology, not just of neurolinguistics.

Another high point of LiOB are the two chapters on language development. The first half of Chapter 5 is dedicated to bilingual language acquisition and processing; the second makes a case for critical or sensitive periods in first- and second-language acquisition. White matter maturation is viewed as the prime constraint on developmental trajectories, while learning results in the fixation and strengthening of white matter pathways—possibly accompanied by a decrease in gray matter (pp. 155–6). Using evidence from sign language processing, Friederici argues for the universality of the language network: a common set of brain areas supporting language, largely independent of input modality—spoken, written, or signed.

In Chapter 6, Friederici discusses the ontogeny of the language network. The review of the experimental literature is well balanced and insightful. The emerging picture indicates that children become sensitive to semantic relations in sentences by the end of the second year of life, as suggested by N400 research (p. 174), although N400 effects to incongruent words and pictures are observed earlier on. Syntactic processes leave traces in ERPs only between 2 and 3 years of life: the P600 at 24 months, and the ELAN at 32 months (pp. 180–2). It is unlikely, however, that Friederici would readily accept a Meaning Before Grammar account of language development. On the one hand, children’s ability to establish semantic relations in sentences arises earlier than the ability to compute syntactic relations (Fig. 6.14; p. 198). On the other hand, the SF hypothesis has a counterpart in ontogeny, in the notion that ‘morphosyntactic categorization’ and ‘phrase structure reconstruction’ appear during the first year of life, at about the same time as ‘lexical semantic categorization’ and ‘lexical access and retrieval’. Therefore, Friederici’s conclusion is that infants can compute aspects of phrase structure before they can relate words semantically.

In Chapter 7, Friederici briefly discusses evolutionary evidence for the linguistic advantage of humans over nonhuman primates, focusing on two components of the language system: sensory-motor learning—which is shared with some other species, for example, songbirds—and the capacity to process hierarchical structures—which is instead argued to be uniquely human. To account for the latter, Friederici calls on human neuroanatomical specificity: the relative strength of dorsal white matter, connecting the inferior frontal cortex to the pSTG/STS (the ‘slight rewiring of the brain’), and the core language network’s leftward asymmetry.

In Chapter 8, Friederici summarizes the discussion and formulates a cohesive picture of the biological foundations of language, starting from the assumption that ‘syntax is at the core of human language, as it provides the rules and computational mechanisms that determine the relation between words in a sentence’ (p. 223). This syntactocentrism is both the main strength and the main weakness of LiOB. On the one hand, it gives the book a clear focus on syntax in our brain, and a sharp angle of attack on the problems and phenomena addressed. On the other hand, it relegates semantics—which also provides mechanisms that determine relations between words—to a secondary role. Friederici’s argument for this choice is thin, however. She warns us that ‘our feeling—when considering language—may rather tend to favor semantics and meaning as the most important aspects of language’ (p. 231), and she remarks that ‘conceptual semantics’—what is meant is rather world knowledge—specifies representations that are highly variable, idiosyncratic, and contingent: their brain bases are ‘less confined than the linguistic-semantic representation’ (p. 4). Yet, ease of localization is not necessarily a good guide to what is worth investigating empirically. Moreover, the wall between formal and cognitive semantics has fallen (Hamm et al. 2006), and the autonomy of generative operations in semantics, from the lexicon to discourse, has been demonstrated (Jackendoff 2002). Giving semantics its due weight would greatly enrich our understanding of language in the brain (for a sustained argument, see Baggio 2018).

LiOB endorses the latest version of generative grammar: the Minimalist Program (Chomsky 1995). Specifically, the thesis is that ‘the posterior portion of Broca’s area [BA 44] together with the adjacent frontal operculum represent a local circuit subserving the basic syntactic computation of Merge’ (p. 224). Friederici’s idea, that Merge should have a cognitive basis, is not new—it was originally put forward nearly 20 years ago by Townsend and Bever (2001, p. 82: ‘we should consider examining the psychological basis for the two primary and ever-present operations, merge and move’). Nor is Friederici alone in believing that Merge itself can be—in fact, has been—localized or otherwise identified in the brain (e.g. see Nelson et al. 2017). But what is Merge formally? And does it make sense to look straightforwardly for its implementation in the brain? Merge is set formation: given two syntactic objects, X and Y (elements of the lexicon, such as words, or sets resulting from Merge), Merge forms a novel, hierarchically structured object, that is, the set {X, Y} (Berwick and Chomsky 2016, p. 10). A few remarks are in order. Linguistic theory has often been a moving target for experimentalists (Townsend and Bever 2001, pp. 38–9): how to assess the psychological and neural reality of operations that may be abandoned or radically revised in newer versions of the formalism? Might this be the case for Merge, too? A more serious concern is this: it is not clear just how syntactic Merge really is. First, Merge does not label objects syntactically—that is done by a separate ‘labeling algorithm’. Secondly, Merge yields unordered sets, not strings or sequences: it does not give word order, for example. Merge as such cannot support sequence processing as Friederici envisages it (p. 4). Perhaps an intrinsically temporal operation like Unification (Hagoort 2003; Hagoort et al. 2009) would come closer to hitting the spot?

Empirically minded readers can fully appreciate LiOB even ignoring its allegiance to Merge. Theoretically minded readers can find in it an opportunity to imagine how semantics could be reassessed within the framework of LiOB, without altering that framework significantly. As a strictly syntactic operation, Merge appears so abstract and general that it requires the theorist to describe a set of composition operations at the semantic level, which may or may not correspond one-to-one (as is required by Compositionality) to applications of Merge at the syntactic level. Without these additional semantic composition operations, it is doubtful that set formation can produce complex meanings. Friederici—perhaps inspired by work in Liina Pylkkänen’s lab—links ‘elementary combinatorics’ to the anterior temporal lobe. But elsewhere in LiOB, this region is claimed to subserve merely the ‘identification’ of adjacent elements, which are then merged into a ‘local syntactic hierarchy’ in posterior BA (p. 225). Are syntactic and semantic composition distinct? And how are they realized in the brain? Providing definite answers to these questions could much increase the explanatory and predictive powers of the model presented in LiOB.

In Chomsky’s more recent theorizing, Merge is tightly interfaced with semantics at both the input and output levels, and meaning is given a more prominent role than in LiOB. Merge is thus defined as ‘an operation that takes human concepts as computational atoms and yields structured expressions that, semantically interpreted by the conceptual system, provide a rich language of thought’ (Berwick and Chomsky 2016, p. 87; emphasis added). Berwick and Chomsky argue that ‘language evolved for thought and interpretation: it is fundamentally a system of meaning. Aristotle’s classic dictum that language is sound with meaning should be reversed. Language is meaning with sound (or some other externalization, or none)’ (p. 101). Even within Friederici’s preferred ‘linguistic philosophy’, one may find the arguments required to restore semantics to its proper place. An extended version of LiOB’s theory that takes meaning fully into account would represent an exciting new development.

LiOB is a masterful discussion of language in the brain, and a massive effort at unifying key empirical results on language processing, acquisition, and evolution into a coherent whole. Friederici deserves credit for the cohesiveness and breadth of the end result. Moreover, LiOB stands out for bringing such extensive knowledge into one place, facilitating researchers in their continuous endeavor to understand brain and language. Finally, LiOB is an excellent illustration of how ‘linguistic philosophy’ can orient and drive empirical research, offering a framework for systematizing large swathes of seemingly inconsistent experimental results. Friederici’s book is a must read for students of psycholinguistics, a resource for researchers interested in developing models of language processing informed by generative syntax, and an essential reference for those striving to formulate alternative theories.

Acknowledgements

Elena Morgan is supported by the Research Council of Norway (NFR) under the FRIHUMSAM grant ‘Origins of Semantic Composition in Early Cognitive Development’ (project no. 251219). Giosuè Baggio is grateful to Andrea Moro for providing initial feedback on some of the ideas discussed here.

References