Segregating the core computational faculty of human language from working memory

Abstract

In contrast to simple structures in animal vocal behavior, hierarchical structures such as center-embedded sentences manifest the core computational faculty of human language. Previous artificial grammar learning studies found that the left pars opercularis (LPO) subserves the processing of hierarchical structures. However, it is not clear whether this area is activated by the structural complexity per se or by the increased memory load entailed in processing hierarchical structures. To dissociate the effect of structural complexity from the effect of memory cost, we conducted a functional magnetic resonance imaging study of German sentence processing with a 2-way factorial design tapping structural complexity (with/without hierarchical structure, i.e., center-embedding of clauses) and working memory load (long/short distance between syntactically dependent elements; i.e., subject nouns and their respective verbs). Functional imaging data revealed that the processes for structure and memory operate separately but co-operatively in the left inferior frontal gyrus; activities in the LPO increased as a function of structural complexity, whereas activities in the left inferior frontal sulcus (LIFS) were modulated by the distance over which the syntactic information had to be transferred. Diffusion tensor imaging showed that these 2 regions were interconnected through white matter fibers. Moreover, functional coupling between the 2 regions was found to increase during the processing of complex, hierarchically structured sentences. These results suggest a neuroanatomical segregation of syntax-related aspects represented in the LPO from memory-related aspects reflected in the LIFS, which are, however, highly interconnected functionally and anatomically.DTI ͉ fMRI ͉ hierarchical structure L anguage appears to be a trait specific to humans-at least in its core computational component, that is, grammar. Defining language as a sequence of symbols, Chomsky (1) proposed a hierarchy of grammars as language production mechanisms with increasing generative powers. The lowest-level grammar is finite state grammar (FSG). FSG can be fully specified by transition probabilities between a finite number of states (e.g., words), being not powerful enough to generate structures of natural human languages. Phrase structure grammar (PSG) has more generative power than FSG. A key difference between FSG and PSG is that only PSG can generate the sequence A n B n , where A and B denote symbols and n the number of repetitions. The ability to process the sequence A n B n is crucial for the processing of center-embedded sentences, such as ''The man the boy the dog bit greeted is my friend.'' where subjects (i.e., the man, the boy, and the dog) are A-symbols and the verbs (bit, greeted, and is) are B-symbols. Surprisingly, tests on monkeys (2) and on songbirds (3) showed that whereas songbirds can process A n B n sequences, monkeys cannot. However, even if the birds could correctly discriminate A n B n sequences from A n B m , (4 Ͼ n, m Ͼ0, n m),...

What artificial grammar learning reveals about the neurobiology of syntax

“…Friederici and colleagues (2006) argue that their FMRI findings (p. 2458) ''show that the processing of these two sequence types [generated by a ''finite state" and a ''phrase structure" grammar, respectively] is supported by different areas in the human brain" and that their results (p. 2460) ''indicate a functional differentiation between two cytoarchitectonically and phylogenetically different brain areas in the left frontal cortex ["frontal operculum" and ''BA 44"]". Similar reasoning is found in Friederici, Bahlmann et al (2006), Friederici, Fiebach, Schlesewsky, Bornkessel, and von Cramon (2006), Bahlmann et al (2008), and Makuuchi et al (2009) 1 . In the context of these FMRI findings and suggestions, we raise the question whether Broca's region (or subregions) is specifically related to syntactic movement operations or the processing of hierarchically nested non-adjacent dependencies?…”

“…In fact, dynamic functional Table 2 Overlap between the activated clusters in the listed studies and the clusters that we found activated in this region for the NG vs. G contrast. Columns 1-3: The [x, y, z] coordinates and the function labeling are taken from the studies: Makuuchi et al (2009), Bahlmann et al (2008, Friederici, Bahlmann et al (2006), Friederici, Fiebach et al (2006), Santi and Grodzinsky (2007a), Ben-Shachar et al (2003, 2004. Santi and Grodzinsky (2007b) reported the coordinates [À50 32 6] and [À50 33 8], but since this is a fixed-effects study we did not further analyze this here.…”

“…In particular the syntactic feature of center-or nested embedding has been the focus of recent research (Bahlmann, Schubotz, & Friederici, 2008;Fitch & Hauser, 2004;Friederici, Bahlmann, Heim, Schubotz, & Anwander, 2006;Makuuchi, Bahlmann, Anwander, & Friederici, 2009). In the linguistic and psycholinguistic literature, the Chomsky hierarchy is most often formulated in terms of formal grammars.…”

Language processing in the occipital cortex of congenitally blind adults

“…Therefore, like the LPFC, occipital areas may support selection and manipulation of linguistic representations in working memory (47,48). Alternatively, the left occipital cortex might combine linguistic units into structured wholes (49) or support recursive structure building during sentence processing (12,33,50). Furthermore, just as there are distinct functions with the prefrontal cortex, so there may be distinct parts of the occipital cortex that support different high-level cognitive or linguistic functions.…”

“…A key feature of human language is that elementary linguistic units (words and morphemes) combine into higher-level structures, such as phrases and sentences. It has been suggested that language regions, such as Broca's area, have evolved to be uniquely suited for such combinatorial processing (12,33,34). Finding that occipital areas in early blind individuals are capable of combinatorial language processing would suggest that such processing does not require intrinsic biological prop-erties of classic language regions.…”