福井 直樹

Surface linear (left-to-right) arrangements of human languages are actually an amalgam of the core language system and systems that are not inherently related to language. It has been widely recognized that an unbounded array of hierarchically structured linguistic expressions is generated by the simplest combinatorial operation “Merge,” and the notion of Merge-generability has been proposed as a key feature that characterizes structural dependencies among linguistic elements. Here we tested Merge-generable dependencies by using a Subject-Predicate matching task, which required both linguistic capacity and short-term memory. We used three types of dependency: Nesting, Crossing, and Grouping as the control. The Nesting dependency is totally Merge-generable, while the Crossing dependency requires some additional processes for memory-based ordering. In order to identify the regions employed for these two dependencies, we directly compared cortical responses to the sentence stimuli (with noun phrases and an adverb as the first half of stimuli, and with verbs as the latter) using functional magnetic resonance imaging (fMRI), and the following results were obtained. First, for the Nesting – Crossing contrast, significant activations were observed in the bilateral lateral premotor cortices (LPMCs) and inferior frontal gyri, left middle temporal gyrus, and bilateral angular/supramarginal gyri, indicating engagement of the syntax-related networks. In contrast, the Crossing – Nesting contrast showed focal activations in the left fusiform gyrus, lingual gyrus, and middle occipital gyrus (L. FG/LG/MOG). Secondly, for the first half of the Nesting stimuli, signal changes in the bilateral LPMCs were well fitted with the estimates of computational costs to search the workspace and to select items (Σ operations). Moreover, for the latter half of the Crossing stimuli, the signal changes in the L. FG/LG/MOG were differentially fitted with the estimates of loads related to the ordering of elements/words (numbers of Ordering). Thirdly, these fitting models were by far more likely than the exchanged estimates between bilateral LPMCs and L. FG/LG/MOG, confirming a double dissociation for primary processes with Σ and Ordering. In conclusion, these results indicate that separate cortical networks are differentially employed, and their careful elucidation will provide further insights and challenges.

The nature of computational principles of syntax remains to be elucidated. One promising approach to this problem would be to construct formal and abstract linguistic models that parametrically predict the activation modulations in the regions specialized for linguistic processes. In this article, we review recent advances in theoretical linguistics and functional neuroimaging in the following respects. First, we introduce the two fundamental linguistic operations: Merge (which combines two words or phrases to form a larger structure) and Search (which searches and establishes a syntactic relation of two words or phrases). We also illustrate certain universal properties of human language, and present hypotheses regarding how sentence structures are processed in the brain. Hypothesis I is that the Degree of Merger (DoM), i.e., the maximum depth of merged subtrees within a given domain, is a key computational concept to properly measure the complexity of tree structures. Hypothesis II is that the basic frame of the syntactic structure of a given linguistic expression is determined essentially by functional elements, which trigger Merge and Search. We then present our recent functional magnetic resonance imaging experiment, demonstrating that the DoM is indeed a key syntactic factor that accounts for syntax-selective activations in the left inferior frontal gyrus and supramarginal gyrus. Hypothesis III is that the DoM domain changes dynamically in accordance with iterative Merge applications, the Search distances, and/or task requirements. We confirm that the DoM accounts for activations in various sentence types. Hypothesis III successfully explains activation differences between object- and subject-relative clauses, as well as activations during explicit syntactic judgment tasks. A future research on the computational principles of syntax will further deepen our understanding of uniquely human mental faculties.

Our goal of this study is to characterize the functions of language areas in most precise terms. Previous neuroimaging studies have reported that more complex sentences elicit larger activations in the left inferior frontal gyrus (L.F3op/F3t), although the most critical factor still remains to be identified. We hypothesize that pseudowords with grammatical particles and morphosyntactic information alone impose a construction of syntactic structures, just like normal sentences, and that "the Degree of Merger" (DoM) in recursively merged sentences parametrically modulates neural activations. Using jabberwocky sentences with distinct constructions, we fitted various parametric models of syntactic, other linguistic, and nonlinguistic factors to activations measured with functional magnetic resonance imaging. We demonstrated that the models of DoM and "DoM+ number of Search (searching syntactic features)" were the best to explain activations in the L. F3op/F3t and supramarginal gyrus (L. SMG), respectively. We further introduced letter strings, which had neither lexical associations nor grammatical particles, but retained both matching orders and symbol orders of sentences. By directly contrasting jabberwocky sentences with letter strings, localized activations in L. F3op/F3t and L. SMG were indeed independent of matching orders and symbol orders. Moreover, by using dynamic causal modeling, we found that the model with a inhibitory modulatory effect for the bottom-up connectivity from L. SMG to L. F3op/F3t was the best one. For this best model, the top-down connection from L. F3op/F3t to L. SMG was significantly positive. By using diffusion-tensor imaging, we confirmed that the left dorsal pathway of the superior longitudinal and arcuate fasciculi consistently connected these regions. Lastly, we established that nonlinguistic order-related and error-related factors significantly activated the right (R.) lateral premotor cortex and R. F3op/F3t, respectively. These results indicate that the identified network of L. F3op/F3t and L. SMG subserves the calculation of DoM in recursively merged sentences.

現代生成文法がいかにして記述的道具立てから説明理論へ脱皮してきたかを「併合演算」が持つ特性を中心にして論じる。

The importance of abstract syntactic structures and their crucial role in analyzing sentences have long been emphasized in contemporary linguistics, whereas the linear order model, in which next-coming words in a sentence are claimed to be predictable based on lexico-semantic association or statistics alone, has also been proposed and widely assumed. We examined these possibilities with magnetoencephalography (MEG) and measured cortical responses to a verb with either object-verb (OV) or subject-verb (SV) sentence structures, which were tested in a minimal-pair paradigm to compare syntactic and semantic decision tasks. Significant responses to the normal OV sentences were found in the triangular part of the left inferior frontal gyrus (F3t) at 120-140 ms from the verb onset, which were selective for explicit syntactic processing. The earliest left F3t responses can thus be regarded as predictive effects for the syntactic information of the next-coming verb, which cannot be explained by associative memory or statistical factors. Moreover, subsequent responses in the left insula at 150-170 ms were selective for the processing of the OV sentence structure. On the other hand, responses in the left mediofrontal and inferior parietal regions at 240-280 ms were related to syntactic anomaly and verb transitivity, respectively. These results revealed the dynamics of the multiple cortical regions that work in concert to analyze hierarchical syntactic structures and task-related information, further elucidating the top-down syntactic processing that is crucial during on-line sentence processing. (C) 2008 Elsevier Inc. All rights reserved.

現代言語学の基本目標を説明するとともに,20世紀における言語理論の進展を概説した.Saussureが「社会的所産」とみなした言語構造を人間の脳内に実在する機能(再帰的生成システム)ととらえ,それに関する明示的説明理論の構築を目指したのが生成文法理論であるが,「より深い説明」を目指す継続的試みが現在に至る理論的変遷の根本的原動力になっていることを論じた.その上で,言語学と言語障害学が将来さらに有機的に連携するためには,言語脳科学ともいうべき分野の確立が必須であるとともに,今までの生成文法理論には欠如していた「言語の社会性」を厳密に研究するためのモデル構築が望まれるということも指摘した.