Oscillatory neuronal dynamics during language comprehension

Abstract

Language comprehension involves two basic operations: the retrieval of lexical information (such as phonologic, syntactic, and semantic information) from long-term memory, and the unification of this information into a coherent representation of the overall utterance. Neuroimaging studies using hemodynamic measures such as PET and fMRI have provided detailed information on which areas of the brain are involved in these language-related memory and unification operations. However, much less is known about the dynamics of the brain's language network. This chapter presents a literature review of the oscillatory neuronal dynamics of EEG and MEG data that can be observed during language comprehension tasks. From a detailed review of this (rapidly growing) literature the following picture emerges: memory retrieval operations are mostly accompanied by increased neuronal synchronization in the theta frequency range (4–7 Hz). Unification operations, in contrast, induce high-frequency neuronal synchronization in the beta (12–30 Hz) and gamma (above 30 Hz) frequency bands. A desynchronization in the (upper) alpha frequency band is found for those studies that use secondary tasks, and seems to correspond with attentional processes, and with the behavioral consequences of the language comprehension process. We conclude that it is possible to capture the dynamics of the brain's language network by a careful analysis of the event-related changes in power and coherence of EEG and MEG data in a wide range of frequencies, in combination with subtle experimental manipulations in a range of language comprehension tasks. It appears then that neuronal synchrony is a mechanism by which the brain integrates the different types of information about language (such as phonological, orthographic, semantic, and syntactic information) represented in different brain areas.

Introduction

Understanding natural language or, roughly stated, the mapping of sound or orthography onto meaning, is a deceivingly simple task for most of us. The fact that language comprehension is a hard-to-suppress “reflex” is nicely illustrated, for instance, by the word interference effect in a standard color–word Stroop task (e.g., MacLeod, 1991). Yet, understanding how language comprehension is achieved is by all means a much more difficult enterprise. Indeed, studying language comprehension (and other aspects of language processing, such as language production, language acquisition, and more generally the relation between language, thought, and culture) has evolved into a large and active research field, that of psycholinguistics.

Traditionally, psycholinguistic research has made use of empirical methods such as behavioral experiments, computational modeling, the analysis of cross-linguistic differences, and many more. However, a seminal paper by Kutas and Hillyard (1980) showed that some aspects of semantic processing induce reliable responses in the EEG recordings of normal, healthy subjects. Since then, cognitive neuroscientific methods have become increasingly popular in psycholinguistic research (see, e.g., Brown and Hagoort, 1999). The success, or popularity of the cognitive neuroscience of language (although being stimulated by the emergence of new brain imaging techniques like PET, fMRI, and MEG), is for a large part based on the insight that the human brain is the only known system that is able to fluently produce and understand natural language. Therefore, it seems reasonable to assume that a better understanding of the neuronal processes underlying language comprehension will be helpful in shaping the existing functional models of language comprehension (this is sometimes referred to as upward adequacy). Vice versa, the same functional models may be helpful in guiding our understanding of the neuronal processes that are observed during language comprehension (downward adequacy).

In this chapter we concentrate on the rapid dynamics of the neural processes underlying language comprehension. However, before turning to this, let us first briefly delineate at a very general level what is thought to be the cognitive architecture of language comprehension.

It is generally agreed that during language comprehension, incoming sounds or orthographic patterns trigger a cascade of memory retrieval operations that make available the phonologic, syntactic, and semantic properties of individual words. Once available, these different ingredients have to be integrated (unified) at the sentence and/or discourse levels into a meaningful whole, in order to yield a coherent interpretation of the linguistic input (see Hagoort, 2005 for a more detailed elaboration of this framework). Thus, two different cognitive processes, namely memory retrieval operations and unification operations, play a crucial role during language comprehension.

Note that in this formulation it is a very general framework that does not address most of the more detailed — and often hotly debated — issues in language comprehension research (e.g., whether or not syntactic analysis precedes semantic analysis, compare, e.g., Marslen-Wilson and Tyler, 1980 and Friederici, 2002). However, exactly by avoiding such details it provides a common ground for most psycholinguistic researchers — and a good starting point for entering the available neural data into the debate.

Ever since Paul Broca and Carl Wernicke identified separate areas in the brain that are specialized for certain aspects of language processing, it has been evident that there must be a distributed network in the brain that is responsible for language processing. Hemodynamic neuroimaging techniques such as PET and fMRI, with their excellent spatial resolution, are particularly suited for identifying the different brain areas that participate in this distributed network, or stated differently, to unveil the structure of the brain's language network.

Although language is a very well-delineated cognitive function, the neural structures involved in the memory operations involved in retrieving stored linguistic knowledge are likely to have evolved from, and therefore at least show a good deal of overlap with, the structures involved in retrieving domain-general knowledge. A large number of hemodynamic studies have linked the (domain-general) retrieval of declarative (i.e., factual and episodic) information from long-term memory to increased blood oxygenation level dependent (BOLD) activity in a large number of areas including prefrontal, temporal, anterior cingulate, and cerebellar areas (see Cabeza and Nyberg, 2000 for a review). The medial temporal lobe (MTL) system is assumed to play a central role in memory operations, in that it establishes cortico-hippocampal loops that serve the purpose of coactivating distributed cortical areas in which the relevant information is stored (e.g., Miller, 1991; Murre et al., 2001). Among the cortical areas that have been specifically associated with retrieval of lexical information in the context of language comprehension are the central and posterior parts of the left middle and superior temporal gyri, and the posterior inferior frontal gyrus (see Wise, 2003; Indefrey, 2004; Indefrey and Cutler, 2005 for reviews and meta-analyses).

Linguistic unification operations pertaining to the domains of phonology, syntax, and semantics are thought to be carried out in a set of brain regions including the left inferior frontal gyrus (Broca's area and surrounding tissue, roughly BAs 44, 45, 47, and ventral BA 6, see Hagoort, 2005 for details), and possibly the left posterior superior temporal gyrus (Indefrey and Cutler, 2005). In the left inferior frontal gyrus, there appears to be regional specificity for phonology, syntax, and semantics (see the review by Bookheimer, 2002).

The brief literature overview presented above shows that the network of brain structures that contributes to the process of language comprehension is becoming increasingly well understood. However, due to the inherently poor temporal resolution of hemodynamic measures (i.e., in the order of seconds), the resulting picture is a rather static one, emphasizing mainly the structural aspects of the brain's language network. This static view does not do justice to the dynamic properties that any language comprehension device must have. Normal speech has a rate of about three to five words per second, which means that the linguistic retrieval and unification operations must be carried out very rapidly. Therefore, trying to understand the neuronal implementation of language comprehension by relying exclusively on hemodynamic techniques would be analogous to trying to understand a piece of music by making an inventory of the instruments that constitute the orchestra. What is missing in such a static description is each instrument's melody, and the way the different instruments interact. Similarly, in order to capture the fast dynamics of the brain's language network, the information obtained from PET and fMRI studies needs to be complemented with information derived from EEG and MEG, which record neuronal activity on a millisecond time scale.

This chapter aims at providing a literature review of the rapid, oscillatory changes that are present in the EEG or MEG signals while subjects perform a variety of language comprehension tasks. In the following section we briefly indicate why these oscillatory changes are considered to provide a window onto the neuronal dynamics of the brain's language comprehension system. In addition, we delineate a general methodological framework for analyzing oscillatory brain dynamics. In the section “Experimental data,” we review the available experimental data, and the “Discussion” section contains a discussion of the data and some concluding remarks.