Brain and Language 96 (2006) 90–105 www.elsevier.com/locate/b&l EEG theta and gamma responses to semantic violations in online sentence processing Lea A. Hald a,c, Marcel C.M. Bastiaansen a,b,¤, Peter Hagoort a,b a Max Planck Institute for Psycholinguistics, P.O. Box 310, 6500 AH Nijmegen, The Netherlands b F. C. Donders Centre for Cognitive Neuroimaging, Radbout Universiteit Nijmegen, P.O. Box 9101, 6500 HB Nijmegen, The Netherlands c Center for Research in Language, University of California, San Diego, 9500 Gilman Dr., Dept. 0526, La Jolla, CA 92093-0526, USA Accepted 18 June 2005 Available online 3 August 2005 Abstract We explore the nature of the oscillatory dynamics in the EEG of subjects reading sentences that contain a semantic violation. More speciWcally, we examine whether increases in theta (t3–7 Hz) and gamma (around 40 Hz) band power occur in response to sen- tences that were either semantically correct or contained a semantically incongruent word (semantic violation). ERP results indicated a classical N400 eVect. A wavelet-based time-frequency analysis revealed a theta band power increase during an interval of 300– 800 ms after critical word onset, at temporal electrodes bilaterally for both sentence conditions, and over midfrontal areas for the semantic violations only. In the gamma frequency band, a predominantly frontal power increase was observed during the processing of correct sentences. This eVect was absent following semantic violations. These results provide a characterization of the oscillatory brain dynamics, and notably of both theta and gamma oscillations, that occur during language comprehension. 2005 Elsevier Inc. All rights reserved. Keywords: EEG oscillations; Gamma; Language processing; N400; Theta; Wavelets 1. Introduction eVect. We will focus on the modulation of oscillatory neuronal activity recorded in the EEG during the read- Currently, there has been a growing interest in the ing of sentences that contain either a semantically incon- oscillatory dynamics that can be observed in the electro- gruent word (which elicits an N400 in the ERP analysis) physiological activity of the brain. Unlike Event-Related or that are semantically congruent. More speciWcally, we brain Potentials or Welds (ERPs/ERFs; the so-called want to establish whether semantic violations induce evoked activity) that can be extracted from the EEG or oscillatory activity that is qualitatively diVerent from MEG by straightforward averaging of single trials, those induced by syntactic violations (which elicit a P600 event-related changes in oscillatory EEG/MEG activity in the ERP analysis) that have been found in a previous (termed induced activity) that speak on the synchroniza- study (Bastiaansen, van Berkum, & Hagoort, 2002b). tion and desynchronization aspect of neural activity Much of the research on language comprehension has have received little attention. The goal of the present focused on the integration of semantic and syntactic study is to explore the oscillatory dynamics that coincide information. In particular, a large body of research has with the well-studied semantic ERP eVect, the N400 centered on assessing the inXuence of semantic informa- tion on the syntactic analysis of sentences. This has spawned a debate about whether semantic and syntactic * Corresponding author. Fax: +31 24 3610 989. E-mail address: [email protected] (M.C.M. processing are subserved by neuronally distinct net- Bastiaansen). works [i.e., as proposed by Hagoort and Brown (1999)] 0093-934X/$ - see front matter 2005 Elsevier Inc. All rights reserved. doi:10.1016/j.bandl.2005.06.007 L.A. Hald et al. / Brain and Language 96 (2006) 90–105 91 or whether semantics and syntax are processed by a sin- To form a full understanding of language input all of gle network [i.e., as proposed by McClelland, StJohn, the diVerent parts of information that are processed in and Taraban (1989)]. Within the domain of event-related diVerent brain areas must be integrated. How this bind- potential (ERP) research, this debate has resulted in a ing between the distributed nodes of the language net- search for ERP components that are diVerentially sensi- work is implemented still an open question, but it has tive to semantic and syntactic processing in language been suggested that it may be propagated by synchroni- comprehension. Several diVerent ERP proWles appear to zation and desynchronization of oscillatory neural activ- be related to either aspects of semantic or syntactic pro- ity. In this view, synchronization and desynchronization cessing. Two in particular are the N400, which is related link spatially distributed brain areas together to form to semantic processing, and the P600, which is related to transient functional networks (Singer, 1993, 1999). syntactic processing. It is generally agreed upon that analyzing event- The P600 component is a broadly distributed positive related changes in either amplitude, or (phase) coherence shift that begins around 500 ms after the occurrence of a of EEG oscillations provide a window onto the pro- syntactic violation. The P600 has been observed with cesses of synchronization and desynchronization of neu- various syntactical violations, such as phrase structure ronal populations (Tallon-Baudry & Bertrand, 1999; violations, verb argument violations, and morpho-syn- Varela, Lachaux, Rodriguez, & Martinerie, 2001). Two tactic violations such as breaches of number and gender types of increase are studied: an increase in amplitude of agreement (Friederici, Pfeifer, & Hahne, 1993; Hagoort, oscillatory EEG activity is taken to at least partially Brown, & Groothusen, 1993; Neville, Nicol, Barss, For- reXect an increase in synchrony of the underlying local ster, & Garrett, 1991; Osterhout & Mobley, 1995). neuronal population. On the other hand, an increase in A seminal study by Kutas and Hillyard (1980) dem- (phase) coherence between EEG sensors is thought to onstrated that if participants are presented with sen- reXect increased synchrony between spatially distributed tences ending with a semantically inappropriate word neuronal populations. These event-related increases in (such as “He spread the warm bread with socks”) vs. a synchrony in turn are considered to reXect the transient contextually appropriate word, (for example, “He formation of local and spatially distributed functional spread the warm bread with butter”), a signiWcantly networks (for review, see Varela et al., 2001). larger negativity in the ERP waveform occurs from 250 Oscillatory dynamics cannot be studied with standard to 600 ms after the onset of the inappropriate word, with ERP methodology. The reason for this is that although a peak amplitude around 400 ms. Since Kutas and Hill- event-related changes in amplitude of a given oscillation yard (1980), a large body of research has utilized the may be time-locked to an experimental event, the phase N400 to investigate semantic processing. As a result of of the activity at a given point in time will diVer from this research the N400 has been shown to occur to each trial to trial. As a result, oscillations largely cancel out and every word, but the amplitude and latency can be during the averaging process that is used in ERP analy- aVected by a number of factors including word class, sis (see Bastiaansen & Hagoort, 2003; Tallon-Baudry, semantic relatedness to the context, cloze probability,1 Bertrand, Delpuech, & Pernier, 1996; for a more detailed word frequency, presentation modality, and verbal account of the signal-analytic distinction between ERPs working memory load (Brown & Hagoort, 1999; Kutas and event-related oscillatory activity). For a proper & Van Petten, 1994, for reviews; see Chao, Nielsen-Bohl- analysis of oscillatory dynamics, diVerent analytic tools man, & Knight, 1995; Neville, Kutas, Chesney, & have to be used such as wavelet-based time-frequency Schmidt, 1986; Smith, Stapleton, & Halgren, 1986; Stuss, analysis (for quantifying amplitude changes) or event- Picton, & Cerri, 1986; for examples of N400 eVects related coherence analysis (for quantifying changes in related to working memory). In general, a larger ampli- phase coherence between electrodes; for a further review, tude N400 will be found to a word that is not related to see Bastiaansen & Hagoort, 2003). Recent studies the context, has a low cloze probability or a low fre- employing such techniques have clearly demonstrated quency. The N400 eVect has been interpreted as reXect- that synchronous oscillations have functional signiW- ing some aspect(s) of the processes that integrate the cance during the execution of tasks engaging a variety of meaning of a particular word into a higher-order seman- cognitive operations, such as memory encoding and tic representation (Brown & Hagoort, 1999; Osterhout & retrieval (e.g., Burgess & Ali, 2002; Fell et al., 2001; Holcomb, 1992; Rugg, 1990; however see Kutas & Van Klimesch, 1999), working memory (e.g., Jensen & Petten, 1994; for an alternative suggestion). Tesche, 2002; Kahana, Sekuler, Caplan, Kirschen, & Madsen, 1999; Tesche & Karhu, 2000), face perception (Rodriguez et al., 1999), object detection (Tallon-Baudry & Bertrand, 1999), and attentional processes (e.g., Basti- 1 The cloze probability test requires subjects to complete sentence fragments, and the “cloze probability” of a word refers to the propor- aansen & Brunia, 2001; Foxe, Simpson, & Ahlfors, 1998; tion of subjects who completed a particular sentence fragment with Fries, Reynolds, Rorie, & Desimone, 2001;