Localization of the syntactic mismatch negativity in the temporal cortex: An MEG study

Abstract

Recent auditory oddball studies using syntactic stimuli report a syntactic effect on the mismatch negativity (sMMN) around 100–200 ms. For morphosyntactic violations, this sMMN effect has been localized in the left superior temporal cortex. Independently, a recent visual sentence processing study introduced a “sensory hypothesis” which postulates that sensory cortices are sensitive to syntactic violations when these are overtly marked by closed-class morphemes, and thus contribute to early syntax-related effects in EEG and MEG. The present study aimed to test the sensory hypothesis in the auditory modality by localizing the neuronal sources of the sMMN to phrase structure violations. Using whole head magnetoencephalography, two-word utterances which were syntactically correct, uncommon or incorrect due to a word category violation were presented in an auditory oddball paradigm. In the sMMN time window (100–180 ms), incorrect phrases elicited strongest activation in the left Sylvian fissure (including the primary auditory cortex) and in the left superior temporal sulcus. Prior to this, a very early grammaticality effect (40–80 ms), focused in the left Sylvian fissure, was found. An additional grammaticality effect in a later time window (280–380 ms) was localized in the anterior part of the superior temporal gyrus, the planum polare. Processing of an uncommon phrase did not differ from processing a correct phrase in any of the time windows, indicating the genuinely syntactic nature of the sMMN effect. Our results are in line with previous studies localizing the sMMN to morphosyntactic violations and are furthermore compatible with the sensory hypothesis of closed-class morphology based syntactic processes.

Introduction

A number of event-related potential (ERP) studies suggest that syntactic processing commences shortly after 100 ms. For instance, a word that violates the word category expected in the context of a given sentence gives rise to an ERP component termed early left anterior negativity (ELAN) which usually peaks between 100 and 200 ms after the onset of the critical word (Friederici et al., 1993, Neville et al., 1991). The ELAN is assumed to reflect the disruption of phrase structure building, which is regarded as the earliest and most automatic stage of syntactic analysis. Its presumed automaticity is deduced from the findings that the ELAN is independent of task demands (Hahne and Friederici, 2002) and the expectancy of the phrase structure violation (Hahne and Friederici, 1999).

Localizations of the ELAN in response to phrase structure violations revealed activation in the anterior part of the superior temporal gyrus (STG) and in the inferior frontal gyrus (IFG) with a left-hemispheric predominance (Friederici et al., 1999Friederici et al., 2000a, Friederici and Kotz, 2003). In addition, Kubota et al. (2003) found activity in the Sylvian fissure and the superior temporal sulcus to be correlated with local syntactic violations. These findings are consistent with the prevalent notion that processing of local phrase structure is mainly realized in a left temporo-frontal network (Friederici, 2002).

In a magnetoencephalography (MEG) study using visually-presented sentence material, Dikker et al. (2009) recently showed modulations of the visual N100m component in response to unexpected word categories, when these were overtly marked by closed-class morphology. The neuronal sources of this effect were localized in the visual cortex, suggesting that syntactically relevant cues can influence processing at the sensory level. Therefore, the authors referred to this contention as the “sensory hypothesis”. It remains to be tested, however, whether and to what extent this sensory hypothesis holds true for the auditory modality as well.

A sensory ERP component in the auditory modality that has recently been associated with syntactic processing is the mismatch negativity (MMN). The MMN was originally reported in response to infrequent deviant stimuli that are presented among frequent standard stimuli in auditory oddball sequences (Näätänen et al., 1978) and interpreted as an indicator of auditory sensory memory (Näätänen et al., 1989). Over the years, it was shown to reflect not only standard sound representations but also the extraction of abstract rules and the existence of long term memory traces for speech sounds (for a review, see Näätänen et al., 2007), which, together with its attentional independence, makes it an ideal tool for the investigation of automatic language-related processes (Shtyrov and Pulvermüller, 2007). Among others, aspects of speech processing that have been investigated using the MMN comprise the discrimination of vowels (e.g. Cheour et al., 1998) and syllables (e.g. Shtyrov et al., 2000), as well as lexical aspects (e.g. Shtyrov et al., 2005). Furthermore, the MMN paradigm has been developed such that it allows for the investigation of syntactic processing (Pulvermüller and Shtyrov, 2003, Shtyrov et al., 2003). In such a paradigm, usually correct and syntactically incorrect stimuli are contrasted as standard and deviant stimuli in oddball sequences with deviants containing syntactic violations eliciting a larger MMN than syntactically correct deviants. This so-called syntactic mismatch negativity (sMMN) effect, together with its magnetic counterpart labeled sMMNm, has been reported consistently for both morphosyntactic violations realized by incorrect inflection (e.g. “we comes”) (Pulvermüller and Shtyrov, 2003, Shtyrov et al., 2003, Pulvermüller and Assadollahi, 2007, Pulvermüller et al., 2008) and phrase structure violations realized by an incorrect word category (e.g. “a folds”) (Hasting et al., 2007). The sMMN(m) has been interpreted as an indicator of the very early and automatic application of the underlying syntactic rules (Shtyrov et al., 2003, Pulvermüller and Shtyrov, 2006, Pulvermüller and Assadollahi, 2007, Hasting et al., 2007). Although the sMMN paradigm is somewhat limited with respect to its generalizability to natural language processing due to the repetitive presentation of a small number of short utterances, it provides a unique means to study the application of syntactic rules independently of attentional biases or task specific strategies, and even more importantly, to maximally control for confounds of acoustic and linguistic stimulus features (Pulvermüller and Shtyrov, 2006, Shtyrov and Pulvermüller, 2007). The latter aspect is indispensable for the identification of syntactic influences on sensory processing in the auditory modality, which is in the focus of the current investigation.

Using MEG, Shtyrov et al. (2003) and also Pulvermüller and Assadollahi (2007) localized the main generator of the sMMNm to morphosyntactic violations in the left superior temporal cortex. Pulvermüller and Shtyrov (2003) additionally report a frontal source for the sMMN to subject–verb disagreement. These localizations appear quite consistent with the generators reported for the ELAN as elicited by word category violations (Friederici et al., 2000a), which led the authors to suggest that the ELAN and the sMMN share the same sources (Pulvermüller and Assadollahi, 2007, Pulvermüller and Shtyrov, 2006). To substantiate this suggestion, it is mandatory to identify the generators of the sMMN to word category violations which, however, has not yet been done. Hence, the present study aimed to localize the sources underlying the sMMN effect to word category violations, and to compare these to the sources involved in the generation of the ELAN (Friederici et al., 2000a).

Another issue regarding early syntactic ERP effects that awaits further investigation is whether they reflect the genuine application of syntactic rules or rather the violation of serial or structural predictions (e.g. Lau et al., 2006). To distinguish the morphosyntactic MMN effect from serial order processing, the study by Pulvermüller and Assadollahi (2007) included stimuli which were morphosyntactically uncommon but not outright incorrect with regard to the respective grammar. They found no difference between the sMMNm to correct and uncommon deviants, but these two stimulus types differed from outright violations which elicited a stronger sMMNm. This is a strong indication of the genuine grammatical origin of the morphosyntactic MMN effect. Hence, a second goal of the present study was to see whether this finding can be generalized to the phrase structure MMN effect.

To this end, we contrasted two-word utterances that were either correct, uncommon or included a word category violation in acoustically well-controlled oddball sequences designed to elicit the sMMNm. By using a magnetoencephalographic approach, we intended to localize the sources underlying the sMMN to phrase structure violations and test its grammatical nature. The following predictions were made:

• We expected sequences in which the incorrect phrases served as deviants to elicit stronger sMMNm responses as compared to sequences with correct deviants (i.e. the sMMNm effect).

• This difference in neuronal activation should be most pronounced in the left superior temporal cortex, with possible additional activations in the left inferior frontal cortex. In accordance with the sensory hypothesis (Dikker et al., 2009), activations should include the auditory cortex.

• No difference should be found between correct and grammatically correct, but uncommon stimuli. The incorrect stimuli, however, should elicit a greater sMMNm than the uncommon stimuli, indicating the genuinely grammatical nature of the sMMNm effect to phrase structure violations.