Towards a neural basis of auditory sentence processing
Trends in Cognitive Sciences
Listening to connected speech is a task that humans perform effortlessly each day. This is surprising given the short time that the processing system has to deal with different types of information. Segmental phonemes and suprasegmental phonological information (prosody or pitch) as well as syntactic and semantic information must be accessed and coordinated within milliseconds. With respect to syntactic and semantic processes, two alternative views have been proposed in psycholinguistic
... nsion models. One view, which is characterized by serial or syntax-first models, holds that syntax is processed autonomously prior to semantic information [1, 2] . A second view, represented by interactive or constraint-satisfaction models, claims that all types of information interact at each stage of language comprehension [3, 4] . Both classes of models are supported by a number of sentencereading studies that use different behavioral paradigms (for details see 1). None of these models addresses explicitly the role of prosodic information that is available whenever spoken sentences are processed. Unfortunately, the few behavioral studies that have investigated possible interactions between prosodic and syntactic information during auditory language comprehension do not provide a unitary view: although some data indicate an interaction between prosodic and syntactic information [5,6], others do not  . These differences are attributable partly to the fact that different behavioral paradigms tap into different processing aspects (automatic versus Functional dissociations within the neural basis of auditory sentence processing are difficult to specify because phonological, syntactic and semantic information are all involved when sentences are perceived. In this review I argue that sentence processing is supported by a temporo-frontal network. Within this network, temporal regions subserve aspects of identification and frontal regions the building of syntactic and semantic relations. Temporal analyses of brain activation within this network support syntax-first models because they reveal that building of syntactic structure precedes semantic processes and that these interact only during a later stage. Two main classes of models have been proposed to account for the behavioral data on language comprehension: serial, syntax-first and interactive, constraint-satisfaction models [a]. As these models are based on data from reading, they comprise semantic and syntactic processes but ignore prosodic processes. Serial, syntax-first models assume that the parser initially constructs the simplest syntactic structure on the basis of word-category information, independent of lexical-semantic information. The latter information is processed during a second stage that is responsible for thematic-role assignment. If the initial syntactic structure and the thematic structure cannot be mapped onto one another, reanalysis takes place [b-d]. Recent studies, however, indicate that, for ambiguous structures, the initial structure building is not totally independent of nonstructural variables such as the frequency of a particular structure or the semantic plausibility associated with the main verb [e,f]. This has led to constraint-satisfaction models in which it is assumed that, in the case of structural ambiguities, multiple syntactic interpretations are generated and weighted according to nonstructural factors. An influential interactive model that describes processes of auditory comprehension was formulated in 1980 [g]. In this model, syntactic and semantic processes interact from an early stage during auditory language comprehension. Experiments that focus on prosodic aspects indicate that this is also true for syntactic and prosodic processes [h]. Although in both classes of models syntactic and semantic information are integrated during language perception to achieve understanding, interaction takes place at different points during processing: interactive, constraint-based models predict early interaction, whereas serial, syntax-first models predict interaction during a later stage of processing.