Language provides one of the most powerful cues into the brain's storehouse of world knowledge. Language contexts play a crucial role in shaping memory access, as, during comprehension, semantic and structural constraints make some words more likely to occur than others. These context effects have a profound influence on the speed and accuracy of word processing, as manifested not only behaviorally but also neurophysiologically. In particular, the N400 is an event-related potential (ERP) component, arising in part from medial temporal cortical regions, that has been closely linked to language comprehension. It is elicited in response to meaningful stimuli of all kinds and its amplitude is reduced by congruent contextual information.

We have used N400 responses to examine how context information guides the search through long-term memory, both in monolingual speakers (e.g., Federmeier and Kutas, 2001; Federmeier, McLennan et al., 2002; Coulson and Federmeier, in press) and in bilinguals (Moreno, Federmeier, and Kutas, 2002). We have found that the organization of information in long-term memory has a profound impact on language processing and, moreover, that, when it can, the brain uses context information to predict (i.e., to anticipate and prepare for) the semantic and even perceptual features of upcoming items. This stands in contrast to most mainstream language comprehension models, which assume that word access and integration are largely bottom-up processes that initially receive very little guidance from context information. Our data instead suggest that top-down context information impacts even early stages of processing and is an integral part of normal, efficient language comprehension – a part that is, however, susceptible to age- and disease-related deterioration (outlined in aging section). Currently, the Federmeier lab is examining different types of semantic relationships (e.g., associates, opposites) and various levels and types of contextual constraint to uncover more detail about the type of feature information used during comprehension and the mechanisms underlying predictive processing.

In addition to investigating the processes that are used to build meaning representations during language comprehension, we have explored the nature of the semantic representations that are formed on-line and ultimately stored in long-term memory. During comprehension, information derived from different modalities must rapidly come together to yield a coherent conceptual level understanding. There has long been an assumption that such concepts are “amodal” and relatively static in nature. In a series of studies, we have examined how semantic memory is structured – and used on-line – as a function of modality, stimulus type, and mood. The results of these studies suggest that meaning arises from synchronized activity in distributed brain networks that do modality-specific processing, and that factors like representation type, experience, and mood can all change the nature of the semantic information that is accessed in response to the same stimulus.

Visual words, auditory words, and pictures:
Words and pictures are both physical objects that, through experience, have come to be associated with information not explicitly contained in the physical form of the word or picture itself. In this sense, both pictures and words can be thought of as symbols – objects that "stand for" information that they do not directly represent. A long-standing question in psychology asks how that subset of knowledge that can be conveyed by either a word or a picture is stored and processed. To examine this, we used ERPs to compare picture processing (under several conditions; Federmeier and Kutas, 2001) with word processing (Federmeier, McLennan, et al., 2002). We found that N400 responses to pictures were functionally different from those to words in either modality. We also found strong correlations between perceptual- and conceptual-level analyses of a stimulus. Changing the perceptual predictability of a picture altered the ease with which it could be semantically integrated into a sentence (as indexed by the N400). At the same time, perceptual processing (as indexed by early, sensory- and attention-related ERP components) was influenced by top-down information from the sentence context. As a whole, these results show that semantic processing takes place in a structured, distributed system in which modality information is not completely lost.