Sentence Comprehension

During language comprehension, the larger neural response to unexpected versus expected inputs is often taken as evidence for predictive coding—a specific computational architecture and optimization algorithm proposed to approximate probabilistic inference in the brain. However, other predictive processing frameworks can also account for this effect, leaving the unique claims of predictive coding untested. In this study, we used MEG to examine both univariate and multivariate neural activity in response to expected and unexpected inputs during word-by-word reading comprehension. We further simulated this activity using an implemented predictive coding model that infers the meaning of words from their orthographic form. Consistent with previous findings, the univariate analysis showed that, between 300 and 500 ms, unexpected words produced a larger evoked response than expected words within a left ventromedial temporal region that supports the mapping of orthographic word-forms onto lexical and conceptual representations. Our model explained this larger evoked response as the enhanced lexico-semantic prediction error produced when prior top-down predictions failed to suppress activity within lexical and semantic “error units”. Critically, our simulations showed that despite producing minimal prediction error, expected inputs nonetheless reinstated top-down predictions within the model's lexical and semantic “state” units. Two types of multivariate analyses provided evidence for this functional distinction between state and error units within the ventromedial temporal region. First, within each trial, the same individual voxels that produced a larger response to unexpected inputs between 300 and 500 ms produced unique temporal patterns to expected inputs that resembled the patterns produced within a pre-activation time window. Second, across trials, and again within the same 300–500 ms time window and left ventromedial temporal region, pairs of expected words produced spatial patterns that were more similar to one another than the spatial patterns produced by pairs of expected and unexpected words, regardless of specific item. Together, these findings provide compelling evidence that the left ventromedial temporal lobe employs predictive coding to infer the meaning of incoming words from their orthographic form during reading comprehension.

Extracting meaning from speech requires the use of pragmatic, semantic, and syntactic information. A central question is: Does the processing of these different types of linguistic information have common or distinct neuroanatomical substrates? We addressed this issue using functional magnetic resonance imaging (fMRI) to measure neural activity when subjects listened to spoken normal sentences contrasted with sentences that had either (A) pragmatical, (B) semantic (selection restriction), or (C) syntactic (subcategorical) violations sentences. All three contrasts revealed robust activation of the left-inferior-temporal/fusiform gyrus. Activity in this area was also observed in a combined analysis of all three experiments, suggesting that it was modulated by all three types of linguistic violation. Planned statistical comparisons between the three experiments revealed (1) a greater difference between conditions in activation of the left-superior-temporal gyrus for the pragmatic experiment than the semantic/syntactic experiments; (2) a greater difference between conditions in activation of the right-superior and middle-temporal gyrus in the semantic experiment than in the syntactic experiment; and (3) no regions activated to a greater degree in the syntactic experiment than in the semantic experiment. These data show that, while left- and right-superior-temporal regions may be differentially involved in processing pragmatic and lexico-semantic information within sentences, the left-inferior-temporal/fusiform gyrus is involved in processing all three types of linguistic information. We suggest that this region may play a key role in using pragmatic, semantic (selection restriction), and subcategorical information to construct a higher representation of meaning of sentences.

Deficits in language comprehension in schizophrenia were examined using event-related potentials (ERPs). Schizophrenic and healthy participants read sentences in which the first clause ended with a homograph, and the second clause started with a target word that was semantically related to the homograph’s dominant meaning (e.g., 1. Diving was forbidden from the bridge because the river had rocks in it. or 2. The guests played bridge because the river had rocks in it.). Processing of the targets (e.g., “river”) was expected to be primarily influenced by the preceding overall sentence context (congruent in 1; incongruent in 2) in healthy participants, but to be inappropriately affected by the dominant meaning of homographs (e.g., the “structure” meaning of “bridge”) in sentences like 2 in schizophrenic patients. The N400 ERP component that is known to be sensitive to contextual effects during language processing confirmed these predictions. This showed that language abnormalities in schizophrenia may be related to deficient processing of context-irrelevant semantic representations of words from the discourse.

The aim of this study was to determine whether or not the brain distinguishes between two types of conceptual relationships between noun-phrases (NPs) and verbs during online processing of simple, unambiguous English sentences. A total of 15 participants read and made plausibility judgments on sentences that were presented word-by-word. Event-related potentials elicited by critical verbs were measured. In all cases, the critical verb assigned a thematic role of ’agent’ to its subject NP. In non-violated sentences (e.g. "For breakfast the boys would only eat em leader"), the preceding NP was animate ("boys") and was a likely agent for a given verb ("eat") given its preceding context ("For breakfast"). In both types of conceptually violated sentences, the NPs were unlikely agents for the verbs given their preceding contexts. In ’thematic role animacy violations’ (e.g. "For breakfast the eggs would only eat em leader"), the NP was inanimate ("eggs") and was therefore more likely to occupy the role of ’theme’ than ’agent’, i.e. eggs, being inanimate, cannot eat but they can be eaten. In ’non-thematic role pragmatic violations’ (e.g. "For breakfast the boys would only bury em leader"), the thematic role of agent assigned by the verb ("bury") to its preceding NP ("boys") is inherently acceptable (boys can bury), but the sentence is still pragmatically incongruous given the preceding context ("At breakfast"). As expected, the non-thematic role pragmatic violations elicited a significant N400 effect. The thematic role animacy violations elicited a smaller N400 effect that only approached significance across all participants. The thematic role animacy violations, however, elicited a significant P600 effect-an ERP component that is most commonly associated with processing syntactic information during language comprehension. We discuss the possibility that the P600 was elicited by the thematic role animacy violations (but not by the non-thematic role pragmatic violations) because, in the former but not the latter, there was an online attempt to structurally repair and make sense of the sentences by reassigning the thematic role of the NP that preceded the critical verb from ’agent’ to ’theme’. Our findings suggest a qualitative neural distinction in processing these two types of conceptual anomalies within simple, unambiguous sentences.

Event-related potentials to critical verbs were measured as patients with schizophrenia and healthy controls read sentences word by word. Relative to their preceding context, critical verbs were (a) congruous, (b) incongruous and semantically unrelated to individual preceding words (pragmatic-semantic violations), (c) incongruous but semantically related to individual preceding words (animacy-semantic violations), or (d) syntactically anomalous. The N400 was modulated normally in patients, suggesting that semantic integration between individual words within sentences was normal in schizophrenia. The amplitude of the P600 to both syntactic and animacy-semantic violations was reduced in patients relative to controls. The authors suggest that, in schizophrenia, an abnormality in combining semantic and syntactic information online to build up propositional meaning leaves sentence processing to be primarily driven by semantic relationships between individual words.

Recent event-related potential studies report a P600 effect to incongruous verbs preceded by semantically associated inanimate noun-phrase (NP) arguments, e.g., "eat" in "At breakfast the eggs would eat...". This P600 effect may reflect the processing cost incurred when semantic-thematic relationships between critical verbs and their preceding NP argument(s) bias towards different interpretations to those dictated by their sentences’ syntactic structures. We have termed such violations of alternative thematic roles, ’thematic role violations.’ Semantic-thematic relationships are influenced both by semantic associations and by more basic semantic features, such as a noun’s animacy. This study determined whether a P600 effect can be evoked by verbs whose thematic structures are violated by their preceding inanimate NP arguments, even in the absence of close semantic-associative relationships with these arguments or their preceding contexts. ERPs were measured to verbs under four conditions: (1) non-violated ("At breakfast the boys would eat..."); (2) preceded by introductory clauses and animate NPs that violated their pragmatic expectations but not their thematic structures ("At breakfast the boys would plant..."); (3) preceded by semantically related contexts but inanimate NPs that violated their thematic structures ("At breakfast the eggs would eat..."); (4) preceded by semantically unrelated contexts and inanimate NPs that also violated their thematic structures ("At breakfast the eggs would plant..."). Pragmatically non-thematic role violated verbs preceded by unrelated contexts and animate NPs evoked robust N400 effects and small P600 effects. Thematically violated verbs preceded by inanimate argument NPs evoked robust P600 effects but no N400 effects, regardless of whether these inanimate arguments or their preceding contexts were semantically related or unrelated to these verbs. These findings suggest that semantic-thematic relations, related to animacy constraints on verbs’ arguments, are computed online and can immediately impact verb processing within active, English sentences.

Traditionally, event-related brain potential (ERP) studies of language processing have presented words at a fixed rate using rapid serial visual presentation. Recent studies suggest, however, that the processes engaged during sentence comprehension are contingent on word presentation rate. These findings underscore the importance of allowing participants to read at a natural pace. The present study employed simultaneous self-paced reading and ERP methodologies to examine behavioral and neural responses while participants read sentences containing pragmatic or morphosyntactic violations or no violations. ERP and self-paced reading results replicated previous findings. This novel combination of behavioral and ERP methodologies combines the high temporal resolution and direct neural measures offered by ERPs with the more natural reading environment and information about processing load provided by self-paced reading.

To make sense of a sentence, we must compute morphosyntactic and semantic-thematic relationships between its verbs and arguments and evaluate the resulting propositional meaning against any preceding context and our real-world knowledge. Recent electrophysiological studies suggest that, in comparison with non-violated verbs (e.g. "...at breakfast the boys would eat..."), animacy semantic-thematically violated verbs (e.g. "...at breakfast the eggs would eat...") and morphosyntactically violated verbs (e.g. "...at breakfast the boys would eats...") evoke a similar neural response. This response is distinct from that evoked by verbs that only violate real-world knowledge (e.g. "...at breakfast the boys would plant..."). Here we used fMRI to examine the neuroanatomical regions engaged in response to these three violations. Real-world violations, relative to other sentence types, led to increased activity within the left anterior inferior frontal cortex, reflecting participants’ increased and prolonged efforts to retrieve semantic knowledge about the likelihood of events occurring in the real world. In contrast, animacy semantic-thematic violations of the actions depicted by the central verbs engaged a frontal/inferior parietal/basal ganglia network known to mediate the execution and comprehension of goal-directed action. We suggest that the recruitment of this network reflected a semantic-thematic combinatorial process that involved an attempt to determine whether the actions described by the verbs could be executed by their NP Agents. Intriguingly, this network was also activated to morphosyntactic violations between the verbs and their subject NP arguments. Our findings support the pattern of electrophysiological findings in suggesting (a) that a clear division within the semantic system plays out during sentence comprehension, and (b) that semantic-thematic and syntactic violations of verbs within simple active sentences are treated similarly by the brain.

In two event-related potential (ERP) experiments, we determined to what extent Grice’s maxim of informativeness as well as pragmatic ability contributes to the incremental build-up of sentence meaning, by examining the impact of underinformative versus informative scalar statements (e.g. "Some people have lungs/pets, and…") on the N400 event-related potential (ERP), an electrophysiological index of semantic processing. In Experiment 1, only pragmatically skilled participants (as indexed by the Autism Quotient Communication subscale) showed a larger N400 to underinformative statements. In Experiment 2, this effect disappeared when the critical words were unfocused so that the local underinformativeness went unnoticed (e.g., "Some people have lungs that…"). Our results suggest that, while pragmatic scalar meaning can incrementally contribute to sentence comprehension, this contribution is dependent on contextual factors, whether these are derived from individual pragmatic abilities or the overall experimental context.

We used event-related potentials (ERPs) to examine the time-course of processing metaphorical and literal sentences in the brain. ERPs were measured to sentence-final (Experiment 1) and mid-sentence (Experiment 2) critical words (CWs) as participants read and made plausibility judgments about familiar nominal metaphors ("A is a B") as well as literal and semantically anomalous sentences of the same form. Unlike the anomalous words, which evoked a robust N400 effect (on the CW in experiments 1 and 2 as well as on the sentence-final word in experiment 2), CWs in the metaphorical, relative to the literal, sentences only evoked an early, localized N400 effect that was over by 400ms after CW onset, suggesting that, by this time, their metaphorical meaning had been accessed. CWs in the metaphorical sentences also evoked a significantly larger LPC (Late Positive Component) than in the literal sentences. We suggest that this LPC reflected additional analysis that resolved a conflict between the implausibility of the literal sentence interpretation and the match between the metaphorical meaning of the CW, the context and stored information within semantic memory, resulting from early access to both literal and figurative meanings of the CWs.

This study examined the electrophysiological correlates of complement coercion. ERPs were measured as participants read and made acceptability judgments about plausible coerced sentences, plausible noncoerced sentences, and highly implausible animacy-violated sentences ("The journalist began/wrote/astonished the article before his coffee break"). Relative to noncoerced complement nouns, the coerced nouns evoked an N400 effect. This effect was not modulated by the number of possible activities implied by the coerced nouns (e.g., began reading the article; began writing the article) and did not differ either in magnitude or scalp distribution from the N400 effect evoked by the animacy-violated complement nouns. We suggest that the N400 modulation to both coerced and animacy-violated complement nouns reflected different types of mismatches between the semantic restrictions of the verb and the semantic properties of the incoming complement noun. This is consistent with models holding that a verb’s semantic argument structure is represented and stored at a distinct level from its syntactic argument structure. Unlike the coerced complement noun, the animacy-violated nouns also evoked a robust P600 effect, which may have been triggered by the judgments of the highly implausible (syntactically determined) meanings of the animacy-violated propositions. No additional ERP effects were seen in the coerced sentences until the sentence-final word that, relative to sentence-final words in the noncoerced sentences, evoked a sustained anteriorly distributed positivity. We suggest that this effect reflected delayed attempts to retrieve the specific event(s) implied by coerced complement nouns.

Animacy is known to play an important role in language processing and production, but debate remains as to how it exerts its effects: 1) through links to syntactic ordering, 2) through inherent differences between animate and inanimate entities in their salience/lexico-semantic accessibility, 3) through links to specific thematic roles. We contrasted these three accounts in two event related potential (ERP) experiments examining the processing of direct object arguments in simple English sentences. In Experiment 1, we found a larger N400 to animate than inanimate direct object arguments assigned the Patient role, ruling out the second account. In Experiment 2 we found no difference in the N400 evoked by animate direct object arguments assigned the Patient role (prototypically inanimate) and those assigned the Experiencer role (prototypically animate), ruling out the third account. We therefore suggest that animacy may impact processing through a direct link to syntactic linear ordering, at least on post-verbal arguments in English. We also examined processing on direct object arguments that violated the animacy-based selection restriction constraints of their preceding verbs. These violations evoked a robust P600, which was not modulated by thematic role assignment or reversibility, suggesting that the so-called semantic P600 is driven by overall propositional impossibility, rather than thematic role reanalysis.

We aimed to determine whether semantic relatedness between an incoming word and its preceding context can override expectations based on two types of stored knowledge: real-world knowledge about the specific events and states conveyed by a verb, and the verb’s broader selection restrictions on the animacy of its argument. We recorded event-related potentials on post-verbal Agent arguments as participants read and made plausibility judgments about passive English sentences. The N400 evoked by incoming animate Agent arguments that violated expectations based on real-world event/state knowledge, was strongly attenuated when they were semantically related to the context. In contrast, semantic relatedness did not modulate the N400 evoked by inanimate Agent arguments that violated the preceding verb’s animacy selection restrictions. These findings suggest that, under these task and experimental conditions, semantic relatedness can facilitate processing of post-verbal animate arguments that violate specific expectations based on real-world event/state knowledge, but only when the semantic features of these arguments match the coarser-grained animacy restrictions of the verb. Animacy selection restriction violations also evoked a P600 effect, which was not modulated by semantic relatedness, suggesting that it was triggered by propositional impossibility. Together, these data indicate that the brain distinguishes between real-world event/state knowledge and animacy-based selection restrictions during online processing.

The verb “pounce” describes a single, near-instantaneous event. Yet, we easily understand that, “For several minutes the cat pounced…” describes a situation in which multiple pounces occurred, although this interpretation is not overtly specified by the sentence s syntactic structure or by any of its individual words—a phenomenon known as “aspectual coercion.” Previous psycholinguistic studies have reported processing costs in association with aspectual coercion, but the neurocognitive mechanisms giving rise to these costs remain contentious. Additionally, there is some controversy about whether readers commit to a full interpretation of the event when the aspectual information becomes available, or whether they leave it temporarily underspecified until later in the sentence. Using ERPs, we addressed these questions in a design that fully crossed context type (punctive, durative, frequentative) with verb type (punctive, durative). We found a late, sustained negativity to punctive verbs in durative contexts, but not in frequentative (e.g., explicitly iterative) contexts. This effect was distinct from the N400 in both its time course and scalp distribution, suggesting that it reflected a different underlying neurocognitive mechanism. We also found that ERPs to durative verbs were unaffected by context type. Together, our results provide strong evidence that neural activity associated with aspectual coercion is driven by the engagement of a morphosyntactically unrealized semantic operator rather than by violations of real-world knowledge, more general shifts in event representation, or event iterativity itself. More generally, our results add to a growing body of evidence that a set of late-onset sustained negativities reflect elaborative semantic processing that goes beyond simply combining the meaning of individual words with syntactic structure to arrive at a final representation of meaning.

We used event-related potentials (ERPs) to investigate the neurocognitive mechanisms associated with processing light verb constructions such as “give a kiss”. These constructions consist of a semantically underspecified light verb (“give”) and an event nominal that contributes most of the meaning and also activates an argument structure of its own (“kiss”). This creates a mismatch between the syntactic constituents and the semantic roles of a sentence. Native speakers read German verb-final sentences that contained light verb constructions (e.g., “Julius gave Anne a kiss”), non-light constructions (e.g., “Julius gave Anne a rose”), and semantically anomalous constructions (e.g., *“Julius gave Anne a conversation”). ERPs were measured at the critical verb, which appeared after all its arguments. Compared to non-light constructions, the light verb constructions evoked a widely distributed, frontally focused, sustained negative-going effect between 500 and 900 ms after verb onset. We interpret this effect as reflecting working memory costs associated with complex semantic processes that establish a shared argument structure in the light verb constructions.

Since the early 2000s, several ERP studies have challenged the assumption that we always use syntactic contextual information to influence semantic processing of incoming words, as reflected by the N400 component. One approach for explaining these findings is to posit distinct semantic and syntactic processing mechanisms, each with distinct time courses. While this approach can explain specific datasets, it cannot account for the wider body of findings. I propose an alternative explanation: a dynamic generative framework in which our goal is to infer the underlying event that best explains the set of inputs encountered at any given time. Within this framework, combinations of semantic and syntactic cues with varying reliabilities are used as evidence to weight probabilistic hypotheses about this event. I further argue that the computational principles of this framework can be extended to understand how we infer situation models during discourse comprehension, and intended messages during spoken communication.

Background: Schizophrenia is characterized by abnormalities in referential communication, which may be linked to more general deficits in proactive cognitive control. We used event-related potentials (ERPs) to probe the timing and nature of the neural mechanisms engaged as people with schizophrenia linked pronouns to their preceding referents during word-by-word sentence comprehension.Methods: We measured ERPs to pronouns in two-clause sentences from 16 people with schizophrenia and 20 demographically-matched control participants. Our design crossed the number of potential referents (1-referent, 2-referent) with whether the pronoun matched the gender of its preceding referent(s) (matching, mismatching). This gave rise to four conditions: (1) 1-referent matching (“…Edward took courses in accounting but he…”), (2) 2-referent matching (“…Edward and Phillip took courses but he…”), (3) 1-referent mismatching (“…Edward took courses in accounting but she…”), and (4) 2-referent mismatching (“…Edward and Phillip took courses but she…”).Results: Consistent with previous findings, healthy controls produced a larger left anteriorly-distributed negativity between 400-600ms to 2-referent matching than to 1-referent matching pronouns (the “Nref effect”). In contrast, people with schizophrenia produced a larger centro-posterior positivity effect between 600-800ms. Both patient and control groups produced a larger positivity between 400-800ms to mismatching than to matching pronouns.Conclusions: These findings suggest that proactive mechanisms of referential processing, reflected by the Nref effect, are impaired in schizophrenia, while reactive mechanisms, reflected by the positivity effects, are relatively spared. Indeed, patients may compensate for proactive deficits by retro-actively engaging with context to influence the processing of inputs at a later stage of analysis.

BackgroundPeople with schizophrenia process language in unusual ways, but the causes of these abnormalities are unclear. In particular, it has proven difficult to empirically disentangle explanations based on impairments in the top-down processing of higher-level information from those based on the bottom-up processing of lower-level information.MethodsTo distinguish these accounts, we used visual world eye-tracking, a paradigm that measures spoken language processing during real-world interactions. Participants listened to and then acted out syntactically ambiguous spoken instructions (e.g., “tickle the frog with the feather”, which could either specify how to tickle a frog, or which frog to tickle). We contrasted how 24 people with schizophrenia and 24 demographically-matched controls used two types of lower-level information (prosody and lexical representations) and two types of higher-level information (pragmatic and discourse-level representations) to resolve the ambiguous meanings of these instructions. Eye-tracking allowed us to assess how participants arrived at their interpretation in real time, while recordings of participants’ actions measured how they ultimately interpreted the instructions.ResultsWe found a striking dissociation in participants’ eye movements: the two groups were similarly adept at using lower-level information to immediately constrain their interpretations of the instructions, but only controls showed evidence of fast top-down use of higher-level information. People with schizophrenia, nonetheless, did eventually reach the same interpretations as controls.ConclusionsThese data suggest that language abnormalities in schizophrenia partially result from a failure to use higher-level information in a top-down fashion, to constrain the interpretation of language as it unfolds in real time.

We used Magnetoencephalography (MEG) in combination with Representational Similarity Analysis to probe neural activity associated with distinct, item-specific lexico-semantic predictions during language comprehension. MEG activity was measured as participants read highly constraining sentences in which the final words could be predicted. Before the onset of the predicted words, both the spatial and temporal patterns of brain activity were more similar when the same words were predicted than when different words were predicted. The temporal patterns localized to the left inferior and medial temporal lobe. These findings provide evidence that unique spatial and temporal patterns of neural activity are associated with item-specific lexico-semantic predictions. We suggest that the unique spatial patterns reflected the prediction of spatially distributed semantic features associated with the predicted word, and that the left inferior/medial temporal lobe played a role in temporally “binding” these features, giving rise to unique lexico-semantic predictions.

Presuppositions convey information that comprehenders assume to be true, even when it is tangential to the communicator’s main message. For example, a class of verbs called ‘factives’ (e.g. realize, know) trigger the presupposition that the events or states conveyed by their sentential complements are true. In contrast, non-factive verbs (e.g. think, believe) do not trigger this presupposition. We asked whether, during language comprehension, presuppositions triggered by factive verbs are encoded within the comprehender’s discourse model, with neural consequences if violated by later bottom-up inputs. Using event-related potentials (ERPs), we examined neural activity to words that were either consistent or inconsistent with events/states conveyed by the complements of factive versus non-factive verbs while comprehenders read and actively monitored the coherence of short discourse scenarios. We focused on the modulation of a posteriorly-distributed late positivity or P600. This ERP component is produced when comprehenders constrain their discourse model such that it restricts predictions only to event structures that are compatible with this model, and new input violates these event structure predictions. Between 500-700ms, we observed a larger amplitude late posterior positivity/P600 on words that were inconsistent (versus consistent) with the events/states conveyed by the complements of factive verbs. No such effect was observed following non-factive verbs. These findings suggest that, during active discourse comprehension, the presuppositions triggered by factive verbs are encoded and maintained within the comprehender’s discourse model. Downstream input that is inconsistent with these presuppositions violates event structure predictions and conflicts with this prior model, producing the late posterior positivity/P600.

It has been proposed that hierarchical prediction is a fundamental computational principle underlying neurocognitive processing. Here we ask whether the brain engages distinct neurocognitive mechanisms in response to inputs that fulfill versus violate strong predictions at different levels of representation during language comprehension. Participants read three-sentence scenarios in which the third sentence constrained for a broad event structure, e.g. Agent caution animate-Patient. High constraint contexts additionally constrained for a specific event/lexical item, e.g. a two-sentence context about a beach, lifeguards and sharks constrained for the event, Lifeguards cautioned Swimmers and the specific lexical item, “swimmers”. Low constraint contexts did not constrain for any specific event/lexical item. We measured ERPs on critical nouns that fulfilled and/or violated each of these constraints. We found clear, dissociable effects to fulfilled semantic predictions (a reduced N400), to event/lexical prediction violations (an increased late frontal positivity), and to event structure/animacy prediction violations (an increased late posterior positivity/P600). We argue that the late frontal positivity reflects a large change in activity associated with successfully updating the comprehender’s current situation model with new unpredicted information. We suggest that the late posterior positivity/P600 is triggered when the comprehender detects a conflict between the input and her model of the communicator and communicative environment. This leads to an initial failure to incorporate the unpredicted input into the situation model, which may be followed by second-pass attempts to make sense of the discourse through reanalysis, repair, or reinterpretation. Together, these findings provide strong evidence that confirmed and violated predictions at different levels of representation manifest as distinct spatiotemporal neural signatures.

It has been proposed that people can generate probabilistic predictions at multiple levels of representation during language comprehension. We used magnetoencephalography (MEG) and electroencephalography (EEG), in combination with representational similarity analysis, to seek neural evidence for the prediction of animacy features. In two studies, MEG and EEG activity was measured as human participants (both sexes) read three-sentence scenarios. Verbs in the final sentences constrained for either animate or inanimate semantic features of upcoming nouns, and the broader discourse context constrained for either a specific noun or for multiple nouns belonging to the same animacy category. We quantified the similarity between spatial patterns of brain activity following the verbs until just before the presentation of the nouns. The MEG and EEG datasets revealed converging evidence that the similarity between spatial patterns of neural activity following animate-constraining verbs was greater than following inanimate-constraining verbs. This effect could not be explained by lexical-semantic processing of the verbs themselves. We therefore suggest that it reflected the inherent difference in the semantic similarity structure of the predicted animate and inanimate nouns. Moreover, the effect was present regardless of whether a specific word could be predicted, providing strong evidence for the prediction of coarse-grained semantic features that goes beyond the prediction of individual words.