The lexical semantics of adjective–noun phrases in the human brain

Abstract

As a person reads, the brain performs complex operations to create higher order semantic representations from individual words. While these steps are effortless for competent readers, we are only beginning to understand how the brain performs these actions. Here, we explore lexical semantics using magnetoencephalography (MEG) recordings of people reading adjective–noun phrases presented one word at a time. We track the neural representation of single word representations over time, through different brain regions. Our results reveal two novel findings: (a) a neural representation of the adjective is present during noun presentation, but this representation is different from that observed during adjective presentation and (b) the neural representation of adjective semantics observed during adjective reading is reactivated after phrase reading, with remarkable consistency. We also note that while the semantic representation of the adjective during the reading of the adjective is very distributed, the later representations are concentrated largely to temporal and frontal areas previously associated with composition. Taken together, these results paint a picture of information flow in the brain as phrases are read and understood.

Neural basis of basic composition: what we have learned from the red–boat studies and their extensions

“…For example, lexical representations are known to have highly distributed characteristics and thus to understand their time course and evolution during composition we will need techniques such as multivariate pattern analysis. Early work using this approach has characterized, for example, how the representations of adjectives activate and deactivate during composition with subsequent nouns [74]. This work uses vector representations of word semantics, based on the cooccurrence of words in documents across a dataset of millions of webpages [75].…”

Studying language in context using the temporal generalization method

“…This means that in the 2 versus 2 test, equation (2.1), the left side is very often less than than the right, causing the 2 versus 2 test to fail a significant number of times. Because of the regression framework, this systematically incorrect prediction can be traced to MEG data having opposite sign during this time period (see Fyshe et al [26] for a full explanation). That is, the MEG features on which the model depends have changed sign, causing the 2 versus 2 prediction to be systematically wrong.…”

“…In addition, there are significantly below chance regions when training during adjective presentation and testing during noun presentation (near the 'Y' annotation in (b)). From Fyshe et al[26].…”

Combining computational controls with natural text reveals new aspects of meaning composition

“…Even though such studies have been pivotal in beginning to study the processes behind meaning composition, we argue that their findings are related to the process of integrating supra-word meaning, while missing other key components, such as the storage and maintenance of the current supra-word meaning. In a different line of work, neuroscientists build computational models of meaning through Natural Language Processing (NLP) embeddings of words and sentences (Mitchell et al, 2008;Sudre et al, 2012;Wehbe et al, 2014b,a;Huth et al, 2016;Jain and Huth, 2018;Toneva and Wehbe, 2019;Fyshe et al, 2019). Thanks to these studies, we are starting to uncover some properties of meaning representation, such as the fact that neural activity associated with single word meaning is distributed (Mitchell et al, 2008;Huth et al, 2016).…”