Semantic composition of sentences word by word: MEG evidence for shared processing of conceptual and logical elements

Negation mitigates rather than inverts the neural representations of adjectives

Combinatoric linguistic operations underpin human language processes, but how meaning is composed and refined in the mind of the reader is not well understood. We address this puzzle by exploiting the ubiquitous function of negation. We track the online effects of negation ("not") and intensifiers ("really") on the representation of scalar adjectives (e.g., "good") in parametrically designed behavioral and neurophysiological (MEG) experiments. The behavioral data show that participants first interpret negated adjectives as affirmative and later modify their interpretation towards, but never exactly as, the opposite meaning. Decoding analyses of neural activity further reveal significant above chance decoding accuracy for negated adjectives within 600 ms from adjective onset, suggesting that negation does not invert the representation of adjectives (i.e., "not bad" represented as "good"); furthermore, decoding accuracy for negated adjectives is found to be significantly lower than that for affirmative adjectives. Overall, these results suggest that negation mitigates rather than inverts the neural representations of adjectives. This putative suppression mechanism of negation is supported by increased synchronization of beta-band neural activity in sensorimotor areas. The analysis of negation provides a steppingstone to understand how the human brain represents changes of meaning over time.

Modeling Brain Representations of Words' Concreteness in Context Using GPT-2 and Human Ratings

The meaning of most words in language depends on their context. Understanding how the human brain extracts contextualized meaning, and identifying where in the brain this takes place, remain important scientific challenges. But technological and computational advances in neuroscience and artificial intelligence now provide unprecedented opportunities to study the human brain in action as language is read and understood. Recent contextualized language models seem to be able to capture homonymic meaning variation ("bat", in a baseball vs. a vampire context), as well as more nuanced differences of meaning-for example, polysemous words such as "book", which can be interpreted in distinct but related senses ("explain a book", information, vs. "open a book", object) whose differences are fine-grained. We study these subtle differences in lexical meaning along the concrete/abstract dimension, as they are triggered by verb-noun semantic composition. We analyze functional magnetic resonance imaging (fMRI) activations elicited by Italian verb phrases containing nouns whose interpretation is affected by the verb to different degrees. By using a contextualized language model and human concreteness ratings, we shed light on where in the brain such fine-grained meaning variation takes place and how it is coded. Our results show that phrase concreteness judgments and the contextualized model can predict BOLD activation associated with semantic composition within the language network. Importantly, representations derived from a complex, nonlinear composition process consistently outperform simpler composition approaches. This is compatible with a holistic view of semantic composition in the brain, where semantic representations are modified by the process of composition itself. When looking at individual brain areas, we find that encoding performance is statistically significant, although with differing patterns of results, suggesting differential involvement, in the posterior superior temporal sulcus, inferior frontal gyrus and anterior temporal lobe, and in motor areas previously associated with processing of concreteness/abstractness.

Minimal Phrase Composition Revealed by Intracranial Recordings

The ability to comprehend phrases is an essential integrative property of the brain. Here, we evaluate the neural processes that enable the transition from single-word processing to a minimal compositional scheme. Previous research has reported conflicting timing effects of composition, and disagreement persists with respect to inferior frontal and posterior temporal contributions. To address these issues, 19 patients (10 male, 9 female) implanted with penetrating depth or surface subdural intracranial electrodes, heard auditory recordings of adjective-noun, pseudoword-noun, and adjective-pseudoword phrases and judged whether the phrase matched a picture. Stimulus-dependent alterations in broadband gamma activity, low-frequency power, and phase-locking values across the language-dominant left hemisphere were derived. This revealed a mosaic located on the lower bank of the posterior superior temporal sulcus (pSTS), in which closely neighboring cortical sites displayed exclusive sensitivity to either lexicality or phrase structure, but not both. Distinct timings were found for effects of phrase composition (210-300 ms) and pseudoword processing (∼300-700 ms), and these were localized to neighboring electrodes in pSTS. The pars triangularis and temporal pole encoded anticipation of composition in broadband low frequencies, and both regions exhibited greater functional connectivity with pSTS during phrase composition. Our results suggest that the pSTS is a highly specialized region composed of sparsely interwoven heterogeneous constituents that encodes both lower and higher level linguistic features. This hub in pSTS for minimal phrase processing may form the neural basis for the human-specific computational capacity for forming hierarchically organized linguistic structures.SIGNIFICANCE STATEMENT Linguists have claimed that the integration of multiple words into a phrase demands a computational procedure distinct from single-word processing. Here, we provide intracranial recordings from a large patient cohort, with high spatiotemporal resolution, to track the cortical dynamics of phrase composition. Epileptic patients volunteered to participate in a task in which they listened to phrases (red boat), word-pseudoword or pseudoword-word pairs (e.g., red fulg). At the onset of the second word in phrases, greater broadband high gamma activity was found in posterior superior temporal sulcus in electrodes that exclusively indexed phrasal meaning and not lexical meaning. These results provide direct, high-resolution signatures of minimal phrase composition in humans, a potentially species-specific computational capacity.

Functional differentiation in the language network revealed by lesion-symptom mapping

Theories of language organization in the brain commonly posit that different regions underlie distinct linguistic mechanisms. However, such theories have been criticized on the grounds that many neuroimaging studies of language processing find similar effects across regions. Moreover, condition by region interaction effects, which provide the strongest evidence of functional differentiation between regions, have rarely been offered in support of these theories. Here we address this by using lesion-symptom mapping in three large, partially-overlapping groups of aphasia patients with left hemisphere brain damage due to stroke (N = 121, N = 92, N = 218). We identified multiple measure by region interaction effects, associating damage to the posterior middle temporal gyrus with syntactic comprehension deficits, damage to posterior inferior frontal gyrus with expressive agrammatism, and damage to inferior angular gyrus with semantic category word fluency deficits. Our results are inconsistent with recent hypotheses that regions of the language network are undifferentiated with respect to high-level linguistic processing.

Conceptual Combination in the LATL With and Without Syntactic Composition

The relationship among syntactic, semantic, and conceptual processes in language comprehension is a central question to the neurobiology of language. Several studies have suggested that conceptual combination in particular can be localized to the left anterior temporal lobe (LATL), while syntactic processes are more often associated with the posterior temporal lobe or inferior frontal gyrus. However, LATL activity can also correlate with syntactic computations, particularly in narrative comprehension. Here we investigated the degree to which LATL conceptual combination is dependent on syntax, specifically asking whether rapid (∼200 ms) magnetoencephalography effects of conceptual combination in the LATL can occur in the absence of licit syntactic phrase closure and in the absence of a semantically plausible output for the composition. We find that such effects do occur: LATL effects of conceptual combination were observed even when there was no syntactic phrase closure or plausible meaning. But syntactic closure did have an additive effect such that LATL signals were the highest for expressions that composed both conceptually and syntactically. Our findings conform to an account in which LATL conceptual composition is influenced by local syntactic composition but is also able to operate without it.

Composition within and between Languages in the Bilingual Mind: MEG Evidence from Korean/English Bilinguals

The ability of the human brain to build complex expressions from simpler parts is fascinating, but the ability of the bilingual brain to do so is perhaps even more remarkable. When highly proficient bilinguals converse, they can fluidly switch from one language to another even inside sentences. Thus, they build expressions using words from more than one language. How are bilinguals able to compose words across different languages in real time? While robust evidence has implicated the left anterior temporal lobe (LATL) for the composition of words within one language, we do not know how the LATL, or other regions implicated for composition, operates when the language switches. We also do not know whether prefrontal regions associated with language control are recruited for language switching during composition. We addressed these questions with magnetoencephalography measurements in bilinguals who are fluent in two typologically distant languages, English and Korean. We observed early composition effects in the LATL at ∼200 ms that were unaffected by either language or orthography switching, which was also varied (Hangul vs Roman alphabet). Thus, the combinatory mechanism at 200 ms housed in the anterior temporal cortex appears blind to the language in which its input concepts are expressed. However, in later time windows, language and orthography switching interacted both in regions implicated for composition [LATL, ventromedial prefrontal cortex, left inferior frontal gyrus (LIFG)] as well as in regions associated with language control (ACC, LIFG). This establishes a starting point for understanding how bilingual brains code switch: words are initially combined without consideration of which language they come from, but language switching affects later processing.

How the conceptual specificity of individual words affects incremental sentence composition: MEG evidence

While much research has addressed the neural basis of lexical access and the composition of lexical items into larger meanings, little is known about how the semantic properties of individual words affect composition. Research on modifier-noun combinations has, however, shown that composition related activity in the left anterior temporal lobe (LATL) is sensitive to the conceptual specificity of the composing words. Here we tested whether this pattern extends to verb-argument combinations in minimal subject-verb-object sentences. If the LATL specificity effects extend to verb-argument integration, this would suggest a general mechanism that composes not only entity concepts, but also propositions describing events. Results showed an overall similar modulation by conceptual specificity in the verb domain, suggesting a central, category-insensitive, role for the LATL as a conceptual combiner. Additionally, we saw specificity effects in the left mid-superior temporal cortex, but the angular gyrus, often hypothesized as combinatory, showed no effects of composition.

Left posterior temporal cortex is sensitive to syntax within conceptually matched Arabic expressions

During language comprehension, the brain processes not only word meanings, but also the grammatical structure-the "syntax"-that strings words into phrases and sentences. Yet the neural basis of syntax remains contentious, partly due to the elusiveness of experimental designs that vary structure independently of meaning-related variables. Here, we exploit Arabic's grammatical properties, which enable such a design. We collected magnetoencephalography (MEG) data while participants read the same noun-adjective expressions with zero, one, or two contiguously-written definite articles (e.g., 'chair purple'; 'the-chair purple'; 'the-chair the-purple'), representing equivalent concepts, but with different levels of syntactic complexity (respectively, indefinite phrases: 'a purple chair'; sentences: 'The chair is purple.'; definite phrases: 'the purple chair'). We expected regions processing syntax to respond differently to simple versus complex structures. Single-word controls ('chair'/'purple') addressed definiteness-based accounts. In noun-adjective expressions, syntactic complexity only modulated activity in the left posterior temporal lobe (LPTL), ~ 300 ms after each word's onset: indefinite phrases induced more MEG-measured positive activity. The effects disappeared in single-word tokens, ruling out non-syntactic interpretations. In contrast, left anterior temporal lobe (LATL) activation was driven by meaning. Overall, the results support models implicating the LPTL in structure building and the LATL in early stages of conceptual combination.

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

Language is our mind's most powerful generative system for the expression of meaning and thought. What are the neural mechanisms of our ability to compose complex meanings from simpler representations? This question is impossible to answer unless we decompose the notion of 'meaning composition' in some theoretically guided way and then begin to assess the extent to which brain activity tracks the posited subroutines. Here, I summarize results from a body of MEG research that has begun to address this question from the ground up, first focusing on simple combinations of two words. The work sets off with a hypothesis space offered by theoretical linguistics, positing syntactic and logico-semantic composition as the main combinatory routines, but then reveals that the most consistent and prominent reflection of composition, localized in the left anterior temporal cortex at 200-250 ms, cannot be described with this toolkit. Instead, this activity tracks a much more conceptually driven process, robustly sensitive to the density of the conceptual feature space of the composing items. I will describe our functional understanding of this activity and how it may operate within a broader 'combinatory network.' This article is part of the theme issue 'Towards mechanistic models of meaning composition'.

Neural dynamics of semantic composition

Human speech comprehension is remarkable for its immediacy and rapidity. The listener interprets an incrementally delivered auditory input, millisecond by millisecond as it is heard, in terms of complex multilevel representations of relevant linguistic and nonlinguistic knowledge. Central to this process are the neural computations involved in semantic combination, whereby the meanings of words are combined into more complex representations, as in the combination of a verb and its following direct object (DO) noun (e.g., "eat the apple"). These combinatorial processes form the backbone for incremental interpretation, enabling listeners to integrate the meaning of each word as it is heard into their dynamic interpretation of the current utterance. Focusing on the verb-DO noun relationship in simple spoken sentences, we applied multivariate pattern analysis and computational semantic modeling to source-localized electro/magnetoencephalographic data to map out the specific representational constraints that are constructed as each word is heard, and to determine how these constraints guide the interpretation of subsequent words in the utterance. Comparing context-independent semantic models of the DO noun with contextually constrained noun models reflecting the semantic properties of the preceding verb, we found that only the contextually constrained model showed a significant fit to the brain data. Pattern-based measures of directed connectivity across the left hemisphere language network revealed a continuous information flow among temporal, inferior frontal, and inferior parietal regions, underpinning the verb's modification of the DO noun's activated semantics. These results provide a plausible neural substrate for seamless real-time incremental interpretation on the observed millisecond time scales.