Preprint
Refine
Document Type
- Preprint (5) (remove)
Language
- English (5)
Has Fulltext
- yes (5)
Is part of the Bibliography
- no (5)
Institute
- Psychologie (5) (remove)
The outstanding speed of language comprehension necessitates a highly efficient implementation of cognitive-linguistic processes. The domain-general theory of Predictive Coding suggests that our brain solves this problem by continuously forming linguistic predictions about expected upcoming input. The neurophysiological implementation of these predictive linguistic processes, however, is not yet understood. Here, we use EEG (human participants, both sexes) to investigate the existence and nature of online-generated, category-level semantic representations during sentence processing. We conducted two experiments in which some nouns – embedded in a predictive spoken sentence context – were unexpectedly delayed by 1 second. Target nouns were either abstract/concrete (Experiment 1) or animate/inanimate (Experiment 2). We hypothesized that if neural prediction error signals following (temporary) omissions carry specific information about the stimulus, the semantic category of the upcoming target word is encoded in brain activity prior to its presentation. Using time-generalized multivariate pattern analysis, we demonstrate significant decoding of word category from silent periods directly preceding the target word, in both experiments. This provides direct evidence for predictive coding during sentence processing, i.e., that information about a word can be encoded in brain activity before it is perceived. While the same semantic contrast could also be decoded from EEG activity elicited by isolated words (Experiment 1), the identified neural patterns did not generalize to pre-stimulus delay period activity in sentences. Our results not only indicate that the brain processes language predictively, but also demonstrate the nature and sentence-specificity of category-level semantic predictions preactivated during sentence comprehension.
How is semantic information stored in the human mind and brain? Some philosophers and cognitive scientists argue for vectorial representations of concepts, where the meaning of a word is represented as its position in a high-dimensional neural state space. At the intersection of natural language processing and artificial intelligence, a class of very successful distributional word vector models has developed that can account for classic EEG findings of language, i.e., the ease vs. difficulty of integrating a word with its sentence context. However, models of semantics have to account not only for context-based word processing, but should also describe how word meaning is represented. Here, we investigate whether distributional vector representations of word meaning can model brain activity induced by words presented without context. Using EEG activity (event-related brain potentials) collected while participants in two experiments (English, German) read isolated words, we encode and decode word vectors taken from the family of prediction-based word2vec algorithms. We find that, first, the position of a word in vector space allows the prediction of the pattern of corresponding neural activity over time, in particular during a time window of 300 to 500 ms after word onset. Second, distributional models perform better than a human-created taxonomic baseline model (WordNet), and this holds for several distinct vector-based models. Third, multiple latent semantic dimensions of word meaning can be decoded from brain activity. Combined, these results suggest that empiricist, prediction-based vectorial representations of meaning are a viable candidate for the representational architecture of human semantic knowledge.
Most current models assume that the perceptual and cognitive processes of visual word recognition and reading operate upon neuronally coded domain-general low-level visual representations – typically oriented line representations. We here demonstrate, consistent with neurophysiological theories of Bayesian-like predictive neural computations, that prior visual knowledge of words may be utilized to ‘explain away’ redundant and highly expected parts of the visual percept. Subsequent processing stages, accordingly, operate upon an optimized representation of the visual input, the orthographic prediction error, highlighting only the visual information relevant for word identification. We show that this optimized representation is related to orthographic word characteristics, accounts for word recognition behavior, and is processed early in the visual processing stream, i.e., in V4 and before 200 ms after word-onset. Based on these findings, we propose that prior visual-orthographic knowledge is used to optimize the representation of visually presented words, which in turn allows for highly efficient reading processes.
To characterize the left-ventral occipito-temporal cortex (lvOT) role during reading in a quantitatively explicit and testable manner, we propose the lexical categorization model (LCM). The LCM assumes that lvOT optimizes linguistic processing by allowing fast meaning access when words are familiar and filter out orthographic strings without meaning. The LCM successfully simulates benchmark results from functional brain imaging. Empirically, using functional magnetic resonance imaging, we demonstrate that quantitative LCM simulations predict lvOT activation across three studies better than alternative models. Besides, we found that word-likeness, which is assumed as input to LCM, is represented posterior to lvOT. In contrast, a dichotomous word/non-word contrast, which is assumed as the LCM’s output, could be localized to upstream frontal brain regions. Finally, we found that training lexical categorization results in more efficient reading. Thus, we propose a ventral-visual-stream processing framework for reading involving word-likeness extraction followed by lexical categorization, before meaning extraction.
Across languages, the speech signal is characterized by a predominant modulation of the amplitude spectrum between about 4.3-5.5Hz, reflecting the production and processing of linguistic information chunks (syllables, words) every ∼200ms. Interestingly, ∼200ms is also the typical duration of eye fixations during reading. Prompted by this observation, we demonstrate that German readers sample written text at ∼5Hz. A subsequent meta-analysis with 142 studies from 14 languages replicates this result, but also shows that sampling frequencies vary across languages between 3.9Hz and 5.2Hz, and that this variation systematically depends on the complexity of the writing systems (character-based vs. alphabetic systems, orthographic transparency). Finally, we demonstrate empirically a positive correlation between speech spectrum and eye-movement sampling in low-skilled readers. Based on this convergent evidence, we propose that during reading, our brain’s linguistic processing systems imprint a preferred processing rate, i.e., the rate of spoken language production and perception, onto the oculomotor system.