Neurocomputational Models of Language Processing

Literary reception is a special case of language processing. The judgment of literature reveals deep social patterns with embodied cognition. In this study, we investigate how differences in literary quality resonate in the human brain. Modifying a series of stimuli previously used in studies of the emotional potential of Harry Potter, we alternate passages from the original novels with passages from imitative and intentionally poorly written fanfiction. EEG data shows how the three text types are processed differently by the brain. Comparing the brain activity of the readers for the various text types, we see a difference in the absolute power but not in the relative power of the frequency bands. Reading badfiction evokes the lowest activity. However, the functionality of this activity is the same for all texts since the relative power of the frequency bands does not differ. When comparing the participant groups, we observe the opposite situation. Here, different relative powers of the frequency bands reflect different judgments and reading habits of participants. For example, fans of Harry Potter, regular readers of fantasy texts, and generally frequent readers read the texts more attentively, which is reflected in a pronounced relative activity of the theta and alpha frequency bands. Non-frequent readers and readers who are not devoted to Harry Potter and fantasy in general have increased activity in the delta frequency band. This suggests their saliency detection is more prominent because they are less familiar with reading or the subject matter. To support our findings, we use the EEG data without averaging over stimuli and participants, capturing the participants' responses on the level of individual stimuli. A Kohonen self-organizing map trained on this more extensive data finds reliably detectable differences in the responses to passages from the original Harry Potter novels and fan- and badfiction. Our study allows for an interpretation of an adaptive brain response. Readers who enjoy Harry Potter or have experience with the fantasy genre show different reactions from those who do not. Thus, badfiction appears to be processed differently by the human brain, but not for all readers in the same way.

Word frequency is a strong predictor in most lexical processing tasks. Thus, any model of word recognition needs to account for how word frequency effects arise. The Discriminative Lexicon Model (DLM) models lexical processing with mappings between words' forms and their meanings. Comprehension and production are modeled via linear mappings between the two domains. So far, the mappings within the model can either be obtained incrementally via error-driven learning, a computationally expensive process able to capture frequency effects, or in an efficient, but frequency-agnostic solution modeling the theoretical endstate of learning (EL) where all words are learned optimally. In the present study we show how an efficient, yet frequency-informed mapping between form and meaning can be obtained (Frequency-informed learning; FIL). We find that FIL well approximates an incremental solution while being computationally much cheaper. FIL shows a relatively low type- and high token-accuracy, demonstrating that the model is able to process most word tokens encountered by speakers in daily life correctly. We use FIL to model reaction times in the Dutch Lexicon Project by means of a Gaussian Location Scale Model and find that FIL predicts well the S-shaped relationship between frequency and the mean of reaction times but underestimates the variance of reaction times for low frequency words. FIL is also better able to account for priming effects in an auditory lexical decision task in Mandarin Chinese, compared to EL. Finally, we used ordered data from CHILDES to compare mappings obtained with FIL and incremental learning. We show that the mappings are highly correlated, but that with FIL some nuances based on word ordering effects are lost. Our results show how frequency effects in a learning model can be simulated efficiently, and raise questions about how to best account for low-frequency words in cognitive models.