Search

Over the last years, advancements in deep learning models for computer vision have led to a dramatic improvement in their image classification accuracy. However, models with a higher accuracy in the task they were trained on do not necessarily develop better image representations that allow them to also perform better in other tasks they were not trained on. In order to investigate the representation learning capabilities of prominent high-performing computer vision models, we investigated how well they capture various indices of perceptual similarity from large-scale behavioral datasets. We find that higher image classification accuracy rates are not associated with a better performance on these datasets, and in fact we observe no improvement in performance since GoogLeNet (released 2015) and VGG-M (released 2014). We speculate that more accurate classification may result from hyper-engineering towards very fine-grained distinctions between highly similar classes, which does not incentivize the models to capture overall perceptual similarities.

Research into second language (L2) reading is an exponentially growing field. Yet, it still has a relatively short supply of comparable, ecologically valid data from readers representing a variety of first languages (L1). This article addresses this need by presenting a new data resource called "MECO L2" (Multilingual Eye Movements Corpus), a rich behavioral eye-tracking record of text reading in English as an L2 among 543 university student speakers of 12 different L1s. MECO L2 includes a test battery of component skills of reading and allows for a comparison of the participants' reading performance in their L1 and L2. This data resource enables innovative large-scale cross-sample analyses of predictors of L2 reading fluency and comprehension. We first introduce the design and structure of the MECO L2 resource, along with reliability estimates and basic descriptive analyses. Then, we illustrate the utility of MECO L2 by quantifying contributions of four sources to variability in L2 reading proficiency proposed in prior literature: reading fluency and comprehension in L1, proficiency in L2 component skills of reading, extralinguistic factors, and the L1 of the readers. Major findings included (a) a fundamental contrast between the determinants of L2 reading fluency versus comprehension accuracy, and (b) high within-participant consistency in the real-time strategy of reading in L1 and L2. We conclude by reviewing the implications of these findings to theories of L2 acquisition and outline further directions in which the new data resource may support L2 reading research.

People with schizophrenia spectrum disorders (SSD) show anomalies in language processing with respect to "who is doing what" in an action. This linguistic behavior is suggestive of an atypical representation of the formal concepts of "Agent" in the lexical representation of a verb, i.e., its thematic grid. To test this hypothesis, we administered a silent-reading task with sentences including a semantic violation of the animacy trait of the grammatical subject to 30 people with SSD and 30 healthy control participants (HCs). When the anomalous grammatical subject was the Agent of the event, a significant increase of Gaze Duration was observed in HCs, but not in SSDs. Conversely, when the anomalous subject was a Theme, SSDs displayed an increased probability of go-back movements, unlike HCs. These results are suggestive of a higher tolerability for anomalous Agents in SSD compared to the normal population. The fact that SSD participants did not show a similar tolerability for anomalous Themes rules out the issue of an attention deficit. We suggest that general communication abilities in SSD might benefit from explicit training on deep linguistic structures.

How quantifiers are represented in the human mind is still a topic of intense debate. Seminal studies have addressed the issue of how a subclass of quantifiers, that is, number words, is spatially coded displaying the Spatial-Numerical Association of Response Codes (SNARC) effect; yet, none of these studies have explored the spatial representation of nonnumerical quantifiers such as "some" or "many." The aim of the present study is to investigate whether nonnumerical quantifiers are spatially coded in the human mind. We administered two typical comparison tasks to 52 participants: the first task involved nonnumerical quantifiers; the second task involved number words. Results showed a response-side compatibility effect for both number words and nonnumerical quantifiers, suggesting that both types of quantifiers are encoded in a spatial format; quantifiers referring to "small" quantities are responded to faster with the left hand and quantifiers referring to "large" quantities are responded to faster with the right hand. We labeled this effect for nonnumerical quantifiers as the Spatial-Linguistic Association of Response Codes (SLARC) effect. Notably, we found that the SNARC and the SLARC effects were strictly related to each other, namely the more a participant was sensitive to the SNARC effect in the number--word task, the more a SLARC effect was detectable in the nonnumerical quantifier task. These findings add evidence to the tendency of humans to align magnitude information on a mental line that is coded from left to right.

In this study, we use temporally aligned word embeddings and a large diachronic corpus of English to quantify language change in a data-driven, scalable way, which is grounded in language use. We show a unique and reliable relation between measures of language change and age of acquisition ("AoA") while controlling for frequency, contextual diversity, concreteness, length, dominant part of speech, orthographic neighborhood density, and diachronic frequency variation. We analyze measures of language change tackling both the change in lexical representations and the change in the relation between lexical representations and the words with the most similar usage patterns, showing that they capture different aspects of language change. Our results show a unique relation between language change and "AoA," which is stronger when considering neighborhood-level measures of language change: Words with more coherent diachronic usage patterns tend to be acquired earlier. The results support theories positing a link between ontogenetic and ethnogenetic processes in language.

The preference for simple explanations, known as the parsimony principle, has long guided the development of scientific theories, hypotheses, and models. Yet recent years have seen a number of successes in employing highly complex models for scientific inquiry (e.g., for 3D protein folding or climate forecasting). In this paper, we reexamine the parsimony principle in light of these scientific and technological advancements. We review recent developments, including the surprising benefits of modeling with more parameters than data, the increasing appreciation of the context-sensitivity of data and misspecification of scientific models, and the development of new modeling tools. By integrating these insights, we reassess the utility of parsimony as a proxy for desirable model traits, such as predictive accuracy, interpretability, effectiveness in guiding new research, and resource efficiency. We conclude that more complex models are sometimes essential for scientific progress, and discuss the ways in which parsimony and complexity can play complementary roles in scientific modeling practice. [ABSTRACT FROM AUTHOR]

Large language models (LLMs) are being increasingly incorporated into scientific workflows. However, we have yet to fully grasp the implications of this integration. How should the advancement of large language models affect the practice of science? For this opinion piece, we have invited four diverse groups of scientists to reflect on this query, sharing their perspectives and engaging in debate. Schulz et al. make the argument that working with LLMs is not fundamentally different from working with human collaborators, while Bender et al. argue that LLMs are often misused and overhyped, and that their limitations warrant a focus on more specialized, easily interpretable tools. Marelli et al. emphasize the importance of transparent attribution and responsible use of LLMs. Finally, Botvinick and Gershman advocate that humans should retain responsibility for determining the scientific roadmap. To facilitate the discussion, the four perspectives are complemented with a response from each group. By putting these different perspectives in conversation, we aim to bring attention to important considerations within the academic community regarding the adoption of LLMs and their impact on both current and future scientific practices. [ABSTRACT FROM AUTHOR]

Effects of semantic transparency, reflected in processing differences between semantically transparent ("teabag") and opaque ("ladybird") compounds, have received considerable attention in the investigation of the role of constituents in compound processing. However, previous studies have yielded inconsistent results. In the present article, we argue that this is due to semantic transparency's often being conceptualized only as the semantic relatedness between the compound and constituent meanings as separate units. This neglects the fact that compounds are inherently productive constructions. We argue that compound processing is routinely impacted by a compositional process aimed at computing a compositional meaning, which would cause compositional semantic transparency effects to emerge in compound processing. We employ recent developments in compositional distributional semantics to quantify relatedness- as well as composition-based semantic transparency measures and use these to predict lexical decision times in a large-scale data set. We observed semantic transparency effects on compound processing that are not captured in relatedness terms but only by adopting a compositional perspective.

This article introduces the Children and Young People's Books-Lexicon (CYP-LEX), a large-scale lexical database derived from books popular with children and young people in the United Kingdom. CYP-LEX includes 1,200 books evenly distributed across three age bands (7–9, 10–12, 13+) and comprises over 70 million tokens and over 105,000 types. For each word in each age band, we provide its raw and Zipf-transformed frequencies, all parts-of-speech in which it occurs with raw frequency and lemma for each occurrence, and measures of count-based contextual diversity. Together and individually, the three CYP-LEX age bands contain substantially more words than any other publicly available database of books for primary and secondary school children. Most of these words are very low in frequency, and a substantial proportion of the words in each age band do not occur on British television. Although the three age bands share some very frequent words, they differ substantially regarding words that occur less frequently, and this pattern also holds at the level of individual books. Initial analyses of CYP-LEX illustrate why independent reading constitutes a challenge for children and young people, and they also underscore the importance of reading widely for the development of reading expertise. Overall, CYP-LEX provides unprecedented information into the nature of vocabulary in books that British children aged 7+ read, and is a highly valuable resource for those studying reading and language development. [ABSTRACT FROM AUTHOR]

Objective: We aimed to develop a machine learning model to infer OCEAN traits from text. Background: The psycholexical approach allows retrieving information about personality traits from human language. However, it has rarely been applied because of methodological and practical issues that current computational advancements could overcome. Method: Classical taxonomies and a large Yelp corpus were leveraged to learn an embedding for each personality trait. These embeddings were used to train a feedforward neural network for predicting trait values. Their generalization performances have been evaluated through two external validation studies involving experts (N = 11) and laypeople (N = 100) in a discrimination task about the best markers of each trait and polarity. Results: Intrinsic validation of the model yielded excellent results, with R2 values greater than 0.78. The validation studies showed a high proportion of matches between participants' choices and model predictions, confirming its efficacy in identifying new terms related to the OCEAN traits. The best performance was observed for agreeableness and extraversion, especially for their positive polarities. The model was less efficient in identifying the negative polarity of openness and conscientiousness. Conclusions: This innovative methodology can be considered a "psycholexical approach 2.0," contributing to research in personality and its practical applications in many fields. [ABSTRACT FROM AUTHOR]

Pseudowords such as "knackets" or "spechy"—letter strings that are consistent with the orthotactical rules of a language but do not appear in its lexicon—are traditionally considered to be meaningless, and used as such in empirical studies. However, recent studies that show specific semantic patterns associated with these words as well as semantic effects on human pseudoword processing have cast doubt on this view. While these studies suggest that pseudowords have meanings, they provide only extremely limited insight as to whether humans are able to ascribe explicit and declarative semantic content to unfamiliar word forms. In the present study, we utilized an exploratory-confirmatory study design to examine this question. In a first exploratory study, we started from a pre-existing dataset of words and pseudowords alongside human-generated definitions for these items. Using 18 different language models, we showed that the definitions actually produced for (pseudo)words were closer to their respective (pseudo)words than the definitions for the other items. Based on these initial results, we conducted a second, pre-registered, high-powered confirmatory study collecting a new, controlled set of (pseudo)word interpretations. This second study confirmed the results of the first one. Taken together, these findings support the idea that meaning construction is supported by a flexible form-to-meaning mapping system based on statistical regularities in the language environment that can accommodate novel lexical entries as soon as they are encountered. [ABSTRACT FROM AUTHOR]