Search

A central issue in the design of curricula for all school levels is the development of the learners' high-order thinking skills and metacognitive skills. Among such required skills is the ability to solve problems. The literature dealing with the development of problem-solving skills is vast and primarily addresses the scientific disciplines, even though the importance of integrating such skills into all disciplines is widely mentioned. The current study aims to compare problem-solving questions used in two seemingly different disciplines: linguistics and computer science. Our premise was that the basis for imparting problem-solving skills is asking problem-solving questions. We analysed problem-solving questions posed in various teaching and learning school scenarios for keywords, and identified nine categories of problem-solving question keywords common to both disciplines. We present those categories and extend our study of two categories to demonstrating the ability to follow the problem-solving process in linguistics is similar to the exact sciences.

Aim/Purpose: Research shows that students encounter difficulties in identifying the structure of argumentation texts and in understanding the main message of the argument. The research examined the effect that learning Logic Programming (LP), while applying logic inference, has on students' understanding of argumentation texts. Background: Understanding an argumentation text means exposure to its structure, which requires the ability to identify the argument presented and to distinguish between the argument and its justifications. Argumentation is an important cognitive capacity for handling conflicting information, viewpoints, and opinions. Students' lack of ability to identify the structure of argumentation texts, and to understand its' main message, affects the understanding of texts in general, the writing of texts, and the presentation of oral arguments. Since Logic Programming is based on inference that is similar to the way in which people commonly believe that human inferential thinking is performed, our research approach was to investigate how learning LP in Computer Science affects the understanding of argumentation texts in Linguistics. Methodology: The research population included 319 11th-grade students from five high schools, divided into a study group and a control group. Students' understanding was tested using knowledge questionnaires after completing their language studies, before (pre-study) and after (post-study) a year of learning LP. The knowledge questionnaires included argumentation paragraphs where students were asked to give each paragraph a title and to analyze the argument structure. In addition, an attitudes questionnaire was administered at the end of the school year in order to examine the students' attitudes towards the connection between the two disciplines. The research applied a mixed method approach, combining both qualitative and quantitative methods. Contribution: The research and its' findings contribute to the previous body of knowledge with relation to students difficulties in understanding argumentation texts in Linguistics studies. Moreover, it suggests a new approach of using argumentation in the framework of inference as apply in LP to scaffold students' conceptions. The use of an interactive computerized system (like the logic programming language Prolog) can scaffold students in constructing their knowledge, develop their computational thinking skills, and also enables to vary the teaching methods. Findings: Findings show that the students' understanding of argumentation texts improved after learning LP. The study group students' achievements were explicitly better compared with the control group students, who did not learn LP, though this was not always reflected with significant statistics. Students' attitudes questionnaire revealed that students did not identify on their own the connections between the two disciplines and so could not explicitly use it to promote their understanding. Recommendations for Practitioners: Creative educators, who value challenges, can greatly benefit their students if they collaborate in aim for applying interdisciplinary learning while combining those two disciplines. The research conclusions shows that it is possible to improve students' understanding if teachers explicitly mediate and guide students in drawing analogies. Recommendations for Researchers: The analysis tool we developed and apply can be used by educators and researchers to evaluate the understanding of argumentative texts by learners. It can be used in language classes at all levels as well as by educators in other disciplines in which the understanding of the argumentative structure is fundamental. Impact on Society: Developing argumentation skills and computational thinking skills. Future Research: Vary future possible research can follow the presented approach: examining how LP teachers expose the logical structure of an argumentation paragraph when they write logic programs that describe the inference represented in texts; examining how language teachers coupe with learning and using LP; examines the knowledge and skills of students that experienced a mediate learning process in the two disciplines in parallel.

Largely due to technological advances, methods for analyzing readability have increased significantly in recent years. While past researchers designed hundreds of formulas to estimate the difficulty of texts for readers, controversy has surrounded their use for decades, with criticism stemming largely from their application in creating new texts as well as their utilization of surface-level indicators as proxies for complex cognitive processes that take place when reading a text. This review focuses on examining developments in the field of readability during the past two decades with the goal of informing both current and future research and providing recommendations for present use. The fields of education, linguistics, cognitive science, psychology, discourse processing, and computer science have all made recent strides in developing new methods for predicting the difficulty of texts for various populations. However, there is a need for further development of these methods if they are to become widely available.

This is the first study to use self-reported ratings of student learning, instructor recommendations, and course recommendations as the outcome measure to estimate class size effects, doing so across 24 disciplines. Fixed-effects models controlling for heterogeneous courses and instructors reveal that increasing enrollment has negative and significant effects on student satisfaction in Sociology, Political Science, Computer Science and Engineering, and Mechanical and Aerospace Engineering. Educational outcomes in Linguistics, Psychology, Biological Sciences, Chemistry, Math, Physics, Cognitive Sciences, Visual Arts, History, and Philosophy are unaffected by class size. Other disciplines, including Economics, have inconclusive findings. No discipline benefits from increasing enrollment. (Contains 6 tables and 1 figure.)

The term "recursion" is used in at least four distinct theoretical senses within cognitive science. Some of these senses in turn relate to the different levels of analysis described by David Marr some 20 years ago; namely, the underlying competence capacity (the "computational" level), the performance operations used in real-time processing (the "algorithmic" level), and the neural implementation (the "hardware" level). It is demonstrated here that the current focus of the artificial grammar learning literature on recursion blurs the different levels, resulting in three main corollaries: (1) the literature cannot tell us anything about the first level, as the explanation for the role of recursion therein lies elsewhere; (2) it has not studied the possible recursive processing of artificial strings properly, as studies have so far focused on the correct processing of the internal hierarchy of recursive structures, but none have attempted to probe the underlying processing operations; and (3) as a consequence, much cannot in fact be known about the neural basis of recursion. It is furthermore argued that, in general, these studies can actually tell us very little about the language faculty, and some programmatic remarks are described and defended towards a more coherent study. (Contains 2 figures, 1 table and 16 footnotes.)

Ways to use personal computers to simplify the task of identifying and describing the phonology, morphology, and syntax of a language are outlined. The actual programing is viewed as simple once a phonetic alphabet is defined and symbols to indicate parts of speech are chosen. (MP)

In-context learning refers to the ability of a neural network to learn from information presented in its context. While traditional learning in neural networks requires adjusting network weights for every new task, in-context learning operates purely by updating internal activations without needing any updates to network weights. The emergence of this ability in large language models has led to a paradigm shift in machine learning and has forced researchers to reconceptualize how they think about learning in neural networks. Looking beyond language models, we can find in-context learning in many computational models relevant to cognitive science, including those that emerge from meta-learning.

We conduct a large-scale study of online community variation in language. We show that factors of efficient communication, which have been shown to drive crosslinguistic variation in lexical semantic systems, also play a role in within-language variation across 1926 English-language Reddit communities. We study variation in stancetaking behaviour, a domain where efficient communication may be influenced by social motivations for language use. We find that communities indeed have efficient stancetaking systems, particularly with respect to their own communicative needs. However, contrasting with crosslinguistic work, we find that communities are often not optimized for their needs. Moreover, we find that community-level social factors correlate with how optimized they are. These results highlight the importance of accounting for social pressures for language use when studying how efficient communication drives variation.

Large scale neural models show impressive performance across a wide array of linguistic tasks. Despite this they remain, largely, black-boxes - learning vector-representations of their input that prove difficult to interpret. This limits our ability to understand what they learn, and when the learn it, or characterise why they often fail to generalise systematically. To address this we introduce a novel approach to interpretability that looks at the mapping a model learns from sentences to representations as a kind of language in its own right. In doing so we introduce a set of information-theoretic measures that quantify how structured a model's representations are with respect to its input, and when during training that structure arises. Our measures are fast to compute, grounded in linguistic theory, and can predict which models will generalise best based on their representations. We use these measures to describe two distinct phases of training a transformer: an initial phase of in-distribution learning which reduces task loss, then a second stage where representations becoming robust to noise. Generalisation performance begins to increase during this second phase, drawing a link between generalisation and robustness to noise. Finally we look at how model size affects the structure of the representational space, showing that larger models ultimately compress their representations more than their smaller counterparts.

This paper explores the use of linguistic strategies, specifically discourse markers like 'but', to express contrasts between expectations and reality when faced with unexpected events. The study concentrates on Denial of Expectation (DofE), themost powerful form of contrast, which arises when the expected value based on background assumptions is not met. The main focus of this paper is to model DofE as a weighted homogeneous relationship between object properties. The aim is to predict DofE for numerical properties in specific contexts. I aim to address a gap in previous models by considering the role of context. This is achieved by analyzing contrastive sentences from German car and motorcycle reviews. The research presents the concept of expectation intervals for scalar properties. These intervals align with expectations and exceeding them triggers a potential contrast. The study incorporates causality, expected behavior, and a shift function in selecting contrastive pairs, transforming the conditions into an algorithm.Keywords: contrast; computational and cognitive modeling; discourse analysis

Word senses rise and fall due to a variety of causes. Previous research has explored how words grow novel senses, but the opposite problem of word sense decline is much less studied. Inspired by recent work on word decline, we investigate the cognitive factors that might explain the historical decline of word senses. We formalize a set of eight psycholinguistic predictors and assess their roles in discriminating declining senses from stable ones over the past two centuries in English. We find that semantic density, change in usage frequency in the semantic neighbourhood, and contextual diversity all predict word sense decline. Our study elucidates the cognitive underpinnings of word sense decline as the lexicon evolves.

Compositionality is considered a key hallmark of humanlanguage. However, most research focuses on item-level compositionality, e.g.,to what extent the meanings of phrases are composed of the meanings of theirsub-parts, rather than on language-level compositionality, which is the degreeto which possible combinations are utilized in practice during language use.Here, we propose a novel way to quantify the degree of language-levelcompositionality and apply it in the case of English adjective-nouncombinations. Using corpus analyses, large language models, and humanacceptability ratings, we find that (1) English only sparsely utilizes thecompositional potential of adjective–noun combinations; and (2) LLMs struggle topredict human acceptability judgments of rare combinations. Taken together, ourfindings shed new light on the role of compositionality in language andhighlight a challenging area for further improving LLMs.

This study explores the relationship between predictability, as measured by surprisal, and processing difficulty in code comprehension. We investigate whether similar mechanisms govern the processing of programming and natural languages. Previous research suggests that programmers prefer and produce more predictable code, akin to natural language patterns. We utilize eye-tracking data from the Eye Movements in Programming (EMIP) dataset to examine the impact of surprisal on various eye movement measures. Contrary to expectations based on natural language processing, our results reveal that surprisal does not significantly influence fixation metrics. Additionally, regressions in code reading show an unexpected inverse relationship with surprisal, suggesting that readers have different reasons for making regressions while reading code versus natural text. These findings contribute insights into the unique dynamics of code comprehension and opens avenues for further research in this domain.

The idea of curriculum learning, whereby a model is first exposed to simpler examples before an increase in complexity, has long fascinated the AI community. Unfortunately, the experimental successes of curriculum learning have been mixed, particularly applied to natural language, where a vast body of literature appears to evidence its failures. However, recent work has shown that language models trained on transcribed-child-directed-speech (CDS) learn more grammar compared to those trained on Wikipedia. To a lesser extent, the same trend has been observed through training on transcribed speech and simple text data. Motivated by these findings, we revisit the idea of curriculum learning starting from CDS, before moving to simple data, and finally finishing with complex long form text. Unfortunately, through experimentation with an array of models and training step sizes, only in the smallest models trained for the least steps does curriculum learning show any advantage over random sampling.

We propose a generative modeling framework of impaired language production and an inference framework that models rational comprehension of impaired language. Given a task (e.g. picture-description), we approximate the prior distribution over intended sentences using a language model trained on unimpaired speakers' utterances. We define a generative model of operations (e.g., semantic and phonological errors, retracing, filled pauses) that intervene on the intended sentence to yield an utterance. The model is implemented in the Gen probabilistic programming language, with data from AphasiaBank's ‘Window' picture-description task. Given observed utterances, a particle filter estimates posterior probabilities for latent variables (e.g. the speaker's intended sentence or sequence of errors). Our framework models comprehension as inference on a generative model of production, and provides a way to quantify incremental processing difficulty for impaired language in a way that combines a language model prior with explicit reasoning about errors.

Language models have great potential as cognitive models for studying human language acquisition, but current models are far less data-efficient than human learners. Children acquire language from 100 million words or less, but large language models are trained on trillions of words. We discuss the prospects for improving language models' developmental plausibility through a meta-analysis of results from the 2023 BabyLM Challenge. BabyLM was a competition that invited participants to train a language model on a 100 million-word corpus including transcribed speech and child-appropriate texts. Results from over 30 submissions showed that new machine learning techniques and increased training iterations yielded models that outperformed leading large language models in grammar, language understanding, and linguistic generalization, while cognitively plausible approaches such as curriculum learning were less effective. We discuss the implications of these and other findings for computational cognitive modeling and explore ideas to ensure future competitions' contributions to cognitive science.

Does grammatical gender influence the meaning of inanimate nouns? We examined word embeddings from distributional semantics models, representing meanings in a vector space. In 26 gendered languages and non-gendered English, we measured the meaning similarity of inanimate nouns to gendered anchor nouns like 'male' and 'female.' In gendered languages, noun meanings aligned more with the anchor noun congruent with grammatical gender. This effect persisted when comparing the same nouns across languages (e.g., 'cucchiaio' vs 'cuchara' vs 'spoon'). We propose that grammatical gender introduces a gender bias into lexical semantics through distributional similarities with anchor words, revealing masculine/feminine features even without direct sensorimotor experience. This suggests that embodiment in language processing may become statistically embedded in word usage patterns.

One of the many tasks facing the typically-developing child language learner is learning to discriminate between distinctive sounds that make up words in their native language. We investigate whether multimodal information---specifically adult speech coupled with video frames of speakers' faces---benefits a computational model of phonetic learning. We introduce a method for creating high-quality synthetic videos of speakers' faces for an existing audio corpus. Our learning model, when trained and tested on audiovisual inputs, achieves 8.1% relative improvement on a phoneme discrimination battery compared to a model trained and tested on audio-only input. It outperforms the audio model by 3.9% when tested on audio-only data, suggesting that visual information facilitates the acquisition of acoustic distinctions. In noisy audio environments, our audiovisual model recovers 67% of the loss in performance of the audio model relative to non-noisy environments. These results demonstrate that visual information benefits an ideal learner and illustrate multiple ways that children might leverage visual cues when learning to discriminate speech sounds.

This paper explores the hypothesis of the historical evolution of Semitic morphology from biconsonantal (2C) etymons, to triconsonantal (3C) roots, which make up the majority of words in Biblical Hebrew as well as in other Semitic languages such as modern Hebrew, Arabic, etc. The rules for reducing the 3C roots to their 2C etymons are provided in detail.We use BHSA, a manually annotated corpus of the Hebrew Bible, and Word2Vec, a method for converting words to a vector representing their semantic meaning, to study the hypothesis of evolution from 2C etymons to 3C roots in biblical Hebrew.Namely, we show that words in Hebrew with different roots, that might have originated from the same 2C etymons form a denser cluster than random sets of words of the same size. These differences are statistically significant and strongly support the hypothesis of evolution from 2C etymons to 3C roots.

Conducting experiments with diverse participants in their native languages can uncover insights into culture, cognition, and language that may not be revealed otherwise. However, conducting these experiments online makes it difficult to validate self-reported language proficiency. Furthermore, existing proficiency tests are small and cover only a few languages. We present an automated pipeline to generate vocabulary tests using text from Wikipedia. Our pipeline samples rare nouns and creates pseudowords with the same low-level statistics. Six behavioral experiments (N=236) in six countries and eight languages show that (a) our test can distinguish between native speakers of closely related languages, (b) the test is reliable (r=0.82), and (c) performance strongly correlates with existing tests (LexTale) and self-reports. We further show that test accuracy is negatively correlated with the linguistic distance between the tested and the native language. Our test, available in eight languages, can easily be extended to other languages.

This study describes the experimentally developed technique of sentiment analysis in an automatic semantic parser that processes daily news input blocks. The 2022 time period was used to analyze news in Telegram channels published by twelve official news agencies and five influential non-official news sources. In the advanced analytical system that we present, automatic learning is combined with a frame-based approach to identify patterns that influence judgments based on a frame representation of the text. The parser analyzes the syntactic structure of each sentence input, constructs its semantic representation, and associates it with one of 3800 semantic frames extracted from a large text corpus. The research revealed that the relative frequency and combination of frames varied between the two groups of news sources. The results prove that news is conceptualized and represented by various media sources in very different ways with the potential to affect the recipient's beliefs and emotions.

Recent advancements in Large Language Models (LLMs) harness linguistic associations in vast natural language data for practical applications. However, their ability to understand the physical world using only language data remains a question. After reviewing existing protocols, we explore this question using a novel and tightly controlled reasoning test (ART) and compare human norms against versions of GPT-3. Our findings highlight the categories of common-sense relations models that could learn directly from data and areas of weakness. GPT-3 offers evidence for verbal reasoning on a par with human subjects for several relations including Synonymy, Antonymy, and Default inheritance, Without reinforcement learning from human judgements, it appears GPT-3 performs at the lower end of the reference interval for Has-part and Contained-in. Weaknesses were observed also in affordance characteristics through Necessary-quality, Order-of-size and Order-of-intensity. Combining LLMs with symbolic world grounding is a promising direction to address associative learning.

Modelling of early language acquisition aims to understand how infants bootstrap their language skills. The modelling encompasses properties of the input data used for training the models, the cognitive hypotheses and their algorithmic implementations being tested, and the evaluation methodologies to compare models to human data. Recent developments have enabled the use of more naturalistic training data for computational models. This also motivates development of more naturalistic tests of model behaviour. A crucial step towards such an aim is to develop representative speech datasets consisting of speech heard by infants in their natural environments. However, a major drawback of such recordings is that they are typically noisy, and it is currently unclear how the sound quality could affect analyses and modelling experiments conducted on such data. In this paper, we explore this aspect for the case of infant-directed speech (IDS) and adult-directed speech (ADS) analysis. First, we manually and automatically annotated audio quality of utterances extracted from two corpora of child-centred long-form recordings (in English and French). We then compared acoustic features of IDS and ADS in an in-lab dataset and across different audio quality subsets of naturalistic data. Finally, we assessed how the audio quality and recording environment may change the conclusions of a modelling analysis using a recent self-supervised learning model. Our results show that the use of modest and high audio quality naturalistic speech data result in largely similar conclusions on IDS and ADS in terms of acoustic analyses and modelling experiments. We also found that an automatic sound quality assessment tool can be used to screen out useful parts of long-form recordings for a closer analysis with comparable results to that of manual quality annotation.

Virtual assistants have become fixtures in everyday settings, but most research focuses on their development rather than their use following deployment. To facilitate study of their use in office settings, we introduce OfficeDial, a multimodal dataset containing audio recordings, transcriptions, eye tracking data, and screen recordings from conversations between humans and virtual assistants in office environments. Conversations are paired with physical and behavioral measures of cognitive load. We study the associations between verbal behavior and noise level and reveal key relationships between verbal redundancy, disfluency, and noise level. We make our new dataset available to interested researchers to inspire further exploration.

In this paper, we study human sequential behavior by integrating cognitive, evolutionary, and computational approaches. Our work centers around the emergence of shared vocabularies in the Embodied Communication Game (ECG). Here, participant pairs solve a shared task without access to conventional means of communication, enforcing the emergence of a new communication system. This problem is solved typically by negotiating a shared set of sequential signals that acquire meaning through interactions. Individual differences in Personal Need for Structure (PNS) have been found to influence how this process develops. We trained deep neural networks to mimic the emergence of new communicative systems and used hyperparameter optimization to approximate latent human cognitive variables to explain human behavior. We demonstrate that models based on bidirectional LSTM networks are better at capturing human behavior than unidirectional LSTM networks. This suggests that human sequence processing in the ECG is influenced by expected future states. The approximated variables cannot explain the differences in PNS, but we do provide evidence suggesting that random and uncertainty-directed exploration strategies are combined to develop optimal behavior.

Probabilistic models of language understanding are valuable tools for investigating human language use. However, they need to be hand-designed for a particular domain. In contrast, large language models (LLMs) are trained on text that spans a wide array of domains, but they lack the structure and interpretability of probabilistic models. In this paper, we use chain-of-thought prompts to introduce structures from probabilistic models into LLMs. We explore this approach in the case of metaphor understanding. Our chain-of-thought prompts lead language models to infer latent variables and reason about their relationships in order to choose appropriate paraphrases for metaphors. The latent variables and relationships chosen are informed by theories of metaphor understanding from cognitive psychology. We apply these prompts to the two largest versions of GPT-3 and show that they can improve performance in a paraphrase selection task.

In Mandarin Chinese, abbreviation happens commonly to compound words across different syntactic categories. What is the motivation behind this shortening of words? This paper presents an investigation of this phenomenon from an information-theoretic point of view. A corpus study was con- ducted to measure the average amount of information contained in the full (long) form and the abbreviated (short) form of words given certain contexts. The amount of information was then compared between the long and short forms of a word, revealing that the short one usually contains less information, and therefore is more likely to be used in more predictive contexts. This result indicates that speakers of Chinese can choose to use shorter words when the context is more predictive, in accordance with considerations of efficiency.

That cognition is embodied is a claim that virtually no cognitive scientist today will deny: after all, even the researcher who models cognition in terms of entirely abstract, “medium-independent” states and processes will concede that particular instances are always necessarily realized in some body (of some kind) or other. The same is true for the theme of this year’s CogSci meeting, “Cognition in Context”: even if you think that there are cases in which the context plays merely a peripheral role in cognitive processing, you cannot deny that cognition always occurs in some context or other. This symposium is motivated by the realization, on the one hand, that the concept of embodiment means different things to different researchers in different contexts (see, e.g., Wilson 2002; Wilson and Golonka 2013; Crippen and Schulkin 2020), just as, on the other hand, the concept of context means different things to different researchers with different views on body and mind (see, e.g., Clancey 1997; Mesquita, Barrett and Smith 2010; Ibáñez and García 2018).