Top Research Papers on Language Learning

The digital competencies teachers and learners require to use this chatbot ethically and effectively to support language learning are presented.

It is shown that Chronos models significantly outperform other methods on datasets that were part of the training corpus; and have comparable and occasionally superior zero-shot performance on new datasets, relative to methods that were trained specifically on them.

The Educators must place their students as people who naturally have experience, knowledge, desires, and thoughts that can be used to learn, both individually and in groups. All language learning strategies have the main goal of increasing one's ability to speak. The choice of teaching and learning strategies language learning strategies must be based on the consideration of placing students as subjects of learning who not only passively accept what is conveyed by educators. Teaching Japanese especially kanji using various learning strategies makes it easy for Japanese learners especially in l...

Reflexion is a novel framework to reinforce language agents not by updating weights, but instead through linguistic feedback, which obtains significant improvements over a baseline agent across diverse tasks (sequential decision-making, coding, language reasoning).

The ability to speak a second language opens up a world of possibilities for people at all stages and all levels of their careers. This is especially true in the modern 'connected' global business world where companies of all shapes and sizes have interests, offices or operations in multiple countries; and this is what this research paper aims at. Learning agility is a set of complex skills that enable the learner to learn something new in one place and then apply what he has learned elsewhere, in a wholly different situation. Agile learners have the mindset that allows them to continually dev...

O-LoRA is proposed, a simple and efficient approach for continual learning in language models, effectively mitigating catastrophic forgetting while learning new tasks, and excels in preserving the generalization ability of LLMs on unseen tasks.

Dynalang is an agent that learns a multimodal world model to predict future text and image representations, and learns to act from imagined model rollouts, and can be pretrained on text-only data, enabling learning from offline datasets, and generate language grounded in an environment.

This paper presents a unified framework for pre-training models that are universally effective across datasets and setups, and proposes Mixture-of-Denoisers (MoD), a pre-training objective that combines diverse pre-training paradigms together.

Learning a foreign language has become a norm to most of the worlds’ population as the world becomes globalised and the need to become bilingual/multilingual becomes a fundamental aspect for international communication. Despite the need to learn foreign languages, it has however triggered fear or anxiety in learners. For some, fear can be a driving force towards success in language learning yet, for most learners, it tremendously affects their learning process. This study is conducted to explore the fear of foreign language learning based on the Social Learning Theory (Bandura, 1977) and frame...

Voltron is introduced, a framework for language-driven representation learning from human videos and associated captions that outperform the prior state-of-the-art, especially on targeted problems requiring higher-level features.

If one is interested in enhancing language proficiency these apps really help and make a great difference in the teaching and learning process.

This is a case study using an interactive digital game to teach grammar via Mobile Assimilation Language Learning (MALL) and shows that the game has somewhat improved students’ grammar knowledge.

It is suggested that the most recent generation of Large Language Models (LLMs) might finally provide the computational tools to determine empirically how much of the human language ability can be acquired from linguistic experience.

Transformers significantly outperform neural sequence models with recurrent or convolutional representations on ICLL tasks, and there is evidence that their ability to do so relies on specialized"n-gram heads"(higher-order variants of induction heads) that compute input-conditional next-token distributions.

Purpose: Although autistic perspectives are paramount for evidence-based practice and the neurodiversity paradigm, they are often underrepresented in research and have only recently started to get attention from professionals in different fields, including communication sciences and disorders (CSD) and foreign language teaching (FLT). To alleviate this situation, we conducted an exploratory study on the experiences of self-identified autistic bloggers with foreign language learning (FLL). Methods: We used deductive thematic analysis to characterize the bloggers and establish the perceived effe...

This paper surveys recent works on continual learning for LLMs and catalogs continue learning techniques in a novel multi-staged categorization scheme, involving continual pretraining, instruction tuning, and alignment.

More than 25 years have passed since CLIL was first introduced in schools all over Europe with objectives, methods and organizational structures that often also vary from one country to another. Time has passed, and still without a precise framework of reference, CLIL shows its great potential as well as its drawbacks. Using data from research conducted in Italy with four CLIL teachers who teach physics through a foreign language, this article aims to highlight the possibility that CLIL can serve to make every teacher, regardless of the subject, aware of the role that language can play in lea...

A new transfer learning framework called AbMAP is proposed, which fine-tunes foundational models for antibody-sequence inputs by supervising on antibody structure and binding specificity examples, and provides compelling evidence for the hypothesis that antibody repertoires of individuals tend to converge towards comparable structural and functional coverage.

Language learning can be very emotional, as anyone who has ever tried to learn or use another language (L2) will attest. The range of emotions varies widely in both type and intensity, from the thrill of successfully articulating yourself, for example, to the anxiety of navigating a high-stakes encounter in an L2. It is not surprising, therefore, that there is a longstanding tradition of research on emotion in the context of L2 learning (e.g., Horwitz et al., 1986). In fact, more than 40 years ago, Scovel (1978) reviewed the accumulated evidence on the role of just one emotion: Anxiety.

It is shown that LLMs are unable to learn concepts beyond the first level of the Borel Hierarchy, which imposes severe limits on the ability of LMs, both large and small, to capture many aspects of linguistic meaning.

The findings of the study revealed that ChatGPT and AI were considered to be useful tools for language learning and teaching, although they presented some challenges that can be overcome with the advancement of technology.

A unified framework for pre-training models that are universally effective across datasets and setups is presented and Mixture-of-Denoisers (MoD) is proposed, a pre- Training objective that combines diverse pre- training paradigms together.

Conditional Context Optimization (CoCoOp), which extends CoOp by further learning a lightweight neural network to generate for each image an input-conditional token (vector), and yields stronger domain generalization performance as well.

This practice report describes a 12-week stand-alone course designed to address the challenges university students face in foreign language classes. Adopting principles of transformative language learning, course content, activities and resources were designed and implemented to dispel myths and preconceptions regarding language instruction, promote self-directed, independent learning, and raise awareness of the cognitive and socio-emotional processes involved in language learning. Students’ feedback indicate that the course had a significant impact on their perceptions of adult language learn...

In this article, I advocate for an enriched view of children's linguistic input, with the aim of building sustainable and tangible links between theoretical models of language development and families' everyday experiences. Children's language experiences constrain theoretical models in ways that may illuminate universal learning biases. However, more than that, these experiences provide a staggering array of test cases and demonstrate the stage-setting effects of situational, familial, and societal context on language use and uptake. Centering on activity type as an entry point into context, ...

PurposeGrounded in second-language acquisition (SLA) field, with a particular focus on the positive psychology (PP) theoretical perspective, this study examined the potential interplay between learning engagement (LE) and language learning strategies (LLSs), and their impact on language learning achievement of Saudi English as a foreign language (EFL) learners.Design/methodology/approachThis quantitative study adopted a cross-sectional design using an online questionnaire distributed to 168 Saudi EFL college-level students in Saudi Arabia. Various statistical analyses (descriptive analyses, co...

My research focuses on perceptual approaches to second language speech learning, or pronunciation training. Two main approaches to perceptual training have been investigated in the literature: (1) perceptual fading in which learners are presented stimuli from opposite ends of a synthetic continuum in which the differences between the members of the non-native phonemic contrast have been exaggerated (Jamieson and Morosan, 1986), and (2) High-Variability Pronunciation Training (HVPT) in which learners are presented minimal pairs containing the non-native phonemic contrasts in forced-choice ident...

This work introduces Flamingo, a family of Visual Language Models (VLM) with this ability to bridge powerful pretrained vision-only and language-only models, handle sequences of arbitrarily interleaved visual and textual data, and seamlessly ingest images or videos as inputs.

ABSTRACT This secondary analysis study examined patterns, predictors, and outcomes of dual language learners’ (DLLs’) development in Spanish and English. Research Findings: Latent profile analysis identified three groups of DLLs: Balanced Bilinguals (54%), English-Dominant Bilinguals (25%), and Spanish-Dominant Bilinguals (21%). These groups had fairly comparable Spanish skills at Head Start (HS) entry but showed distinct learning patterns in English and Spanish during the 2.5 years in HS and kindergarten (K), which resulted in group differences in dual-language skills at K. When their home an...

A new multiple-choice question dataset is introduced to evaluate the effectiveness of LLMs in the realm of education, specifically in the areas of spoken language learning which encompass phonetics, phonology, and second language acquisition and preliminary findings on conversational communication are explored.

A taxonomy of tasks inspired by the steps involved in the study of ancient documents, spanning over three and a half millennia of civilizations around the ancient world is introduced, promoting and supports the continued collaborative impetus between the humanities and machine learning.

This work investigated how reinforcement learning (RL) can be used to finetune the downstream performance of pretrained Language Models (LMs) and implemented an off-policy and value-based algorithm: Deep-Q Learning (DQN), which can be applied to finetuning langauge models for downstream applications.

Shortcuts often hinder the robustness of large language models.

This approach transforms the continuous, complex space of protein structures into a manageable, discrete format with a codebook ranging from 4096 to 64000 tokens and sets a foundation for seamless, multi-modal integration, enhancing the capabilities of computational methods in protein design.

Researchers provided intensive guidance to grade VII teachers on the application of the whole language approach in learning to improve the writing skills of grade VII students who were still low. Therefore, this researcher can be said to be collaborative classroom action research. The subjects of this study were all Grade 7 Senior High School students and teachers. The data collected was in the form of data concerning the implementation of experiential writing skills learning in grade VII students. Data collection techniques used were observation, interviews, and tests. Test the validity of th...

A systematic study on two representative prompt tuning methods, namely text prompt tuning and visual prompt tuning, and proposes a simple approach called Unified Prompt Tuning (UPT), which essentially learns a tiny neural network to jointly optimize prompts across different modalities.

The Prompt-Transformer (P-Former) is introduced, a model that predicts these ideal prompts to align with visual features, which is trained exclusively on linguistic data, bypassing the need for image-text pairings.

It is demonstrated that the simple pre-training task of predicting which caption goes with which image is an efficient and scalable way to learn SOTA image representations from scratch on a dataset of 400 million (image, text) pairs collected from the internet.

This cross-sectional and longitudinal study sought to refine the measurement of foreign language learning boredom (FLLB) and examine its links with overall/skill-specific second and foreign language (L2) achievement (curriculum-based course exam scores) and proficiency (Cambridge English test scores). In Substudy 1, we developed and validated an 11-item Foreign Language Learning Boredom Scale–Short Form among secondary and tertiary English as a foreign language (EFL) learners in China (n1 = 2,223, n2 = 504, n3 = 934, n4 = 1,109). The scale showed sound psychometric properties (i.e., construc...

Evaluated instruction-tuned models are evaluated and it is found that instruction tuning strengthens both the use of semantic priors and the capacity to learn input-label mappings, but more of the former.

This work introduces Progressive Prompts - a simple and efficient approach for continual learning in language models that allows forward transfer and resists catastrophic forgetting, without relying on data replay or a large number of task-specific parameters.

This work introduces MiniGPT-v2, a model that can be treated as a unified interface for better handling various vision-language tasks and proposes using unique identifiers for different tasks when training the model to better distinguish each task instruction effortlessly.

This article reviews David Singleton’s books and articles published during the period 2014–2020. The first section concerns a popular book which he co-authored with Vivian Cook; the second gives an account of articles covering questions about the concept of language aptitude; the third deals with articles on the manner in which a learner’s competencies in different languages interact; the fourth section then summarizes his recent age-related work on second language learning in childhood, adolescence and midlife; and the fifth deals with his contributions on language learning in senior adulth...

The results show that this work can achieve strong performance while adding additional control to ensure diverse, tailored content for individual users, and develop novel methods for controlling items for factors relevant in language learning.

: Language learning strategies plays an important role in the second language acquisition (SLA). Studies on learning strategies adopt a series of methods to investigate various variables, for instance, language proficiency, gender and cultural backgrounds, which may affect language learners’ strategy choice and use. Raising learners’ metacognitive awareness of strategy use becomes crucial in the learning process, because less proficient learners may not know how and when to use learning strategies. Thus, training on how to employ strategies attracts researchers’ attention through comparisons b...

This paper addresses the issue of identifying the most informative demonstrations for few-shot learning by approaching it as a pool-based Active Learning (AL) problem over a single iteration, and unambiguously demonstrates that in-context example selection through AL prioritizes high-quality examples that exhibit low uncertainty and bear similarity to the test examples.

This work proposes a more general method that dynamically skips the execution of a layer (or module) for any input token with a binary router and demonstrates that the proposed method can significantly improve the 1-shot performance compared to other competitive baselines only at mild extra cost for inference.

Second language acquisition theory has traditionally focused on the cognitive and psycholinguistic processes involved in additional language (L2) learning. In addition, research on learner psychology has primarily centered on learners’ cognitive abilities (e.g., aptitude and working memory) and internal traits or states (e.g., dispositions, motivations, and affect). Language learning behavior, however, has remained largely neglected and under‐theorized. To address this gap, this paper proposes the proactive language learning theory, which delineates the agentic and strategic behaviors that lea...

It is seen that different domains give similar importance values, which indirectly shows that the proxy can approximately identify the common general knowledge.

Policy Learning with a Language Bottleneck is introduced, a framework enabling AI agents to generate linguistic rules that capture the strategies underlying their most rewarding behaviors and can also share the learned rules with human users, enabling more effective human-AI coordination.