Hasil untuk "Language acquisition"

Wenhong Zhu, Weinan Zhang, Rui Wang

Post-training alignment has increasingly become a crucial factor in enhancing the usability of language models (LMs). However, the strength of alignment varies depending on individual preferences. This paper proposes a method to incorporate alignment control into a single model, referred to as CLM. This approach adds one identity layer preceding the initial layers and performs preference learning only on this layer to map unaligned input token embeddings into the aligned space. Experimental results demonstrate that this efficient fine-tuning method performs comparable to full fine-tuning. During inference, the input embeddings are processed through the aligned and unaligned layers, which are then merged through the interpolation coefficient. By controlling this parameter, the alignment exhibits a clear interpolation and extrapolation phenomenon.

Lujun Li, Yewei Song, Lama Sleem et al.

Grammar refers to the system of rules that governs the structural organization and the semantic relations among linguistic units such as sentences, phrases, and words within a given language. In natural language processing, there remains a notable scarcity of grammar focused evaluation protocols, a gap that is even more pronounced for low-resource languages. Moreover, the extent to which large language models genuinely comprehend grammatical structure, especially the mapping between syntactic structures and meanings, remains under debate. To investigate this issue, we propose a Grammar Book Guided evaluation pipeline intended to provide a systematic and generalizable framework for grammar evaluation consisting of four key stages, and in this work we take Luxembourgish as a case study. The results show a weak positive correlation between translation performance and grammatical understanding, indicating that strong translations do not necessarily imply deep grammatical competence. Larger models perform well overall due to their semantic strength but remain weak in morphology and syntax, struggling particularly with Minimal Pair tasks, while strong reasoning ability offers a promising way to enhance their grammatical understanding.

M. Kannan, S. Meenakshi

This paper discusses the literature on language acquisition in individuals with autism spectrum disorders (ASD), highlighting rapid changes in the field. Researchers in psycholinguistics are exploring language acquisition theories due to ASD’s significant differences across language, social, and cognitive domains. The study highlights areas where knowledge is lacking and explores potential future directions. While pragmatic deficits are commonly associated with ASD, clinicians and researchers should consider phonological, morph syntactic differences and rehabilitation to change the condition of phonological errors, which impact language comprehension and production.

Roberto B. Sileo, Luca Cilibrasi, Julia Heine et al.

This study investigates pronominal reference assignments across sentences that contain English verbs of transfer in monolingual English speakers and second-language (L2) learners having German as a first language and English as an L2. In a forced-choice task, participants were presented with sentences in perfective or imperfective aspect, like “Elizabeth took/was taking a meal to Mary” (adapted from Ferretti et al., 2009). They were then shown sentences that contained gender-matching pronouns, as in “She breathed in the smell of fresh basil”, and they were finally asked to choose who performed the relevant actions: “Who breathed in the smell of fresh basil? Elizabeth or Mary?”. We found that both groups preferred more often goal-oriented interpretations in the perfective condition, while in the imperfective condition only English monolingual speakers preferred more often source-oriented interpretations. The pattern observed in the perfective condition is consistent with previous studies and indicates that perfective aspect creates a strong bias towards end-states. For the imperfective condition, we argue that the different pattern observed in L2 learners may be due to some features of German, where an overall bias for end-states was previously observed. This indicates an effect of first-language strategies on L2 processing, consistent with previous research on different languages.

Wanru Zhao, Vidit Khazanchi, Haodi Xing et al.

Large language model (LLM) services have recently begun offering a plugin ecosystem to interact with third-party API services. This innovation enhances the capabilities of LLMs, but it also introduces risks, as these plugins developed by various third parties cannot be easily trusted. This paper proposes a new attacking framework to examine security and safety vulnerabilities within LLM platforms that incorporate third-party services. Applying our framework specifically to widely used LLMs, we identify real-world malicious attacks across various domains on third-party APIs that can imperceptibly modify LLM outputs. The paper discusses the unique challenges posed by third-party API integration and offers strategic possibilities to improve the security and safety of LLM ecosystems moving forward. Our code is released at https://github.com/vk0812/Third-Party-Attacks-on-LLMs.

Anastasia Zhukova, Christian E. Matt, Bela Gipp

Domain-specific languages that use a lot of specific terminology often fall into the category of low-resource languages. Collecting test datasets in a narrow domain is time-consuming and requires skilled human resources with domain knowledge and training for the annotation task. This study addresses the challenge of automated collecting test datasets to evaluate semantic search in low-resource domain-specific German language of the process industry. Our approach proposes an end-to-end annotation pipeline for automated query generation to the score reassessment of query-document pairs. To overcome the lack of text encoders trained in the German chemistry domain, we explore a principle of an ensemble of "weak" text encoders trained on common knowledge datasets. We combine individual relevance scores from diverse models to retrieve document candidates and relevance scores generated by an LLM, aiming to achieve consensus on query-document alignment. Evaluation results demonstrate that the ensemble method significantly improves alignment with human-assigned relevance scores, outperforming individual models in both inter-coder agreement and accuracy metrics. These findings suggest that ensemble learning can effectively adapt semantic search systems for specialized, low-resource languages, offering a practical solution to resource limitations in domain-specific contexts.

Alessandro Benati, Mable Chan

This study investigated the possible effects of motivational factors on the positive effects generated by structured input on the acquisition of English causative passive forms. This investigation builds upon the work carried out within the structured input research framework with the intention to measure online effects utilising a self-paced reading test. The self-paced reading test is a reliable measurement of language processing. Fifty Chinese (L1) subjects participated in the current study. They were all learning English in a university in the United Kingdom. After receiving two motivation questionnaires three groups were formed: structured input low-motivated (n = 15); structured input high-motivated (n = 20); and a control group (n = 15). Pre-tests and post-tests (immediate and delayed) were administered before and after the instructional treatment period which lasted for 3 h over a two-day period. The main findings from this experimental study confirmed the positive effects of structured input in facilitating the correct processing of English causative passive forms (accuracy and response time). The structured input activities groups improved equally from pre-to post-tests and they both retained the positive instructional effects over a 3-week period. Motivation was not a factor influencing the positive results generated by structured input. The present study contributes to the current call in the field for more empirical research to investigate the role of instruction and individual differences and the use of online tests to measure in-depth language processing.

Benni Ichsanda Rahman Hz, Rita Seroja Br Ginting

Each individual possesses a distinct essence that sets them apart, even in the case of identical twins. In language acquisition, learners’ personality type has emerged as one of the most determining constructions for students’ learning strategy. Through a phenomenological case study, intricate interplay between personality types and language learning strategies of two Indonesian identical twin girls studying English language education were investigated, supported by one triangulator participant. By employing the Oxford (1990) Strategy Inventory for Language Learning and the MBTI test, the researchers gathered the extensive data on their language learning strategy distinctions and personality differences. The results indicate that despite sharing the same upbringing, their diverse personalities - one an ENFJ and the other an ENTJ - have contributed to vastly different cognitive, memorization, compensation, metacognitive, social learning, and affective learning approaches. The findings provide intriguing insights into the intricacies of language acquisition and highlight the significance of individual differences in shaping the learning styles.

Spandan Dey, Md Sahidullah, Goutam Saha

This work addresses the cross-corpora generalization issue for the low-resourced spoken language identification (LID) problem. We have conducted the experiments in the context of Indian LID and identified strikingly poor cross-corpora generalization due to corpora-dependent non-lingual biases. Our contribution to this work is twofold. First, we propose domain diversification, which diversifies the limited training data using different audio data augmentation methods. We then propose the concept of maximally diversity-aware cascaded augmentations and optimize the augmentation fold-factor for effective diversification of the training data. Second, we introduce the idea of domain generalization considering the augmentation methods as pseudo-domains. Towards this, we investigate both domain-invariant and domain-aware approaches. Our LID system is based on the state-of-the-art emphasized channel attention, propagation, and aggregation based time delay neural network (ECAPA-TDNN) architecture. We have conducted extensive experiments with three widely used corpora for Indian LID research. In addition, we conduct a final blind evaluation of our proposed methods on the Indian subset of VoxLingua107 corpus collected in the wild. Our experiments demonstrate that the proposed domain diversification is more promising over commonly used simple augmentation methods. The study also reveals that domain generalization is a more effective solution than domain diversification. We also notice that domain-aware learning performs better for same-corpora LID, whereas domain-invariant learning is more suitable for cross-corpora generalization. Compared to basic ECAPA-TDNN, its proposed domain-invariant extensions improve the cross-corpora EER up to 5.23%. In contrast, the proposed domain-aware extensions also improve performance for same-corpora test scenarios.

Ahmed Abdelali, Hamdy Mubarak, Shammur Absar Chowdhury et al.

Recent advancements in Large Language Models (LLMs) have significantly influenced the landscape of language and speech research. Despite this progress, these models lack specific benchmarking against state-of-the-art (SOTA) models tailored to particular languages and tasks. LAraBench addresses this gap for Arabic Natural Language Processing (NLP) and Speech Processing tasks, including sequence tagging and content classification across different domains. We utilized models such as GPT-3.5-turbo, GPT-4, BLOOMZ, Jais-13b-chat, Whisper, and USM, employing zero and few-shot learning techniques to tackle 33 distinct tasks across 61 publicly available datasets. This involved 98 experimental setups, encompassing ~296K data points, ~46 hours of speech, and 30 sentences for Text-to-Speech (TTS). This effort resulted in 330+ sets of experiments. Our analysis focused on measuring the performance gap between SOTA models and LLMs. The overarching trend observed was that SOTA models generally outperformed LLMs in zero-shot learning, with a few exceptions. Notably, larger computational models with few-shot learning techniques managed to reduce these performance gaps. Our findings provide valuable insights into the applicability of LLMs for Arabic NLP and speech processing tasks.

Jionghao Lin, Danielle R. Thomas, Feifei Han et al.

Research demonstrates learners engaging in the process of producing explanations to support their reasoning, can have a positive impact on learning. However, providing learners real-time explanatory feedback often presents challenges related to classification accuracy, particularly in domain-specific environments, containing situationally complex and nuanced responses. We present two approaches for supplying tutors real-time feedback within an online lesson on how to give students effective praise. This work-in-progress demonstrates considerable accuracy in binary classification for corrective feedback of effective, or effort-based (F1 score = 0.811), and ineffective, or outcome-based (F1 score = 0.350), praise responses. More notably, we introduce progress towards an enhanced approach of providing explanatory feedback using large language model-facilitated named entity recognition, which can provide tutors feedback, not only while engaging in lessons, but can potentially suggest real-time tutor moves. Future work involves leveraging large language models for data augmentation to improve accuracy, while also developing an explanatory feedback interface.

Mitja Nikolaus, Afra Alishahi, Grzegorz Chrupała

AbstractRecent computational models of the acquisition of spoken language via grounding in perception exploit associations between spoken and visual modalities and learn to represent speech and visual data in a joint vector space. A major unresolved issue from the point of ecological validity is the training data, typically consisting of images or videos paired with spoken descriptions of what is depicted. Such a setup guarantees an unrealistically strong correlation between speech and the visual data. In the real world the coupling between the linguistic and the visual modality is loose, and often confounded by correlations with non-semantic aspects of the speech signal. Here we address this shortcoming by using a dataset based on the children’s cartoon Peppa Pig. We train a simple bi-modal architecture on the portion of the data consisting of dialog between characters, and evaluate on segments containing descriptive narrations. Despite the weak and confounded signal in this training data, our model succeeds at learning aspects of the visual semantics of spoken language.

Adela Böhmerová

This study is devoted to tracing, presenting and linguo-culturally interpreting some of the aspects of the early history of the contacts of Slovaks with the English language. Although English in Slovakia started to be of interest to several men of letters already in the 18th century, the need for it as means of communication only arose in the US in the second half of the 19th century among Slovak immigrants. The paper focuses above all on Janko Slovenský’s book as the first material assisting Slovaks in the acquisition of English, and analyses its content, educational merit and cultural value. Also surveyed is the history of the first dictionaries contrasting English and Slovak. The final part introduces the beginnings of English studies in Slovakia dating from the early 1920s, and their early development. The study offers insight into an educationally important subject that so far has only marginally received scholarly attention.

Youssef Rehouma, Mohamed Abd El basset Mahboub, Aicha Degla et al.

We created a real-time acquisition system to track the voltage, current and temperature changes of the solar panel as we installed it in a charging regulator with a battery. The system consists of an Arduino Uno board, the controllership, which is programmed by the Arduino IDE application, based on the C language, and sensors to capture the variables, we put the SD card to save the data and the LCD to see it currently and can be monitoring the data by connecting the Arduino Uno board to the computer and processing it with the Excel application.

Barack Wanjawa, Lilian Wanzare, Florence Indede et al.

Indigenous African languages are categorized as under-served in Natural Language Processing. They therefore experience poor digital inclusivity and information access. The processing challenge with such languages has been how to use machine learning and deep learning models without the requisite data. The Kencorpus project intends to bridge this gap by collecting and storing text and speech data that is good enough for data-driven solutions in applications such as machine translation, question answering and transcription in multilingual communities. The Kencorpus dataset is a text and speech corpus for three languages predominantly spoken in Kenya: Swahili, Dholuo and Luhya. Data collection was done by researchers from communities, schools, media, and publishers. The Kencorpus' dataset has a collection of 5,594 items - 4,442 texts (5.6M words) and 1,152 speech files (177hrs). Based on this data, Part of Speech tagging sets for Dholuo and Luhya (50,000 and 93,000 words respectively) were developed. We developed 7,537 Question-Answer pairs for Swahili and created a text translation set of 13,400 sentences from Dholuo and Luhya into Swahili. The datasets are useful for downstream machine learning tasks such as model training and translation. We also developed two proof of concept systems: for Kiswahili speech-to-text and machine learning system for Question Answering task, with results of 18.87% word error rate and 80% Exact Match (EM) respectively. These initial results give great promise to the usability of Kencorpus to the machine learning community. Kencorpus is one of few public domain corpora for these three low resource languages and forms a basis of learning and sharing experiences for similar works especially for low resource languages.

Devansh Mehta, Harshita Diddee, Ananya Saxena et al.

The primary obstacle to developing technologies for low-resource languages is the lack of representative, usable data. In this paper, we report the deployment of technology-driven data collection methods for creating a corpus of more than 60,000 translations from Hindi to Gondi, a low-resource vulnerable language spoken by around 2.3 million tribal people in south and central India. During this process, we help expand information access in Gondi across 2 different dimensions (a) The creation of linguistic resources that can be used by the community, such as a dictionary, children's stories, Gondi translations from multiple sources and an Interactive Voice Response (IVR) based mass awareness platform; (b) Enabling its use in the digital domain by developing a Hindi-Gondi machine translation model, which is compressed by nearly 4 times to enable it's edge deployment on low-resource edge devices and in areas of little to no internet connectivity. We also present preliminary evaluations of utilizing the developed machine translation model to provide assistance to volunteers who are involved in collecting more data for the target language. Through these interventions, we not only created a refined and evaluated corpus of 26,240 Hindi-Gondi translations that was used for building the translation model but also engaged nearly 850 community members who can help take Gondi onto the internet.

Przemyslaw Joniak, Akiko Aizawa

Language model debiasing has emerged as an important field of study in the NLP community. Numerous debiasing techniques were proposed, but bias ablation remains an unaddressed issue. We demonstrate a novel framework for inspecting bias in pre-trained transformer-based language models via movement pruning. Given a model and a debiasing objective, our framework finds a subset of the model containing less bias than the original model. We implement our framework by pruning the model while fine-tuning it on the debiasing objective. Optimized are only the pruning scores - parameters coupled with the model's weights that act as gates. We experiment with pruning attention heads, an important building block of transformers: we prune square blocks, as well as establish a new way of pruning the entire heads. Lastly, we demonstrate the usage of our framework using gender bias, and based on our findings, we propose an improvement to an existing debiasing method. Additionally, we re-discover a bias-performance trade-off: the better the model performs, the more bias it contains.

Simin Meykadeh, Simin Meykadeh, Arsalan Golfam et al.

While most studies on neural signals of online language processing have focused on a few—usually western—subject-verb-object (SVO) languages, corresponding knowledge on subject-object-verb (SOV) languages is scarce. Here we studied Farsi, a language with canonical SOV word order. Because we were interested in the consequences of second-language acquisition, we compared monolingual native Farsi speakers and equally proficient bilinguals who had learned Farsi only after entering primary school. We analyzed event-related potentials (ERPs) to correct and morphosyntactically incorrect sentence-final syllables in a sentence correctness judgment task. Incorrect syllables elicited a late posterior positivity at 500–700 ms after the final syllable, resembling the P600 component, as previously observed for syntactic violations at sentence-middle positions in SVO languages. There was no sign of a left anterior negativity (LAN) preceding the P600. Additionally, we provide evidence for a real-time discrimination of phonological categories associated with morphosyntactic manipulations (between 35 and 135 ms), manifesting the instantaneous neural response to unexpected perturbations. The L2 Farsi speakers were indistinguishable from L1 speakers in terms of performance and neural signals of syntactic violations, indicating that exposure to a second language at school entry may results in native-like performance and neural correlates. In nonnative (but not native) speakers verbal working memory capacity correlated with the late posterior positivity and performance accuracy. Hence, this first ERP study of morphosyntactic violations in a spoken SOV nominative-accusative language demonstrates ERP effects in response to morphosyntactic violations and the involvement of executive functions in non-native speakers in computations of subject-verb agreement.

Ozgun Ozalay, Irmak Polat Nazli, Sebnem Tunay et al.

INTRODUCTION: Major depressive disorder (MDD) is one of the most prevalent psychiatric disorders, which leads to significant social and economic burden across the world. One way to increase knowledge about pathophysiology is to understand how the pre-disease state processes to the disease state. The pre-disease state can be best studied in the high-risk populations who have higher chance of proceeding to disease. First degree relatives especially the offspring of patients are accepted as high-risk groups. Researchers were able to observe structural and functional abnormalities in the high-risk groups. Miskowiak et al observed increased anterior cingulate and dorsomedial prefrontal cortexes, pre-supplementary motor area with parietooccipital areas activation to happy or fearful faces in healthy twins with a co-twin history of depression1 . On the other hand, Chen et al reported reduced hippocampal volume in girls of depressed mothers2 . The primary aim of this study was to investigate the cortical grey matter volume alterations in young women who were at risk for familial depression. The secondary aim was to perceive if those alterations were parallel with their MDD diagnosed mothers. METHOD: Subjects: After approval by the Ethics Committee of Ege University and the recruitment via Internet advertisements and by invitation from the hospital database between 2009-2013, we screened 53 pairs of depressed mothers with their daughters. Among these pairs, 24 women with the diagnosis of MDD with recurrent episodes (mean age: 46.2±3.9 years) and their healthy daughters (the high-risk for familial depression group; HRFD) aged between 18-26 (mean age: 22.3±2.1 years) were included. The recurrent MDD diagnoses weremconfirmed by Structured Clinical Interview for DSM-4 (SCID). iInclusion criteria for the MDD mothers were; being free from clinically significant depression (Hamilton Depression Rating Scale (HAM-D)<16), having no other axis I diagnoses, having at least one healthy daughter with no history of depression between the ages 18-25, having a first-degree relative with an MDD diagnosis, having no history of psychotic symptoms. Patients with chronic medical illness and those who had relatives with bipolar disorder or schizophrenia were excluded. The control group was composed of age similar 24 healthy mothers (mean age: 47.3±5.6 years) who had healthy daughters of similar ages (mean age: 22.1±2.1 years) to the high-risk group. The healthy daughters of the control group constituted the low-risk for familial depression group (LRFD). Exclusion criteria for control groups were; any current or past psychiatric disorder confirmed with SCID, having any first-degree relative diagnosed with major depression, bipolar disorder or schizophrenia, having a chronic medical illness, having significant childhood trauma (e.g. sexual or physical abuse). Mothers and daughters gave their written informed consent after receiving a full explanation of the study’s purpose and procedures. Mothers in the MDD group continued their ongoing treatment and no treatment modification was done for this study. Depressive symptoms were assessed by HAM-D and Beck Depression rating scales on the same week for MRI scan. MRI Acquisition: 3D T1 weighted, sagittal, magnetization prepared rapid gradient echo (MPRAGE) scans of the head, 2D T2 weighted axial, Turbo Spin Echo (TSE) scans of the whole brain and 3D coronal fluid attenuated inversion recovery (FLAIR) scans were acquired on a Siemens Magnetom Verio, Numaris/4, Syngo MR B17, 3T MR scanner. TSE and FLAIR scans were used for clinical evaluation and MPRAGE scans were used in the region of interest (ROI) analysis. Image Processing: For image processing we used FreeSurfer Software and it’s recommended mainstream pipeline to segment gray matter volumes. Statistics: Sociodemographic and clinical variables were compared by t- or chi-square tests for MDD vs Controls and HRFD vs LRFD separately. Analyses of co-variance (ANCOVA) were used to compare the grey matter volume in each ROI. Total Intracranial Volume (ICV) was accepted nuisance variable for ANCOVA analyses. Due to multiple comparisons, we reported both uncorrected and false discovery rate (FDR) corrected p values adjusted to 0.05. The correlation between grey matter volumes in ROIs with clinical parameters were assessed by Pearson correlation test. RESULTS: Sociodemographic and Clinical Variables: There were no differences in age or education status among the MDD vs controls and HRFD vs LRFD. We observed higher Beck and HAM-D scores not only in MDD but also in the HRFD compared to controls and LRFD. Neuroimaging Variables: We observed greater grey matter volumes in right anterior cingulate and entorhinal cortexes in HRFD compared to LRFD. The difference in the right anterior cingulate cortex was preserved after FDR correction. We observed smaller ICV in the control group compared to MDD (t=3.2 df=46 p<0.01). ANCOVA relieved that MDD had greater grey matter volume in left inferior frontal cortex. However, this difference was disappeared after FDR correction. There were no correlation between any of the clinical variables and the GM values obtained from ROIs. DISCUSSION: This study investigated the cortical GMV of a group of young women who were at risk for familial depression and was able to confirm altered GMV in the right anterior cingulate and entorhinal cortex in this population. Our secondary aim of comparing GMV alterations observed in HRFD with their depressed mothers was partially reached because of weak differences between MDD and control groups. Despite our expectations, we found increased GMV in MDD group at left inferior frontal cortex, which was lost with multiple comparisons correction. The most prominent finding of this study is increased GMV at the right anterior cingulate cortex (ACC) in HRFD while there was no volume alteration in the depressed mothers. ACC has connections with various brain areas including limbic system and dorsolateral prefrontal cortex. These two regions with ACC have significant role in executing working memory, language, attention, and information processing and affect regulation. It is well known that those cognitive functions and affect regulation are impaired in depressive patients. Therefore, it is thought that ACC is one of the major dysfunctional areas in MDD. While the functional neuroimaging studies have consistently supported dysfunctional ACC in MDD patients, such consistency has not emerged from the studies using structural neuroimaging methodology. One study investigated the rostral ACC in healthy children with depressed mood and found significant reduction only in male children3 . In a three generation study, Peterson et al showed cortical thinning in the lateral surface of right hemisphere while thickening in the subgenual, anterior and posterior cingulate cortex4 . With the current non-depressed status of HRFD group, our findings of increased GMV in ACC and entorhinal cortex -which were absent in their depressed mothers- might be related to resilience to depression. We also observed increased GMV in the right entorhinal cortex of HRFD group, which is the main interface between neocortex and hippocampus. It plays an important role in memory formation and consolidation that are impaired in MDD. It has been proposed that volumetric alterations in entorhinal cortex lead to impairment of the cortical-hippocampal circuit and that these structural changes have been implicated in the etiology of depression We found no major GMV difference between MDD and healthy controls in selected brain regions. In this study, we used ROI based approach that gives the mean values of the whole GMV in the selected regions and smaller local alteration might be missed by this method. Despite this disadvantage of ROI approach, it has advantages of better registration over voxel-based approach. It should be kept in mind that we investigated certain areas of brain to test our hypothesis. Therefore, it would be premature to conclude that there is no regional GMV differences between MDD and controls. One strength of this study is to include the mothers and their daughters as couples in both arm of the comparison. On the other hand, the main limitation was small subject number in each group. Obtaining our data from female subjects also limits us to generalize our findings for male high-risk population. As a conclusion, we provide evidence for GMV alterations in high-risk populations for familial depression. As those regional GMV alterations were not observed in depressed mothers, these regional alterations might help the young women at risk to be depressed free. Longitudinal follow-up studies are needed to interpret the high-risk data and determine the anatomical alterations related to vulnerability and resilience to disease.

Nataliya Lavrova, Elena Nikulina

Foreign language acquisition is notoriously constrained by learners’ lack of awareness of the systemic relations that are obtained among stable multiple-unit lexical items. This results in learners’ inability to variegate their performance (both written and oral) with idioms that stand in complementary (synonymy) or contrastive (antonymy) distribution to one another. Nor are learners typically able to distinguish between the multiple senses of English idioms. Given these impedimenta, the present research investigates the degree of entrenchment of idiomatic synonymy, antonymy, and polysemy and, on the back of it, sets the agenda for partial revision of the practice of exposing learners to English idioms. Data were collected to investigate the knowledge of idiomatic synonymy, antonymy, and polysemy amongst Russian EFL learners. The results of the ANOVA analysis revealed that the degree of awareness of the major types of idiomatic paradigmatic relations significantly differed between the groups, with learners being more aware of synonymy and polysemy than antonymy. The findings suggest that current EFL materials and dictionaries need to be updated and revisited with a view to exposing foreign learners to an extended network of paradigmatic idiomatic relations.