This paper presents a novel multistage fine-tuning strategy designed to enhance automatic speech recognition (ASR) performance in low-resource languages using OpenAI’s Whisper model. In this approach we aim to build ASR model for languages with limited digital resources by sequentially adapting the model across linguistically similar languages. We experimented this on the Malasar language, a Dravidian language spoken by approximately ten thousand people in the Western Ghats of South India. The development of speech recognition for Malasar language faces significant barriers due to its lack of written form and the absence of digital speech datasets needed for training. Working in collaboration with Wycliffe India and Malasar community members, we created a spoken Malasar corpus paired with transcription in Tamil script, a closely related major language. In our approach to build ASR model for Malasar, we first built an intermediate Tamil ASR, leveraging higher data availability for Tamil annotated speech. This intermediate model is subsequently fine-tuned on Malasar data, allowing for more effective ASR adaptation despite limited resources. The multistage fine-tuning strategy demonstrated significant improvements over direct fine-tuning on Malasar data alone, achieving a word error rate (WER) of 51.9%, which is 4.5% absolute reduction when compared to the direct fine-tuning method. Further a WER reduction to 47.3% was achieved through punctuation removal in post-processing, which addresses formatting inconsistencies that impact evaluation. Our results underscore the effectiveness of sequential multistage fine-tuning combined with targeted post-processing as a scalable strategy for ASR system development in low-resource languages, especially where linguistic similarities can be leveraged to bridge gaps in training data.

Advancements in natural language processing (NLP) have markedly improved paraphrase generation, an essential task for numerous applications. However, current methods face limitations due to model and constraint specificity, which hinder their flexibility and practical deployment. In this work, we introduce a unified prompt-driven approach to paraphrase generation that leverages diverse prompts, enabling fine-grained user control over aspects such as syntax and sentiment. Moreover, we incorporate translation to enable sophisticated cross-lingual text controls. Our system employs a data-centric paradigm which organizes prompts with natural language instructions. The proposed method is compatible with various sequence-to-sequence architectures and utilizes a novel training strategy to address the versatility of prompt combinations. Empirical results show that our approach not only demonstrates its capacity to adhere to multiple user-defined constraints but also maintains high performance in generation tasks without prompts. Moreover, extensive analysis shows that the model exhibits robustness to prompt variance such as language and quantity.

Metaphor, as a common type of linguistic expression, helps people intuitively understand complex concepts in communication, writing, and cognition. Metaphor components, including source-domain words and target-domain words, are critical elements for metaphor identification and interpretation. This paper focuses on metaphor components and proposes a metaphor components identification framework employing F eedback-enhanced F eature-driven I n- C ontext L earning (FF-ICL) based on the large language model (LLM). Specifically, in-context learning and feedback mechanisms inspired by human learning are integrated. Firstly, a machine feedback mechanism is designed to perform prior predictions on training samples, constructing a candidate demonstration pool enriched with prediction results and feedback information. Secondly, a multi-head graph attention network (GAT) is introduced to capture the linguistic and structural information embedded in metaphorical expressions, producing feature-rich representations and establishing a vector repository. Based on the repository, the framework retrieves demonstrations most relevant to the input query across different feature dimensions, incorporating in-context prompts to effectively fine-tune the LLM. Experiments and analyses on public datasets demonstrate the superiority of FF-ICL. Furthermore, the metaphor concept mapping experiment validates the crucial role of metaphor components in downstream computational metaphor tasks. Relevant data and codes are available at https://github.com/WXLJZ/FF-ICL.

State-of-the-art multilingual Automatic Speech Recognition (ASR) models produce systematic errors when applied to low-resource languages like Rajasthani, for which they lack dedicated training data. This article addresses this challenge by introducing a post-ASR correction framework that leverages the complementary error patterns in the outputs (termed as views) from two distinct models: Whisper-large-v3 and MMS-1B-All. We propose a multi-view, 1 character-level sequence-to-sequence (Seq2Seq) model that uses a gated fusion mechanism to dynamically weigh information from the two ASR outputs. On a new benchmark created from the IndicTTS Rajasthani corpus, our gated model achieves a Character Error Rate (CER) of 7.86% and a Word Error Rate (WER) of 29.98%. This outperforms the best single-view baselines (8.01% CER and 30.33% WER), simple multi-view concatenation (8.21% CER and 30.05% WER), as well as Llama-3.2-3B and mBART-50-large, both fine-tuned on Whisper and MMS inputs. It also surpasses powerful Large Language Models (LLMs) like GPT-4o and Gemini 2.5 Pro in a zero-shot setting. This work establishes the first baseline for post-ASR correction in Rajasthani, demonstrating that a compact, specialized model is more effective than general-purpose LLMs for this targeted, low-resource task.

Emotion classification is a critical component in understanding human expression and interaction in textual data. However, achieving high accuracy in emotion classification, particularly in complex languages like Hindi, presents several challenges. This article proposes a hybrid method for emotion classification in Hindi text, combining a knowledge-based approach for stability and a statistical approach for accuracy. Machine learning (ML) techniques, while powerful, require extensive training data and may not generalize well across different domains. Conversely, lexicon-based methods offer consistent performance across domains but often lack precision. To leverage the advantages of both methods, a hybrid model for emotion analysis in Hindi is presented. Results indicate that the hybrid model outperforms both individual classifiers and the simple lexicon approach in terms of accuracy. Specifically, incorporating Multinomial Naïve Bayes within the hybrid framework yields significant performance improvements. The proposed hybrid model presents a viable and superior approach for Hindi emotion analysis, balancing the need for stability and accuracy across varied textual domains

Metonymy resolution (MR) is a crucial challenge in natural language understanding and information retrieval. Recent large-scale pre-trained language models have shown promising results in various natural language processing (NLP) tasks, including MR. Despite these achievements, current models still struggle in many real-world scenarios. Since these models rely heavily on contextual information and ignore entity information, they are prone to extract irrelevant features and overfit when fine-tuned with less training data. In this article, we propose a mixed information bottleneck framework to address the above issues, which learns optimal data representations based on the principle of minimal sufficiency. Our model can effectively mitigate irrelevant features in context and entity by using different types of information bottlenecks for entity and context information separately while reducing the dimensionality of latent representations. We show that our approach achieves state-of-the-art performance on three benchmark datasets for location MR, outperforming previous Bert-based methods by a large margin. Ablation studies and qualitative analysis show the effectiveness of our models in reducing dimensionality while extracting more relevant features.

In the current development of large language models (LLMs), it is important to ensure the accuracy and reliability of the underlying data sources. LLMs are critical for various applications, but they often suffer from hallucinations and inaccuracies due to knowledge gaps in the training data. Knowledge graphs (KGs), as a powerful structural tool, could serve as a vital external information source to mitigate the aforementioned issues. By providing a structured and comprehensive understanding of real-world data, KGs enhance the performance and reliability of LLMs. However, it is common that errors exist in KGs while extracting triplets from unstructured data to construct KGs. This could lead to degraded performance in downstream tasks such as question-answering and recommender systems. Therefore, anomaly detection in KGs is essential to identify and correct these errors. This article presents an anomaly detection algorithm in knowledge graphs with dual-channel learning (ADKGD). ADKGD leverages a dual-channel learning approach to enhance representation learning from both the entity-view and triplet-view perspectives. Furthermore, using a cross-layer approach, our framework integrates internal information aggregation and context information aggregation. We introduce a kullback-leibler (KL)-loss component to improve the accuracy of the scoring function between the dual channels. To evaluate ADKGD’s performance, we conduct empirical studies on three real-world KGs: WN18RR, FB15K, and NELL-995. Experimental results demonstrate that ADKGD outperforms state-of-the-art anomaly detection algorithms. The source code and datasets are publicly available at https://github.com/csjywu1/ADKGD .

Arabic, as a language, encompasses numerous dialectical variations. Despite sharing a common vocabulary, these variants exhibit significant differences across territories, even within the same country. The Arabic dialects identification (ADI) is important for many NLP tasks and can be conducted as a preliminary step. In this work, we focus on the fine-grained ADI task, investigating the effectiveness of various feature extraction techniques and classification algorithms through a comparative study across three datasets: MADAR, NADI and NADAR. The first two are available online and well known in the literature. While we construct the third one by merging the first two datasets. We experimented with three groups of models: machine learning models, deep learning models and pre-trained language models. Our goal is to analyze the efficiency of these models for a fine-grained ADI task from different perspectives. Specifically, we analyzed the variation in classifier performance based on various criteria: the training set size, the considered dialect, and the dataset type (whether it is parallel, balanced or not). We employed the Explainable AI (XAI) technique LIME to investigate the interpretability of the models. Our analysis of the results yielded several key insights that guided our conclusions : (1) classical ML models have competitive performances compared to sophisticated neuronal models in terms of training time and F1-score. (2) The size of the dataset has a positive impact on the performance of the classifiers but in some experiments, larger datasets resulted in decreased performance. (3) The confusion errors mainly concern the dialects of geographically close regions. (4) Parallel corpora are more adapted to training ADI models.

The Media is stated as the “fourth pillar of democracy" that influences day to day life of each individual. The falling rank of India in the World Press Freedom Index (WPFI) portrays biasness in news articles, especially for political news articles. 1 The biased headlines morph the perception of the people regarding the topic of interest and fill them up with baseless prejudice. Currently, thousands of Hindi newspaper publishers in India have more than 100 million subscribers. 2 Political bias detection from news is a fresh research area, researchers did not considers contextual analysis yet. This is the first political news bias detection model that considers neutrosophic sets for contextual word analysis to address inconsistency and indeterminacy, newly enhanced Hindi SentiWordNet and MuRil language model for encoded output and multiclass classification. This proposed novel model transforms input words into a neutrosophic domain and is characterized by degree of truth, indeterminacy, and false membership components. This set applies the concept of indeterminacy that improves the biasness detection accuracy for each news headline. The experimentation and evaluation of proposed model shows the better accuracy than the state-of-the-art methods, MuRil language model combined with neutrosophic sets improves 37.5% to 50% accuracy.

Word sense disambiguation (WSD) is the process of determining the correct meaning of a word based on its context in a sentence, a task that remains one of the core challenges in natural language processing (NLP) despite the advancements made by LLMs. Although LLMs have improved WSD performance, challenges remain–especially in nuanced contexts, low-resource languages, and domain-specific terminology. These gaps can hinder the accuracy of LLMs in high-stakes applications and multilingual settings. In response to the need for WSD, our study introduces a novel approach to WSD data collection through gamified crowdsourcing, which, to our knowledge, has not been previously applied in this field. A messaging bot method has been used to engage a wide, diverse audience to generate high-quality WSD data. Using a multiplayer game format, native speakers provide examples for different senses of ambiguous words and rate others’ contributions. Together with the suggested enhancements, this platform attracted a diverse range of participants–spanning ages, backgrounds, and genders–20 times more than a similar crowdsourcing method applied to a different NLP task, and sustained their engagement over an extended period. Unlike conventional academic crowdsourcing pools, this approach focuses on drawing individuals from varied backgrounds beyond academia or AI-focused communities. It not only gathered data but also encouraged participants to evaluate each other’s contributions, creating a rich and reliable dataset. Our findings suggest that gamified crowdsourcing can be a powerful tool for creating WSD corpora. Our approach not only supports future WSD research but also contributes valuable training data for developing more precise and resilient language models.