ACM Transactions on Asian and Low-Resource Language Information Processing最新文献

Manchu language, a minority language of China, is of significant historical and research value. An increasing number of Manchu documents are digitized into image format for better preservation and study. Recently, many researchers focused on identifying Manchu words in digitized documents. In previous approaches, a variety of Manchu words are recognized based on visual cues. However, we notice that visual-based approaches have some obvious drawbacks. On one hand, it is difficult to distinguish between similar and distorted letters. On the other hand, portions of letters obscured by breakage and stains are hard to identify. To cope with these two challenges, we propose a visual-language framework, namely the Visual-Language framework for Manchu word Recognition (VLMR), which fuses visual and semantic information to accurately recognize Manchu words. Whenever visual information is not available, the language model can automatically associate the semantics of words. The performance of our method is further enhanced by introducing a self-knowledge distillation network. In addition, we created a new handwritten Manchu word dataset named (HMW), which contains 6,721 handwritten Manchu words. The novel approach is evaluated on WMW and HMW. The experiments show that our proposed method achieves state-of-the-art performance on both datasets.

Primary school English teaching resources play an important role in primary school English teaching. The information age requires that primary school English teaching should strengthen the use of multimedia resources and gradually realize the diversification of teaching content. Expanded reality innovation is a sort of mixture picture handling innovation, which is one of the significant innovations that would influence the improvement of fundamental schooling in the following five years. It can seamlessly output virtual objects to the real environment, which is convenient for this paper to obtain and absorb information. It can also help students to participate in exploration and cultivate their creativity and imagination. It can strengthen the cooperation between students and teachers and create various learning environments. It has an immeasurable prospect of development in the field of education. The primary school English teaching resources based on augmented reality create a realistic learning situation from two-dimensional plane to three-dimensional three-dimensional display, and enrich the presentation of primary school English teaching content. It can stimulate students’ interest in learning English and promote the transformation of English teaching methods. It is a useful attempt in the field of education. This paper made statistics on the test results of the experimental class and the control class. Most of the scores of the experimental group were between 71 and 100, a total of 27, accounting for 67.5%. The score distribution of the control class was relatively balanced, with the highest number between 61-70, and the number was 10, accounting for 25%. Therefore, it can be seen that hybrid image processing technology is important for interactive English teaching.

Social media is a widely used platform that includes a vast amount of user-generated content, allowing the extraction of information about users’ thoughts from texts. Individuals freely express their thoughts on these platforms, often without constraints, even if the content is offensive or contains hate speech. The identification and removal of offensive content from social media are imperative to prevent individuals or groups from becoming targets of harmful language. Despite extensive research on offensive content detection, addressing this challenge in code-mixed languages remains unsolved, characterised by issues such as imbalanced datasets and limited data sources. Most previous studies on detecting offensive content in these languages focus on creating datasets and applying deep neural networks, such as Recurrent Neural Networks (RNNs), or pre-trained language models (PLMs) such as BERT and its variations. Given the low-resource nature and imbalanced dataset issues inherent in these languages, this study delves into the efficacy of the syntax-aware BERT model with continual pre-training for the accurate identification of offensive content and proposes a framework called Cont-Syntax-BERT by combining continual learning with continual pre-training. Comprehensive experimental results demonstrate that the proposed Cont-Syntax-BERT framework outperforms state-of-the-art approaches. Notably, this framework addresses the challenges posed by code-mixed languages, as evidenced by its proficiency on the DravidianCodeMix [10,19] and HASOC 2109 [37] datasets. These results demonstrate the adaptability of the proposed framework in effectively addressing the challenges of code-mixed languages.

Time-sensitive question answering is to answer questions limited to certain timestamps based on the given long document, which mixes abundant temporal events with an explicit or implicit timestamp. While existing models make great progress in answering time-sensitive questions, their performance degrades dramatically when a long distance separates the correct answer from the timestamp mentioned in the question. In this paper, we propose a Context-enhanced Adaptive Graph network (CoAG) to capture long-distance dependencies between sentences within the extracted question-related episodes. Specifically, we propose a time-aware episode extraction module that obtains question-related context based on timestamps in the question and document. As the involvement of episodes confuses sentences with adjacent timestamps, an adaptive message passing mechanism is designed to capture and transfer inter-sentence differences. In addition, we present a hybrid text encoder to highlight question-related context built on global information. Experimental results show that CoAG significantly improves compared to state-of-the-art models on five benchmarks. Moreover, our model has a noticeable advantage in solving long-distance time-sensitive questions, improving the EM scores by 2.03% to 6.04% on TimeQA-Hard.

Predicting information cascades holds significant practical implications, including applications in public opinion analysis, rumor control, and product recommendation. Existing approaches have generally overlooked the significance of semantic topics in information cascades or disregarded the dissemination relations. Such models are inadequate in capturing the intricate diffusion process within an information network inundated with diverse topics. To address such problems, we propose a neural-based model (named ICP-TMAN) using Topic-Aware Masked Attentive Network for Information Cascade Prediction to predict the next infected node of an information cascade. First, we encode the topical text into user representation to perceive the user-topic dependency. Next, we employ a masked attentive network to devise the diffusion context to capture the user-context dependency. Finally, we exploit a deep attention mechanism to model historical infected nodes for user embedding enhancement to capture user-history dependency. The results of extensive experiments conducted on three real-world datasets demonstrate the superiority of ICP-TMAN over existing state-of-the-art approaches.

Aspect-based sentiment classification (ABSC) is a crucial subtask of fine-grained sentiment analysis (SA), which aims to predict the sentiment polarity of the given aspects in a sentence as positive, negative, or neutral. Most existing ABSC methods based on supervised learning. However, these methods rely heavily on fine-grained labeled training data, which can be scarce in low-resource domains, limiting their effectiveness. To overcome this challenge, we propose a low-resource cross-domain aspect-based sentiment classification (CDABSC) approach based on a pre-training and fine-tuning strategy. This approach applies the pre-training and fine-tuning strategy to an advanced deep learning method designed for ABSC, namely the attention-based encoding graph convolutional network (AEGCN) model. Specifically, a high-resource domain is selected as the source domain, and the AEGCN model is pre-trained using a large amount of fine-grained annotated data from the source domain. The optimal parameters of the model are preserved. Subsequently, a low-resource domain is used as the target domain, and the pre-trained model parameters are used as the initial parameters of the target domain model. The target domain is fine-tuned using a small amount of annotated data to adapt the parameters to the target domain model, improving the accuracy of sentiment classification in the low-resource domain. Finally, experimental validation on two domain benchmark datasets, restaurant and laptop, demonstrates that significant outperformance of our approach over the baselines in CDABSC Micro-F1.

Token-level data augmentation generates text samples by modifying the words of the sentences. However, data that are not easily classified can negatively affect the model. In particular, not considering the role of keywords when performing random augmentation operations on samples may lead to the generation of low-quality supplementary samples. Therefore, we propose a supervised contrast learning text classification model based on data quality augment (DQA). First, dynamic training is used to screen high-quality datasets containing beneficial information for model training. The selected data is then augmented with data based on important words with tag information. To obtain a better text representation to serve the downstream classification task, we employ a standard supervised contrast loss to train the model. Finally, we conduct experiments on five text classification datasets to validate the effectiveness of our model. In addition, ablation experiments are conducted to verify the impact of each module on classification.

This paper studies entity linking (EL) in Web tables, which aims to link the string mentions in table cells to their referent entities in a knowledge base. Two main problems exist in previous studies: 1) contextual information is not well utilized in mention-entity similarity computation; 2) the assumption on entity coherence that all entities in the same row or column are highly related to each other is not always correct. In this paper, we propose NPEL, a new Neural Paired Entity Linking framework, to overcome the above problems. In NPEL, we design a deep learning model with different neural networks and an attention mechanism, to model different kinds of contextual information of mentions and entities, for mention-entity similarity computation in Web tables. NPEL also relaxes the above assumption on entity coherence by a new paired entity linking algorithm, which iteratively selects two mentions with the highest confidence for EL. Experiments on real-world datasets exhibit that NPEL has the best performance compared with state-of-the-art baselines in different evaluation metrics.

During the last decade, social media has gained significant popularity as a medium for individuals to express their views on various topics. However, some individuals also exploit the social media platforms to spread hatred through their comments and posts, some of which target individuals, communities or religions. Given the deep emotional connections people have to their religious beliefs, this form of hate speech can be divisive and harmful, and may result in issues of mental health as social disorder. Therefore, there is a need of algorithmic approaches for the automatic detection of instances of hate speech. Most of the existing studies in this area focus on social media content in English, and as a result several low-resource languages lack computational resources for the task. This study attempts to address this research gap by providing a high-quality annotated dataset designed specifically for identifying hate speech against religions in the Hindi-English code-mixed language. This dataset “Targeted Hate Speech Against Religion” (THAR)) consists of 11,549 comments and has been annotated by five independent annotators. It comprises two subtasks: (i) Subtask-1 (Binary classification), (ii) Subtask-2 (multi-class classification). To ensure the quality of annotation, the Fleiss Kappa measure has been employed. The suitability of the dataset is then further explored by applying different standard deep learning, and transformer-based models. The transformer-based model, namely Multilingual Representations for Indian Languages (MuRIL), is found to outperform the other implemented models in both subtasks, achieving macro average and weighted average F1 scores of 0.78 and 0.78 for Subtask-1, and 0.65 and 0.72 for Subtask-2, respectively. The experimental results obtained not only confirm the suitability of the dataset but also advance the research towards automatic detection of hate speech, particularly in the low-resource Hindi-English code-mixed language.