Search

In the realm of legal artificial intelligence (AI), the spotlight has been cast on its remarkable precision and efficiency, especially in tasks such as similar case retrieval where the identification of pertinent cases in response to a given query is of paramount importance. This task, distinct from traditional text retrieval, presents a set of unique challenges that necessitate the availability of high-quality, annotated datasets to facilitate efficient model training. The intricacies of handling extended queries and candidate documents, coupled with the varied interpretations of similarity, further compound the complexity of this endeavor. This study introduces an innovative training approach, combining dense and sparse retrieval methods. Utilizing a sparse retrieval model, we extract unlabeled data from extensive legal cases. Subsequently, a dense retrieval model screens this data, merging it with labeled data to create pseudo-labeled data, iteratively training until convergence. The results demonstrate exceptional performance in the Chinese law retrieval task dataset, showcasing a notable 3.66% precision enhancement and a substantial 3.62% improvement in mean average precision (MAP). However, the dataset’s imbalance across different charges of cases poses a challenge, potentially affecting retrieval performance for long-tailed legal cases. Nonetheless, these outcomes signify accelerated and more efficient retrieval of similar cases for legal professionals. Additionally, they provide high-quality references for non-legal individuals lacking expertise in the field.

The explanation of legal judgments is crucial for the transparency, fairness, and trustworthiness, aiming to provide rationales for decision-making. While previous works have focused on improving the accuracy of legal judgment prediction, the lack of explainability seriously limits the practical application of these methods. Many researchers have dedicated effort to extracting or generating rationales as explanations for legal judgments, ignoring the factual consistency of these rationales. Inconsistencies between rationales and fact descriptions severely impact their applicability. To address these issues, we investigate Event Chain – ordered sequences of events related to criminal behavior – as an intermediate representation in the fact descriptions, to enhance the representation of causal relationships among events, and focus on crucial events within the fact description. Specifically, we propose a multi-task learning approach, dubbed LegalMind, that introduces Event Chain as an auxiliary task and jointly models Event Chain and rationale in a unified decoder, to improve factual consistency in rationale generation. The experiment results show that our model outperforms the state-of-the-art methods, achieving improvements of 7.65% and 6.65% in AVG-BLEU on the CJO dataset and LAIC2021 dataset respectively compared to BART-C3VG. Furthermore, in comparison to BART-C3VG, our model demonstrated superior performance with increases of 6.17% and 8.33% in factual consistency on the CJO and LAIC2021 datasets, respectively.

Emotion recognition constitutes a pivotal research topic within affective computing, owing to its potential applications across various domains. Currently, emotion recognition methods based on deep learning frameworks utilizing electroencephalogram (EEG) signals have demonstrated effective application and achieved impressive performance. However, in EEG-based emotion recognition, there exists a significant performance drop in cross-subject EEG Emotion recognition due to inter-individual differences among subjects. In order to address this challenge, a hybrid transfer learning strategy is proposed, and the Domain Adaptation with a Few-shot Fine-tuning Network (DFF-Net) is designed for cross-subject EEG emotion recognition. The first step involves the design of a domain adaptive learning module specialized for EEG emotion recognition, known as the Emo-DA module. Following this, the Emo-DA module is utilized to pre-train a model on both the source and target domains. Subsequently, fine-tuning is performed on the target domain specifically for the purpose of cross-subject EEG emotion recognition testing. This comprehensive approach effectively harnesses the attributes of domain adaptation and fine-tuning, resulting in a noteworthy improvement in the accuracy of the model for the challenging task of cross-subject EEG emotion recognition. The proposed DFF-Net surpasses the state-of-the-art methods in the cross-subject EEG emotion recognition task, achieving an average recognition accuracy of 93.37% on the SEED dataset and 82.32% on the SEED-IV dataset.

Legal judgment prediction (LJP) automatically predicts the judgment results of a legal case based on its fact description, which has excellent prospects in judicial assistance systems and consultation services for the public. Most previous studies either focused on enhancing LJP’s performance while ignoring the issue of confusing charges and law articles, or only used law articles to improve the judgment of confusing verdicts, resulting in the limited model performance. This paper introduces legal charge knowledge as a type of knowledge to enhance the representation of fact descriptions and incorporates it into deep neural networks. We then propose a Knowledge-enriched Multi-Cross Attention Network (KEMCAN) to improve LJP’s performance, and resolve legal cases involving confusing charges and law articles. Specifically, a cross-attention mechanism is proposed to model the relationship between legal charge knowledge and fact description in a unified model. The experimental results demonstrate that our model outperforms the state-of-the-art methods on two real-world datasets, achieving an average improvement of 3.95% in macro-F1 for charge prediction and 1.98% for law article prediction.

Electroencephalogram (EEG) signals have emerged as an important tool for emotion research due to their objective reflection of real emotional states. Deep learning-based EEG emotion classification algorithms have made encouraging progress, but existing models struggle with capturing long-range dependence and integrating temporal, frequency, and spatial domain features that limit their classification ability. To address these challenges, this study proposes a Bi-branch Vision Transformer- based EEG emotion recognition model, Bi-ViTNet, that integrates spatial-temporal and spatial-frequency feature representations. Specifically, Bi-ViTNet is composed of spatial-frequency feature extraction branch and spatial-temporal feature extraction branch that fuse spatial-frequency-temporal features in a unified framework. Each branch is composed of Linear Embedding and Transformer Encoder, which is used to extract spatial-frequency features and spatial-temporal features. Finally, fusion and classification are performed by the Fusion and Classification layer. Experiments on SEED and SEED-IV datasets demonstrate that Bi-ViTNet outperforms state-of-the-art baselines.

In this technological era, smart and intelligent systems that are integrated with artificial intelligence (AI) techniques, algorithms, tools, and technologies, have impact on various aspects in our daily life. Communication and interaction between human and machine using speech become increasingly important, since it is an obvious substitute for keyboards and screens in the communication process. Therefore, numerous technologies take advantage of speech such as Automatic Speech Recognition (ASR), where human natural speech for many languages is used as the means to interact with machines. Majority of the related works on ASR concentrate on the development and evaluation of ASR systems that serve a single language only, such as Arabic, English, Chinese, French, and many others. However, research attempts that combine multiple languages (bilingual and multilingual) during the development and evaluation of ASR systems are very limited. This paper aims to provide comprehensive research background and fundamentals of bilingual ASR, and related works that have combined two languages for ASR tasks from 2010 to 2021. It also formulates research taxonomy and discusses open challenges to the bilingual ASR research. Based on our literature investigation, it is clear that bilingual ASR using deep learning approach is highly demanded and is able to provide acceptable performance. In addition, many combinations of two languages such as Arabic-English, Arabic-Malay, and others, are still limited, which can open new research opportunities. Finally, it is clear that ASR research is moving towards not only bilingual ASR, but also multilingual ASR.

Approaches to automatic speech recognition have transited from Hidden Markov Model (HMM)-based ASR to deep neural networks. The advantages of deep neural network approaches are that they can be developed quickly and perform better given large language resources. Nevertheless, dialect speech recognition is still challenging due to the limited resources. Transfer learning approaches have been proposed to improve speech recognition for low resources. In the first approach, the model is pre-trained on a large and diverse labeled dataset to learn the acoustic and language patterns from the speech signal. Then, the model parameters are updated with a new dataset, and the pre-trained model is fine-tuned on a low-resource language dataset. The fine-tuning process is usually completed by freezing the pre-trained layers and training the remaining layers of the model on the low-resource language corpus. Another approach is to use a pre-trained model to capture the compact and meaningful features as input to the encoder. Pre-training in this approach usually involves using unsupervised learning methods to train models on a corpus of large amounts of unmarked data. It enables the model to learn the general patterns and relationships between the input speech signals. This paper proposes a training recipe using transfer learning and Standard Malay models to improve automatic speech recognition for Kelantan and Sarawak Malay dialects. [ABSTRACT FROM AUTHOR]

Parallel texts corpora are essential resources in linguistics and natural language processing, especially in translation and multilingual information retrieval. The publicly available parallel text corpora are limited to certain genres, types and domains. Furthermore, the parallel dialect text is scarce, even though they are important in the analysis and study of a dialect. Collecting parallel dialect text is challenging because dialects typically appear in the form of speech and very limited dialectic texts exist. Moreover, there is no standard orthography in most dialects. The contributions of this paper are threefold. First, the paper describes a methodology in acquiring a parallel text corpus of Standard Malay and Malay dialects, particularly Kelantan Malay and Sarawak Malay. Second, we propose a hybrid of distancebased and statistical-based alignment algorithm to align words and phrases the parallel text. The results show that the precision and recall values of the proposed alignment algorithm are more than 95% and better than the state-of-the-art GIZA++. Third, the alignment obtained were compared to find out the lexical similarities and differences between Standard Malay and the two studied Malay dialects, contributing valuable insights into the linguistic variations within the Malay language family. [ABSTRACT FROM AUTHOR]