Sentence Similarity Research Articles

Design of a turbo-based deep semantic autoencoder for marine Internet of Things

With the rapid growth of the global marine economy and flourishing maritime activities, the marine Internet of Things (IoT) is gaining unprecedented momentum. However, current marine equipment is deficient in data transmission efficiency and semantic comprehension. To address these issues, this paper proposes a novel End-to-End (E2E) coding scheme, namely the Turbo-based Deep Semantic Autoencoder (Turbo-DSA). The Turbo-DSA achieves joint source-channel coding at the semantic level through the E2E design of transmitter and receiver, while learning to adapt to environment changes. The semantic encoder and decoder are composed of transformer technology, which efficiently converts messages into semantic vectors. These vectors are dynamically adjusted during neural network training according to channel characteristics and background knowledge base. The Turbo structure further enhances the semantic vectors. Specifically, the channel encoder utilizes Turbo structure to separate semantic vectors, ensuring precise transmission of meaning, while the channel decoder employs Turbo iterative decoding to optimize the representation of semantic vectors. This deep integration of the transformer and Turbo structure is ensured by the design of the objective function, semantic extraction, and the entire training process. Compared with traditional Turbo coding techniques, the Turbo-DSA shows a faster convergence speed, thanks to its efficient processing of semantic vectors. Simulation results demonstrate that the Turbo-DSA surpasses existing benchmarks in key performance indicators, such as bilingual evaluation understudy scores and sentence similarity. This is particularly evident under low signal-to-noise ratio conditions, where it shows superior text semantic transmission efficiency and adaptability to variable marine channel environments.

A representation learning-based approach to enhancing manufacturing quality for low-voltage electrical products

In low-voltage electrical product manufacturing, resolving quality issues is heavily reliant on engineering experience, and can be time-consuming and error-prone. Through quality management systems, a large number of historical defect cases can be consolidated for analysis along with relevant causes. However, these defect descriptions are often casually described with a mix of Chinese and English language, containing domain-specific terms. Additionally, defect product features have varieties and complex relationships. Therefore, historical defect cases have not been effectively utilized to support manufacturing quality issues. To address this challenge, this study proposes a representation learning-based approach to enhance manufacturing quality. Key research contributions include: (1) A two-stage word embedding technique based on the pre-trained language model. First, TSDAE is utilized for unsupervised pre-training on a large amount of unlabeled data. Then, Sentence-BERT is utilized for fine-tuning on a small set of labeled similar sentence pairs. This process yields a pre-trained language model specific to low-voltage electrical product defects. (2) NSHPSAGE graph embedding model based on the constructed product feature knowledge graph. We select more valuable neighboring nodes during sampling and explore different aggregation functions to enhance graph embedding performance. This model effectively aggregates product feature information into “Defect_Case” nodes, yielding graph embedding vectors. The model exhibits good Weighted-Precision and Weighted-Recall with a short training duration, and it can handle new nodes, addressing the issue of heterogeneous graph embedding. (3) A defect case recommendation technique that fuses word embedding and graph embedding. We use Multi-Head Attention Fusion in the late-fusion to obtain defect case vectors. This approach comprehensively considers defect description semantic knowledge and complex product feature relationships, enabling accurate defect case recommendation with the prototype system.

Unveiling the clinical incapabilities: a benchmarking study of GPT-4V(ision) for ophthalmic multimodal image analysis

PurposeTo evaluate the capabilities and incapabilities of a GPT-4V(ision)-based chatbot in interpreting ocular multimodal images.MethodsWe developed a digital ophthalmologist app using GPT-4V and evaluated its performance with a dataset (60...

Compositionality and Sentence Meaning: Comparing Semantic Parsing and Transformers on a Challenging Sentence Similarity Dataset

Abstract One of the major outstanding questions in computational semantics is how humans integrate the meaning of individual words into a sentence in a way that enables understanding of complex and novel combinations of words, a phenomenon known as compositionality. Many approaches to modelling the process of compositionality can be classified as either ‘vector-based’ models, in which the meaning of a sentence is represented as a vector of numbers, or ‘syntax-based’ models, in which the meaning of a sentence is represented as a structured tree of labelled components. A major barrier in assessing and comparing these contrasting approaches is the lack of large, relevant datasets for model comparison. This paper aims to address this gap by introducing a new dataset, STS3k, which consists of 2,800 pairs of sentences rated for semantic similarity by human participants. The sentence pairs have been selected to systematically vary different combinations of words, providing a rigorous test and enabling a clearer picture of the comparative strengths and weaknesses of vector-based and syntax-based methods. Our results show that when tested on the new STS3k dataset, state-of-the-art transformers poorly capture the pattern of human semantic similarity judgments, while even simple methods for combining syntax- and vector-based components into a novel hybrid model yield substantial improvements. We further show that this improvement is due to the ability of the hybrid model to replicate human sensitivity to specific changes in sentence structure. Our findings provide evidence for the value of integrating multiple methods to better reflect the way in which humans mentally represent compositional meaning.

Scan pattern similarity predicts the semantic similarity of sentences across languages above and beyond their syntactic structures.

Scan pattern similarity predicts the semantic similarity of sentences across languages above and beyond their syntactic structures.

B - 51 Are my Constructs Bass-Ackwards? A Hierarchical Approach to Neuropsychological Factors in Traumatic Brain Injury

Abstract Objective This study examined neuropsychological measures within a hierarchical framework in a traumatic brain injury (TBI) sample. Methods 270 individuals (M = 32 years old, SD = 14.8; 70% male, 94% white) with mixed TBI severity (73.7% mild TBI) completed the Meyers Neuropsychological Battery (MNB). Principal Components Analyses (PCAs) were conducted using Goldberg’s (2006) “bass-ackwards” hierarchical approach commonly used to evaluate “stratum” in personality and psychopathology. This involves an iterative top-down approach extracting a single general factor, then using the same data to extract two factors, then three factors, and continues as long as theoretically feasible. Extracted factors are saved as factor scores and correlated to reveal pathways between stratum. Results Iterative PCAs revealed a 4-stratum model. The first unrotated principal component (FUPC) was defined by loadings for Arithmetic, AVLT, RCFT, and Block Design. The second stratum factors were defined by AVLT and BNT (factor 1) and RCFT and Block Design (factor 2). The third stratum was defined by RCFT, Block Design, and Finger Localization nondominant (factor 1); Information and BNT (factor 2); and AVLT, Coding, and Finger Tapping dominant (factor 3). Stratum 4 was RCFT and Block Design (factor 1); AVLT (factor 2); Sentence Repetition, Similarities, and Information (factor 3); and Finger Tapping bilateral, Coding, and Trails Making Test (factor 4). Correlational pathways between factors ranged from 0.50 to 0.98. Conclusion Factors at lower stratum generally aligned with functions for cortical lobes. For example, Stratum 4 can be conceptualized as right temporoparietal (factor 1), left temporal (factor 2), left frontotemporal (factor 3), and bilateral frontal (factor 4).

Cross-lingual dependency parsing for a language with a unique script

Abstract Syntactic parsing is one of the areas in Natural Language Processing. The development of large-scale multilingual language models has enabled cross-lingual parsing approaches, which allows us to develop parsers for languages that do not have treebanks available. However, these approaches rely on the assumption that languages share orthographic representations and lexical entries. In this article, we investigate methods for developing a dependency parser for Xibe, a low-resource language that is written in a unique script. We first investigate lexicalized monolingual dependency parsing experiments to examine the effectiveness of word, part-of-speech, and character embeddings as well as pre-trained language models. Results show that character embeddings can significantly improve performance, while pre-trained language models decrease performance since they do not recognize the Xibe script. We also train delexicalized monolingual models, which yield competitive results to the best lexicalized model. Since the monolingual models are trained on a very small training set, we also investigate lexicalized and delexicalized cross-lingual models. We use six closely related languages as source language, which cover a wide range of scripts. In this setting, the delexicalized models achieve higher performance than lexicalized models. A final experiment shows that we can increase performance of the cross-lingual model by combining source languages and selecting the most similar sentences to Xibe as training set. However, all cross-lingual parsing results are still considerably lower than the monolingual model. We attribute the low performance of cross-lingual methods to syntactic and annotation differences as well as to the impoverished input of Universal Dependency Part-of-Speech tags that the delexicalized model has access to.

Rate of speech affects the comprehension of pronouns in children with developmental language disorder.

This study examined whether children with Developmental Language Disorder (DLD) have knowledge of binding principles (i.e., linking pronouns to their structurally licensed antecedent) during real-time sentence processing (cross-modal priming, real-time) and overt comprehension (sentence-picture matching, interpretative) and whether rate of speech impacted access to that knowledge. Fourteen children with DLD participated in two experiments, with sentences presented auditorily at either a regular or slow speech rate. Sentences were matched except to contain a pronoun, reflexive, or noun phrase (control) in the same syntactic position. Experiment (1) used a cross-modal picture priming paradigm to test real-time pronoun-antecedent linking abilities at both rates of speech. Children were instructed to make a binary decision during the uninterrupted auditory presentation of a sentence to a visually presented image (of the antecedent) at the offset of a pronoun, a reflexive, or a control noun. Response times between conditions (e.g., pronoun vs. control noun) were compared to determine whether participants showed evidence of facilitative priming (faster response times in the pronoun than control noun condition) at either speech rate. Experiment (2) used an auditory sentence-picture-matching task to test final comprehension of similar sentences containing a pronoun or reflexive. Accuracy was compared across both speech rates. For Experiment (1), children with DLD did not show evidence of real-time pronoun-antecedent priming at the regular speech rate. However, when sentences were slowed, they showed facilitative priming for the pronoun condition. For experiment (2), children with DLD performed at-chance when interpreting sentences with pronouns regardless of speech rate. While children with DLD have been shown to have difficulty processing sentences containing anaphors (such as pronouns), results suggest that this is not due to loss of intrinsic knowledge of binding principles. By slowing the rate of speech input, we showed that children with DLD do have access to that knowledge and can make the correct link during real-time processing between a pronoun and its structurally licensed antecedent (Experiment 1) but need more time to do so. However, the effect of slowed speech input does not extend to final comprehension (Experiment 2).

Market intelligence applications leveraging a product-specific Sentence-RoBERTa model

Market intelligence, which collects and analyzes market trends and competitive landscape, is crucial for business success in the market. In particular, defining market scope is important because the market analysis outcomes vary depending on where the market boundaries are. However, this has been challenging due to the difficulty in scoping a number of similar products into the target market appropriately. To address this challenge, this study devised a novel method for acquiring reliable market intelligence utilizing the Sentence-RoBERTa (SRoBERTa) model, a pre-trained language model (PLM) specialized in computing sentence similarity. The SRoBERTa model was fine-tuned with 26,248,771 data on product relations and built a product-specific SRoBERTa model (named Pro-SRoBERTa) for effectively identifying the products that fall within the market scope. It outperformed the untrained model by up to 82.8 % on the hierarchical product categorization task. Finally, the Pro-SRoBERTa model was applied to determine the scope of the market followed by obtaining diverse market intelligence. The analysis module was successfully implemented in the market information system using a large-scale, real-world dataset.

Mixture-of-languages Routing for Multilingual Dialogues

We consider multilingual dialogue systems and ask how the performance of a dialogue system can be improved by using information that is available in other languages than the language in which a conversation is being conducted. We adopt a collaborative chair-experts framework, where each expert agent can be either monolingual or cross-lingual, and a chair agent follows a mixture-of-experts procedure for globally optimizing multilingual task-oriented dialogue systems. We propose a mixture-of-languages routing framework that includes four functional components, i.e., input embeddings of multilingual dialogues, language model, pairwise alignment between the representation of every two languages, and mixture-of-languages. We quantify language characteristics of unity and diversity using a number of similarity metrics, i.e., genetic similarity, and word and sentence similarity based on embeddings. Our main finding is that the performance of multilingual task-oriented dialogue systems can be greatly impacted by three key aspects, i.e., data sufficiency, language characteristics, and model design in a mixture-of-languages routing framework.

Text document clustering using mayfly optimization algorithm with k-means technique

Text clustering is a subfield of machine learning (ML) and natural language processing (NLP) that consists of grouping similar sentences or documents based on their content. However, insignificant features in the documents minimize the accuracy of information retrieval which makes it challenging for the clustering approach to efficiently cluster similar documents. In this research, the mayfly optimization algorithm (MOA) with a k-means approach is proposed for text document clustering (TDC) to effectively cluster similar documents. Initially, the data is obtained from Reuters-21678, 20-Newsgroup, and BBC sports datasets, and then pre-processing is established by stemming and stop word removal to remove unwanted phrases or words. The data imbalance approach is established using an adaptive synthetic sampling algorithm (ADASYN), then term frequency-inverse document frequency (TD-IDF) and WordNet features are employed for extracting features. Finally, MOA with the K-means technique is utilized for TDC. The proposed approach achieves better accuracy of 99.75%, 99.54%, and 98.24% when compared to the existing techniques like fuzzy rough set-based robust nearest neighbor-convolutional neural network (FRS-RNN-CNN), TopicStriker, Modsup-based frequent itemset, and rider optimization-based moth search algorithm (Modsup-Rn-MSA), hierarchical dirichlet-multinomial mixture, and multi-view clustering via consistent and specific non-negative matrix (MCCS).

Challenges on Translating the Elements of Traditional Chinese Medicine in Hong Lou Meng

This study aimed to investigate the challenges associated with translating elements of Traditional Chinese Medicine (TCM) within the context of Hong Lou Meng. Building upon the identified challenges, the study delved into solutions aimed at employing effective strategies in the translation of diverse Traditional Chinese Medicine. This study employed a qualitative method to investigate the translation of TCM terms in the English version translated by Hawkes and Minford (2014) from the perspective of translation aesthetics. The study carefully chose representative examples from different facets of TCM in the source text, encompassing TCM theories, disease diagnosis and syndromes, and terminologies and formulas. The translated content was juxtaposed with the source text to scrutinize the translation methods employed in the translation. Based on this analysis, the study assessed both the achievements and shortcomings in translating various TCM terms from the perspective of translation aesthetics. Ultimately, the study consolidated solutions and compensation methods on how to retain the aesthetic and cultural value of Chinese medicine elements in the translation of classical literature. The study findings indicate that the challenges of translating TCM lie in conveying the difficult TCM theories, the recreation of the strict and beautiful language structures in disease diagnosis and syndromes, and the accurate reproduction of TCM terminologies. It shows that to overcome the challenges, translators should adopt various methods flexibly, such as annotating the TCM theories, recreating the content using similar sentence structures by adopting domestication plus transliteration, or foreignization plus annotation to translate Chinese medicine terminologies and formulas.

Shared representations of human actions across vision and language

Humans can recognize and communicate about many actions performed by others. How are actions organized in the mind, and is this organization shared across vision and language? We collected similarity judgments of human actions depicted through naturalistic videos and sentences, and tested four models of action categorization, defining actions at different levels of abstraction ranging from specific (action verb) to broad (action target: whether an action is directed towards an object, another person, or the self). The similarity judgments reflected a shared organization of action representations across videos and sentences, determined mainly by the target of actions, even after accounting for other semantic features. Furthermore, language model embeddings predicted the behavioral similarity of action videos and sentences, and captured information about the target of actions alongside unique semantic information. Together, our results show that action concepts are similarly organized in the mind across vision and language, and that this organization reflects socially relevant goals.

A BERT-GRU Model for Measuring the Similarity of Arabic Text

Semantic Textual Similarity (STS) aims to assess the semantic similarity between two pieces of text. As a challenging task in natural language processing, various approaches for STS in high-resource languages, such as English, have been proposed. In this paper, we are concerned with STS in low resource languages such as Arabic. A baseline approach for STS is based on vector embedding of the input text and application of similarity metric on the embedding space. In this contribution, we propose a cross-encoder neural network (Cross-BERT-GRU) to handle semantic similarity of Arabic sentences that benefits from both the strong contextual understanding of BERT and the sequential modeling capabilities of GRU. The architecture begins by inputting the BERT word embeddings for each word into a GRU cell to model long-term dependencies. Then, max pooling and average pooling are applied to the hidden outputs of the GRU cell, serving as the sentence -pair encoder. Finally, a softmax layer is utilized to predict the degree of similarity. The experiment results show a Spearman correlation coefficient of around 0.9 and that Cross-BERT-GRU outperforms the other BERT models in predicting the semantic textual similarity of Arabic sentences. The experimentation results also indicate that the performance improves by integrating data augmentation techniques.

Editorial

Dear Readers, It gives me great pleasure to announce the sixth regular issue of 2024. In this issue, 6 papers cover various topical aspects of computer science by 19 authors from 7 countries: Brazil, Chile, France, India, Saudi Arabia, Tunisia, and Turkey. In an ongoing effort to further strengthen our journal, I would like to expand the editorial board: If you are a tenured associate professor or above with a strong publication record, you are welcome to apply to join our editorial board. We are also interested in high-quality proposals for special issues on new topics and trends. As always, I would like to thank all authors for their sound research and the editorial board and our guest reviewers for their extremely valuable review effort and suggestions for improvement. These contributions, together with the generous support of the consortium members, help to maintain the quality of our journal.  In this regular issue, I am very pleased to introduce the following 6 accepted articles: George Marsicano, Edna Dias Canedo, Glauco V. Pedrosa, Cristiane S. Ramos, and Rejane M. da C. Figueiredo from Brazil look in their study into digital transformation of public services in a startup-based environment by 23 focus groups and 175 participants in total. Mauricio Solar and Pablo Aguirre from Chile discuss their research on 3D chest CT processes applying a ResNet-50 model to which a new dimension of information has been added, namely a simple autoencoder. In a collaborative work between researchers from Tunisia and France, Rakia Saidi, Fethi Jarray, and Didier Schwab propose in their article a cross-encoder neural network (Cross-BERT-GRU) to deal with the semantic similarity of Arabic sentences that benefits from both the strong contextual understanding of BERT and the sequential modeling capabilities of GRU. Also in a collaborative research between institutions from Tunisia and Saudi Arabia, Nozha Jlidi, Sameh Kouni, Olfa Jemai, and Tahani Bouchrika present their research on MediaPipe with GNN for human activity recognition. G.V.Vidya Lakshmi and S. Gopikrishnan from India look into missing values research for the IoT domain and in particular present IMD-MP technique that improves imputation accuracy for big data analysis in IoT applications based on spatial-temporal correlations. Last but not least, F. Didem Alay, Nagehan İlhan, and M. Tahir Güllüoğlu address in their article a comparative study of data mining methods for solar radiation and temperature forecasting models. Enjoy Reading Cordially,  Christian Gütl, Managing Editor Graz University of Technology, Graz, Austria

Fairness Testing of Machine Translation Systems

Machine translation is integral to international communication and extensively employed in diverse human-related applications. Despite remarkable progress, fairness issues persist within current machine translation systems. In this article, we propose FairMT, an automated fairness testing approach tailored for machine translation systems. FairMT operates on the assumption that translations of semantically similar sentences, containing protected attributes from distinct demographic groups, should maintain comparable meanings. It comprises three key steps: (1) test input generation, producing inputs covering various demographic groups; (2) test oracle generation, identifying potential unfair translations based on semantic similarity measurements; and (3) regression, discerning genuine fairness issues from those caused by low-quality translation. Leveraging FairMT, we conduct an empirical study on three leading machine translation systems–Google Translate, T5, and Transformer. Our investigation uncovers up to 832, 1,984, and 2,627 unfair translations across the three systems, respectively. Intriguingly, we observe that fair translations tend to exhibit superior translation performance, challenging the conventional wisdom of a fairness-performance tradeoff prevalent in the fairness literature.

The association between discourse production and schizotypal personality traits

Discourse abnormalities are a prominent feature in schizophrenia spectrum disorders, including poor lexical diversity, and have been found to differentiate patients from healthy subjects. However, discourse processing in individuals with high schizotypal personality traits is less understood and is often overshadowed by research on clinical psychotic symptoms. In the present study, we examined schizotypal traits at a non-clinical threshold and their association with lexical diversity and discourse coherence using two automated Natural Language Processing (NLP) tools - Type-Token-Ratio (TTR) measures from the Tool for the Automatic Analysis of Lexical Diversity (TAALED) and discourse coherence using sentence-level average cosign similarity with FastText to assess sentence similarity. 276 college students completed the full assessment including the Schizotypal Personality Questionnaire- Brief Revised (SPQ-BR) and recorded a speech sample describing a detailed painting. Results revealed that high schizotypal traits, specifically positive traits, were associated with lower lexical diversity and higher sentence similarity. Our findings suggest that even when clinically significant symptoms are not present, discourse abnormalities are present in healthy populations with high ST. The stronger association with positive traits suggests that theories of perseveration and top-down processing may warrant further investigation in schizophrenia-spectrum disorders.

A Sentence-Matching Model Based on Multi-Granularity Contextual Key Semantic Interaction

In the task of matching Chinese sentences, the key semantics within sentences and the deep interaction between them significantly affect the matching performance. However, previous studies mainly relied on shallow interactions based on a single semantic granularity, which left them vulnerable to interference from overlapping terms. It is particularly challenging to distinguish between positive and negative examples within datasets from the same thematic domain. This paper proposes a sentence-matching model that incorporates multi-granularity contextual key semantic interaction. The model combines multi-scale convolution and multi-level convolution to extract different levels of contextual semantic information at word, phrase, and sentence granularities. It employs multi-head self-attention and cross-attention mechanisms to align the key semantics between sentences. Furthermore, the model integrates the original, similarity, and dissimilarity information of sentences to establish deep semantic interaction. Experimental results on both open- and closed-domain datasets demonstrate that the proposed model outperforms existing baseline models in terms of matching performance. Additionally, the model achieves matching effectiveness comparable to large-scale pre-trained language models while utilizing a lightweight encoder.

Multi-view graph-based interview representation to improve depression level estimation

Depression is a serious mental illness that affects millions worldwide and consequently has attracted considerable research interest in recent years. Within the field of automated depression estimation, most researchers focus on neural network architectures while ignoring other research directions. Within this paper, we explore an alternate approach and study the impact of input representations on the learning ability of the models. In particular, we work with graph-based representations to highlight different aspects of input transcripts, both at the interview and corpus levels. We use sentence similarity graphs and keyword correlation graphs to exemplify the advantages of graphical representations over sequential models for binary classification problems within depression estimation. Additionally, we design multi-view architectures that split interview transcripts into question and answer views in order to take into account dialogue structure. Our experiments show the benefits of multi-view based graphical input encodings over sequential models and provide new state-of-the-art results for binary classification on the gold standard DAIC-WOZ dataset. Further analysis establishes our method as a means for generating meaningful insights and visual summaries of interview transcripts that can be used by medical professionals.

Exploring climate change discourse on social media and blogs using a topic modeling analysis

Climate change is one of the most pressing global issues of our time, and understanding public perception and awareness of the topic is crucial for developing effective policies to mitigate its effects. While traditional survey methods have been used to gauge public opinion, advances in natural language processing (NLP) and data visualization techniques offer new opportunities to analyze user-generated content from social media and blog posts. In this study, a new dataset of climate change-related texts was collected from social media sources and various blogs. The dataset was analyzed using BERTopic and LDA to identify and visualize the most important topics related to climate change. The study also used sentence similarity to determine the similarities in the comments written and which topic categories they belonged to. The performance of different techniques for keyword extraction and text representation, including OpenAI, Maximal Marginal Relevance (MMR), and KeyBERT, was compared for topic modeling with BERTopic. It was seen that the best coherence score and topic diversity metric were obtained with OpenAI-based BERTopic. The results provide insights into the public's attitudes and perceptions towards climate change, which can inform policy development and contribute to efforts to reduce activities that cause climate change.